Unstable Housing and Mortality Among US Veterans Receiving Dialysis

Key Points Question Among US veterans receiving dialysis, is unstable housing associated with mortality and does risk differ according to age? Findings In this cohort study of 25 689 veterans receiving dialysis, unstable housing was associated with an increased risk of all-cause mortality, and this risk increased with age. Meaning These findings suggest that unstable housing may contribute to socioeconomic disparities in mortality among US veterans receiving dialysis, with older adults being particularly vulnerable.


Introduction
Unstable housing is variably defined but includes homelessness and housing that one might lose because it is unaffordable, overcrowded, or dangerous. 12][3] Unstable housing has the potential to negatively affect nearly every aspect of dialysis care and increase complications, and it limits access to home dialysis modalities and kidney transplantation. 34][5] However, little is known about unstable housing among individuals receiving dialysis.
Since 2010, the Office of the President of the United States and the US Department of Veterans Affairs have made it a priority to eliminate homelessness among veterans. 6Comprehensive efforts have included investing in veteran-specific housing programs that use a housing-first approach, increasing affordable housing for veterans, and funding programs that prevent veterans from becoming homeless. 6Screening for unstable housing at US Veterans Health Administration (VHA) clinics began in 2012 as part of these national efforts; a positive screen for unstable housing triggers a timely referral to veteran-specific housing programs. 7With this initiative, the number of homeless veterans decreased from 74 087 to 33 129 between 2010 and 2022. 6e goals of this study were (1) to identify characteristics associated with unstable housing among veterans receiving dialysis and (2) to evaluate potential associations between unstable housing before dialysis initiation and mortality.There is evidence that socioeconomic disparities in mortality differ according to age for individuals receiving dialysis, 8 so we sought to examine whether the association between unstable housing and mortality differed according to age.We hypothesized that unstable housing would be associated with increased risk of all-cause mortality and risks would increase with age.

Study Population
This cohort study was approved by the institutional review boards of the University of Texas Health Science Center at San Antonio, Audie L. Murphy Veterans Memorial Hospital, and University of Texas at Austin Dell Medical School.Informed consent was waived in accordance with the Common Rule because data were deidentified.The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
We developed a retrospective cohort examining veterans who initiated dialysis between October 1, 2012, and December 31, 2018 (Figure 1).Entry into the cohort was defined by the presence or absence of an International Classification of Diseases, Ninth Revision (ICD-9), or

International Statistical Classification of Diseases and Related Health Problems, Tenth Revision
(ICD-10), diagnostic code for end-stage kidney disease and the presence of a screener for unstable housing.The Veterans Information Resource Center identified 36 626 veterans receiving dialysis by linking with US Renal Data System (USRDS) data for veterans who had obtained health care at VHA clinics, were screened for unstable housing, and started dialysis within the designated time period.
We chose a start date of October 1, 2012, since that was when housing screening began at VHA locations.We excluded individuals who were not aged between 18 and 85 years at dialysis initiation, had housing screen data that were not within the 3-year period before starting dialysis, were missing the medical evidence form, lived in a US territory (eg, Puerto Rico or Guam), recovered kidney function during the study period, or discontinued dialysis in less than 90 days due to death, kidney transplant receipt, or loss to follow-up (n = 10 937).We excluded individuals aged older than 85 years, given the contribution of age to mortality.In addition, we excluded those who discontinued dialysis in less than 90 days to study associations with unstable housing among individuals with kidney failure instead of acute kidney disease or kidney injury.The final analytic cohort consisted of (eTable 1 in Supplement 1).

Exposure Variable
0][11][12] The clinical reminder includes 2 questions: "In the past two months, have you been living in stable housing that you own, rent, or stay in as part of a household?"and "Are you worried or concerned that in the next two months you may not have stable housing that you own, rent or stay in as part of a household?"Development and validation of the clinical reminder was described previously, and the screener has demonstrated good internal consistency (reliability coefficient, 0.85). 12The Homelessness Screening Clinical Reminder is administered annually in outpatient primary care and mental health settings. 11Our sample included veterans with at least 1 complete housing screen; thus, there was no missingness.A positive screen increases screening frequency to every 6 months, and 3 consecutive negative screens extend screening to every 2 years.Using all available screens within 3 years before dialysis initiation, we defined unstable housing (yes or no) as answering "no" to having stable housing within the past 2 months and/or "yes" to being worried about not having stable housing in the upcoming 2 months. 10We used the screen most proximal to the date of dialysis initiation.We did not evaluate housing screens after dialysis initiation.

