Thornley-Final

84
NZMJ 19 August 2016, Vol 129 No 1440
ISSN 1175-8716 © NZMA
www.nzma.org.nz/journal
Lack of housing, hospital
treatment and premature
mortality: a cohort study of
people in Counties Manukau
district
Simon Thornley, Roger Marshall
ABSTRACT
AIM: We considered risk factors for mortality in people admitted to Counties Manukau inpatient facilities,
who were also identified by medical staff to have insufficient housing.
METHOD: A cohortstudyof peopleaged15to75yearsadmittedtoCountiesManukauinpatientfacilitieswere
selected between 2002 and 2014, with ICD-10 codes for insufficient housing. Diagnostic records identified
people with substance use and other clinical conditions. Mortality records were used to track survival.
RESULTS: During the study period, 1,182 individuals were identified, 126 (10.7%) of whom died during a
medianfollow-upof5.7years.Mediansurvivalofthecohortwas63.5years(95%confidenceinterval(CI):58.7
to 69.9) which is about 20 years less than the general population. Of the cohort, the strongest associations
with premature mortality were among people with cannabis-related disorders (adjusted hazard ratio [aHR]
2.15; 95% CI: 1.10 to 4.22), diabetes (aHR 1.75; 95% CI: 1.05 to 2.93) and Maaori, compared to European and
other ethnic groups, except Asian and Pacific (aHR 1.80; 95% CI: 1.14 to 2.85).
CONCLUSION: This population has high mortality. Within this group, Māori and people diagnosed with
substance use and diabetes are at even higher risk of premature death.
A
ttention has recently focused on
the effect of housing on health.
Improving warmth and reducing
damp inside houses, reducing fall-related
hazards and overcrowding are important
interventions that are likely to influence
the health of residents.1
But the absence
of a home is itself an issue that is likely
to affect health.2
The definition of “home-
lessness” in published research has a wide
scope, from the most restrictive defini-
tion of ‘sleeping rough’ to more inclusive
versions that include being a temporary
resident in a permanent private dwelling.3
People living in camping grounds, shelters
and temporary commercial accommoda-
tion, such as boarding houses, are also
often included in the definition.3
With this
more inclusive definition, an estimate of
the prevalence of homelessness is less than
1% in New Zealand.3
Studies from the US2
and Sweden4
high-
light the high prevalence of alcohol-related
harm, drug addiction, severe mental illness
and increased mortality risk in homeless
populations. The extent to which lack of
housing is a cause, or an effect, of these
conditions is difficult to estimate, yet
published research clearly indicates that
homelessness and these conditions co-exist.
Provision of shelter is often managed by
the state (in the form of social services) or
by privately run charities. Since it is likely
that lack of housing adversely affects these
conditions, and that homelessness is likely
to complicate the recovery from medical
disorders, we conducted a study, using
healthcare data, to describe the population
who attended Counties Manukau inpatient
services and who were identified as having
inadequate housing. This study estimates
the burden and risk factors associated with
mortality among this group.
ARTICLE

85
ISSN 1175-8716 © NZMA
Methods
This study was carried out with a cohort
of people with inadequate housing, aged 15
to 75 years, who were admitted (a presen-
tation to hospital lasting three hours or
more) to Counties Manukau inpatient
facilities in South Auckland, between
January 2002 and December 2014 for any
condition, including psychiatric, emergency
department presentations and elective
surgery. “Inadequate housing” was iden-
tified by ICD-10 additional codes Z59x
(Table 1), which indicate a degree of insuf-
ficient housing. These codes encompass
a range of issues that include housing-re-
lated and economic hardship, which affect
their diagnosis. The National Minimum
Dataset was used to identify subjects. Text
searches of all diagnosis codes for the index
admission were used to identify people who
smoked tobacco or cannabis, had diabetes,
alcohol-related conditions, psychotic
disorders, depression, suicidality, obesity or
cancer. These variables were chosen since
other studies had identified these conditions
as common among homeless populations.
The service from which the patient was
discharged was also included: mental
health crisis (inpatient), adult medicine or
paediatrics; emergency medicine; intensive
care; surgical, gynaecological and dental;
maternity or neonates; other mental health
and geriatrics specialties. Discharge diag-
nosis data was then linked with national
mortality records using an encrypted
unique identifier derived from the National
Health Index (NHI).
The statistical analysis centred on esti-
mating factors associated with premature
mortality. Cox’s proportional hazard model
was used to estimate the magnitude of
the association between baseline factors
and survival. Age—rather than time-on-
study—was used as the time scale, as is
now recommended, so that observations
were considered left truncated and right
censored.5
The age of the subject at the date
of first hospital discharge records during
the study period was considered the entry
time point, and survival was estimated for
each individual, based on whether they
died during follow-up or reached the end of
follow-up (December 2014) alive. Gender,
ethnicity, tobacco smoking and diabetes
were considered important confounders
and were adjusted for. People with cancer
at baseline were excluded from the
analysis, since they had a very high rate of
premature death.
Ethical approval was not sought as all
data was de-identified using encrypted
national health identifier and only
aggregate measures are reported.
Results
During the period of the study, 1,182
individuals were identified, and 126 (10.6%)
died during a median follow-up of 5.7 years
(Table 2). The cohort had marginally more
men than women (607/1,182; 51.4%), and
Table 1: ICD-10 diagnoses used to indicate a hospital admission for which
homelessness has been a likely contributor.
Code Description
Z590 Homelessness
Z591 Inadequate housing
Z592 Discord with neighbours, lodgers and landlord
Z593 Problems related to living in residential institution
Z594 Lack of adequate food
Z596 Low income
Z597 Insufficient social insurance and welfare support
Z598 Other problems related to housing and economic circumstances
Z599 Problem related to housing and economic circumstances, unspecified
ARTICLE

