2024 IDM Annual Symposium

We will focus on introducing working groups to one another, short presentations of works in progress, Q&A, and future paths of development, new data opportunities.

Join other EMOD modelers to find out what is happening in the world of EMOD – a new Slack group, running EMOD without an HPC, emodpy 2.0 and the overall future of EMOD.

Mr. Paul Ansah, Albright Stonebridge Group

Paul C. Ansah is a Partner and leads ASG’s Africa practice. He draws on his extensive commercial experience and deep network across the continent to assist clients in entering new markets, building partnerships, and navigating complex business, regulatory, and policy environments. His expertise includes market entry, growth strategy, acquisitions, finance, due diligence, deal structuring, and negotiations.

He was most recently the founder and CEO of Answer Group, advising clients on the development of international partnerships in Africa.
Previously, Mr. Ansah was Vice President of International Development for Marriott International, where he was responsible for the growth of all Marriott brands across 30 countries in Africa and the Indian Ocean. In this role, he led a team of executives in the successful sourcing, negotiation, and execution of dozens of hotel transactions with assets valued in excess of $2.5 billion.

Prior to his Africa leadership role, Mr. Ansah spent several years underwriting and structuring complex transactions with Marriott across the world, including the growth of Ritz-Carlton’s global branded residential portfolio. He began his career as a civil engineer, building mixed-use commercial real estate in North America.

Mr. Ansah received his MBA in Finance and Real Estate from The Wharton School at the University of Pennsylvania, and his B.S. in Civil Engineering from the University of Maryland.

View Presentation for

Ms. Anuradha Das Mathur, Albright Stonebridge Group

Anuradha Das Mathur is a Partner and co-leads DGA-ASG’s South Asia practice. She works extensively at the intersection of the policy, business, and development sectors, and offers strategic advice to clients to help navigate the policy environment.  She brings a unique blend of intellectual insight and sound judgment to her work by leveraging her advocacy experience with strong professional networks across the country. 

She is a Director at IMA India, a niche economic, business, and market research firm that provides insights and analysis to top management audiences in India. IMA manages the country’s largest CXO peer group representing more than 2,500 Indian and global business/functional heads from over 1,500 member companies. 

She is the Founder and Dean of a highly acclaimed all-women’s management and leadership institute. She is recognized widely for her work in higher education, gender studies and her commitment to ensuring financial independence for women. She also founded Samarth Life Management Pvt Ltd which is India’s largest community of active seniors and amongst its most trusted eldercare service providers. 

Ms. Mathur ran Businessworld, India’s most widely-read business magazine. Prior to this, she spent over 12 years with the Indian affiliate of the Economist Intelligence Unit (now IMA India) where she led the research and advisory business, conceptualized and managed a series of businesses for CFOs and CEOs.

Ms. Mathur was selected, along with twenty-five other women globally, for the prestigious Global Women’s Mentoring Partnership program, a joint initiative by Fortune and the U.S. Department of State. In 2016, she was elected as one of India’s ‘100 Women Achievers,’ by the Government of India. She is a part of the Yale World Fellows Program, which enables 16 extraordinary individuals annually, from across the globe and diverse disciplines. to increase their capacity to make the world a better place. 

Dr. Bethany DiPrete, UNC Chapel Hill: Combining epidemiologic and genomic data to better understand cholera transmission in Africa

Regional approaches to cholera control have long been targeted, though epidemiologically derived cholera regions have not been well described. The observed synchrony of reported cholera cases between countries can provide some insights into connected regions, but contemporaneous case reports in two settings can be a result of transmission between populations, or similar exposures to extrinsic factors that increase cholera risk (e.g., weather/climate). Molecular data (e.g., whole genome sequences) can augment reported case data to provide a more robust understanding of the epidemiologic connectivity between populations.

In this work, we use a novel statistical framework to combine case counts and whole genome sequence data to estimate the proportion of reported cases attributable to each 7th pandemic lineage and then estimate the ‘cholera connectivity’ between countries. The first stage of this framework uses a hidden Markov model (HMM) to estimate three targets of interest: the strength of connectivity driving cholera transmission between countries, the underlying occurrence of lineages in each country across time, and the proportion of cases attributable to each lineage through time.

The resulting estimates of connectivity in Africa are then used to define epidemiologically connected cholera regions, which may be helpful to enhance regional cholera control planning and assessment of risk to connected areas during outbreaks. Further, the results from this work can be used to help target resources for sequencing in unsampled areas where additional genomic data would improve the inferential capacity of our model.

View Presentation for Combining epidemiologic and genomic data to better understand cholera transmission in Africa

Dr. Joshua Levy, Scripps Research: Establishing an international wastewater-clinical integrated virus genomic surveillance system

Most modern infectious disease monitoring is focused on testing of infected individuals, requiring costly and labor intensive sample collection, testing, and sequencing. Over the COVID-19 pandemic, wastewater-based surveillance has emerged as a key tool to cost-effectively and comprehensively track virus evolutionary and spread dynamics in the community, without many of the sampling biases associated with clinical sampling.

Here, we show how wastewater genomic surveillance, when performed at sentinel public health labs and research institutes around the globe, has enhanced and has now become fundamental to pathogen monitoring efforts across broad economic and infrastructural contexts. We analyze wastewater collected and sequenced by consortium members, spanning 6 different countries across five continents, along with matched clinical surveillance data, as made available via the outbreak.info platform.

We find that real-time outbreak monitoring using wastewater data often enables early detection of emerging lineages, uncovers regional pathogen trends with weeks to months of lead time, and enables this to be done cost-effectively, providing a much needed supplement to clinical surveillance. We demonstrate that multi-stream integrated surveillance, using all publicly available wastewater and clinical data, vastly expands the information available to guide public health responses and provides a promising model for pandemic intelligence. 

View Presentation for Establishing an international wastewater-clinical integrated virus genomic surveillance system

Dr. Frank Aarestrup, Technical University of Denmark: Opportunities and challenges with metagenomic based waste water surveillance

Wastewater surveillance have become increasingly popular as a potential catch-all methodology for monitoring the occurrence of bacterial and viral pathogens, as well as antimicrobial resistance. When using metagenomics it has been the hope that we in the future can continuously monitor all pathogens, as well s potentially detect entirely novel once even before they are detected in clinical microbiology.

It has however, also become clear that, while showing great promises, especially when using metagenomics approaches, wastewater surveillance comes with a large number of methodological challenges including considering on sampling, handling, DNA and RNA purification, sequencing and both bioinformatics and sub-sequent statistical analyses that may greatly affect the results.

At the Technical University of Denmark we have since 2015 conducted longitudinal sampling and metagenomics analyses of wastewater collected from more than 100 countries around the world, as well as conducted different studies to determine the importance of the above-mentioned factors. Our experience and suggestions for how to establish a global wastewater surveillance system will be presented.

View Presentation for Opportunities and challenges with metagenomic based waste water surveillance

Dr. Scott Olesen, US CDC: Modeling and forecasting using genomic surveillance: Lessons from wastewater and COVID-19 variants

Wastewater surveillance is a promising platform methodology, ideally enabling early warning via parallel quantification of dozens of pathogens and pathogen variants from a single, population-level sample. In practice, wastewater signals have relatively high intrinsic noise and require novel analytical and interpretative approaches to extract actionable information. Dynamical modeling of SARS-CoV-2 variant prevalences provides multiple examples and themes that might guide other wastewater applications.

First, integrating multiple signals, such as genomics data from clinical isolates and wastewater genomics, might produce more accurate predictions of SARS-CoV-2 variant prevalences. However, these kinds of signal fusion models are complex, and different genomic signals cannot be naively combined. Second, the relevant entities to be modeled change through time. Not only do variants of concern emerge and disappear, the bioinformatic definition of a variant will necessarily come some time after the emergence of that variant. Third, the relevant forecasting targets will differ by application. Variant proportions are the most common forecasting target, but they do not immediately inform decision-making. For example, the deployment of medical countermeasures depends on which variants will become the dominant variant and when that will happen, which may not be immediately apparent from current variant prevalences. Finally, a model of relative variant prevalences might have greater utility when combined with an epidemiological model of disease transmission, so that the emergence of a new variant could better predict whether an associated wave of infections will occur. 

View Presentation for Modeling and forecasting using genomic surveillance: Lessons from wastewater and COVID-19 variants

Ms. Aasli Nur, University of Oxford: Modeling contraceptive method skew in Ethiopia using FPsim

The skewed distribution of contraceptive methods limits women’s and couples’ ability to exercise agency over their reproductive choices. In Ethiopia, the dramatic increase in contraceptive use and the dominance of a single method, injectables, raises significant concerns about method skew and women’s contraceptive choice. In five of eleven regions, injectables account for over 50% of total use, while implants and injectables combined represent over 50% in all regions. The field of global family planning does not endorse an “ideal” method mix; however, the predominance of one or two methods calls into question the extent to which a skewed market can meet the demand for family planning, which exists at the individual- and couple-level.

Applying individual-level modeling approaches helps us to understand how micro-level family planning dynamics contribute to macro-level trends in fertility, which are closely monitored by researchers and policymakers. The Family Planning Simulator, or FPsim, is an agent-based model designed to engage with the dynamic nature of women’s reproductive health. We use FPsim to test a series of hypothetical scenarios that explore the relationship between contraceptive choice and method mix in Ethiopia at the national and subnational level.

We find that improving women’s access to more effective contraceptive methods not only increases overall use but also accelerates the decline in method skew towards a more balanced method mix. We present these findings to demonstrate how FPsim can be used by researchers to investigate their own research questions and promote the development of more effective, women-centered programming and policy interventions.

Dr. Vincent Huang, Surgo Health: An end-to-end approach to enhancing family planning uptake in Madhya Pradesh, India through causal AI- guided intervention design to a cluster random control trial

Despite providing significant advantages across health, economic, and social domains, modern contraception uptake remains stagnant in many lower- and middle-income areas. However, efficient identification, design and validation of effective interventions remains a challenge. Here, we propose that use of appropriate observational data and machine learning approaches could systematically inform optimal intervention design through causal inference.

With the Surgo-CHAI 2019 Survey data from non-sterile women with at least one child (n=6190) in Madhya Pradesh (MP), India, we trained a causal Bayesian network model to identify the structural causal relationships between 39 potential factors. From these, 29 were directly or indirectly causal to modern, effective, temporary methods (MET) uptake, with postpartum family planning advice, facility awareness, and method awareness being direct influences. This single model allowed for a systematic comparison of 29 inferred interventional outcomes. We found the top intervention was ensuring family planning advice after birth, with an estimated net lift of 9.0 people switching to MET per 100 people advised.

To validate, in late 2022 we conducted a cluster randomized controlled trial in MP (n=1767), where uptake had remained at 3 to 6% in the past decade. We observed that the intervention resulted in tripling MET uptake to 17.5% when compared to the controls (9.0%). Notably, the net difference of 8.5 percentage points closely echoes the model estimate.
Taken together, these results suggest that machine learning of a causal Bayesian network derived from observational data can effectively inform interventions for modern contraception uptake and simplify the RCT required for validation.

View Presentation for An end-to-end approach to enhancing family planning uptake in Madhya Pradesh, India through causal AI- guided intervention design to a cluster random control trial

Dr. Annie Haakenstad, Institute for Health Metrics and Evaluation: Decomposition analysis for adolescent sexual and reproductive health and rights exemplars

Background
In 2019, there were 11.5 million births among adolescents (aged 15-19) in low- and middle-income countries, with an estimated half of those births unintended. Cameroon, Ghana, Malawi, Nepal, and Rwanda stand out as exemplar countries experiencing some of the largest declines in adolescent pregnancy globally between 2000 and 2022. Our study decomposes the declines in adolescent pregnancy in these five countries and quantifies the drivers of the observed change.
 
Methods
Our analysis leverages Demographic and Health Surveys (DHS) and Multiple Indicator Cluster Surveys data. We implement an Oaxaca-Blinder decomposition analysis of pregnancy in the last two years comparing the earliest and latest years of data available for each country over 2000-2022. Additionally, we conduct a Cox proportional hazards analysis that tracks women over time from age 15 to 19 and estimates the association of explanatory variables with pregnancy using contraceptive calendar data from the DHS.
 
Results
The Oaxaca-Blinder decomposition revealed that marital status, education level, and sexual activity are the primary factors associated with changes in adolescent pregnancy rates across all countries. The Cox proportional hazard model results corroborated these findings and further highlighted the significant association of contraceptive use with pregnancy. 
 
Conclusions
Reduced adolescent pregnancy rates in Cameroon, Ghana, Malawi, Nepal, and Rwanda are associated with lower marriage rates and higher educational attainment. Sexual activity and contraceptive use also emerge as critical elements in declining pregnancy rates. Our results underscore the importance of supporting educational opportunities, sexual health awareness, and access to contraception to reductions in adolescent pregnancy.

View Presentation for Decomposition analysis for adolescent sexual and reproductive health and rights exemplars

Dr. Robyn Stuart, Gates Foundation (independent consultant): Priority data gaps for quantifying the impact of novel syphilis interventions: A mathematical modeling analysis

The global burden of syphilis, a bacterial sexually transmitted infection (STI), is both sizable and increasing. Each year, nearly 700,000 cases are transmitted from pregnant women to their babies; these congenital syphilis cases account for an estimated 8% of all stillbirths, despite being easily preventable. Improving testing coverage among pregnant women is hampered by weaknesses in the current diagnostic landscape, and the absence of a viable vaccine.

Mathematical modeling is needed to determine optimal characteristics and delivery mechanisms of new diagnostics/vaccine candidates, especially in the face of considerable uncertainty about the transmissibility and natural history of syphilis. In this work, we develop an agent-based syphilis-HIV coinfection model using the Starsim modeling framework, with structured sexual networks calibrated to diverse settings with high burden of syphilis. We use a value-of-information approach to demonstrate that understanding the transmissibility of latent syphilis would be the most important priority for reliable quantification of the impact of new diagnostics or vaccines.

View Presentation for Priority data gaps for quantifying the impact of novel syphilis interventions: A mathematical modeling analysis

Dr. Arminder K. Deol, CEPI: Global South leaders in epidemic analytics and response network

Infectious disease modeling is a valuable tool for informing and guiding policy-making and strategic decision-making in public health. Most global leaders in infectious disease modellers are concentrated in high income countries, leaving a significant gap in local expertise withing regions most affected by outbreaks and endemic disease in the global south – this was made evident during the Covid-19 pandemic. It is widely agreed that developing local context driven models and nurturing local-based modellers is important and needed for fast and efficient responses to epidemics and pandemics to inform country stakeholders and policy decision makers.
 
To address these challenges and fill in the existing gap, there is a need to invest in and expand scientific expertise in computational skills and infectious disease modeling within the global south and enhance preparedness and response against future outbreaks, epidemics pandemics, and management of endemic diseases.  We propose to establish a collaborative network of global south researchers and institutions to facilitate knowledge sharing, skills development leading to more effective infectious disease modeling to ensure that public health decisions are informed by robust, context-specific, and timely modelling evidence. The network will also enable creation and fostering of global partnerships between organisations.

View Presentation for Global South leaders in epidemic analytics and response network

Dr. Britta Lassmann, BEACON: Biothreats Emergence, Analysis and Communications Network

The COVID-19 pandemic illustrated in stark and searing terms the need to improve the ability of communities globally to detect emerging infectious diseases (EIDs) and coordinate their responses to these threats. Global early detection, rapid transparent reporting, and open information sharing regarding potential biological incidents are critical.
 
The Biothreats Emergence, Analysis and Communications Network (BEACON) is a new program dedicated to the rapid collection, vetting, reporting, and analysis of information on emerging threats affecting humans, domestic animals, wildlife, plants and the environment globally. BEACON combines emerging infectious diseases surveillance, a global network of moderators who are subject matter experts and the power of artificial intelligence and large language models.
 
Data generated by BEACON will be made available open access, in near real-time, as a global public good. This will allow researchers, health professionals, and policymakers worldwide to access and utilize this information without barriers, facilitating global collaboration and innovation in public health monitoring and responses. Additionally, the core technological components of BEACON, including the LLM and predictive algorithm code, will be hosted on a GitHub repository for open access to enable the scientific community to review, adapt, and enhance these tools.
 
