News and Events

April 20, 2018

2018 IDM Symposium panel addresses missing populations in health data

At the 2018 IDM Modeling Symposium, a panel of experts discussed why certain populations are missing from public health data and how disease modelers can address this data gap to make better health policy recommendations. When populations are missing from public health data and the data is applied to the population at large, the wrong conclusions are likely to be drawn. Disease modelers generally analyze data far from the field, so it’s essential to consider how the data was gathered when attempting to identify who might be missing.

For example, a common problem in developing countries is difficulty in accessing some rural areas, causing rural inhabitants to be missed by surveillance and health care systems. Dr. Thomas Smith of the Swiss Tropical and Public Health Institute was skeptical of statistical reports showing improvements in mortality rates in Papua New Guinea were similar to those in the rest of the world, despite his first-hand experience with data systems there and very poor transport networks making many rural areas inaccessible. "Frankly, the quality of data what's being collected on the ground is possibly not even as good as what it was a generation ago […] it could be that we are coming to false conclusions about a lot of things.”

When data is missing randomly, it doesn’t cause bias and the solution is straightforward: gather more data. However, this is rarely the case in the real world, where the people who are missed usually aren’t representative of the population at large. “People missing from the data are almost by definition the most marginalized,” said Dr. Smith.

Depending on the reasons why people are missing, there are different solutions to address the data gap. Whatever the solution, says Dr. Sahar Zangeneh of the Fred Hutchinson Cancer Research Center, “If you have missing data, you can't just ignore it. You have to do something about it.” Statistical solutions, such as filling in missing data or weighting it during analysis, can be applied post hoc. Solutions aimed at reaching missed populations in the field are often creative.

Health care may also be inaccessible for individuals due to economic reasons. A successful method for reducing cost is task-switching: transferring many health care tasks to workers with less training. For example, distributing vaccines using community health workers instead of nurses. This also enables strategies such as in-home testing and treatment, which opens care to poor people who cannot afford to miss work to travel to the clinic.

Even when health care is accessible and affordable, certain populations may not access it. For example, men generally seek out health care less often than women do, so one effort involved going from home to home providing TB testing. “But even with health care workers going directly to the household, there were always members of the household that we could not locate,” said Dr. Adrienne Shapiro of the University of Washington. Those absent were often men away from the home for work, a population that was already at higher risk. One effort that was successful in South Africa at reaching a 50/50 gender balance was setting up HIV testing facilities at soccer matches, transit hubs, construction sites, workplaces and other areas. “We needed to think differently to address that gap,” said Dr. Ruanne Barnabas of the University of Washington.

When individuals are reached, they may not provide reliable data. Adherence to treatment may be less than reported, particularly when it involves taking medication when one feels well. People may not provide accurate data about their health or behavior, particularly in HIV research. People may not feel safe disclosing their sexual orientation due to stigma or criminalization, so LGBT populations are not well-represented in the data.

Usually the rate of health care among missing populations is so low that targeted interventions to increase that rate show some impact on disease transmission. It can be especially valuable to reach populations than have a disproportionate effect on disease transmission. For example, peer networks using mobile applications could be a good approach to reach sex workers. “The opportunity costs of missing these people is hard to estimate because they are missing,” said Dr. Shapiro. Working with hidden populations can also help you identify other helpful interventions, such as integrating opioid substitution and HIV testing/treatment for IV drug users, said Dr. Barnabas.

However, even when modelers have done everything they can to improve the quality of their data and reach those hidden populations, there can be political resistance to implementing their health care recommendations. Sanctions may prevent importing necessary medications or discussing certain topics. Policymakers may oppose evidence for unpopular interventions. “This country is an excellent example. Things like needle exchange, which were illegal for years, or certain politicians opposed to evidence-based care,” said Dr. Shapiro.

The panel emphasized the importance of disease modelers considering the limitations of the data gathered in the field and taking steps to address those limitations. Many options are available to address gaps in data, including creative methods for reaching missing populations. While work is needed to validate some of these new methods, even imperfect methods are better than ignoring those gaps. It’s also important to keep in mind how to communicate findings to policymakers so effective health care policies can be implemented.