Patterns of health care interactions of individuals with alcohol use disorder: A latent class analysis

Introduction: Given the high rates at which individuals with alcohol use disorder (AUD) utilize health care for co-existing conditions, health systems are promising venues for interventions that will facilitate access to AUD treatment. However, how individuals with AUD interact with such systems and, thus, how systems should intervene is unclear. In this study, we seek to identify patterns in how individuals diagnosed with AUD within an academic health system interacted with the system prior to diagnosis.Methods: We use electronic health records from a single academic health system in a major US metropolitan area to create a deidentified retrospective cohort including all individuals age 18+ diagnosed with AUD 2010-2019 (n = 26,899). Latent class analysis (LCA) identified subgroups defined by aspects of previous system interaction and health status, including having an in-system primary care provider, previous utilization of primary and specialty care, diagnosis setting, payer, and presence of other chronic conditions. We then assessed subgroup differences in demographics and associations with in-system AUD treatment receipt in the year following diagnosis, adjusting for demographics.Results: The population was on average 38.6 years old (standard deviation = 15.4) and predominantly male (66.1 %), White (64.5 %), and not of Hispanic/Latino ethnicity (87.8 %). Only 4.7 % received in-system treatment following diagnosis. We deemed the four-class model the optimal LCA model. This model identified subgroups that can be described as 1) average utilization (20.7 % of population), 2) low utilization (54.5 %), 3) high health burden and low utilization (14.2 %), and 4) high health burden and high utilization (10.6 %). Predicted membership in the high health burden and high utilization subgroup and low utilization subgroup were associated with higher and lower odds of treatment receipt, respectively, compared with predicted membership in the average utilization subgroup (odds ratio (OR) for high/high subgroup = 1.21, 95 % confidence interval (CI) = 1.01, 1.27; OR for low subgroup = 0.29 95 % CI = 0.24, 0.34). Conclusion: Individuals diagnosed with AUD within a health system interact with that system in markedly different ways and are unlikely to benefit uniformly from system-based interventions to facilitate treatment. Group-tailored interventions are more likely to have impact and provide returns on investments for systems.