The program is based at Boston University’s Center on Emerging Infectious Diseases (CEID) and operated in partnership with the Hariri Institute for Computing and Data Sciences at Boston University and HealthMap at Boston Children’s Hospital/Harvard Medical School. BEACON is supported by the US National Science Foundation, the Noyce Foundation and the Bill & Melinda Gates Foundation. Technical partners include the World Health Organization, the Food and Agriculture Organization of the United Nations, and the World Organization for Animal Health.
 
At its launch, BEACON will be the only open-source global surveillance platform of its kind, linking public health authorities, practitioners, researchers, and the general public, rapidly and transparently sharing data and contextual knowledge about new threats. By providing early warnings of sentinel cases, clusters and outbreaks, BEACON will enable early public health response.

View Presentation for Biothreats Emergence, Analysis and Communications Network

Dr. Justin Lessler, University of North Carolina: Scenario Modeling Hub: Update and future work

Even the best models of emerging infections struggle to give accurate forecasts at time scales greater than 3-4 weeks due to unpredictable drivers such as a changing policy environment, behavior change, the development of new control measures, pathogen evolution, and stochastic events. However, policy decisions and planning in reaction to infectious diseases often require projections on the time frame of months. The goal of long-term scenario projections is to compare outbreak trajectories under different assumptions governing key features of interventions, pathogens, and populations that drive disease dynamics. This is in contrast to forecasts, which offer unconditional estimates of what “will” happen.

Here we will discuss the achievements of the US Scenario Modeling Hub (SMH) which has generated 18 rounds of operational scenario projections for COVID19 at the state and national levels in the US since Dec 2020, and has gradually expanded to include influenza and RSV. We will review the impact of SMH on policy and highlight progress in the evaluation of scenario projections. We will also discuss the recent evolution of the hub to address more research-oriented topics in disease modeling, including disease disparities and the cryptic phase of a pandemic.

View Presentation for Scenario Modeling Hub: Update and future work

Dr. Cliff Kerr, IDM

Starsim is a new open-source agent-based modeling framework for simulating the spread of diseases. It is designed to be easy to use, making agent-based modeling more accessible, while also being fast and flexible enough to model detailed, realistic policy scenarios. This session will provide a quick overview of Starsim, including its methodology, usage, and several example applications. It is suitable for anyone interested in learning a new disease modeling tool, or for anyone who wants a preview of Starsim before deciding whether to attend the full-day Starsim session on October 3rd.

Mr. Jeremy Bingham, SACEMA: Modelling the impact of noise on wastewater-based environmental surveillance

Wastewater-based environmental surveillance (WBES) has gained attention as a tool for monitoring infectious diseases at a population level. However, the complex nature of wastewater systems and the presence of various sources of noise can impact the usefulness of WBES as a data source for informing public health decisions. While some pathogens and questions – such as presence/absence monitoring for polioviruses – are well suited to WBES, others are only useful in certain contexts, and many are beyond the scope of WBES.

In this study, we develop a comprehensive modelling framework to simulate the entire WBES process, from epidemic dynamics to sample collection and testing. By considering different epidemic scenarios and noise levels in shedding, flow, dilution, and measurement, we evaluate the robustness of WBES in detecting differences between populations and identifying epidemiologically relevant signals. We focus on SARS-COV-2 as a use-case, but the approach we take can be adapted to other pathogens, given estimates of parameters related to viral shedding and the precision of detection methods. Our results provide insights into the conditions under which WBES can effectively support public health decision-making and highlight the need for further research to optimize WBES systems in real-world settings.

View Presentation for Modelling the impact of noise on wastewater-based environmental surveillance

Dr. Yuke Wang, Emory University: Biological Shedding Hub: A science portal for existing pathogen shedding data and models

Quantitatively well-characterized shedding information for pathogens and biomarkers can be critical for interpreting wastewater-based epidemiology (WBE) results (e.g., disease nowcasting/forecasting from the concentration of pathogens in wastewater), the strategic sampling design for wastewater surveillance (e.g., selecting sensitive sampling sites), and inferring infectious disease dynamics (e.g., estimating effective reproductive numbers). Though many studies collected and analyzed fecal and other biological samples from patients, limited efforts have been made to systematically cumulate this important information and quantitatively describe shedding dynamics.

The goal of the Biological Shedding Hub is to gather shedding data that are publicly available or contributed by individual groups and provide quantitative shedding models. We are interested in pathogens (e.g., SARS-CoV-2, influenza, RSV, Norovirus, etc.), human fecal biomarkers (e.g., PMMoV, crAssphage, mtDNA), and potentially beyond (e.g., dietary markers prescription and illicit drug metabolites). We plan to utilize shedding information in different formats and build hierarchical models accounting for both population-level (e.g., adults vs. children, vaccinated vs. unvaccinated) and patient-level variability.

The Biological Shedding Hub consists of an open data repository, a modeling code repository, tutorials on modeling for educational purposes, and a dashboard showing the estimated shedding dynamics and its impact on the interpretation of WBE results. This portal provides information and toolkits that allow researchers, practitioners, and stakeholders to make decisions related to WBE based on the best information currently available. This hub will be the basis for launching a community of practice to promote better understanding, interpretation, and use of valuable wastewater information from a public health perspective.

View Presentation for Biological Shedding Hub: A science portal for existing pathogen shedding data and models

Dr. Andrew Brouwer, University of Michigan: A mechanistic model and web-based tool for estimating the potential impact of water, sanitation, and hygiene interventions, accounting for contextual and intervention factors

Diarrheal disease is a leading cause of morbidity and mortality in young children. Water, sanitation, and hygiene (WASH) improvements have historically been responsible for major public health improvements, but many individual interventions have failed to consistently reduce diarrheal disease burden. Analytical tools that can estimate the potential impacts of individual WASH improvements in specific contexts would support program managers and policymakers to set targets that would yield health gains.

We developed a disease transmission model to simulate an intervention trial and made it available as a web-based app. We accounted for contextual factors, including preexisting WASH conditions and baseline disease prevalence, as well as intervention WASH factors, including community coverage, compliance, efficacy, and the intervenable fraction of transmission. We illustrated the sensitivity of intervention effectiveness to the contextual and intervention factors in each of two scenarios.

Achieving disease elimination depended on more than one factor, and factors that could be used to achieve disease elimination in one scenario could be ineffective in the other scenario. Community coverage interacted strongly with both the contextual and intervention factors. For example, the positive impact of increasing intervention community coverage increased non-linearly with increasing intervention compliance. Additionally, counterfactually improving the contextual preexisting WASH conditions could have a positive or negative effect on the intervention effectiveness, depending on the values of other factors. When developing interventions, it is important to account for both contextual conditions and the intervention parameters. Our mechanistic modeling approach can provide guidance for developing locally specific policy recommendations.

View Presentation for A mechanistic model and web-based tool for estimating the potential impact of water, sanitation, and hygiene interventions, accounting for contextual and intervention factors

Dr. Till Hoffmann, Harvard T. Chan School of Public Health: Fecal shedding models for SARS-CoV-2 RNA and implications for wastewater-based epidemiology

The concentration of many biomarkers in feces is not well characterized, posing challenges for quantitative wastewater-based epidemiology (WBE). Reliable estimates of biomarker shedding with principled uncertainty quantification are essential for assessing community prevalence of pathogens, per-capita drug consumption, and informing public health policy. We developed Bayesian hierarchical models for fecal shedding applicable to a wide range of biomarkers. For SARS-CoV-2, we fitted these models to RNA copy concentrations in fecal samples that were collected from hospitalized patients in six studies. We found a mean concentration of 1.9×10⁶ mL⁻¹ (95% credible interval: 2.3×10⁵–2.0×10⁸) among unvaccinated inpatients, not considering differences in shedding between viral variants. Model comparison indicated that limits of quantification could explain negative samples, suggesting that all infected individuals shed some RNA copies. Inpatients represented the tail of the shedding profile with a half-life of 34 hours (95% credible interval: 28–43), suggesting that WBE can be a leading indicator for clinical presentation. We demonstrated that shedding among inpatients cannot explain the high RNA concentrations found in wastewater, consistent with more abundant shedding during the early infection course. Current data are neither sufficient to inform the onset and early stages of shedding nor the effects of viral variants and demographic factors. Future studies should collect fecal samples from a representative sample of patients over the entire infection course to refine quantitative estimates of fecal shedding of SARS-CoV-2 RNA.

View Presentation for Fecal shedding models for SARS-CoV-2 RNA and implications for wastewater-based epidemiology

Dr. Kristen Aiemjoy, University of California Davis: Serocalculator: An open-source R package for estimating seroincidence from cross-sectional serosurveys

Public health scientists worldwide lack reliable data on the burden of many infectious diseases, which is a barrier to effective control measures. To address this gap, we developed an open-source R package, serocalculator, and an accompanying Shiny app to estimate incidence rates from serosurveys. Our method leverages within-host models of antibody decay from confirmed cases to interpret quantitative antibody responses in cross-sectional population samples.

After exposure to a pathogen, the immune system rapidly produces antibodies, which are slowly downregulated after the infection is controlled. The pattern of antibody decay acts like a stopwatch counting time since last exposure. Unlike traditional methods that dichotomize responses into binary outcomes, our approach uses quantitative antibody levels and accounts for both measurement noise and non-specific binding. The serocalculator package includes functions for data loading, preprocessing, exploratory analysis, incidence estimation, visualization, and simulations.

We have validated this tool to estimate the seroincidence of typhoid fever using serosurveys in communities with blood culture surveillance in Bangladesh, Nepal, and Pakistan. Our package has since been applied to estimate the seroincidence of enteric fever in populations without blood culture surveillance, including in South Sudan, Afghanistan, Sierra Leone, Côte d’Ivoire, and Niger. It has also been used to estimate the seroincidence of scrub typhus in India and Nepal and is being extended to other infectious diseases, including melioidosis, cholera, shigella, and dengue. Serocalculator meets a critical need in infectious disease surveillance, offering a reliable tool for generating accurate estimates of the force of infection from serologic data.

View Presentation for Serocalculator: An open-source R package for estimating seroincidence from cross-sectional serosurveys

Ms. Sylvia Lutz, Institute for Health Metrics and Evaluation: Introducing the SONIC tool for optimal allocation of antenatal and young child nutrition interventions with microsimulation

Supporting maternal and child health is important for communities and individuals worldwide. With new trial data for nutritional interventions such as SQ-LNS and combined SAM and MAM treatment, and new guidance from the WHO on targeting existing interventions, there is heightened interest in determining the most effective and feasible approaches to implement and scale interventions, particularly in resource-limited settings. 

Our team has developed an individual-based simulation to find the health impact of different antenatal and young child nutritional interventions. We consider iron and folic acid (IFA), multiple micronutrient supplements (MMN), and balanced energy protein (BEP) for antenatal care, and small-quantity lipid-based nutrient supplements (SQ-LNS) and SAM and MAM supplementation for the prevention and treatment of child malnutrition. Our unique mother/child dyad approach allows us to understand how different interventions interact at the individual level, such as how averting stillbirths impacts subsequent malnutrition treatment needs, and how prevention and treatment for malnutrition overlap. 

With this data, we have designed an optimization algorithm to provide the greatest impact on any health metric for a given budget size. Our team designed an online, publicly accessible tool, which allows for customization of cost estimates, health metrics to optimize for, and other parameters. 

In this talk, we will outline the simulation used to generate this result, along with comparing and contrasting our tool to existing models in this space, such as the as the Optima Nutrition model built on the Lives Saved Tool (LiST). 

Dr. Ezra Gayawan, Federal University of Technology, Akure, Nigeria: An assessment of the impact of women’s empowerment on childhood vaccination coverage in Nigeria: a spatio-temporal analysis

Women play a central role as primary caregivers and their level of health literacy, decision-making power, and access to resources can influence their ability to ensure immunization for themselves and their children. Thus, addressing inequality in women’s empowerment at sub-national levels can help improve vaccination coverage, equity, and health outcomes for all individuals, contributing to more effective public health interventions.

Focusing on two aspects of women’s empowerment: healthcare utilization and decision making, we assess how different levels of  the two empowerment indicators exact influence on various childhood vaccinations in Nigeria. We considered BCG vaccine, zero-dose coverage, measles containing vaccine (MCV), DPT-containing vaccine, and all basic vaccinations sourced from four waves of the Nigeria Demographic and Health Survey conducted in 2003, 2008, 2013 and 2018.  Responses to a number of empowerment related questions were combined through factor analysis to create indices for each of the two empowerment indicators and a Bayesian spatially varying coefficient model was used to determined how these indicators exact influence across small-scale spatial units, the lowest administrative levels of Nigeria.

The findings indicate that for many places in central Nigeria, women who were moderately empowered regarding decision making were less likely to have zero-dose children whereas for complete dose of DPT-containing vaccine and the other vaccinations considered, inequality in coverage appear to be more among the highly empowered women. 

View Presentation for An assessment of the impact of women’s empowerment on childhood vaccination coverage in Nigeria: a spatio-temporal analysis

Mr. Alen Kinyina, Prime Health Initiative Tanzania : The impact of group antenatal care on provision of prenatal services in Geita, Tanzania

In Tanzania, implementation of group antenatal care (G-ANC) model contributed to improving maternal and child health outcomes for over 5,000 pregnant women. This approach transforms the traditional one-on-one ANC visits into group sessions, where expectant mothers with similar gestational age (GA) come together under the guidance of healthcare providers and receives health education, individual assessments, and peer support, fostering a sense of engagement among pregnant women.

GANC has potential to address the significant challenges faced by healthcare systems in resource-constrained settings by consolidating ANC visits into group sessions and optimizes the use of limited healthcare resources, such as staffing and facilities, while ensuring that essential antenatal care services are delivered effectively. This efficiency is crucial for improving access to care, especially in rural or underserved areas where healthcare infrastructure may be lacking.

Group antenatal care holds great promise for improving maternal and child health outcomes in LMICs through leveraging the advantages of peer support, health education, and efficient resource utilization. This innovative model has the potential to enhance access to quality ANC services, empower women, and contribute to healthier pregnancies and childbirth experiences. However, further research and implementation efforts are needed to scale up group ANC initiatives and integrate them into routine maternal healthcare delivery in LMIC settings.

View Presentation for The impact of group antenatal care on provision of prenatal services in Geita, Tanzania

Dr. Alyssa Sbarra, Johns Hopkins Bloomberg School of Public Health: Comparing cross-sectional versus longitudinal data to model waning tetanus antibody titers 

Serologic data for vaccine-preventable diseases (VPDs) can be used to assess population- and individual-level immunity. For pathogens, such as tetanus, with known waning antibodies, this data can provide insights into decay rates or time until sero-reversion. However, while some serosurveys occasionally include longitudinal samples from a cohort, they often only contain an cross-section. This study aimed to assess the tetanus antibody waning rate and investigate whether it varied by malaria exposure using different analytical approaches from both ecological and longitudinal interpretations of the same data.

Plasma samples were collected quarterly from enrolled participants aged 0-to-10-years from the PRISM cohort study from 2013 to 2015 in Tororo District, Uganda. Samples were tested for tetanus anti-toxin IgG using a Luminex multiplex bead assay and converted to international units. From each included participant, the “ecological” design was assumed as the first tested sample (n=326) and the longitudinal design included all tested samples (n=1414). Rates of antibody decay and time until sero-revision were estimated using both ecological and longitudinal datasets while accounting for malaria exposure.

Results suggest substantial waning by age and varying by levels of malaria exposure, with antibody levels dropping below potential protective units. Ecological examinations provided valuable insights into waning and are similar to those observed from longitudinal analyses. However, when considering levels of malaria, ecological analyses are unable to disentangle nuanced patterns. Broadly, this illustrates a key use case of serologic data to identify time until sero-reversion across VPDs and highlights optimal situations for both cross-sectional and longitudinal data.

