Upcoming Presentations

Introduction: Recognizing that housing is an important social determinant of health, US states have increasingly offered housing supports as part of their Medicaid benefits. These policies aim to prevent and reduce homelessness and have led to the need to identify people experiencing homelessness and housing insecurity using large-scale administrative datasets. While much work has been done using administrative data to predict homelessness (e.g., Pourat et al., 2023), less research has focused on predicting housing insecurity. Housing insecurity refers to lacking a stable, safe, and/or affordable place to live due to overcrowding, frequent moves, and financial issues like the inability to pay rent or spending most or all of household income on housing costs (Frederick et al., 2014). Housing insecurity on its own can have deleterious effects on health and wellbeing (Meltzer & Schwartz, 2016), emphasizing the need to prevent not just homelessness but all insecure housing situations. This study aimed to leverage machine learning to identify risk factors for housing insecurity using Medicaid data for Hawaiʻi Medicaid beneficiaries.

Methods: The model used Medicaid claims data to extract the predictor variables to predict the likelihood of a member becoming housing insecure within a year of receiving a health service. Predictor variables included demographic characteristics, mental health diagnoses, skin conditions, and Z-codes indicating housing instability and difficulty meeting basic needs. Housing insecurity was defined by non residential and out-of-state addresses. Models were trained on health care encounters from July – December 2015, and the optimal model was temporally validated on health care encounters from January – June 2016.

Results: Housing insecurity in the follow-up period was identified for 5.8% of the sample; among these, 99% also had housing insecurity at the time of the encounter. Since the goal of the model was to identify individuals at risk for future housing insecurity, the remainder of analyses were restricted to individuals without housing insecurity at the time of encounter (N = 7,492,597). The optimal model was a distributed random forest, which had an AUC = 0.74 in the temporal validation sample. Demographic characteristics, prior housing insecurity, incarceration, anxiety and substance use disorders were among the strongest predictors of future housing insecurity. Ongoing research is being conducted to determine whether accuracy can be improved with the incorporation of additional predictors.

Conclusions: Preliminary results suggest that future episodes of housing insecurity can be predicted with acceptable accuracy at the time of an encounter with healthcare services. Ultimately, this work suggests that machine learning can be used in healthcare settings to identify people at risk for homelessness and housing insecurity and can be leveraged to connect people to preventative services, including Medicaid housing supports.

Like many states in the US, Hawaiʻi has seen a steady increase in unsheltered homelessness since 2015 (PIC, 2024; Okada & Pruitt, 2024). Many of these individuals have complex health needs that require levels of care unavailable in traditional shelters and housing programs (Gerrits-Goh et al., 2023). In response, the State of Hawaiʻi developed the “Kauhale Initiative” modeled after a Native Hawaiian houseless community’s response to homelessness in their own backyard. In this instance, “kauhale” refers to intentional community living, kuleana (responsibility to each other), and a “community first” approach to housing (Pakele, n.d.). In 2023, the State opened a pilot kauhale, “Pūlama ʻOla” for unsheltered individuals in need of medical respite. This presentation will present findings from a Photovoice project conducted with Pūlama ʻOla residents and staff that aimed 1) to understand what it means to heal through community and 2) to develop a holistic healing model to be used to guide the Kauhale Initiative. A total of 11 residents (90% of all residents enrolled at the time), 3 staff members, and two traditional researchers participated. The group met for two hours weekly for eight weeks between September and November 2023, with 300 photographs taken and analyzed. We engaged in participatory thematic coding of photographs and arranged these codes to develop an instructive model for future kauhale. Findings were shared with policymakers, and since, 16 additional kauhales have been built by the State. This presentation will describe the project and its findings, which have implications for other locales hoping to implement community-driven solutions to housing and related crises. The presentation will also discuss issues related to mainstream adoption of indigenous approaches.

Learning objectives—consider innovative community-driven solutions to economic problems like homeless; learn about Hawaii’s kauhale (indigenous community living) for housing insecure individuals; learn how to use participatory methods to create a model for community living