Hwang SW. Mortality Among Men Using Homeless Shelters in Toronto, Ontario. JAMA. 2000;283(16):2152-2157. doi:10.1001/jama.283.16.2152
Author Affiliation: F. K. Morrow Inner City Health Research Unit, St Michael's Hospital, University of Toronto, Toronto, Ontario.
Caring for the Uninsured and Underinsured Section
Editors: William A. Roper, MD, MPH, University of North Carolina at
Chapel Hill; Carin M. Olson, MD, Contributing Editor, JAMA.
Context Homeless persons in US cities have high mortality rates. However, few
comparison data exist for death rates among homeless persons in other developed
Objectives To compare mortality rates among men using homeless shelters and the
general population in Toronto, Ontario, and to determine whether mortality
rates differ significantly among men using homeless shelters in Canadian and
Design Cohort study conducted from 1995 through 1997, with a mean follow-up
of 2.6 years.
Participants Men aged 18 years or older who used homeless shelters in Toronto in
Main Outcome Measure Mortality rate ratios comparing age-specific mortality rates among men
using homeless shelters in Toronto with those of men in the general population
of Toronto and of men using homeless shelters in New York, NY; Boston, Mass;
and Philadelphia, Pa.
Results Men using homeless shelters in Toronto were more likely to die than
men in the city's general population. Mortality rate ratios were 8.3 (95%
confidence interval [CI], 4.4-15.6) for men aged 18 to 24 years, 3.7 (95%
CI, 3.0-4.6) for men aged 25 to 44 years, and 2.3 (95% CI, 1.8-3.0) for men
aged 45 to 64 years. In most cases, however, the risk of death was significantly
lower for men using homeless shelters in Toronto than for those in US cities.
For men aged 25 to 44 years using homeless shelters, mortality rate ratios
were 0.52 (95% CI, 0.41-0.65) for Toronto compared with Boston and 0.61 (95%
CI, 0.44-0.85) for Toronto compared with New York City. For men aged 35 to
54 years using homeless shelters, the mortality rate ratio was 0.42 (95% CI,
0.27-0.66) for Toronto compared with Philadelphia.
Conclusions Mortality rates among men who use homeless shelters in Toronto, while
higher than in the general population of Toronto, are much lower than mortality
rates observed among men using homeless shelters in 3 major US cities. Further
study is needed to identify the reasons for this disparity.
Homeless persons suffer from a high prevalence of physical disease,
mental illness, and substance abuse.1- 11
Homelessness is associated with exposure to the elements12
and an increased risk of infections such as tuberculosis and human immunodeficiency
virus (HIV) disease.13- 18
Among the homeless, access to health care is often suboptimal.19- 23
Homeless persons also experience severe poverty and often come from disadvantaged
minority communities, factors that are independently associated with poor
The finding that mortality among homeless persons is much higher than among
their counterparts in the general population is therefore not surprising.
Data on deaths among homeless people are available from a number of
US cities. Early studies from Atlanta, Ga,30
and San Francisco, Calif,31 reported causes
of death but not mortality rates. Three more recent studies have reported
mortality rates in homeless populations. In Philadelphia, Pa, homeless adults
had an age-adjusted mortality rate 3.5 times higher than that of the general
population.32 In a study of clients of the
Health Care for the Homeless Program in Boston, Mass, mortality rates were
5.9, 3.0, and 1.6 times higher than in the general population among men aged
18 to 24, 25 to 44, and 45 to 64 years, respectively.33
The leading causes of death were homicide among young men, the acquired immunodeficiency
syndrome (AIDS) among 25- to 44-year-old men, and cancer and heart disease
among older men. A study of homeless shelter users in New York City found
age-adjusted death rates 2 to 3 times higher than the city's general population.34 Mortality among older men and women of all ages was
higher in New York City's homeless population than in the Boston or Philadelphia
How do homeless persons fare in the United States compared with those
in other developed Western countries? A cross-national comparison of mortality
rates would shed light on how societal factors affect the health of marginalized
segments of the population. However, little information is available on death
rates among homeless persons outside the United States. A search of the literature
reveals a single study conducted in Stockholm, Sweden, during 1969 through
1971.35 Among 6032 homeless men, 327 deaths
occurred, corresponding to a standardized mortality ratio of 3.8; absolute
mortality rates were not reported. High levels of excess mortality were observed
due to accidents, poisonings, and violence. Comparisons of these findings
with US data are difficult due to the lapse of more than 2 decades between
the Swedish and US studies.