Mr. Mark Owusu, University of Cambridge: Modelling stochastic die-out and re-introduction of meningococcal serogroup A in Ghana

Meningitis, an infection of the brain and spinal cord, remains a significant global health concern, particularly in the African meningitis belt, which spans 26 countries from Ethiopia to Senegal, including northern Ghana. Before 2012, meningococcal serogroup A (MenA) was the primary cause of meningitis in Ghana. The introduction of the MenAfriVac vaccine in 2012, a protein-polysaccharide conjugate vaccine targeting MenA, resulted in a dramatic decline in MenA cases, with no reported cases since 2017.
Aligned with the World Health Organization’s (WHO) goal to “Defeat Meningitis by 2030,” meningococcal multivalent conjugate vaccines (MMCVs) have been developed and pre-qualified for use in the meningitis belt, providing a boosting advantage against MenA.

A deterministic age-structured dynamic model was used to investigate the impact of MMCVs in Ghana, suggesting that their implementation could prevent any resurgence of MenA for at least 15 years, maintaining an extremely low carrier population (<1 in 4.7 million). Building on these findings, an age-structured stochastic dynamic model (odin.dust stochastic package) was employed to explore the potential extinction of MenA in Ghana. Data and model parameters were sourced primarily from northern Ghana.
Results indicate a 99% chance of MenA extinction in Ghana if both Ghana and neighboring countries follow WHO Strategic Advisory Group of Experts on Immunization (SAGE) recommendations for MMCV implementation. Conversely, there is only a 2% chance of MenA die-out if further vaccine interventions are not introduced.

This study underscores the critical importance of coordinated intercountry efforts to eliminate meningococcal serogroup A in Ghana and its neighbors.

View Presentation for Modelling stochastic die-out and re-introduction of meningococcal serogroup A in Ghana

Mr. Dominic Delport, Burnet Institute: Estimating the historical impact of outbreak response immunization programs across 210 outbreaks in LMICs

Background
Outbreaks of vaccine-preventable diseases continue to occur in low- and middle-income countries (LMICs), requiring outbreak response immunization (ORI) programs for containment. To inform future investment decisions, this study aimed to estimate the cases, deaths, disability-adjusted life years (DALYs), and societal economic costs averted by past ORI programs. Outbreaks of measles, Ebola, yellow fever, cholera, and meningococcal meningitis in LMICs between 2000-2023 were considered.

Methods
210 outbreaks (51 measles, 40 cholera, 88 yellow fever, 24 meningitis, 7 Ebola) were identified with sufficient data for analysis. Agent-based models were developed using the Starsim framework, and calibrated for each disease such that after controlling for baseline vaccine coverage, ORI initiation time, speed of vaccine delivery, environmental variables, or endemic prevalence of the disease, observed outbreaks were within the distribution of simulated outbreaks. A status-quo and no ORI scenario were compared for each outbreak.

Findings
Across 210 outbreaks, ORI programs are estimated to have averted 5·81M [95% uncertainty interval 5.75M–5.87M] cases (4.01M measles, 283K cholera, 1.50M yellow fever, 21.3K meningitis, 820 Ebola), 327K [317K–338K] deaths (20.0K measles, 5215 cholera, 300K yellow fever, 1599 meningitis, 381 Ebola), 14·6M [14.1M–15.1M] DALYs (1.27M measles, 220K cholera, 13.0M yellow fever, 113K meningitis, 16.6 Ebola), and US$35.5B [32.5B–38.4B] (US$1.14B measles, US$156M cholera, US$34.0B yellow fever, US$97.6M meningitis, US$6.72M Ebola) in societal economic costs. In general, the more rapidly the ORI was initiated the greater the impact.

Interpretation
ORI programs are critical for reducing the health and economic impacts of outbreaks of vaccine-preventable diseases.

View Presentation for Estimating the historical impact of outbreak response immunization programs across 210 outbreaks in LMICs

Dr. Emily Nightingale, London School of Hygiene and Tropical Medicine: Sub-national estimation of surveillance sensitivity to inform declaration of disease elimination

The global eradication of wild poliovirus is now in its final stages. As this initiative approaches its goal, in remaining endemic countries we see extended periods of absence of detected virus, prompting discussion of criteria for certification. It is uncertain after how long without detection we can be confident that the virus is no longer circulating, and premature declaration and withdrawal of vaccination efforts could result in catastrophic resurgence.

Our confidence in elimination is dependent on collection of surveillance data, and on our understanding of that system’s sensitivity to detect circulating virus. We present a statistical framework to estimate time-varying sensitivity of two key components of polio surveillance – wastewater sampling and clinical cases of acute flaccid paralysis – for detecting infection on a sub-national level. We estimate the probability of freedom from infection (FFI) at a critical prevalence level, given absence of virus in collected samples. With this we explore how variability in the volume and quality of surveillance data influences our interpretation of elimination on a national level.

We applied this framework to the period of absence in Nigeria from 2013-2016, ascertaining the FFI probability that would have accumulated by the point that cases of WPV1-poliomyelitis were again detected in July 2016. We find substantial spatial heterogeneity in surveillance coverage and sensitivity and, given this, conclude that re-observing WPV1 infection after 23 months without detection was not improbable. This analysis demonstrates that timelines for certifying elimination should be informed by local, time-updated estimates of surveillance sensitivity.

View Presentation for Sub-national estimation of surveillance sensitivity to inform declaration of disease elimination

Dr. Judith Bouman, University of Bern: Understanding the population-level impact of mass vaccination campaigns against cholera in Uvira, South Kivu, Democratic Republic of the Congo

Cholera vaccines have become a routine part of cholera control in both endemic and epidemic settings.  While the individual level effectiveness of the vaccine has been well described, the population level impacts of mass vaccination have not been well demonstrated, largely due to the absence of robust clinical surveillance systems documenting cases both before and after vaccination. 

Uvira, a cholera endemic city of 300,000 is a unique setting where suspected cholera cases have been systematically counted and tested for almost a decade. In 2020 the DRC Ministry of Health organised an emergency mass cholera vaccination campaign targeting all people ≥1 years old. After vaccination, Uvira had an unprecedented year-long hiatus from cholera cases, followed by a large epidemic and eventually a return to more regular seasonal outbreaks and the overall impact of vaccination is unclear.

In this study, we leverage these unique data to fit mechanistic models to reproduce the observed data, taking into account locally-measured health seeking behavior, vaccine coverage dilution, drinking water access, seroincidence and diagnostic test performance. We fit these data in a Bayesian framework using STAN. To quantify the impact of vaccination, we compare counterfactual incidence scenarios (without vaccination) to the observed incidence and describe shifts in the seasonality of cases that may be attributable to mass immune synchronization.

View Presentation for Understanding the population-level impact of mass vaccination campaigns against cholera in Uvira, South Kivu, Democratic Republic of the Congo

Mr. Limbani Makawa, Ministry of Health Malawi: Cholera outbreak in Mchinji district, Malawi: Epidemiological trends, public health response, and lessons learned (2022-2023)

Introduction
Mchinji district in Malawi experienced a severe outbreak of Cholera from 2022 to 2023, marking one of the most significant public health challenges in recent years. We reviewed the outbreak’s epidemiological trends, public health response, and lessons learned.

Methods
We reviewed data collected from Cholera treatment centers. Data was analyzed using Excel to account for time, place, and person variables.

Results
The outbreak began on 2nd November 2022, by May 2023, the outbreak had spread to 104 villages, and internationally to Mozambique and Zambia, resulting in over 405 cases (M=199, F=206) and 16 fatalities (3.7% CFR). Highest incidence rates were observed in villages with inadequate access to safe water and poor sanitation facilities. Epidemiological analysis revealed a seasonal pattern, with peaks following the rainy season.
The district implemented a multi-faceted public health strategy including promoting safe water usage, establishing cholera treatment centers, and public health education campaigns to promote sanitation.
Despite these efforts, challenges including limited healthcare infrastructure, and community resistance to interventions were encountered. However, the response strategies significantly reduced the outbreak’s spread in the later months, demonstrating the effectiveness of coordinated public health efforts.

Conclusion
The outbreak proves the necessity of improving water and sanitation infrastructure, the importance of early and sustained public health interventions, and the need for coordinated cross-border healthcare delivery systems to manage health emergencies.

View Presentation for Cholera outbreak in Mchinji district, Malawi: Epidemiological trends, public health response, and lessons learned (2022-2023)

Dr. Javier Perez, Johns Hopkins Bloomberg School of Public Health: Challenges and opportunities in modeling the effect of weather and climate on cholera dynamics and control

Cholera is often considered an exemplar of climate-sensitive diseases due to the link between climate, weather, and Vibrio cholerae ecology in natural waters. However, in addition to environmental exposure to naturally-occurring toxigenic bacteria, cholera is transmitted through different fecal-oral routes, including direct human-to-human contacts, fomites, and free-standing fecal material.

As such, weather and climate may impact transmission in multiple ways depending on local transmission settings, beyond controls on bacterial ecology. In addition, climate-induced disruptions to Water, Sanitation and Hygiene (WASH) may also play a key role in changes in cholera transmission. Finally, immune dynamics may further mask cholera-climate relations due to the delays these impose on transmission dynamics through changes in the number of individuals susceptible to infection, with important implications on the role of vaccination on disease dynamics. Taken together, these factors pose challenges in modeling the effect of weather and climate on cholera transmission and control, thus limiting efforts to project potential climate change impacts.

We will here provide an overview of existing evidence of cholera-climate relations in the prescriptive of incorporating climatic drivers in mechanistic models of cholera transmission, highlighting challenges to do so. We will then present a scoping review of past efforts to model cholera-climate relations in the face of the above challenges and conclude with prospects for future avenues to build climate change impact scenarios on cholera transmission and burden.

View Presentation for Challenges and opportunities in modeling the effect of weather and climate on cholera dynamics and control

Dr. Cecilia Mbae, Kenya Medical Research Institute: Measurement of human-environment behaviours and fecal contamination for assessing risk of cholera in urban Kenya

Background
Cholera is caused by the toxigenic Vibrio cholera bacterium.  Several cholera outbreaks have previously been reported in Kenya with case fatality rates of >2.5%, the most recent beginning October 2022. Developing effective strategies to reduce cholera transmission requires identifying hotspots and investigating the relationship between the environment, fecal contamination in the environment, and human exposure behavior. We aimed to identify the pathways of exposure to V. cholerae within slums in Nairobi County.

Methodology
This study was carried out in Mukuru slum, an HDSS 15km, East of Nairobi. The SaniPath exposure assessment method was used to detect the prominent exposure pathways from the environment. Nine different sample types were subjected to culture and qPCR for the detection of V. cholerae. Household surveys measured exposure to environmental transmission pathways such as food, drinking water, and flood water.

Results
Microbial analyses of 791 environmental samples indicate that open drains (97%, 57%), surface water (100%, 67%), flood waters (97%, 43%), and raw produce (80%, 23%) were frequently contaminated with E. coli and V. cholerae, respectively through qPCR test. Comparison of V. cholerae detection in the samples between the two neighbourhoods was almost equal.

Conclusion
Detection of V. cholerae in all pathways is an indication of poor water, sanitation and hygiene conditions in Mukuru slums. The presence of V. cholerae in drinking water, street food and shaved ice can lead to epidemics. This information, combined with results from sequencing of cholera strains, will guide on WaSH and vaccine interventions for cholera prevention and preparedness.

Ms. Megan Knight, Institute for Health Metrics and Evaluation: The sex composition of human resources for health in 204 countries and territories: What role do female health workers play in health systems?    

Background
Better information on the characteristics of human resources for health (HRH) is needed to design inclusive and effective health workforce policies. This includes sex-disaggregated data and analysis, which is critical to accurately recognize women’s contributions. The present study uses standardized data to assess the composition and availability of health workers by sex for 26 cadres across 204 countries and territories between 1990 and 2021. 

Methods
We estimated HRH densities per 10,000 population with 1,738 country-years of surveys and censuses and 4,016 country-years from the World Health Organization. We mapped all data to International Standard Classification of Occupations 2008 categories. Finally, we applied spatiotemporal Gaussian process regression to estimate HRH densities by sex over time and across locations.  
Results
In 2021, 78.0% of the health workforce was women, up from 67.8% in 1990.  Women comprised a larger share of nurses (86.5%) as compared to doctors (46.6%) in 2021, indicating differences in the leadership and responsibilities among health workers by sex. There are high concentrations of female nurses and midwives in Central Europe, Eastern Europe and Central Asia and a predominance of male physicians in sub-Saharan Africa.   

Conclusions
Persistent sex differences in health worker cadres reflect longstanding occupational segregation in the health sector.  To ensure gender-based discrimination does not constrain health system performance, countries should consider gender-responsive policies that promote equity in recruitment, training and retention of HRH;  address disparities in leadership and specialization; and tackle high rates of workplace violence and harassment, among other barriers to women’s advancement as leaders in the health workforce.  

View Presentation for The sex composition of human resources for health in 204 countries and territories: What role do female health workers play in health systems?    

Dr. Betsy Rono, Jomo Kenyatta University of Agriculture and Technology: Patient care continuum link: A real-time spatial linked communication and referral system for patients by healthcare workers in Kenya

Introduction
Patient care continuum is critical in maximizing use and utilization of medical care. Referral system linking community to higher level health facilities and back to the community upon discharge; enhances adherence to reduce readmissions and disease complications. Patient Care Continuum Link is a real-time referral decision making platform employing bottoms-up approach, that integrates detailed Clinical algorithms guided referral directions at each level of care, while leveraging on spatial reporting and route optimization. Receipted message are priority ranked to prompt attention.

Methodology
Patient Care Continuum Link development used agile methodology phases: healthcare professionals were engaged to understand communication and referral needs to design user-friendly interface that integrates real-time communication, spatial reporting, and route optimization. The application was built using cutting-edge technologies for real-time data transmission and analysis. Rigorous testing was conducted for reliability, validity and security. Patient link shall be applied and deployed within healthcare settings guided by training for users. Feedback collected will guide iterative system improvements.

Results
Patient Care Continuum Link was fully developed and tested in May 2024. Four-thousand eight hundred (4800) health facilities are linked into the user-friendly interface platform with minimal learning curve and employs secure login to registered healthcare workers for patient data privacy. Agent based modelling has been fitted to predict and alert referral delays likelihood built onto daily morbidity and mortality data for prompt interventions. The link is connected to other application; queuing, enhanced health and discharge education system etc.

Conclusion
Patient Care Continuum Link reduces communication and referral delays.

View Presentation for Patient care continuum link: A real-time spatial linked communication and referral system for patients by healthcare workers in Kenya

Mr. Sid Zadey, Association for Socially Applicable Research: The IndoHealMap Project: A geospatial modeling study mapping timely access to healthcare in India

Background
Inability to reach the healthcare facility in time can lead to mortality, especially for emergency medical conditions. Hence, timely access to care is important in vast geographies with sociodemographic diversity like India. Using a novel geospatial modeling approach and GIS dataset, we provide novel high-resolution maps of travel times to and population coverage of various healthcare facility types using motorized and walking modes of travel.   

Methods
We used multiple government data sources to compile data for healthcare facilities at primary (primary healthcare centers or PHCs), secondary (community healthcare centers or CHCs), and tertiary (district hospitals or DHs and medical college hospitals or MCHs) levels of care from 2018 to 2023. We geocoded and validated the facility addresses and used an in-house analytical pipeline for mapping travel time from each 1 km2 grid to its nearest facility. Median travel times and population percentage within 30, 60, and 120 minutes of the nearest facility (i.e., coverage) were estimated as access outcomes at the national, state, and district levels and in rural-urban areas.

Results
For 46002 PHCs and 6476 CHCs, the median (interquartile range) travel time to reach the nearest facility by motorized transport was 14.34 (5.96, 30.97) minutes while that for walking was 63.23 (36.83, 108.01) mins. For 707 DHs, the median travel time was 59.09 minutes (35.93, 96.10). 97.30% of people were covered within 120 minutes of their nearest DH. Rural coverage was lower than urban areas  (93.47% vs. 99.13%). For 648 MCHs, only 4.23% of people were within 30 minutes of walking from their nearest facility, while 71.79% were within 60 minutes by motorized transport. We observed major heterogeneities across 36 states and 640+ districts with severely limiting access in northeastern states and certain tribal/rural parts of central and north India.  

