Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Gerontologist . 2025 Apr 14;65(5):gnaf027. doi: 10.1093/geront/gnaf027 Search in PMC Search in PubMed View in NLM Catalog Add to search Identifying and Addressing Housing Insecurity in Older Patients: Trends, Referrals, and Inequities in a California Medical System Erin L Ferguson Erin L Ferguson , MPH 1 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA Find articles by Erin L Ferguson 1, # , Shivani Mehta Shivani Mehta , MPH 2 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA Find articles by Shivani Mehta 2, # , Silvia Miramontes Silvia Miramontes , MS 3 Bakar Computational Health Sciences Institute, University of California, San Francisco, San Francisco, California, USA Find articles by Silvia Miramontes 3 , Minhyuk Choi Minhyuk Choi , MPH 4 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA Find articles by Minhyuk Choi 4 , Ye Ji Kim Ye Ji Kim , PhD, MPH 5 Department of Epidemiology, Boston University, Boston, Massachusetts, USA Find articles by Ye Ji Kim 5 , Tanisha G Hill-Jarrett Tanisha G Hill-Jarrett , PhD 6 Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, California, USA 7 Global Brain Health Institute, University of California, San Francisco, San Francisco, California, USA Find articles by Tanisha G Hill-Jarrett 6, 7 , Nicolas Cevallos Nicolas Cevallos , BS 8 School of Medicine, University of California, San Francisco, San Francisco, California, USA Find articles by Nicolas Cevallos 8 , Yulin Yang Yulin Yang , PhD 9 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA Find articles by Yulin Yang 9 , Scott C Zimmerman Scott C Zimmerman , MPH 10 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA Find articles by Scott C Zimmerman 10 , Ruijia Chen Ruijia Chen , PhD 11 Department of Epidemiology, Boston University, Boston, Massachusetts, USA Find articles by Ruijia Chen 11 , Min Hee Kim Min Hee Kim , PhD 12 Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA Find articles by Min Hee Kim 12 , Kendra D Sims Kendra D Sims , PhD 13 Department of Epidemiology, Boston University, Boston, Massachusetts, USA Find articles by Kendra D Sims 13 , Gabriel L Schwartz Gabriel L Schwartz , PhD 14 Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA 15 Urban Health Collaborative and Department of Health Management and Policy, Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania, USA Find articles by Gabriel L Schwartz 14, 15, ✉ Editor: Benjamin Henwood Author information Article notes Copyright and License information 1 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA 2 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA 3 Bakar Computational Health Sciences Institute, University of California, San Francisco, San Francisco, California, USA 4 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA 5 Department of Epidemiology, Boston University, Boston, Massachusetts, USA 6 Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, California, USA 7 Global Brain Health Institute, University of California, San Francisco, San Francisco, California, USA 8 School of Medicine, University of California, San Francisco, San Francisco, California, USA 9 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA 10 Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA 11 Department of Epidemiology, Boston University, Boston, Massachusetts, USA 12 Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA 13 Department of Epidemiology, Boston University, Boston, Massachusetts, USA 14 Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA 15 Urban Health Collaborative and Department of Health Management and Policy, Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania, USA # Erin L. Ferguson and Shivani Mehta contributed equally to this work and share first authorship. ✉ Address correspondence to: Gabriel L. Schwartz, PhD. E-mail: [email protected] Roles Benjamin Henwood : PhD, MSW , Decision Editor Received 2024 May 21; Collection date 2025 May. © The Author(s) 2025. Published by Oxford University Press on behalf of the Gerontological Society of America. All rights reserved. For commercial re-use, please contact [email protected] for reprints and translation rights for reprints. All other permissions can be obtained through our RightsLink service via the Permissions link on the article page on our site—for further information please contact [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model ( https://academic.oup.com/pages/standard-publication-reuse-rights ) PMC Copyright notice PMCID: PMC11994243 PMID: 40222811 Abstract Background and Objectives Housing insecurity is rising among older adults, especially for racially minoritized people. Few studies have evaluated whether healthcare institutions are meeting that challenge. Using data from a large California medical system, we examined how often older patients are (A) identified as housing insecure and then (B) referred to social services, as well as inequities in those rates. Research Design and Methods We analyzed electronic health records (2013–2022) for 119,127 older adults (55+) receiving primary or emergency care. We used a natural language processing model to identify housing insecurity and social services referrals/connections from unstructured notes, with referrals also captured via structured data. Trends in identification were compared to eviction and homelessness trends in the general population. Racial inequities in referrals were evaluated using logistic regression. Results 0.6% of encounters ( n = 6,253) screened positive for housing insecurity. Positive screening trends were nonlinear, with initial increases followed by declines to baseline (roughly tracking regional eviction trends). Only 7% of patients identified as housing insecure were referred to social services, and connections were more likely in primary than emergency care (odds ratio [OR] = 2.04, 95% confidence interval [95% CI]: 1.41–2.96). Asian patients had lower odds of referral than non-Hispanic White patients (OR = 0.51, 95% CI: 0.28–0.95). Discussion and Implications We identified urgent unmet needs for housing intervention among older patients. Healthcare systems must do more to ensure older patients, especially racially minoritized older adults, are screened for housing insecurity and connected to housing services. Keywords: Homelessness, Housing, Machine learning, Social services Housing insecurity powerfully affects health and is rising among older adults. In California—home to one-third of Americans experiencing homelessness ( Kushel et al., 2023 )—older adults account for nearly half of the homeless population and are its most rapidly growing demographic, with skyrocketing needs for housing services. From 2017 to 2021, the number of older adults in California increased only 7%, whereas the number of older adults accessing homelessness services increased 84% (vs 43% across all ages; Ibarra, 2023 ). Roughly 40% of homeless older adults in California had never experienced homelessness before age 50 ( Kushel, 2020 ), a problem expected to grow in the aftermath of the coronavirus disease 2019 (COVID-19) pandemic. This reflects a housing crisis across all