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

Persistent Disparities: Trends in Rates of Sheltered Homelessness Across Demographic Subgroups in the USA

CONTEXT

Homelessness is a public health crisis affecting millions of Americans every year, with severe consequences for health ranging from infectious diseases to adverse behavioral health outcomes to significantly higher all-cause mortality. A primary constraint of addressing homelessness is a lack of effective and comprehensive data on rates of homelessness and who experiences homelessness. While other types of health services research and policy are based around comprehensive health datasets to successfully evaluate outcomes and link individuals with services and policies, there are few such datasets that report homelessness.

METHODS

Gathering archived data from the US Department of Housing and Urban Development, we created a unique dataset of annual rates of homelessness, nationally, as measured by persons accessing homeless shelter systems, for 11 years (2007-2017, including the Great Recession and prior to the start of the 2020 pandemic). Responding to the need to measure and address racial and ethnic disparities in homelessness, the dataset reports annual rates of homelessness across HUD selected, Census-based racial and ethnic categories.

FINDINGS

Between 2007 and 2017, across all types of sheltered homelessness, whether individual, family, or total, Black, American Indian or Alaska Native, and Native Hawaiian and Pacific Islander individuals and families were far more likely to experience homelessness than non-Hispanic White individuals and families. Particularly concerning about the rates of homelessness among these populations is the persistent and increasing nature of these disparities across the entire study period.

CONCLUSIONS

While homelessness is a public health problem, the hazard of experiencing homelessness is not uniformly distributed across different populations. Because homelessness is such a strong social determinant of health and risk factor across multiple health domains, it deserves the same careful annual tracking and evaluation by public health stakeholders as other areas of health and health care.

The value of utility payment history in predicting first-time homelessness

Homelessness is a costly and traumatic condition that affects hundreds of thousands of people each year in the U.S. alone. Most homeless programs focus on assisting people experiencing homelessness, but research has shown that predicting and preventing homelessness can be a more cost-effective solution. Of the few studies focused on predicting homelessness, most focus on people already seeking assistance; however, these methods necessarily cannot identify those not actively seeking assistance. Providing aid before conditions become dire may better prevent homelessness. Few methods exist to predict homelessness on the general population, and these methods use health and criminal history information, much of which may not be available or timely. We hypothesize that recent financial health information based on utility payment history is useful in predicting homelessness. In particular, we demonstrate the value of utility customer billing records to predict homelessness using logistic regression models based on this data. The performance of these models is comparable to other studies, suggesting such an approach could be productionalized due to the ubiquity and timeliness of this type of data. Our results suggest that utility billing records would have value for screening a broad section of the general population to identify those at risk of homelessness.

Identifying Predictors of Homelessness Among Adults in a Large Integrated Health System in Northern California

Introduction Homelessness contributes to worsening health and increased health care costs. There is little published research that leverages rich electronic health record (EHR) data to predict future homelessness risk and inform interventions to address it. The authors' objective was to develop a model for predicting future homelessness using individual EHR and geographic data covariates. Methods This retrospective cohort study included 2,543,504 adult members (≥ 18 years old) from Kaiser Permanente Northern California and evaluated which covariates predicted a composite outcome of homelessness status (hospital discharge documentation of a homeless patient, medical diagnosis of homelessness, approved medical financial assistance application for homelessness, and/or "homeless/shelter" in address name). The predictors were measured in 2018-2019 and included prior diagnoses and demographic and geographic data. The outcome was measured in 2020. The cohort was split (70:30) into a derivation and validation set, and logistic regression was used to model the outcome. Results Homelessness prevalence was 0.35% in the overall sample. The final logistic regression model included 26 prior diagnoses, demographic, and geographic-level predictors. The regression model using the validation set had moderate sensitivity (80.4%) and specificity (83.2%) for predicting future cases of homelessness and achieved excellent classification properties (area under the curve of 0.891 [95% confidence interval = 0.884-0.897]). Discussion This prediction model can be used as an initial triage step to enhance screening and referral tools for identifying and addressing homelessness, which can improve health and reduce health care costs. Conclusions EHR data can be used to predict chance of homelessness at a population health level.

Easy to use and validated predictive models to identify beneficiaries experiencing homelessness in Medicaid administrative data

OBJECTIVE

To develop easy to use and validated predictive models to identify beneficiaries experiencing homelessness from administrative data.

DATA SOURCES

We pooled enrollment and claims data from enrollees of the California Whole Person Care (WPC) Medicaid demonstration program that coordinated the care of a subset of Medicaid beneficiaries identified as high utilizers in 26 California counties (25 WPC Pilots). We also used public directories of social service and health care facilities.

STUDY DESIGN

Using WPC Pilot-reported homelessness status, we trained seven supervised learning algorithms with different specifications to identify beneficiaries experiencing homelessness. The list of predictors included address- and claims-based indicators, demographics, health status, health care utilization, and county-level homelessness rate. We then assessed model performance using measures of balanced accuracy (BA), sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic curve (area under the curve [AUC]).

DATA COLLECTION/EXTRACTION METHODS

We included 93,656 WPC enrollees from 2017 to 2018, 37,441 of whom had a WPC Pilot-reported homelessness indicator.

PRINCIPAL FINDINGS

The random forest algorithm with all available indicators had the best performance (87% BA and 0.95 AUC), but a simpler Generalized Linear Model (GLM) also performed well (74% BA and 0.83 AUC). Reducing predictors to the top 20 and top five most important indicators in a GLM model yields only slightly lower performance (86% BA and 0.94 AUC for the top 20 and 86% BA and 0.91 AUC for the top five).

