Prioritizing Out-of-Care Case Investigations in King County, Washington

Daniel Cockson | 2023

Advisor: Julia Dombrowski

Research Area(s): Infectious Diseases, Social Determinants of Health

Full Text

Health departments need to investigate cases of people with HIV who appear to be out of care but often have insufficient resources to investigate all cases and need a way to prioritize investigations. Methods: In this retrospective cohort analysis, we used classification and regression tree (CART) methodology to develop and validate a decision algorithm for indicating which HIV cases need investigation. The goal is that this algorithm could be used prospectively to confirm out-of-care status and offer assistance relinking to care. The data utilized is from Public Health – Seattle King County’s Comprehensive HIV/AIDS Database (CHARD) which is used to manage HIV case investigations. A “priority” designation is applied to investigations where 1.) the individual was confirmed to be out of care and had been successfully contacted or 2.) the individual was confirmed to be out of care and could not be contacted. We considered 20 potential predictors for priority designation relevant to patient demographics, laboratory result patterns, and reported investigation characteristics. We compared the test characteristics of an optimized algorithm and simple algorithm. Results: During 01/2018 – 12/2022, 4,311 HIV cases were referred for further investigation. Across the validation data, the optimized and simplified algorithms correctly identified 79.9% and 81.3% of priority investigations, respectively. The optimized algorithm had the lowest specificity at 88.0% and the simplified algorithm had the highest specificity at 89.0%. Models did not perform significantly worse across gender, age, and racial ethnic strata except for when applied to Non-Hispanic Asians (poor positive predictive value) and Non-Hispanic Native Hawaiians/Pacific Islanders (poor negative predictive value). Conclusions: We found that the performance of two algorithms (one optimized and one simplified) developed withCART were effective in identifying non-priority investigations and could be used prospectively to triage case investigations.