ages, but one in which older adults contend with unique challenges. Since the 1990s, wages and public housing construction have stagnated while rents have risen, creating a nationwide housing crunch ( Fischer et al., 2021 ). That crunch is pronounced in California, where income inequality has ballooned and housing construction has not kept up with demand ( Brown et al., 2018 ; Thorman & Payares-Montoya, 2024 ). This is a public health problem as well as an economic one: housing insecurity drives an array of poor health outcomes and makes it more difficult to treat complex health problems such as substance use disorders ( Richards & Kuhn, 2023 ; Smith et al., 2024 ). Poor housing quality is similarly pathogenic, increasing the risk of asthma, respiratory infection, unintentional injury, and poor mental health ( Taylor, 2018 ). Older adults face particular vulnerability to these housing and health difficulties, including few job prospects, often fixed incomes, and higher rates of chronic illness and disability. Little research has assessed how well the healthcare sector is meeting this challenge. Given housing insecurity’s impact on health, a growing chorus of community organizations, clinicians, and policymakers have suggested transforming healthcare settings into places where patients’ housing needs are universally detected and addressed via links to social services ( Cusack et al., 2019 ; Moore, 2019 ; Painter et al., 2024 ). California policymakers have gone as far as requiring that emergency departments offer referrals to shelters or residential treatment programs if patients would otherwise be discharged to the street ( Sorelle, 2019 ). Yet many healthcare systems have not adopted clear, widely utilized methods for detecting which patients are housing insecure in electronic health records (EHRs), nor whether patients have been referred to housing services offered by community organizations, making monitoring and evaluation difficult. When such information is collected—such as in patients’ problem lists—it may not be entered into clinical systems in a way that is easily quantitatively analyzed, and may not be collected for every patient. It is further unclear whether all patients who are identified as housing insecure receive the same level of care in housing services referrals. Racially/ethnically minoritized people face multiple overlapping impediments to high-quality healthcare, including discrimination by healthcare providers ( Hamed et al., 2022 ), lower access to well-resourced clinics and hospitals due to residential and medical segregation ( Yearby et al., 2022 ), and socioeconomic and linguistic inequities ( Sentell & Braun, 2012 ). In this paper, we thus examine four questions, using EHRs from a large California medical system. Analyzing 10 years of data, we answer: Can housing insecurity be reliably identified in EHRs among older adults using natural language processing (NLP), a type of machine learning? How often do clinicians identify older adults as housing insecure, and do trends in identification mirror housing insecurity trends regionally? How often do older patients who are identified as housing insecure receive referrals to social services? Are housing services referral rates equitable across race/ethnicity? Method Patient Population We pulled deidentified EHR data for all primary care or emergency department visits for older patients (aged 55+) within the University of California San Francisco (UCSF) healthcare system, from 1/1/2013 to 10/31/2022 ( University of California, San Francisco, Academic Research Systems, 2023 ). Importantly, this includes time both before and after February 2019, when UCSF implemented a new social needs screener for clinicians to evaluate the social needs of their patients, one that newly included housing insecurity. Primary care visits were defined as patient encounters that occurred at any adult primary care department at UCSF, including primary care, general internal medicine, family medicine, geriatric medicine, HIV programs, and executive health. We limited primary care visits to those containing patient encounters with clinicians to only examine visits in which patients’ needs could be assessed. Emergency department visits were identified via medical claims indicating the encounter occurred in an emergency department. Our final data set included 119,127 unique individuals spanning 1,111,823 encounters (72.2% from primary care). Our data did not include patients from Zuckerberg San Francisco General Hospital, San Francisco’s “safety net” hospital, as it is not part of the UCSF healthcare system per se (and its data are not included in the EHR database we analyzed). To protect patient privacy, all dates could have been date-shifted by up to 365 days; date shifting was consistent within an individual, allowing for accurate calculations of ages and intervals between visits. Ethics Approval We analyzed deidentified data and thus were declared exempt from IRB approval by the UCSF IRB. Patient informed consent was waived given that analyses were conducted on secondary data. Identification of Housing Insecurity To determine which patients were identified as housing insecure, we applied NLP to unstructured clinical notes from each visit (including the text of patients’ problem lists), using the open-source Apache clinical Text Analysis and Knowledge Extraction System (cTAKES; Savova et al., 2010 ). This cTAKES algorithm was paired with the Unified Medical Language System’s Concept Unique Identifier (CUI) codes ( Campbell et al., 1998 ) and the Systematized Medical Nomenclature for Medicine (SNOMED) codes for housing insecurity. In essence, the cTAKES algorithm searches a given piece of text for key phrases related to important clinical topics, represented by CUI and SNOMED codes. For example, cTAKES flags text matching the concept of homelessness (e.g., phrases such as “homeless,” “living on the street,” etc.) into CUI code C0237154. We defined housing insecurity as housing instability (e.g., “Homeless”) or poor housing quality (e.g., “House infested”), as CUI codes contained substantial ambiguity between these concepts. Three authors (S. Mehta, K. D. Sims, G. L. Schwartz) separately searched for relevant CUI codes, after which lists were compared and discrepancies discussed and resolved. We considered clinicians to have identified housing insecurity if clinical notes contained text matching any of these codes. For a complete list of codes, see Supplementary Table 1 . Social Worker Referrals We assessed whether patients flagged as housing insecure were subsequently seen by, or referred to, a social worker using both structured and unstructured EHR data. First, visits or contacts with social workers were defined in structured data as (1) encounters of any kind with social workers or with a social work department or (2) referrals to social workers or a social work department. Second, we used cTAKES and CUI/SNOMED codes ( Supplementary Table 2 ) in unstructured clinical notes to identify referrals to housing services (either a social worker or a housing-related community organization). Any indication of social