|
|
| |||||||||||||||||||||||||||||||||||||||
Background Homeless people are at high risk for death from many causes, but age-adjusted death rates for well-defined homeless populations have not been determined.
Methods We identified 6308 homeless persons 15 to 74 years of age who were served by one or both of two agencies for the homeless in Philadelphia between January 1, 1985, and December 31, 1988. Using a data base that contained all deaths in Philadelphia and listings of all Philadelphia residents during the same period, we compared the mortality rate for this homeless population with the rate in the general population of Philadelphia.
Results The age-adjusted mortality rate among the homeless was 3.5 times that of Philadelphia's general population (95 percent confidence interval, 2.8 to 4.5). The age-adjusted number of years of potential life lost before the age of 75 years was 3.6 times higher for the homeless people than for the general population (345 vs. 97 years lost per 1000 person-years of observation). Fifty-one of the 96 deaths of homeless persons (53 percent) occurred during the summer months. Mortality rates were higher among the homeless than in the general population for nonwhites, whites, women, and men. Within the homeless cohort, white men and substance abusers had higher mortality rates than other subgroups, but even homeless people not known to be substance abusers had a threefold higher risk of death than members of the general population. Injuries, heart disease, liver disease, poisoning, and ill-defined conditions accounted for 73 percent of all the deaths among the homeless.
Conclusions Homeless adults in Philadelphia have an age-adjusted mortality rate nearly four times that of Philadelphia's general population. White men and substance abusers are at particularly high risk. Matching cohorts of homeless people to death records is a useful way to monitor mortality rates over time, evaluate interventions, and identify subgroups with an increased risk of death.
Homeless people are especially likely to die from a variety of preventable causes. Homelessness is associated with substance abuse and psychiatric illnesses,1,2,8,9,10,11,12,13,14 which are in turn associated with multiple health risks. Many homeless people lack health insurance,15,16 obtain medical care sporadically,15,16,17 and are undertreated for common medical problems18,19,20. They are also at high risk for injuries7,8,9,21. Crowding in shelters may contribute to outbreaks of tuberculosis and bacterial pneumonia22,23,24,25,26,27. Risk factors for hepatitis B and infection with the human immunodeficiency virus may also be common among homeless people23,28,29,30. Many of these conditions can be prevented or improved by timely intervention.
Few reports describe causes of death among the homeless,3,31 and we are aware of no published data on mortality in a well-defined cohort of homeless people. We undertook to compare the mortality rate in a homeless cohort in a large city with that in the general population, since we hypothesized that the rate for the homeless would be higher. We adjusted for age, race, and sex, which are known to affect mortality rates within the general population.
Methods
Sources of Information
We identified homeless adults among recipients of services from the Philadelphia Office of Mental Health-Mental Retardation and the Philadelphia Office of Services for Homeless Adults. Teams of two state-certified mental health case managers from the Office of Mental Health-Mental Retardation regularly search the streets for people who appear to be homeless. Those found on the street between 6 p.m. and midnight who state that they have no place to stay and have no money to pay for lodging are considered homeless. The teams record information about each homeless person's age, sex, and race or ethnic group, provide an assessment of the problem, and make a referral for assistance. The computerized contact records of the mental health program date back to January 198532.
The Office of Services for Homeless Adults is responsible for referring homeless people to a network of shelters. Homeless people must apply for housing at a central office, from which they are transported to shelters with available beds. Applicants are defined as homeless if they state that they lack stable housing and adequate money to obtain housing. Of 4800 shelter beds in Philadelphia in January 1987, 3500 were part of the network of the Office of Services for Homeless Adults33. Each time a person applies for temporary housing, an intake form is completed at the central office. This includes information on age, sex, and race or ethnic group. Computerized records of intake forms date back to January 1987.
We used estimates of the population of Philadelphia in 1987 derived from census data34 as the general population in assessing mortality rates and as a reference for standardization of mortality rates according to age. The Pennsylvania Department of Health provided data on all deaths of Philadelphia residents and all deaths known to have occurred in Philadelphia between January 1, 1986, and December 31, 1988. The data in each case included the social security number, name, age, sex, race, and underlying cause of death35.
