Inference of Intervention Impact and Epidemic Intensity for Infectious Diseases

Abstract

Outbreaks of SARS-CoV-2, HIV, and other infectious pathogens have a profound impact on public health globally. Public health surveillance, defined as “ongoing systematic collection, analysis, and interpretation of health data,” serves two primary purposes in informing public health decisions: Evaluating the impact of infectious disease interventions and estimating epidemic intensity. In Chapter 2, we use surveillance data to estimate vaccine-preventable COVID-19-associated deaths among unvaccinated individuals in the United States. We estimated that hundreds of thousands of deaths could have been directly prevented through vaccination among unvaccinated adults during the 15-month study period. In Chapter 3, we examine the commonly held assumption that due to the indirect effects of vaccination in preventing transmissions, vaccination could have prevented more outcomes overall across the entire population than could have directly among unvaccinated individuals (or that vaccination has prevented more outcomes overall than it has directly among the vaccinated individuals). We demonstrate that the direct impact of vaccination among vaccinated (or unvaccinated) individuals is a lower bound on overall impact across all individuals when indirect effects are non-negative. Using simulations, we illustrate how this lower bound may fail under common vio¬lations to assumptions on time-invariant vaccine efficacy, pathogen properties, or behavioral parameters. In Chapter 4, we explore key considerations when using routinely collected data on HIV diagnosis and status ascertainment among pregnant women attending antenatal care (ANC) as a sentinel population to monitor HIV incidence trends in the population at large in generalized HIV epidemic settings. However, ancillary factors—such as those related to fertility, non-disclosure of status awareness, and testing patterns outside of ANC—may also influence trends in ANC surveillance measures, even though incidence remain unchanged. Using simulations and data from a recent study as an example, we demonstrated that trends in ANC surveillance measures may be explained by changes in these ancillary factors, leading to biased incidence estimates if unaccounted. Our findings highlight the importance of accounting for these ancillary factors when interpreting trends in the ANC measures to infer incidence trends. In addition, we show that when incidence is low, a modest, unaccounted change in non-disclosure proportion produce trends in the new diagnoses rate similar to those caused by a large decrease in incidence. Therefore, true new diagnoses should be distinguished from non-disclosing re-tests for the new diagnoses rate to be a more useful measure for estimating incidence trend.