Conclusion
These are India’s first nationwide findings for timely access to different levels of healthcare facility types depicting regional and rural-urban disparities. Our work has high significance for India’s policymaking and generalizability potential for health systems research in other LMICs. 

View Presentation for The IndoHealMap Project: A geospatial modeling study mapping timely access to healthcare in India

Ms. Valentina Martufi, CIDACS-IGM/FIOCRUZ-BA: Federal transfers for PHC in Brazil and their relation with municipal material deprivation: An SMSN modelling approach

Brazil presents severe variances in the financial resources invested in Primary Health Care (PHC) at the municipal level across its regions. The PAB has been a federal effort to ameliorate such inequalities, with direct, per capita and results-based financial transfers to municipalities. This study aims to explore PAB transfers’ relation with the region and socioeconomic context of each municipality.

Public data from the National Health Fund was used to study per capita PAB transfers to Brazilian municipalities between 2010 and 2017. A simple descriptive analysis of the data’s distributions was carried out. Subsequently, a scale mixtures of skew-normal (SMSN) distributions model – employed in applications with extreme observations – was used to get a sharper image of how the per capita PAB relates to the region in which the municipality is located and the level of material deprivation affecting it. Quintiles of the Brazilian Deprivation Index (IBP) were used to measure material deprivation, contemplating per capita income, alphabetization and housing conditions (water, sewage, indoor bathroom, garbage collection).

The descriptive analysis of the data’s distribution presented extreme values, at either side of the relatively narrow middle 50% of the observations. This was the case for both the distribution according to region and material deprivation level. With the SMSN distributions modelling it was possible to observe an explicit association between per capita federal transfers for PHC to municipalities and the IBP, with per capita transfers progressively increasing as material deprivation increases. Therefore, it seems PAB transfers have a pro-equity effect on PHC financing.

View Presentation for Federal transfers for PHC in Brazil and their relation with municipal material deprivation: An SMSN modelling approach

Ms. Sally Lago, Strathmore University: Can mathematics bridge the divide in Nyanza’s fight against HIV? A mathematical model of the effect of culture HIV transmission dynamics

In the intricate tapestry of the fight against HIV, understanding the subtle yet powerful influence of cultural practices is paramount. Traditional practises such as polygamy, disco matanga, widow cleansing rituals and wife inheritance have been practises that affect HIV transmission and treatment outcomes in the Nyanza region of Kenya. Disco matanga, characterized by communal gatherings typically held after a funeral to celebrate the life of the deceased, often accompanied by risky sexual behaviour, and polygamy, which can lead to complex sexual networks and increased vulnerability to HIV, are particularly noteworthy in their contribution to the HIV epidemic in the region. Widow cleansing compels widows to engage in unprotected sexual intercourse with designated `cleansers` to rid themselves of perceived impurity that one is believed to have gained upon the demise of her husband. Simultaneously, wife inheritance, where a widow is expected to be inherited by a relative, presents unique complexities to HIV prevention and treatment strategies in the region.

This research paper is a comprehensive exploration formulating and analysis of a mathematical compartmental model that delves deeper into the influence of cultural practises on the  HIV transmission dynamics among adults in Nyanza region a recognised HIV hotspot in Kenya. It seeks to quantify the profound effect of culture on HIV transmission, considering the prevalent cultural practices in the broader Nyanza region.

The mathematical narrative in this paper not only captures transition between compartments but also serves as a visionary guide for culturally tailored interventions in a region where cultural ethos intertwines with the pressing need of a HIV free culture. Thus, this research paper serves as a compelling call to action for health care providers, policy makers and researchers alike to embrace a holistic understanding, acknowledging that in the vibrant mosaic of Nyanza region, culture transcends being a mere bystander but a key player in a collective pursuit of a culture emancipated from the scourge of HIV.

View Presentation for Can mathematics bridge the divide in Nyanza’s fight against HIV? A mathematical model of the effect of culture HIV transmission dynamics

Dr. Temidayo Oluwafemi, Newgate University Minna, Niger State, Nigeria: Mathematical modelling of mental health and HIV dynamics

Mental health refers to the overall emotional, social and psychological well-being of an individual. it affects how people feel, act and think, it also influence decisions, developing healthy relations and managing stress. A person can have poor mental health and not have a diagnosed mental illness. Likewise, a person with a mental illness can still enjoy mental well-being. People living with HIV are at a high risk of some mental health conditions (depression) because of the associated stress. In this study, a Mathematical Model is proposed to study mental health and HIV dynamics in the human population. The stability analysis of the model equations is carried out, we compute the reproduction number and carry out sensitivity analysis of the model.

Alisa Hamilton, Center for Systems Science and Engineering, Johns Hopkins University: Gender-responsive dynamic models of COVID-19, HIV, and malaria transmission: A systematic scoping review

Background
Sex and gender differences affect infectious disease outcomes due to biological and social reasons. Gender-associated norms, roles, and behaviors can affect exposure to a pathogen, health-seeking behavior, and access to services; however, many infectious disease models do not incorporate sex/gender into model design.
 
Methods
This ongoing systematic scoping review aims to summarize methods for incorporating sex/gender into dynamic models of three diseases representing different transmission routes: COVID-19 (respiratory), HIV (sexually transmitted), and malaria (vector-borne). Studies must meet inclusion criteria informed by an established gender analysis framework: 1) include a sex/gender-disaggregated population (e.g., women and men), 2) disaggregate the population by an additional social stratifier (e.g., age, race), and 3) consider the needs, rights, and preferences of a gender group as well as gender power relations and systems.
 
Findings
The database search resulted in ~6,000 studies. Among the ~1,000 identified for full-text screening, ~500 of these disaggregated by sex/gender, and ~200 of these included an additional social stratifier. We anticipate 10-20 studies that also consider needs, rights, preferences, and power relations and systems. Additional stratifiers for HIV studies included age, race, and sexual orientation. Additional stratifiers for COVID-19 studies included age. We did not identify any malaria studies that met all inclusion criteria. For each study, we will extract data on reasons for including sex/gender, model type (e.g., compartmental vs ABM), parameterization, and validation.
 
Interpretation
Incorporating sex/gender into infectious models enables decision makers to design more tailored and equitable prevention and response strategies.

Dr. Masabho Peter Milali, New York University, Grossman School of Medicine: Machine learning approaches to classify HIV prevalence of communities using social economic and behavioral data in resource-limited settings

The accurate estimation of HIV prevalence is crucial for public health officials, researchers, and policymakers to effectively monitor the epidemic and evaluate interventions. Traditional methods, which depend on regular HIV testing, encounter challenges due to logistics, limited availability of trained personnel, and individual reluctance, often stemming from stigma, recent testing, feeling sick, and the perception of being at low risk for HIV infection. This study explores the potential of estimating HIV prevalence in communities using socio-economic, behavioral, and other community-level data in the absence of direct HIV biomarkers.

Using Partial Least Squares (PLS) and Random Forest (RF) models, we developed models to predict HIV prevalence based on socio-economic and behavioral variables from Population-based HIV Impact Assessments (PHIA) surveys. Community HIV prevalence, derived from the PHIA biomarkers dataset, served as the dependent variable. Initially, models were trained to classify communities into <10% or ≥10% HIV prevalence categories. This procedure was repeated for prevalence thresholds of 5%, 7%, 15%, and 20%.

PLS and RF achieved 79% and 80.5% accuracy, respectively, in predicting community-level HIV prevalence. In both models, the variables that contributed most to classification included having a first intercourse experience before age 15, being uncircumcised, having a history of not using condoms, being in the lowest wealth quintile, experiencing physical or sexual violence, and having extramarital partners.

The study demonstrates that socioeconomic and behavioral variables can effectively predict community-level HIV prevalence using machine learning models. These insights have the potential to guide the distribution of HIV resources, particularly where direct community testing is infeasible, and to enhance understanding of the HIV epidemic dynamics.

View Presentation for Machine learning approaches to classify HIV prevalence of communities using social economic and behavioral data in resource-limited settings

Dr. Rebecca Clark, London School of Hygiene and Tropical Medicine: Estimating the epidemiological and economic impact of scaling up a nutritional intervention for TB-affected households across India

Approximately 20% of global TB incidence is attributable to undernutrition, which increases risk of developing TB disease and risk of TB death. Despite this, nutritional assessment and support is rarely provided in TB programmes.

We used results from a recent trial of the provision of nutritional support to people with TB (PWTB) and their households to explore the epidemiological implications of expanding the intervention to everyone treated for TB in India. We developed a transmission model of TB infection with explicit body mass index strata linked to disease progression and treatment outcomes. We modelled, separately and together, the provision of nutritional support to PWTB, and the provision of nutritional support to their household contacts.

Compared to a baseline with no nutritional support intervention, at 50% coverage of those on treatment (~15% of all TB-affected households) providing nutritional support to PWTB would avert 1.7% (95% uncertainty interval 1.4–2.3) and 0.1% (0–0.1) of cumulative TB mortality and incidence respectively in India by 2035, with 134,600 (108,200–179,000) fewer people dying of TB and 11,800 (-8,700–32,600) fewer developing disease. Extending this support to household contacts would avert an additional 136,600 (123,100–152,300) TB deaths and 893,400 (817,900–988,000) cases. To prevent one person from developing or dying of TB would require nutritional support to 5.1 and 17.4 TB-affected households, respectively.
A nutritional support intervention for TB-affected households could avert a substantial amount of TB incidence and deaths in India, with a low number needed to support.

Dr. Gabriela Gomes, University of Strathclyde: Remodelling selection to optimise disease forecasts and policies

Mathematical models are increasingly adopted for setting disease prevention and control targets. As model-informed policies are implemented, however, the inaccuracies of some forecasts become apparent, for example overprediction of infection burdens and intervention impacts. Here, we attribute these discrepancies to methodological limitations in capturing the heterogeneities of real-world systems. The mechanisms underpinning risk factors of infection and their interactions determine individual propensities to acquire disease. These factors are potentially so numerous and complex that to attain a full mechanistic description is likely unfeasible. To contribute constructively to the development of health policies, model developers either leave factors out (reductionism) or adopt a broader but coarse description (holism). In our view, predictive capacity requires holistic descriptions of heterogeneity which are currently underutilised in infectious disease epidemiology, in comparison to other population disciplines, such as non-communicable disease epidemiology, demography, ecology and evolution.

View Presentation for Remodelling selection to optimise disease forecasts and policies

Dr. Peter Cegielski, Emory University Rollins School of Public Health: Nutrition and tuberculosis

Updated systematic review of nutritional risk factors for incident TB and critique of the RATIONS trial.

View Presentation for Nutrition and tuberculosis

Dr. Sandip Mandal, John Snow India: The potential impact on tuberculosis of interventions to reduce undernutrition in the WHO South-East Asian Region: A modelling analysis

Background
Undernutrition is a major risk factor for TB incidence in the WHO South-East (SE) Asia Region. We examined the potential impact of addressing undernutrition as a preventive measure, for reducing TB burden in region.

Methods
We developed a deterministic, compartmental mathematical model, capturing undernutrition and its associated excess risk of TB, amongst countries in the Region. We simulated two types of interventions: (i) nutritional rehabilitation amongst all close contacts of TB patients, and (ii) an illustrative, population-wide scenario where 30% of people with undernutrition would be nutritionally rehabilitated each year. We also simulated this impact with additional measures to improve the TB care cascade.

Findings
The impact of nutritional interventions varies by country. For example, in India nutritional rehabilitation of 30% of undernourished population each year would avert 15·9% (95% Uncertainty Intervals (UI) 11·8 – 21·3) of cumulative incidence between 2023 – 2030, contrasting with 4·8% (95% UI 2·9 – 9·5) for Bhutan, which has only 10·9% prevalence of undernutrition. Reductions in cumulative mortality range from 11·6% (95% UI 8·2 – 17·1) for Bhutan, to 26·0% (95% UI 22·4 – 30·8) for India. Comparable incremental reductions in TB burden arise when combined with measures to improve the TB care cascade. Overall, nutritional interventions in the general population would increase incidence reductions by 2-3 fold, and mortality reductions by 5-6 fold, relative to targeting only contacts.

Interpretation
Nutritional interventions could cause substantial reductions in TB burden in the Region. Their health benefits extend well beyond TB, underlining their importance for public health.

View Presentation for The potential impact on tuberculosis of interventions to reduce undernutrition in the WHO South-East Asian Region: A modelling analysis

Benedicta Mensah, Noguchi Memorial Institute for Medical Research, University of Ghana: Training malaria modelers from sub-Saharan Africa: Experiences from the faculty enrichment program in the United States and Africa

The fight against malaria in sub-Saharan Africa requires advanced modeling techniques to inform and optimise intervention strategies. The Faculty Enrichment (FE) Program in Applied Malaria Modeling is a pioneering initiative that enhances the capacity of researchers in low- and middle-income countries (LMICs) in malaria modeling. The 4-month full-time program was held in the US for two years, and this year it was held in Africa in two sites, Ghana and Senegal.

We share the experiences, challenges, and insights gained from the planning and implementation of FE training programs in both HIC and LMIC. The program trains participants in the use of EMOD-Malaria, an advanced agent-based modeling software developed by the Institute for Disease Modeling. The technical component includes the design of an applied modelling project, hands-on tutorial sessions with EMODpy, geospatial modeling, and data visualization. The scientific communication component involves experiential training on preparing a specific aims page, writing an NIH-style biosketch, journal clubs, and presentations.

Conducting FE training in the United States provided access to state-of-the-art  facilities and high-end computational resources. Participants benefited from exposure to a diverse set of experts and a well-established infrastructure for advanced research. However, logistical challenges such as coordinating international travel, accommodating diverse time zones, and extended stays far from family were barriers to full participation. The training program in Africa offered substantial benefits, including cultural familiarity including language for French speakers,  reduced visa challenges, and cost-effectiveness alongside the technical and collaborative benefits observed in the U.S. Implementing the program in Africa also contributed to institutional capacity building for the host institutions, enhancing their ability to conduct and support advanced malaria modeling research​ and the sustainability of the training programs. The challenges faced in the African setting included access to computing resources, and poor internet connectivity.

The FE program represents a significant step forward in the global fight against malaria, enhancing the overall capacity for malaria research and intervention planning in LMICs by empowering local researchers with advanced modeling tools. 

View Presentation for Training malaria modelers from sub-Saharan Africa: Experiences from the faculty enrichment program in the United States and Africa

Dr. John Henry, University of Washington: Infection age as a dynamic predictor of epidemiological metrics for malaria

Malaria infection dynamics are complex and the factors which regulate parasitemia are poorly understood. However, parasitemia is a strong predictor of epidemiologically-relevant latent quantities such as the probability an individual will experience fever or test positive given a particular test, and thus a good model of the distribution of parasitemia at a population level can lead to improved estimates and forecasts of these quantities.

Here we establish a statistical relationship between infection age and parasitemia, develop a semi-Markov model to track the distribution of the age of an infection given a history of exposure, and then compose these models to create a dynamic estimate of the distribution of parasitemia in a cohort of human hosts given a history of exposure. Using the observed statistical relationships between parasitemia and our latent quantities, we can then transform our modeled distribution of parasitemia into dynamic estimators of these latent quantities. This work culminates in a coherent probabilistic framework for modeling malaria epidemiology in a way which is distinct from, and complementary to, standard compartment models.

Dr. Ai-Ling Jiang, University of California Irvine: Balancing irrigation and malaria risk: Integrating hydrologic and malaria modeling to optimize agricultural practices in western Kenya

Irrigation has been found to increase malaria vectors and prevalence in rural Africa. Nevertheless, governments are encouraging the use of irrigation to reduce the reliance on rainfed agriculture and alleviate food insecurity. To balance agricultural needs with public health concerns, it is crucial to implement sustainable irrigation practices that consider their impact on malaria risk.

In this study, we integrated hydrologic modeling with EMOD malaria model to investigate the influence of cropping practices and irrigation type on malaria transmission in Homa Bay, Kenya. Specifically, we considered scenarios involving a single crop and mixed crops as well as the associated irrigation types such as flooded, intermittent and drip irrigation. The approach recognizes differences in spatiotemporal distribution of malaria transmission dynamics resulting from ponding under the influence of crop water requirement and frequency and extent of irrigation.