Identification of the Homeless Cohort
We combined data from the mental health program and the Office of Services for Homeless Adults, first consolidating all data from every contact record into a single file. This file was supplemented whenever possible by records of contacts with mental health care providers and by information from mental health case managers familiar with individual clients.
In order to generate a list of homeless persons that contained no duplications, we grouped records if they shared a common identifying factor. Such factors included the last name, mental health program identification number, birth date, social security number, and client-identifier code (the first three letters of last name, first initial, month and day of birth, and code for sex). In a computerized procedure, all records for which one factor matched were further tested by comparison of the other factors. Records that had matches on two identifying factors were then examined by study workers to confirm that they described the same person. With iterations of this process for each factor, records describing the same person were consolidated and withdrawn from further comparisons.
Matching of Homeless People with Data on Mortality
The process of matching homeless people with data on mortality was also iterative. As in the method described above, each step involved a computer match based on a primary matching factor, with confirmation by a worker's comparison of all available data and a final check based on a comparison of the date of the last contact with a service agency and the date of death. Each step involved a smaller pool of names and more manual checking than the previous step. Primary matching factors (used in the order listed) were the client-identifier code (described above), full name, modified client-identifier code (without the date of birth), social security number, birth date, selected last names (only those occurring fewer than seven times in the combined data bases were used as primary factors) plus sex, and selected first names (selected as above) plus sex. We also compared these files with a list of deaths of homeless people compiled from the Philadelphia Inquirer.
Records for which all variables matched were considered to relate to the same person and were consolidated. Records with partial matches were compared visually according to the same criteria to identify spurious nonmatches, as described above. The date and cause of death were added from the mortality file to the matching individual records.
Age-Adjusted Years of Potential Life Lost
The calculation of years of potential life lost allows the assessment of the extent to which death occurs earlier in one population than in another. To calculate potential life lost before the age of 75 years, we grouped the homeless cohort and Philadelphia's general population according to race or ethnic group and sex into 20-year age groups (15 through 34, 35 through 54, and 55 through 74 years). Years of potential life lost within each stratum were calculated, adjusted according to age, and summed by standard methods36.
Statistical Analysis
Our analyses were performed with the homeless data base stripped of identifiers. For each person in the cohort, follow-up was defined as the time between the date of the first contact with one of the social-service agencies and December 31, 1988, or the date of death, whichever came first. For comparisons between homeless persons and the general population and for calculations of years of potential life lost, only people 15 to 74 years of age were analyzed. For race-specific analyses, the small number of Hispanics was grouped with nonwhites because in terms of their economic circumstances they were thought to resemble nonwhites in Philadelphia more closely than whites. Mantel-Haenszel weighted rate ratios and confidence intervals37 and the two-tailed Fisher's exact test were performed with EpiInfo software (Version 5; Centers for Disease Control and Prevention, Atlanta).
Results
Characteristics of the Homeless Cohort
We identified 10,823 people served by the mental health program or the Office of Services for Homeless Adults between January 1, 1985, and December 31, 1988. The records of 108 people lacked dates of contact and could not be used for the estimation of person-years at risk. The remaining 10,715 people (with valid dates of contact) accounted for a total of 12,481 years of follow-up, for a mean follow-up period of 1.2 years per person.
As Figure 1 shows, 4962 (46 percent) of the members of the homeless cohort first came into contact with the agencies during December, January, February, or March. Among the 10,715 people in the cohort, 6224 (58 percent) had contact only with the mental health program during the study period, 3848 (36 percent) had contact only with the Office of Services for Homeless Adults, and 643 (6 percent) had contact with both agencies. Among the members of the homeless cohort, the age of 6387 (60 percent) was known; of these, 6308 (99 percent) were 15 through 74 years old (mean [±SD] age, 34 ±11). White men were twice as likely as all other cohort members to have unknown age (991 of 1392 vs. 3345 of 9322; rate ratio = 2.0; 95 percent confidence interval, 1.9 to 2.1). Sex was known for 10,173 (95 percent) of the cohort, of whom 6378 (63 percent) were men, and race was known for 8494 (79 percent), of whom 6312 (74 percent) were nonwhite or Hispanic.