This study examines death rates among homeless men in Toronto, Ontario.
We identified a cohort of 8933 men who used shelters in 1995 and ascertained
the number and causes of deaths in this group for 1995 through 1997. Mortality
rates for homeless men in Toronto are compared with rates previously reported
in Boston, New York City, and Philadelphia.
We compiled a database of men who used homeless shelters in Toronto
in 1995 from various sources. The Toronto Hostel Services Division operates
all public shelters in the city and contracts with all private nonprofit organizations
that run full-time shelters. Both publicly and privately operated shelters
provide the Toronto Hostel Services Division with a uniform dataset on every
person admitted, and shelters receive per diem funding based on these reports.
A few small church-run shelters that operate for only 1 night per week during
the winter do not contribute to this dataset. Data on shelter admissions are
compiled into a single master file that is audited by the Hostel Services
Division. Each record includes a unique identifier permanently assigned to
the shelter user (but not their full name), the person's sex, date of birth,
whether or not the person was accompanied by a spouse or children, the shelter
site, and the dates of shelter admission and discharge.
We selected all records from the master file for men aged 18 years or
older who stayed at homeless shelters in 1995. We excluded the small number
of men accompanied by a spouse and/or children who were admitted to shelters
for homeless families. The selected homeless men stayed at any of 10 homeless
shelters with a maximum winter capacity of 1300 beds per night. We matched
the unique identifiers from the master file to full names obtained from the
registration records at each homeless shelter. Women were not included because
registration records were not available at women's shelters at the time this
study was conducted. We compiled a database containing the names of 92% of
the selected men in the 1995 master file, and this group forms the study population.
Persons who were homeless in 1995 but who lived on the street without
ever using a shelter were not represented in the study population. Data are
not available on the size of the street homeless population in Toronto relative
to the shelter homeless population. However, a survey of homeless persons
at daytime meal programs at drop-in centers in Toronto showed that 93% of
these individuals had stayed at a homeless shelter within the last year (P.
Goering, PhD, written communication, January 7, 2000).
The Ontario Office of the Registrar General ascertained deaths in the
study population by comparing the shelter database with provincial death certificate
records for 1995 through 1997. Matches were identified if the records agreed
on (1) last name and exact date of birth or (2) first and last name and at
least 2 of the following: day of birth, month of birth, and year of birth
within 1 year. Names were compared using the Soundex algorithm to allow for
minor differences in spelling. Matches were deemed correct if the first and
last name and date of birth were exactly the same. Three reviewers examined
all other potential matches and reached a consensus as to whether the match
Person-years of observation in the cohort were calculated as follows.
We determined that a single individual could be represented by more than 1
identifier in the master file if he spelled his name differently or gave a
different date of birth at separate shelter admissions. Duplicate identifiers
related to a single individual were combined into a single identifier, using
methods developed for this purpose in a previous study of mortality among
homeless shelter users.32 We then calculated
person-years of observation for each individual, with the period of observation
defined as the time from first shelter admission in 1995 through December
31, 1997, or, in the case of decedents, their date of death. Total person-years
of observation in the homeless cohort were calculated by age groups, with
age determined at first shelter admission.