Covariates
We combined data from the VHA CDW and USRDS to determine baseline demographic characteristics (eTable 3 in Supplement 1).We obtained information on age, sex (female or male), race and ethnicity (analyzed as mutually exclusive categories of Black, Hispanic, White, or other race or ethnicity [specific categories not available]), and urban or rural residence (urban, rural, or highly rural) from the VHA CDW.Race and ethnicity data were obtained from electronic medical records,  which are usually self-report; these data were included due to associations between race and ethnicity and mortality among people receiving dialysis. 13 obtained information from the USRDS on year of dialysis initiation, medical insurance at dialysis initiation (Medicaid, Medicare, group health insurance, other coverage, or no coverage), incident vascular access (arteriovenous fistula, central venous catheter, arteriovenous graft, other, or unknown), predialysis nephrology care (yes, no, or unknown), primary cause of kidney failure (hypertension, diabetes, cystic kidney disease, glomerulonephritis, other, or unknown), and incident dialysis modality (in-center hemodialysis, peritoneal dialysis, home hemodialysis, or unknown).
We extracted information on comorbidities from VHA inpatient and outpatient data tables using ICD-9 and ICD-10 codes (eTable 4 in Supplement 1).We assigned comorbidities if an individual had 2 outpatient visits or 1 inpatient visit with the relevant diagnostic code at any time before dialysis initiation.We looked back up to 3 years from the dialysis start date for the presence of comorbidities and drug dependence variables.

Outcome Assessment
Our primary outcome of interest was time to death from 90 days after dialysis initiation.We obtained mortality, censoring and competing events, and associated dates from the VHA CDW and USRDS files.We followed participants from 90 days after dialysis initiation until death, transplantation, discontinuation of dialysis, loss to follow-up, or end of the follow-up period (December 31, 2018).
Information on transplantation, recovery of kidney function, discontinuation of dialysis, and loss to follow-up was obtained from the USRDS.

Statistical Analysis
We compared baseline characteristics according to unstable housing status, and we present data as means (SDs) or medians (IQRs) for continuous variables and numbers or percentages for categorical variables.We used logistic regression, with unstable housing as the outcome and all covariates included in one model, to assess correlates of unstable housing within 3 years before dialysis initiation.
We estimated the overall cumulative incidence of mortality for veterans with and without unstable housing.We used the Fine and Gray 12 cumulative incidence method to estimate adjusted associations between unstable housing within 3 years before dialysis start and mortality.The Fine and Gray method is conceptually similar to the Cox proportional hazards regression method but allows for competing risks.We incrementally adjusted for the following covariates based on the literature and theoretical considerations: model 1 adjusted for age; model 2 additionally adjusted for sex, race and ethnicity, year of dialysis initiation, and predialysis nephrology care; and model 3 additionally adjusted for dialysis modality, cause of kidney failure, vascular access, medical insurance, urban or rural residence, and comorbidities.We included an age-squared term in all models to account for the nonlinear effects of age.
We tested the proportional hazards assumption using Schoenfeld residuals, which was met for all covariates in models 1 and 2, including age. 14,15The proportional hazards assumption was not met for some of the comorbidities in model 3, so the variables that did not meet the assumption were instead adjusted for as stratified variables. 14,15 tested for effect modification by age by including an interaction term between unstable housing × age in models 2 and 3. We tested for effect modification by race and ethnicity by including interaction terms for unstable housing × race and ethnicity and for unstable housing × age × race and ethnicity in the overall survival analysis.
We stratified the cohort by age (<50, 50-64, 65-74, or 75-85 years) and compared survival between those with and without unstable housing using the Fine and Gray method.We adjusted for the factors previously listed (minus the unstable housing × age interaction term).
In all models, we treated the small numbers of discontinuation and loss to follow-up as censoring events equivalent to the end of the study period, using the date of last known dialysis as the censor date.However, we treated transplantation as a competing risk.Individuals who experience unstable housing are often not considered eligible for kidney transplantation.Thus, the Cox model estimates the risk of death if individuals with and without unstable housing underwent transplantation at equal rates, whereas the Fine and Gray competing risk model better estimates the risk of death on dialysis given the disparity in access to transplantation. 16To confirm whether we should treat transplantation as a competing risk, we estimated associations between unstable housing and transplantation, with death as a competing risk, and adjusted for age, sex, race and ethnicity, year of dialysis initiation, and predialysis nephrology care.
We conducted several sensitivity analyses as follows.We repeated the overall population analysis without the interaction term between unstable housing × age.We repeated the primary analysis and analyzed age as a categorical variable (<50, 50-64, 65-74, or 75-85 years) instead of a continuous variable.We repeated the primary analysis and included individuals aged older than 85 years.We limited the sample to participants who did not have discrepancies between USRDS and VHA data on date of death and repeated the primary analysis (n = 25 617).Finally, we limited the sample to participants who were screened for unstable housing within 1 year of dialysis initiation and repeated the primary analysis (n = 23 644).
Due to the availability of information from 2 data sets and because we excluded patients with missing medical evidence forms, we had only 15 individuals with missing data (in the fields of incident dialysis modality and rurality).

