A Bayesian spatiotemporal approach to analyze publicly available, censored HIV datasets in Mississippi, 2014-2018

Keywords: New HIV diagnosis; Bayesian analysis; spatiotemporal statistics; data suppression
Session Type: Virtual Paper Abstract
Day: Monday
Session Start / End Time: 2/28/2022 09:40 AM (Eastern Time (US & Canada)) - 2/28/2022 11:00 AM (Eastern Time (US & Canada))
Room: Virtual 46

Abstract

Geographical HIV data has been increasingly released at small-area levels for multiple years via online data. These openly accessible HIV datasets, however, have the data suppression issue, which poses notable challenges for statistical modeling of incomplete datasets. Traditional omission approach deletes a substantial amount of information if the suppression rate is high thus potentially, even severely, biasing parameter estimations and distorting geographical patterns. Most imputation approaches such as substituting unknown finer-level values with known coarser-level values and multiple imputation do not directly account for the uncertainty associated with data imputation into statistical inferences.
This study develops a Bayesian spatiotemporal model to analyze censored HIV data at the county-level in Mississippi between 2014 and 2018. The approach uses a truncated Poisson distribution to model the censored values, of which is range is known (i.e., between 1 and 4), and is capable of directly propagating imputation uncertainties into final statistical inferences through posterior distributions. A simulation study indicates that the developed model can accurately identify spatial hotspots as well as global and local trends of the geographical phenomenon with suppressed values. Empirically, new HIV diagnosis in Mississippi remained stable at the state level, but the trends in some local counties departed from the main temporal trend thus warrant further interventions. We demonstrate that publicly available censored HIV data could be analyzed in ways that yield robust results by using a probabilistic approach, which may help health departments and other stakeholders more confidently identify areas that should be prioritized for aggressive HIV preventions.