We calculated age- and cause-specific mortality rates (deaths per 100,000
person-years of observation). Causes of death were obtained from death certificates,
as coded by the Office of the Registrar General according to the International Classification of Diseases, Ninth Revision (ICD-9). We calculated rate ratios by dividing the mortality rate among
shelter users by the corresponding mortality rate for men in the general population
of Toronto in 1995. These values were not adjusted for race, because neither
the race of homeless men in the master file nor race-specific mortality rates
for the general population of Toronto were available. We determined the 95%
confidence intervals (CIs) for rates and rate ratios using standard techniques.36
We compared mortality rates among homeless men in Toronto with mortality
rates reported in studies of homeless men in Boston,33
New York City,34 and Philadelphia.32 The Boston study examined mortality among 11,745
men, almost all of whom were shelter residents, who had contact with a Health
Care for the Homeless Program during 1988 through 1993.33
The New York City study reported mortality rates in a representative sample
of 949 homeless male shelter residents in 1987 through 1994.34
The Philadelphia study, conducted from 1985 through 1988, involved 6378 men
who were homeless shelter users or street-dwelling homeless persons who had
contact with a team of outreach workers.32
Because blacks and Hispanics account for a larger proportion of the
homeless population in US cities than in Toronto, mortality rates were adjusted
for race through the following standardization. We estimated the racial composition
of Toronto's homeless population using data from the Mental Illness and Pathways
into Homelessness Study, a random sample of 300 shelter users in Toronto in
1995 through 1997, stratified by age, sex, and intensity of shelter use. Male
shelter users were 76% white and 24% nonwhite (specifically, 13% black, 6%
Native Indian, and 5% other races) (P. Goering, PhD, written communication,
January 5, 1999). Total mortality rates for homeless men in Boston, New York
City, and Philadelphia were adjusted for race by direct standardization, using
homeless men in Toronto as the standard population. Race-specific mortality
rates for homeless men in Boston (S. W. H., unpublished data, 1997), New York
City (S. Barrow, PhD, and D. Herman, DSW, written communication, July 29,
1999), and Philadelphia32 were used.
Because studies of deaths among homeless men in North America have been
conducted over somewhat different periods, secular trends may have contributed
to differences in observed mortality rates. Trends in deaths due to homicide
and AIDS are particularly notable. Homicide rates among young men in the United
States increased steadily during the mid-1980s and early 1990s, followed by
a sustained decline beginning in 1994.37,38
Similarly, mortality due to AIDS in both the United States and Canada peaked
in 1995, but has fallen dramatically over subsequent years due to highly active
antiretroviral therapy.39,40 To
assess the impact of these factors on differences in mortality between cities,
we performed 2 additional comparisons of mortality rates in Toronto and Boston,
the only other city for which cause-specific mortality rates were available.
In these analyses, we compared total mortality rates excluding deaths from
homicide and excluding deaths from AIDS.
The Research Ethics Board of St Michael's Hospital approved the study
protocol. Because this study analyzed routinely collected administrative data
and posed no significant risk to subjects, informed consent was not obtained.
The study population consisted of 8933 men who stayed at homeless shelters
in Toronto in 1995. Characteristics of the cohort are shown in Table 1. We identified 201 deaths in the cohort during 22,958 person-years
of observation, for a crude mortality rate of 876 per 100,000 person-years.
The mean age at death was 46 years (range, 20-84 years). Death occurred outside
a hospital in 41% of deaths, but the location was not otherwise specified.
Deaths were relatively evenly distributed across months of the year; in particular,
the number of deaths did not increase significantly during winter months.
The coroner's office performed an autopsy on 57% of the decedents.
Age- and cause-specific mortality rates and rate ratios comparing mortality
in the shelter population with that of the general population of Toronto are
shown in Table 2. For almost all
causes of death, rate ratios significantly exceeded 1. For total mortality,
rate ratios were 8.3 among 18- to 24-year-olds (95% CI, 4.4-15.6), 3.7 among
25- to 44-year-olds (95% CI, 3.0-4.6), and 2.3 among 45- to 64-year-olds (95%
CI, 1.8-3.0). Thus, younger homeless men are at greater relative disadvantage,
despite the fact that older homeless men have higher absolute mortality rates.
The leading identified causes of death among men 18 to 24 years old
using shelters in Toronto were accidents (other than poisonings), poisonings,
and suicides. Among men 25 to 44 years old, AIDS, accidents (other than poisonings),
poisonings, and suicide were the most common causes. The most frequently identified
mechanisms of traumatic accidental death were falls from stairways or buildings
and being struck by a motor vehicle, streetcar, or train. Poisoning deaths
included unintentional overdoses of opiates, other drugs, or alcohol. Among
men aged 45 to 64 years, the main identified causes of death shifted to cancer,
heart disease, and cerebrovascular disease, although accidental deaths (other
than poisonings) remained common.