|
Mortality
There were 96 deaths in the homeless cohort during the study period, for a crude mortality rate of 7.7 deaths per 1000 person-years of observation. Expressed in terms of deaths per 1000 person-years of observation, crude mortality was 8.9 among white men, 7.1 among nonwhite men, 6.7 among nonwhite women, and 5.4 among white women. Injuries were the leading cause of death (Table 1).
|
Homeless people of known age, accounting for 6580 person-years of follow-up, had a higher crude mortality than those of unknown age, accounting for 5902 person-years of follow-up (10.6 vs. 4.4 deaths per 1000 person-years; rate ratio = 2.4; 95 percent confidence interval, 1.5 to 3.8). Mortality and years of potential life lost were analyzed for 6308 homeless people known to be 15 to 74 years of age. The age-adjusted annual mortality rate was 3.5 times higher among the homeless people than in Philadelphia's general population (Table 2). Age-adjusted mortality among the homeless was higher than in the general population for each subgroup defined by race and sex; however, the highest difference in mortality rates was found among white men and the lowest among nonwhite men. Homeless white men of known age were more likely to die than all other members of the homeless cohort whose age, sex, and race were known (age-adjusted rate ratio = 1.9; 95 percent confidence interval, 1.0 to 3.6). Mortality among homeless nonwhite women was similar to that among homeless nonwhite men and homeless white women (age-weighted rate ratios = 1.1 and 0.9, respectively).
|
|
Discussion
Our findings support the intuitive hypothesis that homeless people die younger and have a higher mortality rate than the general population. Among the homeless people in our study, the crude annual mortality rate was in the range of 5.6 to 16.8 per 1000 obtained in other studies in which estimated denominators were used without adjustment for age31,38. The nearly fourfold increase in mortality that we observed is an unacceptable human cost of homelessness39 that merits intervention. Excess mortality in this cohort of homeless people is not explained solely by age, race, or sex distribution, nor by any single identifiable risk factor. Interventions aimed at preventing death must therefore be broad-based and must include efforts to address homelessness itself as well as the medical risk factors of homeless people.
The effects of infectious diseases among homeless people are emphasized in many reports,22,23,24,25,26,27,28,29,30 but no infectious disease ranked among the top eight causes of death in our cohort. Although limited by the accuracy of the death certificates we used as a source of data,40,41,42 our results suggest that excess mortality among homeless people results mostly from noninfectious causes. Any purely therapeutic intervention is unlikely to affect the most common causes of death among the homeless.
An earlier study17 found that at least half the deaths of homeless people in Atlanta were due to substance abuse. In line with previous studies,3,17,31 we found that injuries, poisoning, and liver disease -- each of which may be associated with substance abuse -- played a large part in causing deaths among the homeless. In our cohort, substance abuse increased the risk of death, but even homeless people not identified as substance abusers had a much higher mortality rate than the general population. Although substance abuse does not appear to explain most of the excess mortality in this cohort, we predict that successful interventions to treat or prevent substance abuse would have more impact on mortality among the homeless than interventions aimed at infection control.
Our estimate of the mortality rate among the homeless is conservative. Unrecognized data matches within the homeless cohort probably inflated our denominator, whereas unrecognized matches with mortality data would reduce our numerator; both errors would decrease estimated mortality. True matches may have been missed because of data-entry errors, use of aliases, migration out of the area, missing data, and the absence of matching factors other than common names, which we ignored. False matches may have been made when the identifying data were not specific enough to distinguish two people with similar names. Because computer matches were confirmed by visual checks of all available information, and because we limited our final analyses to records judged to reflect definite or highly probable matches, we believe that few false matches occurred. However, we lacked the data to quantify missed and false matches.
Our methods are biased toward detecting deaths among homeless people for whom more information was available for matching, including those with more frequent contacts with social-service agencies32. This limitation probably explains the higher crude mortality among those with known birth dates. Most people with missing birth dates were among those contacted by mental health workers, a fact that decreased the probability of finding matches in that group and removed many of them from the main analysis. Also, by eliminating matches based on common names, we increased the relative likelihood of identifying deaths among people with rare names. Since rare names are more likely among ethnic minorities, we introduced a bias in favor of identifying deaths in these population subgroups. This effect should be small, however, since matches by name were made at the end of the matching process and added few additional records to the list.
