Background: To test data-driven and equity-focused homelessness prevention. Federal policies call for efficient and equitable local responses to homelessness. However, the overwhelming demand for limited housing assistance challenges efforts to prioritize services, and little evidence supports decision-making. We demonstrate a community- and data-driven approach for prioritizing scarce resources.

Methods: Administrative records captured homeless service delivery and outcomes in St. Louis, MO from 2009-2014 (n=10043). Counterfactual machine learning identified services most likely to prevent household-level homelessness within two years, which we aggregated to design group-based service prioritization rules. Simulations re-allocated households to available services and evaluated whether data-driven prioritization reduced community-wide homelessness without excluding marginalized and underrepresented groups.

Findings: Local data showed households with comorbid health conditions avoided homelessness most when provided longer-term supportive housing, and families with children fared best in short-term rentals. Prioritization rules reduced community-wide homelessness and disproportionately benefited marginalized and minoritized populations.

Interpretation: Leveraging local records with machine learning supplements local decision-making and enables ongoing evaluation of data- and equity-driven homeless services. Community- and data-driven prioritization rules more equitably target scarce homeless resources.