These patterns of causes of death are similar to those reported among
homeless men in Boston, with 2 exceptions.33
Homicide was not among the 3 leading causes of death for men aged 18 to 24
years using shelters in Toronto, but was the most common cause of death in
this age group in Boston. Mortality rates due to homicide in these cities
were 38 and 243 per 100,000 person-years, respectively (rate ratio, 0.2; 95%
CI, 0.02-1.4). The proportion of homicides involving firearms was 0% in Toronto
and 13% in Boston. While AIDS was a leading cause of death among men aged
25 to 44 years in both cities, mortality rates due to AIDS were significantly
lower in Toronto than in Boston (115 vs 337 per 100,000 person-years, respectively;
rate ratio, 0.3; 95% CI, 0.2-0.6).
Total mortality rates among men who use homeless shelters in Toronto
are generally lower than race-adjusted mortality rates among homeless men
in Boston, New York City, and Philadelphia (Table 3). Figure 1 shows
the rate ratios and 95% CIs for these comparisons. A disparity is particularly
prominent in the middle age group, with mortality rates 40% to 60% lower in
Toronto compared with US cities. Crude mortality rates (not adjusted for race)
for these comparisons yield similar results (data not shown).
We examined the extent to which deaths due to AIDS and homicide accounted
for the disparity in mortality rates between Toronto and Boston. After excluding
all deaths from AIDS, rate ratios comparing mortality in Toronto with Boston
increased only slightly. Rate ratios rose to 0.78 (95% CI, 0.33-1.83) for
18- to 24-year-olds, 0.58 (95% CI, 0.45-0.75) for 25- to 44-year-olds, and
0.76 (95% CI, 0.57-1.00) for 45- to 64-year-olds. Thus, lower rates of HIV
infection and recent therapeutic advances for this condition can account for
only a small proportion of the overall mortality advantage for shelter users
in Toronto. When deaths from homicide were excluded, the mortality rate ratio
for men aged 18 to 24 years in Toronto compared with Boston rose from 0.75
(95% CI, 0.32-1.75) to 1.51 (95% CI, 0.50-4.53). Both CIs include 1, and the
rate ratios are not significantly different. For men aged 25 to 44 years,
the rate ratio remained unchanged at 0.52 (95% CI, 0.41-0.65), and for men
aged 45 to 64 years it increased minimally to 0.78 (95% CI, 0.59-1.03). Therefore,
the fact that homicide rates were higher in Boston than Toronto did not account
for the overall mortality advantage for homeless men in Toronto, particularly
for men older than 25 years.
Men who use homeless shelters in Toronto experience significant excess
mortality compared with the city's general population. This finding is consistent
with previous studies of homeless persons in major US cities.32- 34
Indeed, rate ratios comparing mortality among homeless men with that of the
general population are remarkably similar in Toronto (8.3, 3.7, and 2.3) and
Boston (5.9, 3.0, and 1.6) for young, middle, and older age groups, respectively.33
Because many deaths among men who used homeless shelters in Toronto
were attributed to unknown or unspecified causes, detailed analyses of cause-specific
death rates must be approached with caution. Despite this limitation, we note
2 causes for which mortality rates are far lower among homeless men in Toronto
compared with Boston: homicide among men aged 18 to 24 years and AIDS among
men aged 25 to 44 years. These differences persist even if all of the deaths
due to unknown or unspecified causes in Toronto are attributed to homicide
(in the case of 18- to 24-year-olds) or AIDS (in the case of 25- to 44-year-olds).
The most striking finding of this study is that men who use homeless
shelters in Toronto had total mortality rates much lower than the corresponding
rates reported among homeless men in Boston, New York City, and Philadelphia.