Within the homeless cohort we found an increased mortality rate among white men. One explanation for this is selection pressure in the general population from which the homeless are drawn. Urban nonwhite men in the United States have mortality rates comparable to those in developing countries, from causes similar to those common among the homeless43. White men in the United States generally have a much lower rate of mortality and are subjected to less stringent selection before becoming homeless. Consistently, white women had the lowest mortality in the general population of Philadelphia and the highest increase in mortality associated with homelessness.
A second possibility is that deaths of homeless white men are more likely to be recorded than deaths of other homeless people. Our methods did not detect deaths of homeless people outside Philadelphia or deaths of those who were not identified as Philadelphia residents, who had anonymous death certificates, or who used an alias. We know no reason why white men would be less likely than others to have one of these characteristics. White men were more likely than other homeless people to have information missing from their records and should therefore have been less likely to be matched with death records.
Third, nonwhites and women may be more likely to become homeless for purely economic reasons, whereas white men may be less likely to become homeless unless they have a predisposing illness. Substance abuse, one condition that we found to be more prevalent among white homeless men, is associated with mortality. If these results are confirmed by other studies, it may be useful to search among the nonwhite homeless population for factors that promote their survival.
Many outreach programs concentrate their efforts during the winter. Most deaths among the members of this homeless cohort, however, occurred in the summer, when contacts between social-service agencies and homeless clients were less frequent. In this cohort, most deaths occurred within six months of a contact with a social-service agency -- in one case, just 24 hours thereafter. Since many deaths in the cohort were from preventable causes, some contacts between the agencies and their clients may have represented missed opportunities for prevention, and other such contacts may in fact have been lifesaving. This possibility should be tested by continuing outreach programs throughout the year.
Opportunities for preventive care in this population are limited. Financial and social barriers prevent homeless people from seeking health care16,44. Poor working conditions and deficient resources hamper those providing care to homeless people45. Therapeutic intervention is inadequate to protect homeless people from injury. Homelessness itself, rather than identifiable medical conditions, appears to be the risk factor that most needs to be eliminated in order to reduce preventable mortality.
Further studies should address the use of screening criteria to identify subgroups of homeless people most likely to die within a year after a contact with a social-service agency. In this cohort, white men and substance abusers appeared to be at particularly high risk. Interventions to prevent death in cohorts such as this one should emphasize the prevention of injury, cardiac disease, and substance abuse. Given the limitations of medical screening in this population, longitudinal follow-up of mortality in defined cohorts of homeless people may be particularly useful for the evaluation of the effects of programs aimed at reducing the duration of exposure to homelessness and its attendant risks.
Source Information
From the Division of Field Epidemiology, Centers for Disease Control and Prevention, Atlanta (J.R.H.); the Bureau of Epidemiology, Pennsylvania Department of Health, Harrisburg (J.R.H.); the Philadelphia Office of Mental Health-Mental Retardation, Philadelphia (L.B., L.K, R.S.); the Philadelphia Office of Services for Homeless Adults, Philadelphia (I.M.); the Division of Disease Control, Philadelphia Department of Public Health, Philadelphia (A.K.M.); R.W. Johnson Pharmaceutical Research Institute, Spring House, Pa. (D.F.); and the Clinical Epidemiology Unit, Department of Medicine, University of Pennsylvania, Philadelphia (D.F.).
Address reprint requests to Dr. Hibbs at the Division of Infectious Diseases, Department of Internal Medicine, University of Minnesota Hospitals, Minneapolis, MN 55455.
References
| |||||||||||||||||||||||||||||||||||||||
Related Letters:
Health Care and the Homeless
Mehal W. Z., Blatt S. D., Meguid V., Hibbs J. R., Redlener I.
Extract |
Full Text
N Engl J Med 1995;
332:64-65, Jan 5, 1995.
Correspondence
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2008 Massachusetts Medical Society. All rights reserved. |
Previous