Results show that rainfall had a significant influence in terms of moderating the difference in adult vector abundance across the scenarios. In addition, growing a variety of crops and assigning them to regions by susceptibility to flooding can help to reduce annual average malaria prevalence by more than 40% compared to predominantly growing rice for self-sufficiency. Lastly, malaria prevalence can be reduced further by more than 70% after enhancing irrigation efficiency through intermittent and drip irrigation as opposed to flooded irrigation. The study is a first step towards making informed decisions on agricultural strategies that can satisfy crop water demand while mitigating its impact on malaria transmission.

View Presentation for Balancing irrigation and malaria risk: Integrating hydrologic and malaria modeling to optimize agricultural practices in western Kenya

Mr. Alexis Martin, Swiss Tropical and Public Health Institute: Climate variability impacts malaria seasonal interventions: An ‘In Silico’ simulation combining two mathematicals models for malaria

Malaria is one of the deadliest infectious diseases and it continues to cause huge public health burden. Climate plays a pivotal role in malaria transmission due to its vector-borne nature. Interventions can be set up to mitigate or even eliminate malaria, but some of them like seasonal malaria chemoprevention and indoor residual spraying are seasonal and their timing and effectiveness relies on climate.

Mathematical modeling is increasingly essential in analysing and forecasting disease transmission dynamics, offering precious insights for effective intervention strategies. However, few mathematical models can assess the effect of interventions under climate variations scenarios. Our research focused on bridging two prominent models, VECTRI and OpenMalaria, to enhance versatility, accuracy and predictive capabilities. VECTRI is known for its ability to capture climate influences on malaria dynamics. OpenMalaria is known for its comprehensive modelling of interventions. By combining VECTRI’s climate modeling capabilities with OpenMalaria’s intervention-focused approach, we created a more holistic understanding of malaria transmission dynamics, considering both environmental factors and intervention strategies.

In the presented work, we used our hybrid model to explore the intricate relationship between seasonal patterns and interannual variations of climate on various malaria interventions. Through extensive simulations, we clarified the true efficacy of these interventions and their nuanced contributions to malaria control strategies. Our findings offer valuable insights into the dynamics between climatic patterns and intervention efforts in combating malaria, providing essential guidance for optimizing disease management practices. These results can be insightful for diseases other than malaria.

This work has not been published yet.

View Presentation for Climate variability impacts malaria seasonal interventions: An ‘In Silico’ simulation combining two mathematicals models for malaria

Dr. Jessica Lundin, IDM: Annotation quality and finetuning LLMs on low-resource languages

The quality of annotations is crucial in finetuning Large Language Models (LLMs), including for low-resource languages. This study investigates the impact of annotation quality on LLM performance on 11 African languages, referred to as low resource languages. With predictive models for quality and appropriateness we find the most salient features including ROUGE-1 scores and languages of the initial and validation translations.  We show how ROUGE1 scores can provide support the annotator and flag lower quality annotations in the moment.  The annotation process involved multiple stages, including initial translation, validation, and quality assessment by independent evaluators.

Our findings suggest that high-quality annotations enhance LLM finetuning performance, and the use of culturally appropriate annotations provides a further lift. These results support the importance of investing in high-quality, culturally sensitive annotation processes, which further close the performance gap in LLMs for low-resource languages in addition to reducing toxic language and bias.

Dr. Katherine Rosenfeld, IDM: Essential AI for translating science to policy

We introduce a framework for the use of large language models (LLMs) in Building Understandable Messaging for Policy and Evidence Review (BUMPER). LLMs are proving capable of providing interfaces for understanding and synthesizing large databases of diverse media. This presents an exciting opportunity to supercharge the translation of scientific evidence into policy and action, thereby improving livelihoods around the world. However, these models also pose challenges related to access, trust-worthiness, and accountability.  The BUMPER framework is built atop a scientific knowledge base (e.g., documentation, code, survey data) by the same scientists (e.g., individual contributor, lab, consortium). We focus on a solution that builds trustworthiness through transparency, scope-limiting, explicit-checks, and uncertainty measures.

LLMs are rapidly being adopted and consequences are poorly understood. The framework addresses open questions regarding the reliability of LLMs and their use in high-stakes applications. We provide a worked example in health policy for a model designed to inform measles control programs. We argue that this framework can facilitate accessibility of and confidence in scientific evidence for policymakers, drive a focus on policy-relevance and translatability for researchers, and ultimately increase and accelerate the impact of scientific knowledge used for policy decisions.

Mr. Dexian Tang, Causal Foundry, Inc.: Adaptive behavioral AI: Reinforcement learning to enhance healthcare services

Digital health is poised to revolutionize healthcare delivery and patient engagement by leveraging Artificial Intelligence (AI) across diverse use cases, from disease self-management to optimizing supply chains and managing patient care. In this talk, I will present an AI-driven operational system that integrates Reinforcement Learning to facilitate adaptive interventions. These interventions are dynamically tailored, continually enhancing their effectiveness through experimentation, real-time monitoring, and adaptive feedback mechanisms. Such advanced predictive capabilities are critical for effectively implementing clinical and behavioral strategies and improving healthcare outcomes by enabling precise, context-sensitive responses.

Our system collects, processes, and transforms logs into AI-ready data from multiple data streams and digital health products, providing a standardized visualization and facilitating comprehensive impact assessments across various mobile health devices. By issuing personalized recommendations grounded in historical data and personalized predictions, we significantly amplify the efficacy of digital interventions in healthcare systems. I will also address the profound implications of this technology in resource-limited settings, where its ability to drive significant health improvements could be particularly impactful. 

Mr. Abel Wilson Walekhwa, University of Cambridge and Dr. Godfrey Madigu, Strathmore University: Agent-based modelling of missed Rift Valley Fever disease cases among cattle in Uganda 

Rift Valley fever disease (RVF) is a notifiable zoonotic infectious disease reported in different parts of Sub-Saharan Africa. The spread of this disease is influenced by temperature, precipitation and land cover. The spread of RVF is facilitated through cattle movement and vector (Aedes and Culex mosquito) distribution. In cattle, RVF is detected through abortions and can be confirmed through real-time Polymerase Chain Reaction (RT-PCR).  Although, majority of cases are asymptomatic, there is available evidence that it leads to animal deaths affecting farmers’ livelihoods. Given its  asymptomatic nature coupled with low testing in Low and Middle Income countries, RVF surveillance remains a challenge. We sought to model missed cases of RVF infections by the time of the first cattle death and to recommend interventions.

We built an agent-based SIS model to simulate RVF transmission among cows in Uganda. The model is built using the “Starsim framework” from IDM.  We use Uganda’s cattle density and movement data, Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and MODIS NASA Earth Data to calibrate the model.  Relative susceptibility of each agent is defined as a function of the environment and acquired immunity. Immunity is acquired when a cow recovers from the disease. A dynamic network of contacts between cattle within a single district is also incorporated into the model.

Preliminary results show that by the time of the first death, we have missed 45 infected cattle. This underscores the need for systematic syndromic surveillance for RVF in endemic or high-risk areas like Uganda.

View Presentation for Agent-based modelling of missed Rift Valley Fever disease cases among cattle in Uganda 

Ms. Joy Nthiwa, CEMA: Modeling the impact of routine treatment on schistosomiasis elimination in Kenya

Kenya aims to eliminate schistosomiasis as a public health problem by 2030. The World Health Organization recommends mass drug administration (MDA) with a 75% therapeutic coverage to achieve this goal. Although Kenya conducts MDA campaigns annually, targeting 5-14 year-olds with a 75% coverage, the country has not yet reached the target. This is because 25% of eligible children are not treated during the campaign, including those who turn five years old within the year and have to wait until the next campaign. These untreated children contribute to the re-establishment of transmission, which hinders progress towards the 2030 goal.

To address this, we propose introducing routine treatment for children who turn five years old during the year after the campaign. To model the dynamics of schistosomiasis and the proposed intervention, we have developed an agent-based model based on the Starsim model developed by the Institute for Disease Modeling. Our model includes a baseline scenario representing the annual MDA campaign with a 75% coverage, as well as three intervention scenarios: annual MDA with routine treatment to maintain a coverage of 75%, 80%, and 85%. The purpose of the last two scenarios is to compare the prevalence levels between 75% coverage and higher coverage rates. We expect that combining the MDA campaign with routine treatment will result in lower prevalence compared to the baseline model without routine treatment. Additionally, we hypothesize that the annual MDA campaign with a 75% coverage and routine treatment will lead to lower prevalence compared to the traditional annual MDA campaign.

View Presentation for Modeling the impact of routine treatment on schistosomiasis elimination in Kenya

Dr. Naffiu Hussaini, Bayero University and Dr. Lawrence Nderu, Jomo Kenyatta University of Agriculture and Technology (JKUAT): Assessing the potential impact of Men5CV meningococcal vaccine on transmission dynamic of meningitis in Nigeria: An agent-based modelling approach

An agent-based model (ABM) was used to evaluate the impact of the Men5CV vaccine on the transmission dynamics of meningococcal meningitis in Nigeria. The objective of the study is to estimate the proportion of the population or sub-group that should be vaccinated for us to achieve herd immunity for meningitis in Nigeria by 2030. The transmission dynamics, vaccination impact, and treatment efficacy were examined. Data from WHO and existing literature informed our model. The research addresses key questions: how different Men5CV vaccination coverage levels affect the transmission dynamics of meningitis, the serogroups that the new vaccines will be able to avert  and cost-effective strategies for administering the Men5CV vaccine. Simulations from the model were run taking into account various vaccination rates and schedule while accounting for model uncertainties and population heterogeneity.

Results indicate that vaccination significantly reduces disease incidence and transmission rates. And vaccination of children aged between 9-18 months in the population did not have a greater impact compared to vaccinating the entire population. The limitation of the study was the data available on the contact patterns of people in Nigeria was not up to date. These results emphasise how critical it is to put in place vaccination campaigns specifically aimed at the Nigerian populace in order to lessen the effects of meningitis. The Menc5CV vaccine should not only target the under 5 but the whole population for Nigeria to achieve elimination of the disease.

View Presentation for Assessing the potential impact of Men5CV meningococcal vaccine on transmission dynamic of meningitis in Nigeria: An agent-based modelling approach

Ms. Eva Akurut, African Centre of Excellence in Bioinformatics and Data Intensive Sciences/Makerere Lung Institute: Modelling the impact of effective malaria vaccination on child mortality in Ugandan children under 5 years old

Introduction
Malaria remains a significant public health challenge, particularly in children under 5 years old, in high-burden countries like Uganda. Vaccines such as malaria can prevent disease severity and deaths. This study aimed to determine the malaria vaccination coverage required to achieve a 90% reduction in child mortality in Uganda, by 2030, as recommended by WHO.

Methods
We utilized an agent-based model and extended the SIR model to SVIRD. The modelling scenarios were baseline vaccine coverage levels (0.2, 0.7, 0.9) and vaccine efficacy rates (0.4, 0.7, 0.9).

Results
At 0.2 malaria vaccine coverage and 0.4 efficacy, approximately 50% of deaths were averted. At 0.7 vaccine efficacy, a 62% reduction in child mortality was achieved. Increasing the parameters to 0.9 resulted in a slight 1% marginal increase. These findings illustrate the role of malaria vaccination in reducing child mortality.

Conclusion
The integration of malaria vaccination into the Ugandan government’s enhanced programme on vaccination is needed. Even with low vaccine coverage and efficacy, there is a huge reduction on malaria mortality of children under 5 years old, as shown in this study. We recommend health authorities to aim for sustainable, moderate malaria vaccine coverage and efficacy which have a higher mortality reduction rate. High vaccination coverage and efficacy despite having a great impact are not cost-effective, other existing malaria interventions are needed for Uganda to reach the WHO 2030 target. The country should also ensure widespread vaccine availability, and enhance public awareness of malaria.

View Presentation for Modelling the impact of effective malaria vaccination on child mortality in Ugandan children under 5 years old

Dr. Verrah Otiende, USIU-Africa: Agent-based simulation of contact networks and mobility patterns for COVID-19 in Kenya

Following the COVID-19 pandemic, understanding the dynamics of infectious disease spread within communities has become imperative for effective public health interventions. This study aims to highlight the importance of leveraging advanced modeling techniques, to address the challenges of the transmission dynamics of infectious disease.

The study utilized Agent-Based Models (ABMs) to simulate contact networks and mobility patterns in Kenya. Like many other low- and middle-income countries, Kenya faced unique challenges in managing the spread of COVID-19. The model assessed the impact of two intervention strategies—vaccination, and movement restrictions—on mitigating the spread of COVID-19. The ABM model offered a flexible framework to capture the heterogeneity of human behavior, interactions, and movement, crucial for accurately representing the transmission dynamics of infectious diseases including COVID-19. Through scenario analysis and sensitivity testing, the model evaluated the complex interplay between individual behavior, contact network, and disease transmission dynamics.

Insights derived from the mathematical simulations revealed that achieving a vaccination coverage of 50% significantly reduced transmission rates, preventing an estimated 20-30% of potential infections, particularly among high-risk demographics. Movement restrictions were projected to yield a further 10-20% reduction in new cases, albeit a noticeable economic impact. Sensitivity analyses reinforced the robustness of the model in predicting transmission dynamics, emphasizing the critical role of evidence-based, context-specific strategies in managing transmission of infectious diseases.

View Presentation for Agent-based simulation of contact networks and mobility patterns for COVID-19 in Kenya

Dr. Evans Omondi, African Population and Health Research Center: Modelling the impact of male involvement on fertility preferences, contraceptive use, and pregnancy outcomes in Nairobi's Urban slums

Background
Kenya has implemented several policies, including the promotion of family planning, to improve overall health and well-being. Despite a decrease in fertility rates, significant disparities in maternal and child health outcomes persist due to varying levels of healthcare access, contraceptive awareness, and income. This study examines the impact of male involvement in fertility preferences and contraceptive use to guide the development of targeted policies and interventions for family planning.

Methods
An agent-based model (ABM), FPsim-NUHDSS, was developed to assess fertility patterns, contraceptive use, and pregnancy outcomes among women in the urban slums of Korogocho and Viwandani in Nairobi. Local data from the Nairobi Urban Health and Demographic Surveillance System facilitated model validation and calibration.

Results
The FPsim-NUHDSS model provides insights into fertility preferences, contraceptive use, and pregnancy outcomes in Korogocho and Viwandani. It highlights the importance of targeted interventions, such as family planning services and prenatal care, in improving maternal and child health outcomes. The results also underscore the relevance of male involvement and local context in decision-making processes to address health disparities effectively.

Conclusion
By utilizing the FPsim-NUHDSS model as a decision support tool, this research contributes to evidence-based policy decisions aimed at achieving Sustainable Development Goals (SDGs) related to poverty alleviation, healthcare access, education, and gender equality. Insights into fertility dynamics in urban slums inform strategies for sustainable and equitable development in Kenya, enhancing understanding of the interaction between socio-economic factors and health outcomes to guide more effective interventions.

View Presentation for Modelling the impact of male involvement on fertility preferences, contraceptive use, and pregnancy outcomes in Nairobi's Urban slums

Dr. Betsy Rono, Jomo Kenyatta University of Agriculture and Technology: Agent-based modelling (ABM) predicts number needed to vaccinate to achieve a 50% reduction in zero-dose vaccination among under-five (5) children in Kenya by 2025

Introduction
Zero-dose proportion remains a challenge in Kenya, with 7% of infants affected, according to Kenya Demographic Health Survey 2022. One in five children lack access to vaccines in Sub Saharan Africa. Globally, 14.2% are penta-zero-dose with 7.5% truly zero dose, indicating a lack of critical vaccinations among under-fives. Achieving WHO global target of 90% immunization coverage is vital to prevent 2.5 million vaccine-related deaths annually, among under-fives.

Methodology
Agent-based modelling was employed to project the reduction needed in zero-dose vaccination rates to achieve 50% decrease by 2025 in Kenya. Currently: diphtheria is eliminated; neonatal tetanus is near elimination; pertussis and measles have marked control; therefore, tetanus was adopted in the model. Agents alive and zero-dose were included in fitting the model. Starsim- SIS model was employed with tetanus incidence rate of 7.8/100,000 (2019), mortality rate of 6/100,000 (2019) and under-five mortality of 41/1000 (2022). Agents were defined as under-five children, interacting with caretakers, health-worker and environment as vaccination and health facilities.