worker contact or referral from either unstructured or structured data is considered evidence of some kind of “referral” to housing resources/social services. To identify care plausibly triggered by the initial housing insecurity flag, our referrals were limited to encounters with social workers within 6 months of being flagged as housing insecure. Validation of NLP Models To validate the cTAKES NLP algorithm identifying housing insecurity in this sample, we conducted a chart review of a random sample of 200 clinical notes flagged for housing instability or poor housing quality and 100 clinical notes without such flags. Each note was reviewed by two of four independent reviewers (S. Mehta, K. D. Sims, M. Choi, Y. J. Kim). We validated our NLP model identifying referrals to housing services similarly, conducting a chart review of 50 notes flagged with a referral to housing services and a random sample of 60 notes negative for housing referral (both only among patients identified as housing insecure). Each note was reviewed by two of three reviewers (S. Mehta, Y. Yang, G. L. Schwartz). Discrepancies were discussed until unanimous agreement by all reviewers, with insights from clinicians on our team (N. Cevallos). Initial agreement between raters was summarized by an intraclass correlation coefficient, and positive predictive value and negative predictive value were calculated after discrepancies were resolved. Covariates Patient characteristics available from EHRs included age in years, sex (male, female, other), race/ethnicity, chronic conditions, and insurance type (Medicaid vs not). Race/ethnicity was defined as per the UCSF Data Equity Taskforce ( Data Equity Taskforce sponsored by the Health Equity Council at UCSF Health, 2021 ): non-Hispanic Asian/Pacific Islander, non-Hispanic Black, Hispanic/Latine, non-Hispanic White, non-Hispanic Other race, or Unknown. We retained individuals with missing sex ( n missing = 127 “Other”), race ( n missing = 90,468 “Unknown”), and insurance type ( n missing = 253,188, considered “not Medicaid”). For descriptive purposes, we defined baseline history of chronic comorbidities using ICD-9 and -10 codes ( Supplementary Table 3 ; diabetes, dementia, cardiovascular disease, stroke, hypertension, and cancer). Statistical Analysis We first graphed housing insecurity trends from 2013 to 2022, overall and by racial/ethnic group. We then visually compared EHR trends to trends in San Francisco housing insecurity calculated using public data on homelessness and evictions ( Department of Homelessness and Supportive Housing, 2022 ). Among patients flagged as housing insecure, we used logistic regression to assess whether core demographic factors (age, race/ethnicity, sex, number of chronic conditions), clinical setting (emergency vs primary), or insurance were associated with the probability that people received referrals to housing services or interacted with UCSF social workers within 6 months of being identified as housing insecure by their clinician. We present both unadjusted (one predictor at a time) and fully adjusted models, as some covariates may be mediators rather than confounders (such as in models predicting racial/ethnic inequities). We fitted two adjusted models: (1) including all demographic/clinical predictors and (2) then adding linear and squared terms for year of visit to account for secular trends. As some individuals were flagged as housing insecure multiple times over the study period, these analyses limited the cohort to the first observation of housing insecurity for each patient. Finally, as a sensitivity analysis, we examined rates of social worker referrals among patients not identified as housing insecure, assessing whether referred patients had higher social worker contact. All analyses were conducted in R (4.1.3; 2022-03-10). Results Validation of NLP Model During the study period (2013–2022), we observed 1,111,823 patient encounters representing 119,127 patients. Individuals had an average of 9.33 encounters ( SD = 17.66) over that time. Clinicians identified housing insecurity in 6,253 of these encounters (0.6%), corresponding to 2,537 unique patients. Among these 2,537, patients were flagged an average of 2.47 times across all years ( SD = 4.05, min = 1, max = 79), and 34.6% (878 patients) were flagged as housing insecure more than once. Validation of our cTAKES NLP model detecting housing insecurity had high inter-rater agreement (intraclass correlation coefficient = 94.5%). Negative predictive value was high, at 97.5%, indicating that NLP-negative encounters were nearly always truly negative according to manual chart review. Positive predictive value was more moderate, at 77.5%, indicating a number of false positives. For our cTAKES NLP model detecting referrals to housing services, initial agreement between raters was substantial (intraclass correlation coefficient = 84.2%). Negative predictive value was high (91.7%), but positive predictive value was low (35%), such that even the very few referrals NLP identified ( n = 21 identified via NLP; <5% of all referrals) were likely an overcount. Characteristics of Encounters Table 1 compares the characteristics of patients in encounters identified versus not identified as experiencing housing insecurity. Identified patients were, on average, younger (70.0 years vs 75.3) and more likely to be male (60.5% vs 43.2%). At the time of their visit, patients experiencing housing insecurity were less likely to have a documented history of chronic conditions, potentially reflecting a healthcare utilization bias (those who are housing insecure may be less able to access care). Table 1. Characteristics of Patients at Time of Their Encounter With the UCSF Healthcare System, by Housing Insecurity Status Characteristic of patient Encounters where housing insecurity was identified ( n = 6,253) Encounters where housing insecurity was not identified ( n = 1,105,570) Age at encounter, years ( SD ) 70.0 (7.53) 75.3 (9.78) Sex Male, n (%) 3,780 (60.5) 478,049 (43.2) Other, n (%) 4 (0.1%) 270 (0.0%) Race/ethnicity Asian, n (%) 477 (7.6) 254,201 (23.0) Black, n (%) 1,466 (23.4) 130,742 (11.8) Latine, n (%) 473 (7.6) 99,782 (9.0) White, n (%) 2,773 (44.3) 466,777 (42.2) Other a , n (%) 458 (7.3) 64,206 (5.8) Unknown, n (%) 606 (9.7) 89,862 (8.1) Primary care encounter, n (%) 2,816 (45.0) 799,410 (72.3) Medicaid, n (%) 2,829 (45.2) 150,036 (13.6) History of diabetes 1,741 (27.8) 377,763 (34.2) History of dementia 322 (5.1) 94,535 (8.6) History of stroke 408 (6.5) 98,567 (8.9) History of CVD 1,449 (23.2) 258,973 (23.4) History of hypertension 4,754 (76.0) 910,717 (82.4) History of cancer 1,355 (21.7) 419,955 (38.0) Open in a new tab Notes : CVD = cardiovascular disease; SD = standard deviation; UCSF = University of California San Francisco. a Including individuals of the following racial groups: Southwest Asian, North African, American Indian, Alaska Native, or self-reported “Other.” There were 8,353 providers represented across all our encounters, of which 1,276 (15.3%) documented housing insecurity in their clinical notes. Residents made up 27% of this population (vs 13% among all