Methods/Statistics
Assessing the Structural Determinants of HIV among Adolescent Girls and Young Women in Malawi: Application of the PC Causal Discovery Algorithm Domonique M. Reed* Domonique M. Reed Daniel Malinsky Jeanette A. Stingone Jessica Justman
Individual, interpersonal, community, and societal factors influence adolescent girls and young women’s (AGYW; 15-24 years) HIV risk, yet their interconnectedness is understudied. The tiered PC (Peter-Clark) algorithm for causal discovery extends the constraint-based PC algorithm to accommodate temporal ordering between variables and uses patterns of conditional independence to determine causal structure. Using this algorithm, we characterized the multi-level paths leading to HIV among AGYW in Malawi.
We integrated data from the Population-based HIV Impact Assessment, Census, and UNAIDS Policy Indicators (2014- 2016). We applied the PC algorithm with a conditional Gaussian likelihood ratio test and test-wise deletion for missingness. We used an alpha level of 0.2 to obtain a graph with reasonable density and power.
We included data on 32 features from 2,644 AGYW in 2,485 households across 479 enumeration areas within 3 administrative regions of Malawi. Of 2,644 AGYW, 4.6% live with HIV. Most AGYW lived in a rural area, completed primary education, married, and lived in a middle wealth quintile household. Figure 1 displays a subset of 9 variables directly connected to HIV status. A region with lower average years of education for women and an average age of marriage <18 years have directed paths to community HIV prevalence and interpersonal relationship variables. HIV prevalence has a directed path to HIV status, as expected, but there are undirected paths between interpersonal relationship factors and HIV. While there is a directed path from transactional sex to HIV status, there are directed paths from HIV to sexual debut age and number of partners, often described as HIV predictors in the literature.
This application of the PC algorithm produced a graph that provides insight into the complex relationship shaping AGYW’s HIV risk in Malawi. These findings are an example of causal discovery and contribute to our understanding of the underlying causal mechanisms of HIV.
