Using Machine Learning to Identify Predictors of Sexually Transmitted Infections Over Time Among Young People Living With or at Risk for HIV Who Participated in ATN Protocols 147, 148, and 149

Abstract

Background: Sexually transmitted infections (STIs) among youth aged 12 to 24 years have doubled in the last 13 years, accounting for 50% of STIs nationally. We need to identify predictors of STI among youth in urban HIV epicenters.

Methods: Sexual and gender minority (gay, bisexual, transgender, gender-diverse) and other youth with multiple life stressors (homelessness, incarceration, substance use, mental health disorders) were recruited from 13 sites in Los Angeles and New Orleans (N = 1482). Self-reports and rapid diagnostic tests for STI, HIV, and drug use were conducted at 4-month intervals for up to 24 months. Machine learning was used to identify predictors of time until new STI (including a new HIV diagnosis).

Results: At recruitment, 23.9% of youth had a current or past STI. Over 24 months, 19.3% tested positive for a new STI. Heterosexual males had the lowest STI rate (12%); African American youth were 23% more likely to acquire an STI compared with peers of other ethnicities. Time to STI was best predicted by attending group sex venues or parties, moderate but not high dating app use, and past STI and HIV seropositive status.

Conclusions: Sexually transmitted infections are concentrated among a subset of young people at highest risk. The best predictors of youth's risk are their sexual environments and networks. Machine learning will allow the next generation of research on predictive patterns of risk to be more robust.

Figure 1:

Kaplan-Meier curves for time to new STI infection in the entire sample (left) and each gender-sexual identity group: cisgender female, cisgender gay, bisexual, and other men who have sex with men (GBMSM), heterosexual male, and transgender/gender-diverse (right).