Homeless persons with AIDS had significantly worse survival than housed persons, and the provision of housing after AIDS improved survival. These findings have important policy implications given that supportive housing is a feasible and affordable intervention.
Several factors associated with homelessness are likely to have contributed to the poorer survival in this group, including drug use, serious mental illness, inadequate use of health care, and poor adherence to medications. Consistent with prior studies, antiretroviral use was lower among homeless compared to housed AIDS cases [17
]. Although we adjusted our model for initiation of antiretroviral use, we did not have data on compliance with regimens, length of time patients were treated, adverse reactions, or development of resistance. Several studies among HIV-infected homeless persons have documented high rates discontinuation of antiretroviral medications and treatment interruption [17
]. In addition, HIV-infected homeless persons have fewer ambulatory care visits than infected persons who are housed [20
]. These unmeasured factors could bias away from the null our estimate of the adverse effect of homelessness.
Injection drug use was more common among homeless than housed persons and was an independent predictor of mortality in our study as well as in others [32
]. Injection drug use has been associated with premature mortality, [34
] due in large part to overdose, suicide, homicide, accidents, and liver disease [34
]. A large proportion of the non-HIV-related deaths among homeless AIDS cases was due to these drug-related causes.
Persons with heterosexually acquired HIV were at increased risk of death. This finding is consistent with some [33
] but not all [32
] studies of AIDS survival. Previous studies of persons with HIV/AIDS in San Francisco and nationally have documented that compared with MSM, heterosexuals are more likely to be diagnosed with HIV within 12 months of AIDS suggesting that they are diagnosed with HIV later in the course of their disease, a factor associated with worse survival [38
Persons lacking private health insurance had an increased risk of death. This finding is supported by other studies of HIV-infected persons [24
] and may reflect differential access to care and/or treatment [42
]. It is likely that persons with private insurance received medical care by private physicians who may be more available to manage adverse reactions to medications and to facilitate uninterrupted use of antiretroviral therapy than providers in public settings.
Prior studies have shown that housing the homeless, including those with mental illness and substance abuse problems, is feasible [46
] and can be accomplished for a modest cost [49
]. Our study advances the field by demonstrating that supportive housing can decrease mortality. This finding may be due to housed persons being better able to keep medical appointments, store their medications, and adhere to their regimen because housing provides a structure that contributes to a steady routine. The medical and case management services that were provided with the housing may have also contributed to the gains in survival. Indeed, case management has been shown to reduce unmet need and improve use of antiretroviral medications among HIV-infected persons [50
] and reduce the number of hospital days and emergency department visits among homeless persons with chronic medical illness [48
]. Housing may also result in decreased substance use among HIV-infected persons [52
This study has several limitations. We defined an individual as homeless at diagnosis based upon information in the medical record. It is possible that persons with unstable housing gave an address at the time they were diagnosed that reflected a temporary housing situation, such as staying with a friend or relative, and would have been misclassified as housed. Also, an individual who was stably housed at diagnosis could have become homeless at a later date. Similarly, someone who was homeless at diagnosis could have subsequently received housing. All of these scenarios would have resulted in minimizing the differences in survival between the housed and homeless groups.
Although few AIDS cases are unreported, we investigated the possibility that differences between reported and unreported cases could have biased our results. The AIDS surveillance system is evaluated annually for completeness of reporting. This is done at selected sites where ICD-10 codes that correspond to HIV, AIDS, and HIV-related conditions are used to identify persons who may have HIV. The list is matched against the HIV/AIDS registry to identify persons who may have been missed. The medical records of persons who did not match with the registry are reviewed to identify persons with HIV/AIDS. Any missed cases are reported and the reporting source is listed as coming from the evaluation. We examined characteristics of cases that were initially not reported but later found during annual evaluations of the surveillance system for the years 1996 through 2002. Missed cases were significantly (p < 0.05) more likely to be housed, white, alive at diagnosis, and to have private health insurance (data not shown). Thus, in order for unreported cases to have biased our results, these cases would have to have worse survival than the reported housed cases, a scenario that we believe is unlikely. Missed cases may impact the generalizability of our findings but because AIDS case reporting is so complete, such an effect would be small.
In our analysis we used all cause mortality. We know that homeless persons are more likely to die from non-AIDS-related causes than are housed persons. To assess the impact of using all cause mortality, we excluded persons whose primary cause of death was not HIV/AIDS and conducted another Cox proportional hazards model with this restricted dataset. Our findings were essentially unchanged (RH for homeless 1.19; 95% CL 1.00, 1.42).
As with any nonrandomized comparison of an intervention, the most significant limitation to our analysis is the possibility that persons who were housed were not equivalent to those who remained homeless in unmeasured, or measured but inadequately modeled, characteristics. In particular, if those who received housing were more likely than those who were not housed to survive even if they did not receive housing, the validity of the comparison would be undermined. Several factors argue against this outcome applying to this study. First, the supportive housing program seeks to house people with the greatest medical and/or psychosocial need. Second, the comparisons between housed and not housed persons showed the groups to be remarkably similar. The strongest difference was that the housed group was more than twice as likely to be aged 50 years or more, a strong predictor of worse survival, and corroborating evidence that the program houses persons in greatest need. Persons who were homeless at diagnosis may have received housing subsequently from sources other than the DAH program, which would have caused underestimation of the effect providing supportive housing has on mortality. The results of the sensitivity analysis indicate that it is unlikely that unmeasured confounding could account for our findings.
Although this study used a population-based sample of persons with AIDS, the findings may not be representative of persons with AIDS outside of this geographic region. In San Francisco, HIV is overwhelmingly a disease of MSM. Although this risk group still accounts for the majority of cases nationwide, heterosexual injection drug users and heterosexual partners of injection drug users account for a larger proportion of AIDS cases elsewhere than occurs in San Francisco.
The annual costs per person housed are comparable to those reported from a multisite study of supportive housing for HIV-infected persons [53
] and the cost of supportive housing per year of life saved ($46,800) was within the range of approved medical interventions [54
]. Because of our small sample size, the upper 95% CL exceeded the usual cut-off for cost-effective interventions. Our cost-effectiveness analysis, however, was very conservative in that we considered only the direct cost of the intervention and did not include any of the expected medical care reductions due to providing supportive housing [46
]. A large study of supportive housing for the mentally ill in New York City found that 95% of the costs of supportive housing are recouped through savings on hospitalizations and shelter stays [56
]. Assuming that only 70% of these costs were recouped, the cost per year of life saved becomes highly cost-effective in our population ($14,040). The cost-effectiveness of housing also would be enhanced if it were quality adjusted, given the gain in quality of life due to being housed versus living on the street.