encounters). Emergency medicine and internal medicine clinicians made up 70% of positively flagged clinical notes (vs 42% of notes overall). Trends in Clinical Identification of Housing Insecurity Figure 1A depicts the overall trend line for identified housing insecurity. The percentage of patients identified as housing insecure rose rapidly between 2013 and 2016, before declining toward baseline from 2017 through 2021 and finally ticking upwards in 2022. This followed, though with a 1–2 year delay, trends in San Francisco eviction filings ( Figure 1B ): eviction filings rose through 2016 before declining after an ordinance required landlords to demonstrate that they had “just cause” to evict a tenant (e.g., nonpayment of rent, as opposed to arbitrarily forcing a tenant to move); they then ticked upwards in 2021–2022 as federal pandemic housing supports expired, though we cannot evaluate whether that was the only factor driving this reversal. (Identification rates for every year from 2014 to 2019 were statistically significantly different from 2013 per a Poisson regression model, whereas rates for 2020–2022 were not; results not shown.) Figure 1. Open in a new tab Housing insecurity trends as identified in UCSF electronic health records versus in the city of San Francisco. Panel A shows housing insecurity in UCSF patients as identified using natural language processing of clinical notes. Panel B shows the annual number of evictions in San Francisco across the general population (not age-specific). The uptick in housing insecurity among UCSF patients from 2021 to 2022 coincides with the end of many pandemic housing assistance programs a year earlier, though it may be due to other factors (e.g., unobserved changes in clinical practice). Data on SF evictions are pulled from the San Francisco Rent Board’s annual eviction reports, posted publicly online by San Francisco’s municipal government. UCSF = University of California San Francisco. Conversely, trends in the percentage of patients identified as housing insecure did not track trends in homelessness among SF older adults, nor trends in older adults accessing homelessness services in the city ( Supplementary Figures 1 and 2 ). Both of these homelessness indicators increased monotonically throughout the study period. The divergence between official homelessness and clinically identified housing insecurity trends may reflect the exclusion of San Francisco’s safety net hospital from our data, such that patients facing eviction were more likely to be present in our sample than those experiencing homelessness. Notably, rates of identification did not appear to dramatically increase following the February 2019 implementation of UCSF’s revised social needs screener, which newly included questions on housing insecurity in the EHR. Figure 2 displays trends in the proportion of older patients identified as housing insecure by racial/ethnic group. Trends for most groups tracked overall trends. The exception was Asian patients, among whom trends were essentially flat. Throughout our study period, rates of clinician-identified housing insecurity were consistently higher among older Black patients. Figure 2. Open in a new tab Trends in NLP-identified housing insecurity among older adults seeking care at UCSF, by race/ethnicity. Housing insecurity was identified using natural language processing applied to older adults’ clinical notes. Racial/ethnic groups are mutually exclusive. NLP = natural language processing; UCSF = University of California San Francisco. Referrals to Housing/Social Services We next assessed whether patients identified as housing insecure were referred by UCSF providers to social services. Of the 6,253 encounters where patients were identified, only 441 (6.6%) were followed by a referral to social services within 6 months. Of these encounters, a total of 21 encounters were identified by NLP, 9 encounters had a documented referral to social services, and 425 encounters had a social worker visit (implying some kind of referral). This represented 167 unique patients with referrals. Referral rates were consistently low. Despite observed increases in housing insecurity over the study period, the absolute number of patients referred to social services remained effectively constant ( Figure 3 ). Further, referral rates among the housing insecure were only marginally higher than among those without housing insecurity (Supplementary Material A). Figure 3. Open in a new tab Trends in connections to social services among patients experiencing housing insecurity at UCSF, 2013–2022. Housing insecurity was identified using natural language processing applied to older adults’ clinical notes. Connections to social services were evaluated via a social worker visit, referral to a social work department, or detection of a housing services referral in unstructured clinical notes using natural language processing. A patient was considered to be connected to social services if any of these indicators were present within 6 months of the encounter during which they were flagged as housing insecure. UCSF = University of California San Francisco. Referrals to social services were also unevenly distributed across demographic groups and clinical settings. Examining referral rates among unique patients experiencing housing insecurity ( n = 2,537), those referred to social services were more likely to be seen in primary care ( Table 2 ). (For detailed characteristics of patients with confirmed visits with social workers, as opposed to merely having some indication of a referral in their EHR, see Supplementary Table 4 .) In multivariable regression, patients seen in primary care had higher odds of later being referred to social services within 6 months compared to those seen in emergency care, both in unadjusted (odds ratio [OR] = 2.24, 95% confidence interval [95% CI]: 1.61–3.12) and adjusted (2.04, 95% CI: 1.41–2.96; Supplementary Table 5 ) models. Across race/ethnicity, Asian patients had much lower odds of receiving a referral in unadjusted (0.72, 95% CI: 0.40–1.30) and adjusted (0.51, 95% CI: 0.28–0.95) models, compared to non-Hispanic White patients. Latine patients were also less likely to receive a referral in fully adjusted models (0.51, 95% CI: 0.25–1.04), though confidence intervals crossed the null. Table 2. Characteristics of Unique Patients Experiencing Housing Insecurity, by Whether They Were Connected to Social Services Characteristic of patient Patients referred to or seen by social services ( n = 167) Patients not seen by social services ( n = 2,370) Referral rates by race/ethnicity Age at encounter, years ( SD ) 65.2 (8.38) 64.3 (7.95) — Male, n (%) 92 (55.1) 1,392 (58.78) — Racial/ethnic designation Asian, n (%) 14 (8.4) 249 (10.5) 5.3% Black, n (%) 41 (24.6) 514 (21.7) 7.4% Latine, n (%) 9 (5.4) 185 (7.8) 4.6% White, n (%) 74 (44.3) 953 (40.2) 7.2% Other, n (%) 15 (9.0) 132 (5.6) 10.2% Unknown, n (%) 14 (8.4) 3,375 (14.2) 0.4% Seen by primary care at index visit, n (%) 110 (65.9) 1,096 (46.2) — Medicaid, n (%) 67 (40.1) 919 (38.8) — Open in a new tab Note : SD = standard deviation. Discussion In this study, we