PrEP retention predictors among key populations in urban areas of Zimbabwe — a machine learning approach

Abstract

Pre-exposure prophylaxis (PrEP) success depends on adherence and hence retention in care. It is vital to find factors that affect PrEP retention to make this HIV-prevention method successful. The main objectives of this study were to identify the major determinants of PrEP retention at six months from day of initiation and to determine PrEP retention clusters. Data of people who were initiated on PrEP care at Population Services International sites between 2017 and 2019 in Zimbabwe were used. Determinants of PrEP retention were identified using multivariable logistic regression and partitioning around medoids (PAM) cluster analysis to determine the number and composition of PrEP retention clusters. Significant PrEP retention determinants were found to be population type, sex, marital status, employment type, age and education level. Female sex workers were most likely to stay on PrEP compared to other users. We found that PrEP users could be grouped into three risk clusters: cluster 1 consisted of married adult males who are high-risk individuals; cluster 2 consisted of the young single men who have sex with men (MSM); and cluster 3 consisted of adult women who are separated or divorced and employed as sex workers. The retention rates were 32%, 24% and 44% for clusters 1 to 3 respectively. We conclude that adults, female sex workers, high-risk individuals, women, people without or with little education and sex workers stay on PrEP better than youths, MSM, men and educated and employed individuals.

Keywords:

• HIV prevention
• HIV risk populations
• multivariable logistic regression
• PAM clustering