This disparity is consistent across cities and age groups and is most significant
in the middle age range, which includes about half of all homeless men in
the United States.41 Differences in mortality
rates could potentially be explained by variations in how each homeless cohort
was defined. Specifically, the Toronto cohort was limited to shelter users;
mortality in the US cohorts might appear higher if they included significant
numbers of street-dwelling homeless people, who presumably have higher mortality
rates than homeless people who use shelters. This may have been a factor in
the Philadelphia study, which included a number of individuals living on the
street. The New York City cohort, however, consisted of a sample of homeless
shelter residents. The Boston cohort was made up of men who had contact with
a Health Care for the Homeless Program, almost all of whom were residing at
homeless shelters. While a group of health service users might have higher
morbidity and mortality than the general shelter population,42
contact with the health care system could also result in lower mortality rates.
In fact, death rates in the Boston cohort were comparable to or lower than
death rates in the general shelter population in New York City. Of note, the
only instance in which death rates were higher in Toronto was for the comparison
against Philadelphia for men aged 55 to 74 years. This finding may be related
to the fact that the mortality rate among older homeless men in Philadelphia
was based on a very small number of observations and may represent an underestimate
of the true rate.
Why are death rates among men who use homeless shelters lower in Toronto
than in US cities? Men entering the shelter system in Toronto may have fewer
baseline comorbidities. Previous studies have shown that risk factors for
death among homeless people include medical conditions such as renal disease,
liver disease, arrhythmias, and seizures.43
Other risk factors include injection drug use and a history of incarceration.34 Shelter users in Toronto may have a lower prevalence
of these risk factors, particularly injection drug use.44
The events that transpire after an individual becomes homeless can also
have important effects on mortality. Preliminary analyses of patterns of shelter
use in our cohort reveal that men who use shelters in Toronto are less likely
to have prolonged episodes of homelessness than persons using shelters in
New York City and Philadelphia.45 Because chronic
homelessness itself may increase the risk of death,34
this difference may contribute to lower mortality among shelter users in Toronto.
Another factor is the dramatically lower incidence of homicide in Canada46 compared with the United States,37
an advantage that extends to homeless persons as well as the general population.
Canada's system of universal health insurance may have contributed to
lower death rates among shelter users in Toronto. Whereas most homeless persons
in the United States lack health insurance,22
homeless persons in Canada do not encounter any financial barriers to obtaining
health care services. Studies have shown that low-income persons in Canada
receive more physician services than their US counterparts.47
Survival rates among low-income patients with cancer are higher in Canada
than the United States, a disparity that has been attributed to better access
to health care in Canada.48 With respect to
mental health services, residents of Ontario with a perceived need for help
are more likely to receive care than comparable individuals in the United
States.49 In view of the consistency of these
findings, a beneficial effect of the Canadian health care system on mortality
among homeless men would not be surprising.
This study has certain limitations. Homeless women were not studied
due to restrictions in data sources. We used death certificates to obtain
causes of death. The accuracy of death certificate data has been questioned,
as significant discrepancies have been found between death certificates and
autopsy reports.50 Other studies have shown
that death certificates reliably document deaths due to coronary heart disease,51 AIDS,52 and injuries.53 Because many deaths in this cohort were attributed
to unknown or unspecified causes, cause-specific death rates should be interpreted
with caution; however, this limitation does not affect analyses of total mortality
rates. Death rates among homeless people in various cities were compared using
data obtained over different time periods. However, the major potential confounding
factors in such an analysis have been discussed and are unlikely to be responsible
for the consistent mortality advantage for the shelter population in Toronto.
In conclusion, this study demonstrates that mortality rates among men
who use homeless shelters in Toronto, while higher than in the general population
of Toronto, are dramatically lower than mortality rates previously observed
among homeless men in Boston, New York City, or Philadelphia. Further investigations
are needed to identify the reasons for this disparity. Possible contributory
factors include the effects of universal health insurance and access to health
care in Canada, lower homicide rates, particularly among young men, and the
differential health effects of short-term vs chronic homelessness. Insights
into the reasons behind lower mortality among homeless men in Toronto could
help direct efforts to reduce deaths among homeless persons and other disadvantaged
populations in the United States.