examined UCSF as a case study of how well healthcare institutions are meeting the challenge of rising housing insecurity among older adults. In particular, we investigated four questions: (1) whether housing insecurity can be identified in older patients’ clinical notes, (2) whether trends in that identification among older patients mirror trends in housing insecurity regionally, (3) how often such identification leads to a referral to housing/social services, and (4) whether referrals are equitably distributed. Below, we discuss each answer, as well as implications for research, policy, and practice. Implications for Research With Natural Language Processing: Identifying Housing Insecurity and Referrals First, we demonstrated that NLP can be used within clinical notes to identify older adults experiencing housing insecurity, with moderate positive predictive value and high negative predictive value. These results are consistent with other studies showing that NLP can be used to identify housing instability (i.e., excluding housing quality) in various populations, including veterans ( Zamora-Resendiz et al., 2024 ), individuals with substance use disorders ( Harris et al., 2023 ), and individuals with serious illness ( Xie et al., 2023 ). Results were also roughly consistent with regional eviction trends in the general population, suggesting that clinicians are capable of detecting differences in housing insecurity in older patients over time (albeit on a 1–2 year delay). Our moderate positive predictive value, however, underscores special challenges for such methods in older populations. False positives were frequently driven by cTAKES flagging discussions about the fit between patients’ housing and their fall risk—discussions about steps in the home, lack of shower handrails, etc. For clinical purposes, it is likely preferable to err on the side of inclusive NLP definitions of unmet housing needs in patient records to ensure linkage to needed services. But researchers seeking to use NLP to identify housing insecurity among older adults must be cautious of this age-specific measurement error. Our investigation into referrals is a key extension to this literature. Our NLP model was unable to identify many referrals in unstructured notes, nor even mentions of interactions with social workers, and false positives were frequent; structured EHR data was a more reliable indicator of patients’ social services connectivity. Whether this is because clinicians irregularly document their social work and housing services referrals or because documentation was uninterpretable to NLP algorithms like cTAKES should be evaluated, including via physician interviews. Future work should also train NLP algorithms to classify the types and quality of social services patients are receiving to ensure they are connected to appropriate and culturally competent care ( Bako et al., 2020 , 2021 ; Kennedy et al., 2023 ). Implications for Practice and Policy Second, our results point to serious gaps between clinical identification of older patients’ housing insecurity and successful referrals to social services. Only 7% of patients identified as housing insecure were seen by or referred to social work within 6 months. This may be due to insufficient training or limited paid time for clinicians to facilitate referrals, but it may also be due to a realization (among clinicians, social workers, or both) that there are simply not enough resources to meet demand, inducing a sense of futility. Social workers, for example, may be overwhelmed with burgeoning caseloads ( Whitaker et al., 2006 ) and lack capacity to pick up referrals for anyone but patients in the most extreme crisis (disincentivizing referrals for all other cases). Alternatively, the housing resources social workers might connect patients to may not exist or have years-long wait times ( Acosta & Gartland, 2021 ). At the end of our study period, San Francisco’s controller found the city had roughly half as many beds available as they had people experiencing homelessness ( White, 2023 ). And across the county, only a quarter of people whose incomes qualify for federal housing assistance actually receive it, given a sustained retreat in federal housing investments relative to need ( DeParle, 2023 ; Human Rights Watch, 2022 ; Matthews, 2014 ). We also found that primary care departments were more likely to connect patients to services than emergency departments, even though EDs are more likely to serve individuals experiencing housing insecurity (and despite a 2019 law requiring housing referrals for ED patients experiencing homelessness; Amato et al., 2019 ; Mayes et al., 2024 ; Sorelle, 2019 ). Emergency rooms treat the majority of the housing insecure population, as these patients face numerous barriers to routine primary care; obtaining basic needs such as food, shelter, and clothing while remaining safe often supersedes primary care needs, especially for those with disabilities ( Gelberg et al., 1997 ; Keene et al., 2018 ; Ku et al., 2010 ; Vohra et al., 2022 ). Yet it was ER patients who experienced the lowest rates of housing referrals. This deficit may be due to ERs lacking resources for effectively managing housing needs—for example, overcrowding limiting providers’ time with each patient ( Morley et al., 2018 ; Salhi et al., 2018 ). In sum, our work highlights a gap between the needs of older patients experiencing housing insecurity and the care clinical systems provide, especially in emergency settings. It also suggests that current California laws meant to encourage housing services referrals in ERs are not accomplishing their goals. More research is needed to evaluate generalizability to other urban communities of older adults, as well as what types of public health interventions would fill this care gap. These problems appear stark for racially/ethnically minoritized older adults. On the identification side, we found distinct levels and trends in housing insecurity by racial/ethnic group. Whether this represents an inequity in terms of clinical care quality or reflects differences in true need (i.e., housing inequities wrought by structural racism, Chatters et al., 2021 ) is unclear. It is further unclear whether clinicians differed in how well they recorded housing insecurity across race/ethnicity, as more or less clearly recorded housing insecurity information for one race or another in their clinical notes could skew racial/ethnic comparisons. What was clear were racial inequities in referrals: once identified, older Asian and Latine patients were substantially less likely to be referred to social services compared to their White counterparts. These results are consistent with research showing racial inequities in access to homelessness interventions, such as shelters and temporary housing ( Fowle, 2022 ; Willison et al., 2024 ). Future studies should investigate the specific causes of this inequity (geographic distributions of health and housing resources, interpersonal discrimination, etc.). Medical systems would be wise, however, to proactively evaluate the adequacy of ameliorating factors under their control—for