A classification model of homelessness using integrated administrative data: Implications for targeting interventions to improve the housing status, health and well-being of a highly vulnerable population
Published on August 20, 2020. Authors: Byrne T, Baggett T, Land T, Bernson D, Hood M-E, Kennedy-Perez C, et al. (2020)
Available on https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0237905#abstract0 as an open-access article distributed under the terms of the Creative Commons Attribution License (unrestricted use, distribution, and reproduction in any medium permitted, provided the original author and source are credited).
Summary of the research:
Homelessness is poorly captured in most administrative data sets making it difficult to understand how, when, and where this population can be better served.
Prior research has shown that homeless individuals have a high burden of medical and mental illnesses, substance use disorders, and health care and human services systems use. By identifying individuals at risk of homelessness, service providers can improve the coordination of services and promote better health outcomes, particularly for conditions such as opioid use disorder that exact a high toll on individuals experiencing homelessness.
Persons experiencing homelessness often interact with multiple publicly-funded systems of care including the emergency shelter, health care, mental health, substance use disorder treatment, and criminal justice systems, thus providing numerous points to address their housing, health care, and other social needs. However, many service systems do not capture information about housing status in a reliable manner.
This issue led to increased interest in developing predictive models to identify persons experiencing homelessness using available data in administrative records.
With this study, the authors sought to develop and validate a classification model of homelessness on a sample including 5,050,639 individuals aged 11 years and older who were included in a linked dataset of administrative records from multiple state-maintained databases in Massachusetts for the period from 2011–2015.
Based on the consensus of a working group of experts in homelessness, the authors used the following criteria to identify the known cases of homelessness: 1) a claim registered in the system with a code indicating homelessness; 2) a record in the Department of Mental Health (DMH) dataset in which individuals were ever identified as experiencing a loss of housing ; 3) an ambulance record in which the word “homeless” or “shelter” appeared in the narrative report; or 4) a prescription record in which the patient’s address matched that of an emergency shelter. Individuals meeting any of these criteria at any point during the 5-year observation period were classified as experiencing homelessness. They also selected 94 possible independent variables from across. "They were classified into several groups, including socio-demographic predictors (e.g. age, gender, race, Medicaid receipt; drug/alcohol use predictors (e.g. presence of drug/alcohol diagnoses, use of substance use disorder treatment services); mental health predictors (e.g. presence of mental health diagnoses, use of mental health services); physical health predictors (e.g. skin disorders); other service use predictors (e.g. history of incarceration in state prison, use of emergency department services). "
Of 5,050,639 individuals in the analytic cohort, 41,457 (0.82%) were identified as experiencing homelessness according to pre-specified indicators of known cases of homelessness.
The present study an example of how large, integrated administrative data from multiple service systems can be used to identify individuals at risk of homelessness to facilitate targeted services or timely intervention.
Keywords: homelessness, access to housing, homeless, shelter, US, data