Variability, Reliability, and Multi-Level Determinants of Routine Alcohol and Drug Use Screening: A Retrospective Cohort Study.

Abstract

Alcohol and drug use screening in primary care is recommended by the US Preventive Services Taskforce, yet multi-level correlates of screening, and measurement properties of screening metrics, are unclear. This study characterized the reliability, variability, and multi-level factors associated with guideline-concordant alcohol and drug screening. Retrospective longitudinal cohort study. A total of 1,097,127 randomly selected visit observations of 479,634 adult primary care patients (2019-2022) from one healthcare system in the US Pacific Northwest. Patients contributed up to one observation per year. The primary outcome was completion of substance use screening on the index visit date or in the prior year. Mixed-effects models estimated the probability of guideline-concordant screening completion. Intraclass correlation coefficients (ICCs) quantified variability at patient, provider, clinic, and county levels, and the Spearman-Brown prophecy formula estimated reliability. Average marginal effects generated from logistic regression models quantified associations between patient-, provider-, clinic-, and community-level characteristics and screening completion. Patients resided in 428 counties and visited 802 primary care providers across 36 clinics. Screening was present for 80% of visits. Most variation in screening completion was present at the patient, provider, and clinic levels (ICCs = 0.075, 0.033, and 0.016, respectively). Achieving a reliability threshold of ≥ 0.80 required modest sample sizes at the provider (n = 117) and clinic levels (n = 246). Significant predictors of screening were identified at the patient (e.g., anxiety and cannabis use disorders), provider (e.g., visit volume), clinic (e.g., visit volume), and community levels (e.g., number of addiction treatment facilities). An analysis of a single health system. Screening measures demonstrated strong performance and meaningful variation. Predictors of screening could be used to target and tailor quality improvement interventions. Not applicable.