Results
According to Kenya Health Information System, there were 130 tetanus cases in April 2024; approximately, 60 cases of tetanus shall be averted; if we reduce zero-dose cases by 50%; through enhanced uptake of vaccinations targeting <70 tetanus cases monthly; by reducing probabilities of infection from 2 to 1.3 and of susceptibility-again-state from 2 to 3 months’ post vaccination supported by SIS model calibration.
Conclusion: The model demonstrated that halving zero-dose vaccination reduces chance of tetanus cases and susceptibility by 2025.

View Presentation for Agent-based modelling (ABM) predicts number needed to vaccinate to achieve a 50% reduction in zero-dose vaccination among under-five (5) children in Kenya by 2025

Prof. Justice Moses Kwaku Aheto, University of Ghana: Spatiotemporal modelling and interactive web-based spatial mapping of malaria risk under Integrated Nested Laplace Approximation to support preventive and control efforts in Ghana

Under-five child malaria is one of the leading causes of morbidity and mortality globally, especially among sub-Saharan African countries like Ghana where malaria is responsible for about 20000 deaths in children annually of which 25% are those aged <5 years. To provide opportunities for efficient under-five malaria (U5M) surveillance and targeted control and elimination efforts amidst limited public health resources in poor-resource settings like Ghana where universal intervention is effectively impossible, there is the need to produce robust small-area (more localized) estimates of U5M risk to support policymakers and other stakeholders responsible for the survival of children.

To achieve this, we demonstrate how spatiotemporal modelling under the novel Integrated Nested Laplace Approximation (INLA) supported with web-based spatial mapping tools can be implemented to analyze U5M risk using routine surveillance health service nationally representative and comprehensive District Health Information Management System II (DHIMS2) data from 2016-2021 in Ghana. Also, we superimposed multiple estimates on the interactive web-based predictive maps for easy visualization. Substantial spatial and temporal differences in U5M risk across the 260 districts were found. The predicted national relative risk was 1.23 (SE: 0.0084) with a range of 0.0012 to 4.8291. Our approaches to the spatiotemporal modelling and mapping of U5M risk can serve as an effective tool to facilitate the identification of high burden U5M risk districts and the development of targeted public health interventions for the identified high risk U5M districts that require urgent attention and further research in a resource-limited setting where universal intervention is practically impossible. 

Dr. Ousmane Diao, MAP: Fine-scale malaria risk maps of routine incidence data in Senegal to inform risk stratification

In malaria endemic areas, identifying spatio-temporal hotspots is becoming an important element of innovative control strategies targeting transmission bottlenecks – especially in countries like Senegal targeting elimination by 2030. Despite significant progress towards elimination in Senegal from 2010 to 2018, there has been an increase in malaria cases between 2019 to 2021 from 359,246 to 547,7731 cases with increasing heterogeneity across the malaria transmission landscape, characterized by localized outbreaks. The necessity for a fine-scale approach to understanding the geographical distribution of malaria transmission is essential for informing priority operational research questions, including but not limited to risk stratification.

Here, we present the bespoke model designed by MAP to support Senegal’s NMCP in partnership with IDM to produce fine-scale spatial risk maps of incidence in Senegal. The model relied on routine surveillance data at health facility levels captured from 2020 to 2022. The risk maps were built using a joint-modeling approach of incidence case data and catchment modeling to allow a more accurate estimation of community-level incidence. A series of annual 1km pixel risk maps were produced and presented to the Senegal NMCP, which further utilized in their annual bulletin. A major limitation to this work was the lack of consensus on a master facility list with validation GPS coordinates. Further work is ongoing to explore the impact of incomplete facility lists have in producing bias within geospatial outputs.  

View Presentation for Fine-scale malaria risk maps of routine incidence data in Senegal to inform risk stratification

Mr. Eric Ibrahim, International Centre of Insect Physiology and Ecology: Spatio-temporal dynamics of malaria vector niche overlaps in Africa 

Malaria remains a significant public health challenge, particularly in Sub-Saharan Africa. Understanding the ecological interactions, especially niche overlaps, between primary and secondary malaria vectors is crucial for targeted malaria control and elimination strategies. This study employed a dynamic Cellular Automata (CA) model to map niche overlaps among primary (Anopheles gambiae complex, Anopheles funestus complex) and secondary (Anopheles pharoensis, Anopheles coustani) malaria vectors across Africa. We utilized environmental data spanning from 1985 to 2021 with a yearly temporal resolution, recognizing this as a limitation and acknowledging that finer temporal scales could potentially enhance the model’s accuracy and detail.

Our model revealed increasing niche overlaps across wide areas of Africa, attributed to vector expansion beyond native regions. These overlaps correspond to increased risks of sustained and prolonged malaria transmission. The model’s validation showed a high alignment with actual vector occurrences, affirming its effectiveness. However, the initial period selection and the yearly resolution of environmental data pose limitations to the model’s comprehensiveness.

The study highlights the growing issue of niche overlaps among malaria vectors in Africa, underscoring the need for vigilant monitoring and targeted vector control interventions in areas identified as overlap hotspots. The findings emphasize the importance of dynamic modeling approaches that incorporate continuous data updates for accurately capturing ecological interactions. Our study provides a framework for malaria control programs to strategize more efficiently, focusing on high-risk areas with significant niche overlaps. Further research incorporating data with finer temporal resolutions and extending historical data range could provide deeper insights into vector dynamics.

View Presentation for Spatio-temporal dynamics of malaria vector niche overlaps in Africa 

Dr. Onyango Sangoro, Ifakara Health Institute: Use of school-based malaria parasiteamia survey (SMPS) data to map the spatio-temporal malaria risk in mainland Tanzania

Background
Effective deployment of malaria interventions requires continuous spatio-temporal surveillance of malaria risk. Current methods of estimating malaria risk in Tanzania predominantly rely on health-facility and household-based malaria surveys that often underestimates the malaria burden in the population as individuals presenting to health facilities exclude asymptomatic cases in the community or are expensive and require extensive logistical undertaking which precludes continuous implementation required to correctly capture malaria trends.

Annual school-based malaria parasiteamia surveys (SMPS) present a potentially more representative, relatively cheaper and logistically feasible alternative for malaria surveillance and can be used to build a novel spatio-temporal approach that explores the malaria risk heterogeneity across Tanzania at a sub-national level.

Methods
We collate cluster level national malaria parasite prevalence surveys from 2011 to 2019. A suite of high-resolution satellite images and spatial covariates are pooled together at 1km resolution and passed through a causal inference algorithm for robust and parsimonious variable selection.  A Bayesian hierarchical spatio-temporal model is applied using R-INLA and Template Model Builder (TMB) to predict malaria prevalence at 1km resolution in unknown locations and timepoints. Prior to prediction, a 5-fold cross validation using random holdouts of 20% was conducted. A total of 18 static and dynamic covariates were selected. 

Findings
The results from the model are expected to inform Tanzania’s malaria stratification strategy and the deployment of interventions at a sub-national level. The uncertainty estimates from the model will allow decision makers to identify areas with high uncertainty driven by low sampling to improve site selection. 

View Presentation for Use of school-based malaria parasiteamia survey (SMPS) data to map the spatio-temporal malaria risk in mainland Tanzania

Mr. James Daniel Harborne, University of Nottingham: Stochastic averaging for a two-strain model of infectious disease epidemiology

We consider an extension to the standard compartmental SIR model of infectious disease epidemiology first presented by Kermack and McKendrick in 1927. Contrary to the standard SIR model with a single infected class, we consider the possibility of two competing strains of disease. We construct the stochastic model as a Continuous Time Markov Chain (CTMC), giving both the generator and the random time change representations. Under certain technical assumptions akin to those made in Quasi-Steady State Approximations (QSSA) literature, wherein we assume that the rates of infection and recovery of one strain is much faster than the other, we can use the stochastic averaging principle to obtain a reduced model. We then use this to prove a Functional Law of Large Numbers (FLLN), showing that the dynamics of our reduced stochastic model behave according to the solutions of a set of deterministic Ordinary Differential Equations (ODEs) in the large number limit. We also present numerical simulations of the solutions of the models.

View Presentation for Stochastic averaging for a two-strain model of infectious disease epidemiology

Dr. David Gurarie, Case Western Reserve University: Immune selection for multi-strain Plasmodium falciparum malaria

Multiple forces drive evolution of malaria parasites in host populations, and host immunity plays an important part. The principal immune evasion strategy adopted by Plasmodium falciparum (Pf) is antigenic variation, whereby parasite switched expressed antigens from its repertoire. This strategy is important for understanding spread, persistence and evolution of Pf in host populations.

To study such multi-strain malaria systems, we developed in-host model that employs genetically structured parasite makeup, and accounts for essential features of Pf biology and immunology. The model allows efficient simulation of infection histories in individual hosts, as well as host-ensembles and communities.

We applied it to explore basic questions of evolutionary biology of Pf, and its implications for parasite control. It includes (i) the meaning of ‘fitness’ and selection, (ii) parasite diversity and its population structure and distribution in host communities over long history (multiple transmission cycles), (iii) the relative importance of competitions vs. cooperation in parasite survival and spread.

View Presentation for Immune selection for multi-strain Plasmodium falciparum malaria

Ms. Anabelle Wong, Max Planck Institute for Infection Biology: Assessing the impact of social contact structure on serotype replacement following pneumococcal conjugate vaccination: A mathematical modeling study

Although pneumococcal conjugate vaccines (PCVs) have greatly reduced invasive diseases caused by vaccine-targeted serotypes (VT) of Streptococcus pneumoniae, vaccine impact may be eroded by the increase in rates of disease caused by non-vaccine serotypes (NVT) – a phenomenon known as serotype replacement. Here, we investigated the effect of social contact patterns on serotype replacement in carriage and the dynamics of indirect effects.

We developed a neutral, age-structured, susceptible–colonized (S–C) model incorporating VT-NVT co-colonization and childhood immunization with PCVs, and verified it against real-world carriage data. After simulating the serotype replacement dynamics with inferred contact matrices from 34 countries, we assessed the impact of contact patterns of different age groups on the time-to-replacement, here defined as the time taken for VT to drop to 5% of the pre-PCV level. Finally, we quantified the contribution of various parameters—such as inter-serotype competition and vaccine efficacy and coverage—to the dynamics.

Our model was able to recapitulate the VT carriage patterns observed in the real-world data and showed that varying the contact structure alone led to different time-to-replacement (range: 3.8 – 6.5 years). We found that higher total contact rate and assortativity in children under 5 were key factors in accelerating serotype replacement. In addition, higher vaccine efficacy and coverage, and to a lesser extent lower inter-serotype competition, led to shorter time-to-replacement.

These findings illuminate the mechanisms controlling the serotype replacement dynamics and may help predict the impact of the higher-valency PCVs in communities with different contact patterns.

View Presentation for Assessing the impact of social contact structure on serotype replacement following pneumococcal conjugate vaccination: A mathematical modeling study

Mr. David Kong, IDM: Installing a verbal autopsy system (EasyVA) in your own country

EasyVA is open source software developed by IDM to support both Physician Coded Verbal Autopsy (PCVA) and Computer Coded Verbal Autopsy (CCVA ). While global health communities are waiting for more sensitive autopsies, governments and researchers in LMICs still rely on verbal autopsies to determine the cause of death for decision making and for research development purposes. In this demonstration, we will explain the components of EasyVA and walk through the entire process of installing it in a local server using container technology.

View Presentation for Installing a verbal autopsy system (EasyVA) in your own country

Mr. Vivek Murali, Johns Hopkins University: Incorporating representative waiting time distributions in epidemiological models via optimized, generalized Erlang-distributions

In standard compartmental models of infectious diseases, assumptions about the duration of time spent in each state (e.g., the latent period, the infectious period) significantly influence model outcomes and interpretation. When represented deterministically as systems of ordinary differential equations or stochastically as Markov chains, these models often assume constant transition rates between states, which implies an exponential distribution of “waiting times” in each state. However, this assumption is not consistent with available data for many diseases. An alternative that remains analytically or computationally tractable is to model each state as a series of multiple identical substates, which results in a gamma (Erlang) distributed overall duration  (a method known as the “linear chain trick”). While well-known, this approach is limited in the distributions it can approximate and few studies have presented formal methods for determining the optimal number of substates.

In this work, we develop a method to help create more realistic models by identifying the smallest number of substates – and the optimal transition rate for each – to match arbitrary empirical distributions of durations using generalized Erlang distributions. Our approach combines maximum likelihood parameter estimation with non-parametric significance testing to prioritize the simplest model structures that produce waiting times consistent with observed data. We demonstrate the utility of our method with examples from multiple diseases, and create a user-friendly tool to use this estimation approach in model construction and testing, facilitating the creation of more accurate and representative infectious disease models.

View Presentation for Incorporating representative waiting time distributions in epidemiological models via optimized, generalized Erlang-distributions

Mrs. Mary Ogunmodimu, Federal University of Technology Akure, Ondo State, Nigeria: An optimal control analysis of COVID-19 and tuberculosis co-infection dynamics in the presence of disease relapse

Patients of Covid-19 are predisposed to aggravated activation of latent infection with tuberculosis due to their weakened immune system. These diseases potentiate their effect on each other and mutually accelerate their progression to a critical stage. This study presents a comprehensive deterministic model, which incorporates the case for disease(s) relapse, for Covid-19 and tuberculosis co-infection dynamics. The positivity and boundedness of the model’s solution were established. The disease-free-equilibrium point of the model was obtained and shown to be stable whenever the effective reproduction number is less than unity.

Numerical simulation of the model, performed by implementing the fourth order Runge Kutta method on MATLAB subroutine, shows that Covid-19 incidence decreases with increase in treatment, vaccination, natural and disease induced death rates. Meanwhile, there exists a prevalence with increase in case of infection after vaccination, contact and recruitment rates. The burden of tuberculosis on the human population increases with an increase in case of infection after vaccination, contact and progression rates from latent to active TB.

An optimal control application was made on the model with the control variables including isolation upon early diagnosis, efficacious treatments and vaccination for both diseases. An optimality system (OS) was derived using the Pontryagin’s maximum principle. The OS, consisting of the model’s state and adjoint co-state equations, was solved and simulated for different combinations of the controls using the forward-backward sweep algorithm on MATLAB subroutine. Graphical results of the simulation were discussed and the most effective combination for the eradication of the co-infection was obtained.

View Presentation for An optimal control analysis of COVID-19 and tuberculosis co-infection dynamics in the presence of disease relapse

Dr. Aurelien Cavelan, Swiss Tropical and Public Health Institute: Improved calibration of OpenMalaria with efficient multi-objective Bayesian optimisation

Infectious diseases, including malaria, continue to be a significant global health concern. Accurate and well-calibrated models are essential for understanding the temporal dynamics of immunity and transmission, as well as estimating the short and long-term effects of interventions on disease burden. However, calibrating these models to diverse data sources is complex and often requires optimizing multiple objective functions over a large parameter space.

This study presents an updated calibration of OpenMalaria, a widely used individual-based model of Plasmodium falciparum malaria in humans. OpenMalaria simulates the dynamics of malaria parasitaemia in the course of an infection, of transmission, of immunity and of the processes leading to illness and death and captures the delivery and impact of many malaria interventions.

The updated calibration workflow includes new curated data from low-incidence settings, a larger parameter space and several model updates including new features geared towards better scenario accuracy, which is key for accurately simulating historical studies. We systematically assess the impact of model parameterizations on key epidemiological relationships at steady-state and under perturbations from interventions, which is an important step of the validation process, resulting in a more reliable, accurate and robust model.

We have developed an efficient multi-objective calibration workflow based on the Trust Region Bayesian Optimization (TuRBO) algorithm that improves the convergence and the performance of the optimization process. We also address issues such as parameter and model identifiability, which are inherent to this problem. Our calibration workflow demonstrates applicability beyond OpenMalaria, supporting diverse modeling efforts. The algorithm and software for the calibration are made publicly available on GitHub, together with OpenMalaria and the updated model parameterization. 

