What helps residents stay longer in recovery housing?

Individuals starting a recovery journey enter recovery housing with diverse sociodemographic backgrounds, substance use histories, and levels of recovery capital, which influence the length of stay in recovery housing. This study examined which admission characteristics correspond to length of stay in recovery residences and used survival mixture clustering to characterise heterogeneity in retention trajectories. This study included data from 2534 residents across 61 U.S. recovery residences between 2019 and 2023. Cox proportional-hazards regression identified baseline correlates of length of stay, with cluster-robust standard errors at the residence level. A complementary survival mixture clustering algorithm jointly modelled latent retention trajectories and survival functions using sociodemographic characteristics, recovery capital, barriers, unmet service needs, and substance use indicators. Older age, higher quality of life, greater recovery group participation, an unmet alcohol treatment need, and criminal legal system involvement associated with longer stays. By contrast, a history of cannabis use, drug use within 90 days before admission, and an unmet drug treatment need associated with shorter stays. Survival mixture clustering supported a nine-cluster solution, with median lengths of stay ranging from 29 to 91 days. Clusters with shorter stays exhibited lower recovery capital, higher unmet needs, and greater recent substance use, whereas longer-stay clusters demonstrated higher quality of life, stronger engagement with support structures, and higher proportions of criminal legal system involvement. Model discrimination was modest overall, indicating that baseline characteristics explain only part of the variation in retention trajectories. Residents reporting recent drug use before admission to recovery housing or an unmet drug treatment need may benefit from proactive support at admission, while peer engagement and structured support pathways may promote longer lengths of stay. Because effect sizes were modest and discharge outcomes were heterogeneous, these findings should inform adaptive and equity-conscious service responses rather than individual risk classification. Future research should disaggregate accrual and loss of recovery capital during initial residence and distinguish between disengagement and recovery-readiness exits.