CONCLUSIONS

Large samples can be used to accurately predict homelessness in Medicaid administrative data if a validated homelessness indicator for a small subset can be obtained. In the absence of a validated indicator, the likelihood of homelessness can be calculated using county rate of homelessness, address- and claim-based indicators, and beneficiary age using a prediction model presented here. These approaches are needed given the rising prevalence of homelessness and the focus of Medicaid and other payers on addressing homelessness and its outcomes.

Understanding Homelessness among Migrants to Thunder Bay using Machine Learning (Preprint)

BACKGROUND

Over the past years, homelessness has become a substantial issue around the globe. The largest social services organization in Thunder Bay, Ontario, Canada, has observed that a majority of the people experiencing homelessness in the city were from outside of the city or province. Thus, to improve programming and resource allocation for people experiencing homelessness in the city, including shelter use, it was important to investigate the trends associated with homelessness and migration.

OBJECTIVE

This study aimed to address 3 research questions related to homelessness and migration in Thunder Bay: What factors predict whether a person who migrated to the city and is experiencing homelessness stays or leaves shelters? If an individual stays, how long are they likely to stay? What factors predict stay duration?

METHODS

We collected the required data from 2 sources: a survey conducted with people experiencing homelessness at 3 homeless shelters in Thunder Bay and the database of a homeless information management system. The records of 110 migrants were used for the analysis. Two feature selection techniques were used to address the first and third research questions, and 8 machine learning models were used to address the second research question. In addition, data augmentation was performed to improve the size of the data set and to resolve the class imbalance problem. The area under the receiver operating characteristic curve value and cross-validation accuracy were used to measure the models' performances while avoiding possible model overfitting.

RESULTS

Factors predicting an individual's stay duration included home or previous district, highest educational qualification, recent receipt of mental health support, migrating to visit family or friends, and finding employment upon arrival. For research question 2, among the classification models developed for predicting the stay duration of migrants, the random forest and gradient boosting tree models presented better results with area under the receiver operating characteristic curve values of 0.91 and 0.93, respectively. Finally, home district, band membership, status card, previous district, and recent support for drug and/or alcohol use were recognized as the factors predicting stay duration.

CONCLUSIONS

Applying machine learning enables researchers to make predictions related to migrants' homelessness and investigate how various factors become determinants of the predictions. We hope that the findings of this study will aid future policy making and resource allocation to better serve people experiencing homelessness. However, further improvements in the data set size and interpretation of the identified factors in decision-making are required.

Hospitalizations for COVID-19 Among US People Experiencing Incarceration or Homelessness

IMPORTANCE

People experiencing incarceration (PEI) and people experiencing homelessness (PEH) have an increased risk of COVID-19 exposure from congregate living, but data on their hospitalization course compared with that of the general population are limited.

OBJECTIVE

To compare COVID-19 hospitalizations for PEI and PEH with hospitalizations among the general population.

DESIGN, SETTING, AND PARTICIPANTS

This cross-sectional analysis used data from the Premier Healthcare Database on 3415 PEI and 9434 PEH who were evaluated in the emergency department or were hospitalized in more than 800 US hospitals for COVID-19 from April 1, 2020, to June 30, 2021.

EXPOSURES

Incarceration or homelessness.

MAIN OUTCOMES AND MEASURES

Hospitalization proportions were calculated. and outcomes (intensive care unit admission, invasive mechanical ventilation [IMV], mortality, length of stay, and readmissions) among PEI and PEH were compared with outcomes for all patients with COVID-19 (not PEI or PEH). Multivariable regression was used to adjust for potential confounders.

RESULTS

In total, 3415 PEI (2952 men [86.4%]; mean [SD] age, 50.8 [15.7] years) and 9434 PEH (6776 men [71.8%]; mean [SD] age, 50.1 [14.5] years) were evaluated in the emergency department for COVID-19 and were hospitalized more often (2170 of 3415 [63.5%] PEI; 6088 of 9434 [64.5%] PEH) than the general population (624 470 of 1 257 250 [49.7%]) (P < .001). Both PEI and PEH hospitalized for COVID-19 were more likely to be younger, male, and non-Hispanic Black than the general population. Hospitalized PEI had a higher frequency of IMV (410 [18.9%]; adjusted risk ratio [aRR], 1.16; 95% CI, 1.04-1.30) and mortality (308 [14.2%]; aRR, 1.28; 95% CI, 1.11-1.47) than the general population (IMV, 88 897 [14.2%]; mortality, 84 725 [13.6%]). Hospitalized PEH had a lower frequency of IMV (606 [10.0%]; aRR, 0.64; 95% CI, 0.58-0.70) and mortality (330 [5.4%]; aRR, 0.53; 95% CI, 0.47-0.59) than the general population. Both PEI and PEH had longer mean (SD) lengths of stay (PEI, 9 [10] days; PEH, 11 [26] days) and a higher frequency of readmission (PEI, 128 [5.9%]; PEH, 519 [8.5%]) than the general population (mean [SD] length of stay, 8 [10] days; readmission, 28 493 [4.6%]).

CONCLUSIONS AND RELEVANCE

In this cross-sectional study, a higher frequency of COVID-19 hospitalizations for PEI and PEH underscored the importance of adhering to recommended prevention measures. Expanding medical respite may reduce hospitalizations in these disproportionately affected populations.