View Presentation for Improved calibration of OpenMalaria with efficient multi-objective Bayesian optimisation

Mr. Tobias Holden, Northwestern University: Bayesian optimization frameworks for recalibration of EMOD’s within-host malaria model

In the complex agent-based malaria transmission model, EMOD, relevant disease processes are governed by equations with parameters that can not be measured and must be inferred through calibration. In calibration, new combinations of model parameter values are tried in simulations, and outputs are compared to reference data to find the best fit. However, EMOD simulations need a lot of time and CPU to run, and the number of possible parameter combinations rises exponentially with the number of parameters under calibration – an obstacle known as the ‘curse of dimensionality’. Our goal was to recalibrate 14 EMOD parameters describing modeled infections, immunity, and parasite dynamics to explore an increase in the maximum concurrent infections per individual and a new custom model of age-related innate immune variation.

We tested two Bayesian optimization frameworks using single- or multi-task Gaussian processes (GPs) as surrogate models to emulate correlations between EMOD parameters and simulation goodness-of-fit to 18 total data objectives describing incidence, prevalence, parasite density, or infectiousness from 8 study sites across Sub-Saharan Africa. We used different trust-region-based sampling methods to strategically sample new parameter sets for further simulation and GP training. The optimization reached convergence and outperformed default EMOD parameters, identifying a recalibrated within-host model with improved fit to in-sample and out-of-sample data. The methods demonstrated are flexible to the incorporation of new within-host datasets as reference targets, and also extensible for calibration of larval habitat, intervention effect sizes, or other EMOD modules.

View Presentation for Bayesian optimization frameworks for recalibration of EMOD’s within-host malaria model

Dr. Daniel Klein, IDM: Noise-free comparison of stochastic agent-based simulations using common random numbers

Random numbers are at the heart of every agent-based computer model of health and disease. By representing each individual in a synthetic population, agent-based models enable detailed analysis of intervention impact and parameter sensitivity. Yet agent-based modeling has a fundamental signal-to-noise problem, small signals cannot be reliably differentiated from stochastic noise resulting from misaligned random number realizations. We introduce novel methodology that eliminates stochastic noise through techniques that achieve common random number alignment, a first for agent-based modeling. Our approach enables meaningful individual-level analysis between ABM scenarios because all differences are driven by mechanisms and “butterfly flaps wings” effects rather than random number noise. We demonstrate benefits of our approach on three disparate examples using the Starsim Framework and discuss limitations.

View Presentation for Noise-free comparison of stochastic agent-based simulations using common random numbers

Mr. Victor Mero, University of California Berkeley: Modeling the impact of gene drive mosquito release on community immunity and malaria resurgence

Malaria remains a major public health challenge in low- and middle-income countries. Gene drive mosquitoes, a novel intervention, are expected to reduce the population of malaria vectors or impair their transmission capability. While initially effective in reducing malaria incidence, this approach may also decrease community immunity, especially among new generations with no exposure to the malaria parasite. The emergence of resistant alleles could allow mosquitoes to bypass genetic modifications, leading to a resurgence of competent vectors over time.

This study, using the MGDrivE 3 modeling software package, explores the implications of diminished immunity due to reduced exposure to malaria vectors. Our simulation analyzes the spread of resistant alleles in gene drive mosquitoes, which could lead to a resurgence of malaria vectors and a subsequent increase in malaria cases due to reduced community immunity. We examine the impact on different age groups and demographics, with a particular focus on younger populations. Strategies to mitigate these risks include renewed mosquito suppression efforts, vaccination programs, and public health campaigns.

The findings underscore the need for continuous monitoring of gene drive projects and the development of contingency plans to address potential resistance. Additionally, the study highlights the importance of integrating genetic control methods with traditional malaria control strategies, such as insecticide-treated nets (ITNs) and indoor residual spraying (IRS), to ensure sustainable malaria vector control.

Dr. Mamadou Alpha Diallo, Cheikh Anta Diop University: Harnessing AI for infectious disease control in Africa

Our project leverages advanced AI, funded by the Bill & Melinda Gates Foundation’s Grand Challenges initiative, to tackle infectious diseases in Africa. Specifically, we aim to enhance decision-making, accelerate evidence-to-policy pathways, and improve communication and resource allocation by analyzing diverse datasets, including environmental factors, health reports, epidemiological data, and mosquito surveillance data. During the pilot phase, our model focused on malaria and utilized ChatGPT-4, showing promising results in these areas.

We aim to improve the speed, accuracy, inclusivity, and quality of decision-making through activities such as data collection and analysis, decision support, policy development, and communication. ChatGPT-4 analyzes and interprets complex data sets, scientific literature, and epidemiological trends, contributing to evidence-based strategies for infectious disease prevention, control, and treatment.

Our goal is to enhance the quality and timeliness of communications related to infectious diseases in Africa. ChatGPT-4 generates context-specific, actionable information for healthcare workers, policymakers, and communities. AI-driven insights optimize the distribution of healthcare resources and interventions through communication planning, resource allocation optimization, and tailored messaging.

Preliminary Results: ChatGPT-4 has demonstrated significant potential in reducing the time required to analyze complex data. It quickly and accurately performed Kaplan-Meier survival analysis using R scripts in analyzing Therapeutic Efficacy Study (TES) data, a task typically requiring specialized expertise. It also extracted valuable information from large Excel datasets about rapid diagnostic tests (RDTs), malaria in pregnancy, malaria in children and medicine use, aiding in effective resource allocation. Furthermore, ChatGPT-4 assisted in analyzing complex genomic data, showcasing its ability to handle sophisticated tasks usually reserved for bioinformaticians.

While still in its early stages, the model’s ability to predict malaria outbreaks could lead to more proactive and targeted interventions, potentially transforming our approach to malaria elimination in Senegal.

Dr. Wesley Wong, Harvard TH Chan School of Public Health: MalKinID (Malaria Kinship Identifier): A likelihood model for identifying parasite genealogy relationships based on genetic relatedness

Pathogen genomics is a potent tool for tracking infectious disease transmission. In malaria, sexual reproduction and recombination in mosquitoes produces genetically related progeny whose genomes contain identical-by-descent (IBD) segments. In theory, IBD can be used to distinguish genealogical relationships (parent-child [PC], full-sibling [FS], etc) and reconstruct transmission history or identify parasites for genotype-to-phenotype quantitative-trait-locus (QTL) experiments.

We developed MalKinID (Malaria Kinship Identifier), a new likelihood model based on genomic data from three laboratory-based genetic crosses (yielding 440 PC and 9060 FS comparisons). MalKinID uses the genome-wide IBD proportion and the per-chromosome max IBD segment block and IBD segment count distributions to identify parasite genealogical relationships. MalKinID was assessed using empirical lab-cross data and simulated importations. MalKinID identified lab-generated F1 progeny with >80% sensitivity and showed that 0.39 (95% CI: 0.28, 0.49) of the second-generation progeny of an NF54 and NHP4026 cross were F1s and 0.56 (0.45, 0.67) were backcrosses with the parental NF54 strain. For simulated, outcrossed point importations, MalKinID reconstructs genealogy history with high precision and sensitivity.

The F1-scores for PC, FS, second-degree, and third-degree relatives were 0.95 (0.84, 1.0), 0.94 (0.72, 1.0), 0.84 (0.60, 1.0) and 0.84 (0.64, 0.94), respectively. When importation involves inbreeding, such as during serial cotransmission, the precision and sensitivity of MalKinID declined, with F1-scores of 0.76 (0.56, 0.92) and 0.23 (0.0, 0.4) for PC and FS and <0.05 for second-degree and third-degree relatives. MalKinID establishes a foundation for using IBD to reconstruct transmission lineages in outcrossed parasite populations or identifying progeny for QTL experiments. 

View Presentation for MalKinID (Malaria Kinship Identifier): A likelihood model for identifying parasite genealogy relationships based on genetic relatedness

Dr. Geoffrey Siwo, University of Michigan Medical School: Grounding large-language models with real-world public health knowledge using Chain of Cause Reasoning

Large language models (LLMs) are a promising tool for democratizing access to knowledge across domains including clinical medicine and global health. Even though LLMs excel in numerous question and answer tasks, they are highly susceptible to providing answers that may sound coherent yet confabulated, an issue that has been commonly referred to as hallucinations. Furthermore, current LLMs provide answers to questions using highly probabilistic deep neural networks (DNNs), whose eventual outputs do not involve reasoning. This reduces their reliability in critical areas like medicine and public health where decisions should be informed by valid reasons. Training LLMs on larger datasets or fine-tuning them are not effective in making them factual or non-hallucinatory.

In this talk, I will introduce Chain of Cause Reasoning (CCR)- a new prompting technique we are developing to enable LLMs to incorporate causal reasoning into their answers. CCR is inspired by human modes of thought based on psychology: System 1 which is fast and involuntary, and System 2 which is slow and involves thoughtful reasoning. Briefly, CCR involves guiding an LLM to incorporate a causal graph between different factors, which for public health can include health, environmental and biological factors. I will provide examples how public health practitioners and researchers can use CCR to explore the impact of different malaria control interventions on malaria transmission, and test and generate hypotheses for malaria drug resistance emergence. CCR provides transparent, causally grounded answers to public health questions and allows practitioners to explore different scenarios customized to their context.  

View Presentation for Grounding large-language models with real-world public health knowledge using Chain of Cause Reasoning

Dr. Scott Olesen, US CDC Center for Forecasting and Outbreak Analytics: Real-time use of a dynamic model to measure the impact of public health interventions on measles outbreak size and duration — Chicago, Illinois, 2024

Measles is a highly infectious, vaccine-preventable disease that can cause severe illness, hospitalization, and death. A measles outbreak associated with a migrant shelter in Chicago occurred during February–April 2024, in which a total of 57 confirmed cases were identified, including 52 among shelter residents, three among staff members, and two among community members with a known link to the shelter.

CDC simulated a measles outbreak among shelter residents using a dynamic disease model, updated in real time as additional cases were identified, to produce outbreak forecasts and assess the impact of public health interventions. As of April 8, the model forecasted a median final outbreak size of 58 cases (IQR = 56–60 cases); model fit and prediction range improved as more case data became available. Counterfactual analysis of different intervention scenarios demonstrated the importance of early deployment of public health interventions in Chicago, with a 69% chance of an outbreak of 100 or more cases had there been no mass vaccination or active case-finding compared with only a 1% chance when those interventions were deployed.

This analysis highlights the value of using real-time, dynamic models to aid public health response, set expectations about outbreak size and duration, and quantify the impact of interventions. The model shows that prompt mass vaccination and active case-finding likely substantially reduced the chance of a large (100 or more cases) outbreak in Chicago. 

View Presentation for Real-time use of a dynamic model to measure the impact of public health interventions on measles outbreak size and duration — Chicago, Illinois, 2024

Ms. Elizabeth Goult, Max Planck Institute for Infection Biology: Estimating the optimal measles vaccination age

Between 2020 and 2023, over 30 countries reported more than 1,000 measles cases per year to the World Health Organization (WHO). The persistence of measles in many countries demonstrates large immunity gaps, resulting from incomplete or ineffective immunization with measles-containing vaccines (MCVs). A key factor affecting MCV impact is age, with infants receiving dose 1 (MCV1) at older ages having a reduced risk of vaccine failure. This results in a trade-off in risks – vaccinating too early risks vaccine failure, while vaccinating too late risks infant contracting measles before vaccination.

Here, we designed a new method—based on a transmission model incorporating realistic vaccination delays and age variations in MCV1 effectiveness—to capture the MCV1 age risk trade-off and estimate the optimal age for recommending MCV1. We predict a large heterogeneity in the optimal MCV1 ages (6–20 months), contrasting the homogeneity of observed recommendations worldwide (89% of countries recommend either 9 or 12 months).

Furthermore, we show that the optimal age depends on the local epidemiology of measles, with a lower optimal age predicted in populations suffering higher transmission. More specifically the social contact structure, MCV1 coverage, and mean age of infection all substantially impact the optimal vaccination age. Overall, our results suggest the scope for public health authorities to tailor the recommended schedule for better measles control.

View Presentation for Estimating the optimal measles vaccination age

Mr. Brian Njuguna, University of Nairobi/CEMA: Measles elimination strategies in Kenya

Background
WHO has set a target to eliminate measles in 80% of African countries by 2030. Africa and Asia contribute to the biggest burden of this disease. For over 60 years, a two-dose vaccination regimen has been the primary intervention, achieving varying levels of success. Kenya continues to report annual measles transmission. To address this, we aim to identify the most effective vaccination strategies for achieving measles elimination.

Methods
We used an agent-based model to assess the impact of routine vaccination, targeting 95% coverage of the eligible population from 2020 to 2040. Using data from the Kenya health integrated system and the 2022 Kenya demographic health survey, we calculated the national and sub-national incidence of measles and vaccination coverage. Our goal was to determine the optimal vaccination scenario for achieving measles elimination by 2030. We compared three scenarios: i) maintaining current vaccination coverage, ii) supplementing current vaccination coverage with supplementary immunization activities (SIA), iii) increasing vaccination to the recommended 95%.

Results
In areas with low measles incidence (1 case per million), elimination could be achieved in 4 years by maintaining 95% coverage for the first dose (MCV1). In areas with moderate incidence (300 cases per million), elimination would be reached in 10 years with 95% MCV1 coverage, and at least 50% coverage for the second dose (MCV2). In high prevalence areas (>400 cases per million), elimination in 11 years would require 95% coverage for both MCV1 and MCV2, supplemented with SIAs every 2 years.

Conclusion
Precision in administering interventions could help us reach our 2030 targets.

View Presentation for Measles elimination strategies in Kenya

Dr. Katherine Rosenfeld, IDM: Spatial modeling in support of measles control and elimination

Measles is a highly infectious disease that can lead to severe outcomes and even death, especially in children. Despite the availability of a safe and effective vaccine, measles remains endemic in many parts of the world, where routine immunization coverage is low and heterogeneous. To achieve measles control and elimination, it is essential to understand the spatial dynamics of transmission and the impact of different vaccination strategies in heterogeneous settings. In this study, we apply a multi-resolution spatial model of measles transmission, calibrated to demographic and epidemiological country data. Our study focuses on establishing benchmarks that can inform policy and underscores the significance of capturing the spatial dynamics of this highly contagious disease.

View Presentation for Spatial modeling in support of measles control and elimination

Dr. Isaac Lyatuu, Prime Health Initiative Tanzania: Leveraging machine learning models to forecast adverse maternal outcome in low-resource settings: An experience from Geita Tanzania

In low and middle-income countries, the persistently high maternal mortality rates (MMR) pose a grave concern, leading to tragic losses of both mothers and infants due to preventable complications. Tanzania, with an MMR of 220 deaths per 100,000 live births, ranks among the highest in the region, exacerbating already challenging living conditions. Studies and maternal death audits have identified several contributing factors to this crisis, including insufficient Antenatal Care (ANC) attendance hindering early screening for adverse maternal outcomes, inadequate ANC screening due to heavy workloads and limited human resources, and a lack of resources for necessary tests and investigations.

Early prediction of maternal complications holds immense potential for preventing these adverse events, thereby saving lives. In response, the Prime Health Initiative Tanzania (PHIT), supported by the Bill & Melinda Gates Foundation (BMGF), initiated a two-year project aimed at predicting adverse maternal outcomes among expectant mothers accessing antenatal care services. This project utilized a tablet-based approach to streamline digital data entry, coupled with a machine learning model for detecting and predicting adverse maternal outcomes. Data collected from 187,438 clients with repeated observations were used to train a predictive model using distributed gradient boosting techniques (XGBoost). Preliminary results indicate a model accuracy and precision of 99% and 43%, respectively, for detecting the risk of gestational hypertension. We aim to refine data collection processes and improve feature selection for more accurate predictions. These early findings underscore the transformative potential of digital health in enhancing maternal outcomes and easing the burden on healthcare providers.

