Equitable Infectious Disease Modeling for Data-Constrained Settings and Data-Overlooked Populations

Abstract

For infectious diseases, social factors have played a long-neglected yet crucial role in modifying infection risk, aggravating disease severity, and determining adverse health, social, and economic post-infection outcomes. This pattern has been observed in a range of settings and for a range of pathogens yet is often disregarded in efforts for risk identification, routine monitoring, forecasting, and intervention impact assessments. Furthermore, because of these social disparities and broader, country-level infrastructural and financial constraints, there has been considerable variation in the strength of infectious disease surveillance programs, where, for some areas, or within some populations, data are limited and statistical modeling efforts are scarce. Consequently, there are notable societal and geographic asymmetries in accurate and timely awareness of current and future trends. Such inequities in both infectious disease dynamics and in how effectively such dynamics are monitored are the subject of this dissertation. In Aim 1, we focused on the malaria monitoring program in Guyana and provided a straightforward and readily translatable tool for estimating current disease incidence (‘nowcasting’) in malaria endemic regions using easily accessed information on prior trends in timeliness. First, to better understand the reporting landscape and inform our nowcasting models, we evaluated potential spatial and time trends in delays in reporting regional malaria cases to the national surveillance office, as well as their possible climatic and demographic cofactors, such as rainfall and the presence of mining sites and Amerindian settlements, two socially disadvantaged populations. We found that the extent of reporting delays varied significantly across malaria endemic regions, although these patterns were relatively consistent over the study period. Additionally, we found evidence for a spatial overlap between high-delay areas and locations with a greater presence of mining communities and Amerindian settlements. For each region and month, we then developed flexible data imputation and network models to estimate that month’s actual total caseload (‘converged cases’). We observed that even the simplest models we implement more accurately estimate monthly converged cases; more detailed models incorporating additional information from other regions provided further, but not dramatic, improvements. In Aim 2, we provided a deterministic modeling framework for accounting for socio-economic differences in disease transmission, mortality, vaccination, and demographics. Specifically, we introduced an income quintile-stratified susceptible-infected-recovered-deceased model for measles applied to Ethiopia, parameterized with quintile-specific rates of birth, transmission, vaccination, and disease-induced and background mortality. We additionally assessed how population-wide immunization programs may differentially avert measles deaths across socio-economic groups and further identified strategies that are most equitable in their impacts. Given that directly relevant empirical information for parameterizing our model was not available to us, we conducted a series of scenario-based analyses to inform these measures. We additionally assessed how changing vaccination coverage across a range of strategies yielded differences in expected mortality proportions for each quintile. Finally, we quantified measles mortality disparities under the different immunization strategies, finding that those leading to the greatest reduction in measles mortality disparities slightly varied under this outcome metric. In Aim 3, we turned to the COVID-19 pandemic, looking specifically at quality of life with long-term sequelae following SARS-CoV-2 infection. In this aim, we addressed gaps in the literature on 1) evaluating the relative role of social correlates, namely educational attainment, employment status, and sex, compared to clinical comorbidities in shaping quality-adjusted life days with post-COVID condition (‘long COVID QALDs’) and 2) quantifying the extent to which the adjusted associations between social factors and long COVID QALDs can be attributed to mediation by major long COVID-predicting clinical comorbidities. To do so, we employed a large, multi-country dataset from a longitudinal COVID-19 follow up study, focusing on the following locations: Norway, the United Kingdom (UK), and Russia. We found that in addition to age and the clinical factors neurological, psychological, and rheumatological conditions, employment status, educational attainment, and female sex were some of the leading predictors of long COVID QALDs. We additionally found evidence to suggest that most of the relationships between each of these social factors and long COVID QALDs were unexplained by the clinical intermediates, a pattern that was observed consistently across cohorts. Thus, we can conclude that factors not under consideration, such as broader societal and structural vulnerabilities, may be just as important in shaping these social disparities and merit increased attention in long COVID-targeting efforts. Through our three aims, we strove to design adaptable methods for tracking the trajectories and impacts of various infectious diseases in diverse contexts where overall data quality is significantly compromised and auxiliary ‘big data’ resources are unavailable, or where important within-population differences in health experiences are underemphasized. Importantly, our models developed for resource-constrained settings facilitate active user buy-in and integration as they can be tailored to context-specific characteristics and are readily implementable. As such, they may serve to support community-centered and community-led epidemiological programs.