Mentored Scientist Award

Exploring the Benefit of Machine Learning to Predict Biomarker-Measured Alcohol Use, Opioid Use, and Co-Use via Routinely Collected Biological Data in the All of Us Research Program

Headshot of Cristina Espinosa da Silva, PhD, MPH
Award mentor
Award date
Award cycle
Award amount - Direct


Hazardous alcohol and opioid use are prevalent among persons with HIV (PWH) in the United States, and contribute to adverse outcomes including HIV disease progression,9,10 various co-morbidities, and mortality. A significant proportion of PWH co-use alcohol and opioids, which worsens HIV-related outcomes. Despite the notable prevalence of these types of substance use and their related negative sequelae, alcohol and opioid use are frequently underdiagnosed among PWH in part due to the widespread use of self-report which is prone to underreporting. Alcohol and opioid use lead to harmful physiological changes (e.g., hematological changes, increased levels of liver-derived enzymes) which are often measured in routine lab tests (e.g., complete blood count, metabolic panel). While these physiological changes may not reliably predict alcohol or opioid use when considered individually, leveraging them in combination with machine learning could develop a predictive tool for detecting alcohol and opioid use among PWH. This mentored RAP research proposes to use routinely collected lab data (e.g., complete blood count, metabolic panel) with machine learning methods to predict biomarker-measured recent alcohol use (Aim 1), biomarker-measured recent opioid use (Aim 2), and biomarker-measured recent alcohol and opioid course (Aim 3) in a nationwide sample of PWH in the All of Us Research Program. This work extends the field beyond previously published machine learning studies that developed models from self-reported substance use assessments in general populations. This research will provide preliminary data for a K01 proposal further investigating improved detection of varying levels of alcohol and opioid use among PWH (e.g., hazardous use, Alcohol Use Disorder, Opioid Use Disorder). The culmination of this work could provide a predictive tool to improve substance use detection in clinical settings and better identify PWH in need of additional support, and also improve substance measurement in research studies unable to collect substance use biomarker data.