View Presentation for Leveraging machine learning models to forecast adverse maternal outcome in low-resource settings: An experience from Geita Tanzania

Ms. Javairia Khalid, Aga Khan University: Using machine learning to determine the association of maternal characteristics and maternal serum-related biomarkers with newborn outcomes

In 2018, around 40% of children under 5 years of age were stunted in South Asia, which makes the prevalence of stunting higher in this region than others worldwide. Stunting contributes towards poor health outcomes later in life and is strongly correlated with impaired cognitive development. In most cases, stunting starts in utero which is why prenatal identification of children at risk for stunting at birth is crucial.
The aim of this study was to identify the maternal characteristics and maternal serum biomarkers that are strong predictors of stunting at birth. , Height-for-age (HAZ), Weight-for-age (WAZ), and Weight-for-height z-scores (WHZ) and presence of small for gestational age (SGA) of children at birth were obtained from two peri urban sites in Karachi. Maternal characteristics and maternal serum biomarkers were measured at the 24-28 weeks pregnancy visit. A descriptive analysis of maternal clinical factors and neonatal growth measurements was performed.

Random forest model was used to assess the importance of different maternal characteristics and maternal serum biomarkers in predicting stunting, wasting, underweight and SGA at birth.

Of the 1059 children, 29.7% children were small for gestational age at birth with no difference between genders. About 20.9% children were stunted, with more males than females. Nearly 12.2% were observed as wasted and distribution of wasting across gender was slightly more for males than females. Only 22.8% of studied population were under weighted with distribution for under weight slightly more in male children than female children. The random forest classifier identified mother’s age, Mid Upper Arm Circumference (MUAC) and Body Mass Index (BMI) as the strongest predictors for stunting, wasting, underweight and SGA at birth. Of the twelve biomarkers, Placental Growth Factor (PLGF) was identified as the strongest predictive maternal serum biomarker for all four phenotypes. The model predicted the growth outcome with 76% accuracy.

We demonstrated that maternal characteristics and systemic pro-inflammatory and inflammatory maternal serum biomarkers are associated with stunting, wasting, underweighting and SGA at birth. Biomarkers such as CRP, ferritin, and pregnancy related hormones such as PLGF, SFLT are strong predictors of malnourishment birth.”

View Presentation for Using machine learning to determine the association of maternal characteristics and maternal serum-related biomarkers with newborn outcomes

Dr. Elzo Pereira Pinto Junior, CIDACS/Fiocruz: Evaluating the quality of maternal and child health services in Brazilian primary health care: A latent transition analysis

Maternal and child health conditions are critically influenced by healthcare services, particularly those provided through Primary Health Care (PHC). This study aimed to evaluate the adequacy of maternal and child health services in PHC using quality standards based on a normative approach and examine changes over eight years using latent transition analysis. Data were sourced from three cycles of a national survey conducted in Brazil between 2011 and 2018, assessing the infrastructure and work processes of PHC teams. The study included 8,776 PHC teams participating in all three evaluation cycles. Quality was measured using indicators for prenatal care (5), child health care (5), and immunization services (4). Conditional and transition probabilities and the prevalence of each class over time were estimated using MPLUS.

The analysis identified two classes (entropy = 0.749): high adequacy and low adequacy. Even within the high adequacy class, some indicators had conditional probabilities below 70%, such as surveillance activities, availability of immunizations, provision of prenatal tests and syphilis treatment, and strategies for postpartum care. The prevalence of high adequacy increased from 29.3% (T0) to 39.2% (T1) and 88.4% (T2). Between the first and second cycles, the probability of transitioning from low to high adequacy was 27.5%. This probability increased significantly to 80.8% between the second and third cycles, indicating substantial improvements in PHC quality over time. These findings underscore the significant advancements in PHC in Brazil and highlight the relevance of using this approach to measure and track the quality of PHC services over time.

View Presentation for Evaluating the quality of maternal and child health services in Brazilian primary health care: A latent transition analysis

Mr. Avik Kumar Sam, Indian Institute of Technology: Association of malaria in northeast India with land-use patterns, meteorology and backward communities 

Malaria incidences across India are steadily declining; however, certain regions, specifically the northeast, contribute a greater burden, driven by the interaction of ecological and climatic factors. The state of Tripura, in northeast India, exhibits a unique type of farming called “Jhum” cultivation by indigenous communities, which was previously associated with an increased malaria risk in Bangladesh, India’s neighbouring country. 

The present study aims to develop a quantitative relationship between ecological and climatic variables and malaria incidence in Tripura, India. 

The response variables are annual parasitic incidences (API) and malaria cases reported in 78 blocks across eight districts. The predictor variables are area under croplands, forest coverage, rubber plantations, Jhum lands, urban settlements, scrublands, temperature, elevation, average precipitation and the minority (ST) population. Due to 41% zero cases, we used regularized zero-inflated Poisson regression for cases and zero-inflated Gaussian regression for API. Lasso, ridge, and elastic net were used as regularization functions to facilitate simultaneous variable selection and model calibration and safeguard the predictions against ill-conditioning and multicollinearity exhibited by the data. 

Malaria transmission is reported in 55 blocks, with the maximum reported in Silachhari block (API = 56.3). No transmission is reported in Sepahijala and West Districts, while high transmission is reported from Dhalai and Gomati. Elastic net penalty is the preferred regularization function in API and case modelling, leading to the smallest mean squared error (MSE=19.3). The model exhibits Jhum lands as the most significant predictor, followed by scrublands, ST, precipitation, temperature and elevation (all significant, p< 0.05).  

Ms. Charlène Naomie Tedto Mfangnia, University of Dschang, Cameroon / icipe, Kenya: System dynamics to support targeted and climate-informed releases of MB-infected mosquitoes for malaria control in Kenya

The emergence of insecticide resistance poses a significant threat to malaria control efforts, emphasizing the imperative of developing innovative and bio-based strategies. One promising approach involves spreading Microsporidia MB (MB), an endosymbiont in Anopheles mosquitoes known to block Plasmodium transmission. MB has the advantage of being naturally present in certain Anopheles mosquitoes and sustaining itself through vertical and horizontal transmission.

This work investigates the effectiveness of controlling malaria by increasing the prevalence of MB-infected mosquitoes. A highly aggregated system dynamics model is developed to capture the interactive dynamics of MB, mosquitoes, and humans, while also exploring the potential for integrated malaria control. Then, the model is calibrated using data collected from Ahero in Kenya, available for 1.25 years in the past. Statistics of fit indicate the model effectively represents the trend for the selected key performance indicators: total mosquito population, MB and malaria prevalence. Simulation results enabled the assessment of the impact of diffusing MB-infected mosquitoes on malaria prevalence, informing optimal release strategies in Ahero. Additionally, integrated control strategies were evaluated for enhanced effectiveness.

To facilitate knowledge exchange, an accessible interface (link-to-interface) was developed, allowing non-modelling researchers to simulate and observe outcomes under different release policies. This interface empowers users to adjust key parameters, tailoring the model to diverse contexts. Furthermore, to aid in identifying relevant target areas for MB-based interventions, we provide maps assessing the climate suitability across years in Kenya. This study provides a tool to guide policymakers, in the strategic implementation of MB-based malaria control interventions.

View Presentation for System dynamics to support targeted and climate-informed releases of MB-infected mosquitoes for malaria control in Kenya

Dr. Hillary Topazian, Imperial College London: Estimating the potential impact of surveillance test-and-treat posts to reduce malaria in border regions in sub-Saharan Africa: A modelling study

The last malaria cases in near-elimination settings are often found in international border regions due to the presence of hard-to-reach populations, conflict, uneven intervention coverage, and human migration. Test-and-treat border posts are an under-researched form of active case detection used to interrupt transmission chains between countries.

We used an individual-based, mathematical metapopulation model of P. falciparum to estimate the effectiveness of border posts on total cases in malaria-endemic sub-Saharan Africa. We estimated that implementation of international border posts across 401 sub-national administrative units would avert a median of 7,173 (IQR: 1,075 to 23,550) cases per unit over a 10-year period and reduce PfPR2-10 by a median of 0.21% (IQR: 0.04% to 0.44%). Border posts were most effective in low-transmission settings with high-transmission neighbors. Border posts alone will not allow a country to reach elimination, particularly when considering feasibility and acceptability, but could contribute to broader control packages to targeted populations.

View Presentation for Estimating the potential impact of surveillance test-and-treat posts to reduce malaria in border regions in sub-Saharan Africa: A modelling study

Ms. Lydia Braunack-Mayer, Swiss Tropical and Public Health Institute: Combining seasonal malaria chemoprevention with new therapeutics for malaria prevention: a mathematical modelling study

In recent years, research and innovation for malaria prevention has led to the development of new vaccines, monoclonal antibodies, and small molecule drugs. If approved, these therapeutics could become important components of a comprehensive malaria prevention strategy. To effectively prioritize new therapeutics for clinical development, evidence is needed to understand the benefits of combining them with malaria chemoprevention.

We developed and applied an individual-based malaria transmission modelling framework to estimate the impact of combining therapeutics with seasonal malaria chemoprevention (SMC) for preventing malaria in children.
In this talk, we present an application of this modelling framework to novel malaria therapeutics with multi-stage activity, focusing in particular on the challenges of modelling products that have not yet entered clinical trials. Our framework combines emulator-based methods with explicit models of intervention dynamics for a range of hypothetical therapeutics, including pre-liver stage vaccines and therapeutics that target multiple stages of the malaria transmission life cycle.

Our findings indicated that, when a pre-liver stage therapeutic was combined with SMC, the new therapeutic needed to provide lasting protection in order to sustain benefit for children after the interventions stopped. Additionally, we found that deploying multi-stage therapeutics with both blood and liver stage activity reduced the burden of severe malaria throughout childhood. Our work thus articulates the benefits of a multi-stage malaria prevention therapeutics across the life-course of a child, strengthening the effectiveness of SMC. Ultimately this will inform the selection of multi-stage candidates in advance of phase two and three clinical trials.

View Presentation for Combining seasonal malaria chemoprevention with new therapeutics for malaria prevention: a mathematical modelling study

Dr. Samuel Mwalili, Center for Health Analytics and Modelling, Strathmore University, Kenya: Estimated HIV risk around funeral practices and mitigation strategies in western Kenya: A mathematical modeling study

Disco matanga, also known as “disco funerals,” are culturally significant celebrations of a deceased person’s life in various parts of Africa. We used agent-based network modeling to estimate how disco matanga impacted HIV transmission, and to explore the impact of relevant biomedical, biobehavioral, and structural interventions to reduce HIV risk.  The model, incorporating cultural assumptions and risk factors, explored interventions including HIV testing and promoting safer sex while addressing structural issues.

We adapted EMOD-HIV, a previously validated network-based model of HIV in the Nyanza region of Kenya. We compared past HIV incidence (1980–2024) with and without incorporating disco matanga, and future HIV incidence (2025–2050) with different interventions for disco matanga attendees: (1) biomedical (HIV prophylaxis), (2) biobehavioral, (3) structural. We estimated the impact of disco matanga on HIV infections and deaths, considering intervention uptake variations. From 1980 to 2024, disco matanga contributed 4.3% (95% CI: 2.1%–5.7%) of all HIV infections, peaking at 9.9% (95% CI: 6.3%–12.0%) in 2004. Implementing biomedical interventions at disco matanga (with 70% coverage) could avert 6.5% (95% CI: 5.7%–7.3%) of adult HIV infections and 1.7% (95% CI: 1.3%–2.0%) of deaths, with bio-behavioral and structural interventions showing lesser impacts.

We conducted the first modelling study that simulated the dynamics among disco matanga, HIV/AIDS, and intervention methods. Biomedical, biobehavioral, or structural strategies directed at disco matanga could substantially reduce HIV transmission and mortality in the Nyanza region. Further investigation is required to assess the feasibility and cultural appropriateness of tailored HIV interventions.

View Presentation for Estimated HIV risk around funeral practices and mitigation strategies in western Kenya: A mathematical modeling study

Dr. Tyler McCormick, University of Washington: From narratives to numbers: Valid inference using language model predictions from verbal autopsy narratives

In settings where most deaths occur outside the healthcare system, verbal autopsies (VAs) are a common tool to monitor trends in causes of death (COD). VAs are interviews with a surviving caregiver or relative that are used to predict the decedent’s COD. Turning VAs into actionable insights for researchers and policymakers requires two steps (i) predicting likely COD using the VA interview and (ii) performing inference with predicted CODs (e.g. modeling the breakdown of causes by demographic factors using a sample of deaths).

In this paper, we develop a method for valid inference using outcomes (in our case COD) predicted from free-form text using state-of-the-art NLP techniques. This method, which we call multiPPI++, extends recent work in “prediction-powered inference” to multinomial classification. We leverage a suite of NLP techniques for COD prediction and, through empirical analysis of VA data, demonstrate the effectiveness of our approach in handling transportability issues. multiPPI++ recovers ground truth estimates, regardless of which NLP model produced predictions and regardless of whether they were produced by a more accurate predictor like GPT-4-32k or a less accurate predictor like KNN.

Our findings demonstrate the practical importance of inference correction for public health decision-making and suggests that if inference tasks are the end goal, having a small amount of contextually relevant, high quality labeled data is essential regardless of the NLP algorithm.

Dr. Abraham Flaxman, Institute for Health Metrics and Evaluation: AI and verbal autopsy: Predicting pregnancy-related causes of death

Understanding the causes of death within a population is crucial for formulating effective health policies and allocating resources, especially in low- and middle-income countries where medical certification is often lacking. Verbal autopsy (VA) provides a method to estimate causes of death through interviews with acquaintances of the deceased. However, translating VA data into precise causes of death remains challenging due to the need for expert knowledge and the scarcity of skilled professionals.

This study explores the potential of large language models (LLMs), specifically ChatGPT 4.0, to improve the accuracy and efficiency of determining causes of death from VA data, focusing on maternal deaths. We used the PHMRC Gold Standard Verbal Autopsy dataset, which includes over 12,000 VA interviews, to evaluate our approach. We assessed the LLM’s performance using metrics such as Chance-Corrected Concordance (CCC) and Cause-Specific Mortality Fraction (CSMF) accuracy.

Our findings show that LLM-prompted approaches, especially those using multiple-choice prompts and comprehensive data encoding, can achieve accuracy comparable to or better than physician-certified VA. Moreover, LLMs demonstrated efficiency in processing VA interviews, suggesting significant cost and time savings over traditional methods.
In addition to quantitative analysis, we conducted a qualitative examination of LLM responses, providing insights into their “reasoning” processes and areas for improvement. Our results highlight the promise of AI-based VA interpretation as a fast, cost-effective, and accurate tool for public health.

With further development and external validation, this approach could revolutionize mortality data collection and health resource allocation, particularly in underserved regions.

View Presentation for AI and verbal autopsy: Predicting pregnancy-related causes of death

Dr. Jamie Perin, Johns Hopkins University: Incorporating calibrated cause-specific mortality from verbal autopsies in cause of death estimation

Estimates of under-five mortality for the predominant causes of death are available for all countries in a time series starting in 2000.  While these estimates of cause-specific mortality have been useful in the planning of health systems where vital registration is not available, these estimates are also based primarily on interviews with caregivers and family members of the deceased child, or verbal autopsies, which are generally limited by the lack of formal medical diagnosis and medical interpretation. 

We aimed to increase the validity of these cause of death estimates and reduce the biases of cause specific mortality introduced by verbal autopsy.  Recently published research related to minimally invasive tissue sampling has provided detailed information on the misclassification of verbal autopsy identified causes of death for different cause assignment algorithms in different settings.  We take advantage of this new information to adjust the publicly available causes of death for children under five as part of our systematic modelling approach to cause of death estimation. 

We use a Bayesian framework with multinomial regression, with health system related covariates for extrapolation to areas without any cause of death data, and random effects for countries with nationally representative cause of death data.  We explore several options for incorporating the misclassification of verbal autopsy determined cause of death into this modeling framework, and present preliminary estimates for the most feasible approach. 

View Presentation for Incorporating calibrated cause-specific mortality from verbal autopsies in cause of death estimation

Training and overview of the agent-based modeling framework Starsim, with the aim of fostering and expanding the user community. This full-day session will provide details on the model design, architecture, and feature roadmap. Participants will have the opportunity for hands-on demonstrations of how to use Starsim to develop a model, analyze data, or explore policy questions.