example, availability of translation services, recruitment and retention of a diverse clinical workforce, and effective training on racial/ethnic biases in patient care. These steps could have wide-ranging impacts. Nearly one in four adults 65 or older in CA report they do not speak English very well ( Migration Policy Institute, 2022 ), and while 38% of CA Medicaid patients identify as Latine, only 5% of CA physicians do, yielding a mere 62 Spanish-speaking physicians for every 100,000 Spanish-speaking patients with limited English proficiency ( Martinez et al., 2019 ; McLellan, 2019 ). Finally, our EHR data lack direct measures of housing insecurity screening. A given clinician may screen every patient, screen occasionally, or never screen. If screening is rare, raising referral rates may only go so far: one cannot be referred to housing services if one’s housing instability is not detected. Future research using physician surveys could clarify policy remedies. Strengths and Limitations Our study has notable strengths. First, we extended the use of NLP to identify housing insecurity in a large sample of over 100,000 older adults, a crucial demographic among individuals experiencing homelessness ( Culhane et al., 2019 ; Kushel, 2020 ). We do so in California, home to one-third of the U.S. homeless population ( Kushel et al., 2023 ). We further extended prior literature by examining connections to social services in addition to housing insecurity, and by examining inequities in those referrals across clinical settings and racial/ethnic groups. This study also has important limitations. First, we utilized a relatively simple NLP model compared to more recent advances ( Yang et al., 2022 ). This is in some ways a strength, given the widespread use of cTAKES in medical research, but misses key technical developments. We hope future studies will replicate our results using more advanced algorithms to improve both the detection of housing insecurity among older adults and referrals to housing services. Second, despite extensive EHR data from UCSF, we observed few referrals to social work, decreasing the precision of our inequity estimates. These social work visits were themselves limited: we do not know if they were in fact targeted toward housing services, and we further do not know whether identification of housing insecurity is what prompted the social worker referral/contact we observed. In addition, our NLP model for detecting referrals had a low positive predictive value. True rates of housing services referrals are thus likely even lower than those we report. Data from a larger health system could improve statistical efficiency. Additional data on factors such as translation service availability would also help identify areas of improvement and proximal causes of racial/ethnic referral inequities. Third, because we used deidentified EHR data, all data are date-shifted by up to a year, which means our annual trends represent general trends. Fourth, our regional eviction data are also limited, given that they do not include information on the age of evictees nor on informal evictions. Lastly, our analysis is limited to care resources within this healthcare system. We miss any undocumented or verbal connections to resources like homeless shelters or nonprofits in the area. UCSF is only one of the many healthcare providers in San Francisco. There are other important providers for this population, including public health practitioners and the Zuckerberg San Francisco General Hospital (the city’s safety net hospital). Our cohort may therefore differ from the larger population of older patients experiencing housing insecurity in the city. In conclusion, this study extends current literature on the identification of housing insecurity among older adults in large healthcare systems. We have demonstrated that NLP can be used to reliably identify older adults experiencing housing insecurity from clinical notes. We have also shown that there is a gap between who is identified as housing insecure and who gets referred or seen by social workers. Only 7% of this cohort experiencing housing insecurity were connected to social services, and these individuals were disproportionately White and treated in primary care. Future work is needed to replicate these results in other urban communities and develop appropriate interventions. Supplementary Material gnaf027_suppl_Supplementary_Tables_1-5_Figures_1-2_Materials_3 gnaf027_suppl_supplementary_tables_1-5_figures_1-2_materials_3.docx (214.6KB, docx) Acknowledgments The authors acknowledge the use of the UCSF Information Commons computational research platform , developed and supported by UCSF Bakar Computational Health Sciences Institute in collaboration with IT Academic Research Services, Center for Intelligent Imaging Computational Core, and CTSI Research Technology Program. Contributor Information Erin L Ferguson, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA. Shivani Mehta, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA. Silvia Miramontes, Bakar Computational Health Sciences Institute, University of California, San Francisco, San Francisco, California, USA. Minhyuk Choi, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA. Ye Ji Kim, Department of Epidemiology, Boston University, Boston, Massachusetts, USA. Tanisha G Hill-Jarrett, Memory and Aging Center, Department of Neurology, University of California, San Francisco, San Francisco, California, USA; Global Brain Health Institute, University of California, San Francisco, San Francisco, California, USA. Nicolas Cevallos, School of Medicine, University of California, San Francisco, San Francisco, California, USA. Yulin Yang, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA. Scott C Zimmerman, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA. Ruijia Chen, Department of Epidemiology, Boston University, Boston, Massachusetts, USA. Min Hee Kim, Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA. Kendra D Sims, Department of Epidemiology, Boston University, Boston, Massachusetts, USA. Gabriel L Schwartz, Philip. R. Lee Institute for Health Policy Studies, University of California, San Francisco, San Francisco, California, USA; Urban Health Collaborative and Department of Health Management and Policy, Dornsife School of Public Health, Drexel University, Philadelphia, Pennsylvania, USA. Funding This work was supported by the National Institutes of Health (NIH) (T32AG049663-06A1 to E. L. Ferguson, K00AG068431 to R. Chen, K99AG078405 to M. H. Kim, K99AG083121 to K. D. Sims, U54CA267735-03 to G. L. Schwartz). The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. Conflict of Interest None. Data Availability For patient privacy, access is restricted to UCSF affiliates. Code is available from the authors by request. Analyses were not preregistered. References Acosta, S., & Gartland, E. (2021). Families wait years for housing vouchers due to inadequate funding: Expanding program would reduce hardship, improve equity. Center on Budget and Policy Priorities. https://www.cbpp.org/research/housing/families-wait-years-for-housing-vouchers-due-to-inadequate-funding [ Google Scholar ] Amato, S., Nobay, F., Amato, D. P., Abar, B., & Adler, D. (2019). Sick and unsheltered: Homelessness as a major risk factor for emergency care utilization. American Journal of Emergency Medicine, 37(3), 415–420. https://doi.org/ 10.1016/j.ajem.2018.06.001 [ DOI ] [ PubMed ] [ Google Scholar ] Bako, A. T., Taylor, H. L., Wiley, K., Zheng, J., Walter-McCabe, H., Kasthurirathne, S. N., & Vest, J. R. (2021). Using natural language processing to classify social work interventions. American Journal of Managed Care, 27(1), e24–e31. https://doi.org/ 10.37765/ajmc.2021.88580 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bako, A. T., Walter-McCabe, H., Kasthurirathne, S. N., Halverson, P. K., & Vest, J. R. (2020). Reasons for social work referrals in an urban safety-net population: A natural language processing and market basket analysis approach. Journal of Social Service Research, 47(3), 414–425. https://doi.org/ 10.1080/01488376.2020.1817834 [ DOI ] [ Google Scholar ] Brown, E., Podesta, A., Metcalf, B., Bates, L., Seeger, J., Kirkeby, M., Coy, M., & Anixter, H. (2018). California’s housing future: Challenges and opportunities. California Department of Housing and Community Development. https://www.hcd.ca.gov/policy-research/plans-reports/docs/sha_final_combined.pdf [ Google Scholar ] Campbell, K. E., Oliver, D. E., & Shortliffe, E. H. (1998). The unified medical language system: Toward a collaborative approach for solving terminologic problems. Journal of the American Medical Informatics Association, 5(1), 12–16. https://doi.org/ 10.1136/jamia.1998.0050012 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Chatters, L. M., Taylor, H. O., & Taylor, R. J. (2021). Racism and the life course: Social and health equity for Black American older adults. Public Policy and Aging Report, 31(4), 113–118. https://doi.org/ 10.1093/ppar/prab018 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Culhane, D., Doran, K., Schretzman, M., Johns, E., Treglia, D., Byrne, T., Metraux, S., & Kuhn, R. (2019). The emerging crisis of aged homelessness in the US: Could cost avoidance in health care fund housing solutions? International Journal of Population Data Science, 4(3), Article 3. https://doi.org/ 10.23889/ijpds.v4i3.1185 [ DOI ] [ Google Scholar ] Cusack, M., Montgomery, A. E., Hunt-Johnson, N., Dichter, M., & True, G. (2019). Making connections: Understanding how screening, triage, and referral processes can promote veteran housing stability. Social Work in Public Health, 34(6), 483–491. https://doi.org/ 10.1080/19371918.2019.1629143 [ DOI ] [ PubMed ] [ Google Scholar ] Data Equity Taskforce sponsored by the Health Equity Council at UCSF Health. (2021). UCSF health’s equity-related variables user’s guide. University of California, San Francisco. [ Google Scholar ] DeParle, J. (2023, December 19). As need rises, housing aid hits lowest level in nearly 25 Years. The New York Times. https://www.nytimes.com/2023/12/19/us/politics/housing-aid-rent-costs.html [ Google Scholar ] Department of Homelessness and Supportive Housing. (2022). Homelessness response system data. https://hsh.sfgov.org/about/research-and-reports/archived-reports/#pit [ Google Scholar ] Fischer, W., Acosta, S., & Bailey, A. (2021). An agenda for the future of public housing. Center on Budget and Policy Priorities. https://www.cbpp.org/research/housing/an-agenda-for-the-future-of-public-housing [ Google Scholar ] Fowle, M. Z. (2022). Racialized homelessness: A review of historical and contemporary causes of racial disparities in homelessness. Housing Policy Debate, 32(6), 940–967. https://doi.org/ 10.1080/10511482.2022.2026995 [ DOI ] [ Google Scholar ] Gelberg, L., Gallagher, T. C., Andersen, R. M., & Koegel, P. (1997). Competing priorities as a barrier to medical care among homeless adults in Los Angeles. American Journal of Public Health, 87(2), 217–220. https://doi.org/ 10.2105/ajph.87.2.217 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Hamed, S., Bradby, H., Ahlberg, B. M., & Thapar-Björkert, S. (2022). Racism in healthcare: A scoping review. BMC Public Health, 22(1), 988. https://doi.org/ 10.1186/s12889-022-13122-y [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Harris, D. R., Anthony, N., Quesinberry, D., & Delcher, C. (2023). Evidence of housing instability identified by addresses, clinical notes, and diagnostic codes in a real-world population with substance use disorders. Journal of Clinical and Translational Science, 7(1), e196. https://doi.org/ 10.1017/cts.2023.626 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Human Rights Watch. (2022, September 27). US: Budget cuts put public housing tenants at risk. https://www.hrw.org/news/2022/09/27/us-budget-cuts-put-public-housing-tenants-risk [ Google Scholar ] Ibarra, A. B. (2023, February 10). The fastest-growing homeless population? Seniors. CalMatters. https://calmatters.org/health/2023/02/california-homeless-seniors/ [ Google Scholar ] Keene, D. E., Guo, M., & Murillo, S. (2018). “That wasn’t really a place to worry about diabetes”: Housing access and diabetes self-management among low-income adults. Social Science & Medicine (1982), 197, 71–77. https://doi.org/ 10.1016/j.socscimed.2017.11.051 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Kennedy, E. E., Davoudi, A., Hwang, S., Freda, P. J., Urbanowicz, R., Bowles, K. H., & Mowery, D. L. (2023). Identifying barriers to post-acute care referral and characterizing negative patient preferences among hospitalized older adults using natural language processing. AMIA Annual Symposium Proceedings, 2022, 606–615. [ PMC free article ] [ PubMed ] [ Google Scholar ] Ku, B. S., Scott, K. C., Kertesz, S. G., & Pitts, S. R. (2010). Factors associated with use of urban emergency departments by the U.S. homeless population. Public Health Reports, 125(3), 398–405. https://doi.org/ 10.1177/003335491012500308 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Kushel, M. (2020). Homelessness among older adults: An emerging crisis. Generations-Journal of the American Society on Aging, 44(2), 1–. http://generations.asaging.org/homelessness-older-adults-poverty-health [ Google Scholar ] Kushel, M., Moore, T., Birkmeyer, J., Dhatt, Z., Duke, M., Knight, K., & Ponder, K. (2023). Toward a new understanding: The California statewide study of people experiencing homelessness. UCSF Benioff Homelessness and Housing Initiative. https://homelessness.ucsf.edu/sites/default/files/2023-06/CASPEH_Report_62023.pdf [ Google Scholar ] Martinez, L., Solis, G., Gomez, V., Mendez-Vargas, J., Diaz, S. F. M., & Hayes-Bautista, D. E. (2019). California’s language concordance mismatch: An analysis of language proficient physicians and limited English proficient individuals who speak Spanish, Tagalog, Thai, Lao or Vietnamese. UCLA Latino Policy & Politics Institute. https://latino.ucla.edu/research/californias-language-concordance-mismatch-an-analysis-of-language-proficient-physiciansand-limited-english-proficient-individuals-who-speak-spanish-tagalog-thai-lao-or-vietnamese/ [ Google Scholar ] Matthews, D. (2014, May 31). Only a quarter of eligible people get housing