Baseline Characteristics of Individuals With Unstable Housing
This study included 25 689 veterans with a median age of 68 (IQR, 62-74) years.Men comprised 98% of the study population and women comprised 2%.Of the included patients, 32% were Black, 7% were Hispanic, 52% were White, and 10% were of other race or ethnicity.There were 771 veterans (3%) with a positive screen for unstable housing within a 3-year period before starting dialysis.Compared with veterans without indication of unstable housing, those with unstable housing were younger (mean [SD] age, 61 [8] vs 68 [10] years); they were also more likely to be women (5% vs 2%), to be Black (45% vs 32%) or Hispanic (9% vs 7%), and to live in urban areas (82% vs 69%) (Table 1).They were less likely to have predialysis nephrology care (16% vs 23%), and they were more likely to start dialysis with a central venous catheter (77% vs 66%), receive in-center hemodialysis (96% vs 91%), and have non-Medicare insurance (53% vs 28%).Covariates associated with higher odds of unstable housing included younger age, female vs male sex, Hispanic vs non-Hispanic ethnicity, lack of predialysis nephrology care, primary insurance (non-Medicare vs Medicare), chronic obstructive pulmonary disease, alcohol dependence, and drug dependence.
Having an arteriovenous fistula vs a central venous catheter and residing in a rural vs urban area were associated with lower odds of unstable housing.

Population-Based Survival Analysis
Among the 25 689 veterans with incident kidney failure receiving dialysis, 9435 (37%) died, 767 (3%) received kidney transplantation, and 15 487 (60%) were administratively censored (219 due to discontinuation or loss to follow-up, and the remainder at the end of follow-up; eTable 5 in Supplement 1).The unadjusted cumulative incidence of death was 68%.After multivariable adjustment, veterans with an indicator of unstable housing had higher risk of death compared with veterans with stable housing (adjusted hazard ratio [AHR], 1.20 [95% CI, 1.04 to 1.37] for a median age of 68 years; P = .03)(Figure 2 and Table 2).The risk of death associated with unstable housing increased with age (eTable 6 in Supplement 1 presents AHRs for unstable housing at various ages due to interaction).There was no interaction between unstable housing and race and ethnicity
International Classification of Diseases (ICD) Codes for Covariables eTable 5. Outcomes at Various Time Points for Individuals With and Without Unstable Housing eTable 6. Hazard of All-Cause Mortality Associated With Unstable Housing for Various Ages eTable 7. Sensitivity Analysis: Analysis Without Age × Unstable Housing Interaction Term eTable 8. Sensitivity Analysis: Including Individuals Aged Older Than 85 Years eTable 9. Sensitivity Analysis: Analysis With Age as a Categorical Variable eTable 10.Sensitivity Analysis: Sample Limited to Participants Who Did Not Have Discrepancies Between US Renal Data System and Veterans Health Administration Data on Date of Death eTable 11.Sensitivity Analysis: Sample Limited to Participants Who Were Screened for Unstable Housing Within 1 Year Before Dialysis Initiation