aid. Vox. https://www.vox.com/2014/5/31/5764262/76-percent-of-people-who-qualify-for-housing-aid-dont-get-it [ Google Scholar ] Mayes, K. D., Cash, R. E., Schiavoni, K. H., Vogeli, C., Thorndike, A. N., Camargo, C. A., Jr., & Samuels-Kalow, M. (2024). Social risk, social need, and use of the emergency department. JAMA Network Open, 7(1), e2352365. https://doi.org/ 10.1001/jamanetworkopen.2023.52365 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] McLellan, H. (2019, June 13). Medical interpreters bridge crucial gaps between patients and providers. California Health Care Foundation. https://www.chcf.org/blog/medical-interpreters-bridge-crucial-gaps-between-patients-and-providers/ [ Google Scholar ] Migration Policy Institute. (2022). State language data—CA. Migration Policy Institute. https://www.migrationpolicy.org/data/state-profiles/state/language/CA [ Google Scholar ] Moore, K. (2019). Social determinants of health: Addressing barriers through screening, referrals, and care coordination. North Carolina Medical Journal, 80(5), 306–311. https://doi.org/ 10.18043/ncm.80.5.306 [ DOI ] [ PubMed ] [ Google Scholar ] Morley, C., Unwin, M., Peterson, G. M., Stankovich, J., & Kinsman, L. (2018). Emergency department crowding: A systematic review of causes, consequences and solutions. PLoS One, 13(8), e0203316. https://doi.org/ 10.1371/journal.pone.0203316 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Painter, H., Parry, E., McCann, L., Dehn Lunn, A., & Ford, J. (2024). Social needs screening in primary care: A tool in the fight for health equity? Public Health in Practice (Oxford, England), 7, 100466. https://doi.org/ 10.1016/j.puhip.2024.100466 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Richards, J., & Kuhn, R. (2023). Unsheltered homelessness and health: A literature review. AJPM Focus, 2(1), 100043. https://doi.org/ 10.1016/j.focus.2022.100043 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Salhi, B. A., White, M. H., Pitts, S. R., & Wright, D. W. (2018). Homelessness and emergency medicine: A review of the literature. Academic Emergency Medicine, 25(5), 577–593. https://doi.org/ 10.1111/acem.13358 [ DOI ] [ PubMed ] [ Google Scholar ] Savova, G. K., Masanz, J. J., Ogren, P. V., Zheng, J., Sohn, S., Kipper-Schuler, K. C., & Chute, C. G. (2010). Mayo clinical Text Analysis and Knowledge Extraction System (cTAKES): Architecture, component evaluation and applications. Journal of the American Medical Informatics Association, 17(5), 507–513. https://doi.org/ 10.1136/jamia.2009.001560 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Sentell, T., & Braun, K. (2012). Low health literacy, limited English proficiency, and health status in Asians, Latinos, and other racial/ethnic groups in California. Journal of Health Communication, 17(Suppl 3), 82–99. https://doi.org/ 10.1080/10810730.2012.712621 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Smith, P. D., Keene, D. E., Dilday, S., Blankenship, K. M., & Groves, A. K. (2024). Eviction from rental housing and its links to health: A scoping review. Health & Place, 86, 103182. https://doi.org/ 10.1016/j.healthplace.2024.103182 [ DOI ] [ PubMed ] [ Google Scholar ] Sorelle, R. (2019). California law setting discharge rules for homeless patients creates tough task for EDs. Emergency Medicine News, 41(3), 1,24–24. https://doi.org/ 10.1097/01.eem.0000554285.25603.8d [ DOI ] [ Google Scholar ] Taylor, L. (2018). Housing and health: An overview of the literature. Health Affairs Health Policy Brief. https://doi.org/ 10.1377/hpb20180313.396577 [ DOI ] [ Google Scholar ] Thorman, T., & Payares-Montoya, D. (2024). Income inequality in California. Public Policy Institute of California. https://www.ppic.org/publication/income-inequality-in-california/ [ Google Scholar ] University of California, San Francisco, Academic Research Systems. (2023). UCSF DeID CDW-OMOP (Version 2023-June). https://data.ucsf.edu/research/deid-data [ Google Scholar ] Vohra, N., Paudyal, V., & Price, M. J. (2022). Homelessness and the use of emergency department as a source of healthcare: A systematic review. International Journal of Emergency Medicine, 15(1), 32. https://doi.org/ 10.1186/s12245-022-00435-3 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Whitaker, T., Weismiller, T., Clark, E., & Wilson, M. (2006). Assuring the sufficiency of a frontline workforce: A national study of licensed social workers. Special report: Social work services in health care settings. National Association of Social Workers. https://www.socialworkers.org/LinkClick.aspx?fileticket=OilZ7p._EEnE%3D&portalid=0 [ Google Scholar ] White, M. (2023, July 27). SF had half as many beds as unhoused people in ’22. San Francisco Examiner. https://www.sfexaminer.com/news/housing/sf-had-half-as-many-shelter-beds-as-unhoused-people-in-2022/article_fe33046a-2cb1-11ee-a0a4-7732f75b78a9.html [ Google Scholar ] Willison, C., Unwala, N., Singer, P. M., Creedon, T. B., Mullin, B., & Cook, B. L. (2024). Persistent disparities: Trends in rates of sheltered homelessness across demographic subgroups in the USA. Journal of Racial and Ethnic Health Disparities, 11(1), 326–338. https://doi.org/ 10.1007/s40615-023-01521-9 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Xie, F., Wang, S., Viveros, L., Rich, A., Nguyen, H. Q., Padilla, A., Lyons, L., & Nau, C. L. (2023). Using natural language processing to identify the status of homelessness and housing instability among serious illness patients from clinical notes in an integrated healthcare system. JAMIA Open, 6(3), ooad082. https://doi.org/ 10.1093/jamiaopen/ooad082 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yang, X., Chen, A., PourNejatian, N., Shin, H. C., Smith, K. E., Parisien, C., Compas, C., Martin, C., Costa, A. B., Flores, M. G., Zhang, Y., Magoc, T., Harle, C. A., Lipori, G., Mitchell, D. A., Hogan, W. R., Shenkman, E. A., Bian, J., & Wu, Y. (2022). A large language model for electronic health records. npj Digital Medicine, 5(1), 1–9. https://doi.org/ 10.1038/s41746-022-00742-2 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Yearby, R., Clark, B., & Figueroa, J. F. (2022). Structural racism in historical and modern US health care policy. Health Affairs (Project Hope), 41(2), 187–194. https://doi.org/ 10.1377/hlthaff.2021.01466 [ DOI ] [ PubMed ] [ Google Scholar ] Zamora-Resendiz, R., Oslin, D. W., Hooshyar, D., & Crivelli, S.; Million Veteran Program Suicide Exemplar Work Group. (2024). Using electronic health record metadata to predict housing instability amongst veterans. Preventive Medicine Reports, 37, 102505. https://doi.org/ 10.1016/j.pmedr.2023.102505 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials gnaf027_suppl_Supplementary_Tables_1-5_Figures_1-2_Materials_3 gnaf027_suppl_supplementary_tables_1-5_figures_1-2_materials_3.docx (214.6KB, docx) Data Availability Statement For patient privacy, access is restricted to UCSF affiliates. Code is available from the authors by request. Analyses were not preregistered. Articles from The Gerontologist are provided here courtesy of Oxford University Press ACTIONS View on publisher site PDF (823.2 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top