86
ISSN 1175-8716 © NZMA
Table 2: People with insufficient housing, aged 15 to 75 years, admitted to Counties Manukau inpa-
tient services, 2002 to 2014, by mortality status at the end of 2014.
No death
(col. %)*
Death
(col. %)*
Total
(col. %)*
P-value
Total 1,056 126 1,182
Gender
Male 530 (50.2) 77 (61.1) 607 (51.4) 0.026
Age (years) <0.001
Median (IQR) 38.0 (27.9, 47.9) 52.6 (41.7, 63.8) 39.2 (28.8, 49.8)
Ethnic Group 0.110
Māori 346 (32.8) 46 (36.5) 392 (33.2)
Pacific 203 (19.2) 30 (23.8) 233 (19.7)
Chinese or ‘Other Asian’ 41 (3.9) 1 (0.8) 42 (3.6)
Indian 38 (3.6) 1 (0.8) 39 (3.3)
NZ European or other 428 (40.5) 48 (38.1) 476 (40.3)
Discharge service <0.001
Mental Health Crisis 387 (36.6) 26 (20.6) 413 (34.9)
Adult medicine or paediatrics 201 (19.0) 54 (42.9) 255 (21.6)
Emergency Medicine 166 (15.7) 13 (10.3) 179 (15.1)
Intensive care 97 (9.2) 16 (12.7) 113 (9.6)
Surgery, gynaecology or
dental
86 (8.1) 9 (7.1) 95 (8.0)
Maternity or neonates 58 (5.5) 0 (0) 58 (4.9)
Other mental health 44 (4.2) 1 (0.8) 45 (3.8)
Geriatrics 17 (1.6) 7 (5.6) 24 (2.0)
Clinical Diagnosis
Diabetes 80 (7.6) 27 (21.4) 107 (9.1) <0.001
Cancer 9 (0.9) 14 (11.1) 23 (1.9) <0.001
Obesity 47 (4.5) 13 (10.3) 60 (5.1) 0.009
Tobacco 487 (46.1) 66 (52.4) 553 (46.8) 0.216
Alcohol 144 (13.6) 20 (15.9) 164 (13.9) 0.582
Cannabinoid 93 (8.8) 12 (9.5) 105 (8.9) 0.919
Other drug use (mainly opioid
or methamphetamine)
142 (13.4) 6 (4.8) 148 (12.5) 0.008
Depression 165 (15.6) 16 (12.7) 181 (15.3) 0.465
Suicidality 18 (1.7) 1 (0.8) 19 (1.6) 0.694
Psychosis 287 (27.2) 26 (20.6) 313 (26.5) 0.142
* Unless otherwise stated. IQR: interquartile range.
ARTICLE

87
ISSN 1175-8716 © NZMA
mean age was 39.2 years (interquartile
range: 29 to 50). In the cohort, during their
first hospital stay, 46.8% (553/1,182) had
a code for cigarette smoking, and 9.7%
(107/1182) were diagnosed with diabetes,
13.9% (164/1182) had a code representing
problem alcohol use, and 8.9% (105/1182)
had a diagnosis related to cannabis use
(mainly associated with drug-induced
psychosis). Other drug use (mostly opioid
and methamphetamine use) was present
in 12.5% (148/1,182) of the population. The
most frequent clinical information obtained
during the baseline admissions were for
tobacco smoking and psychosis (26.5%;
313/1,182).
The largest ethnic group of the cohort
was ‘NZ European or Other’ ethnic groups
(40.2%; 476/1,182), with one third (33%;
392/1,182) Māori, and one fifth Pacific
(19.7%; 233/1,182). Numbers in the Asian
population were small. This contrasts to
the population profile of Counties Manukau
district where 15.7% identify as Māori,
21.5% Pacific and 62.8% as ‘NZ European or
Other’ in 2015.6
The proportion of Māori in
the cohort is about twice that expected from
the ethnic characteristics of people who live
in the district.
Crisis mental health and adult medicine
dominated the services from which this
population were initially discharged. During
the period of the study, the population were
most frequently discharged from the Mental
Health Crisis team (34.9%; 413/1,182).
Adult Medicine was the next most frequent
service, followed by Emergency Medicine.
The principal diagnoses of the index
admission encompassed a wide variety of
different diagnoses (390). Of all the prin-
cipal diagnoses, 46% (544/1,182) were for
mental and behavioural disorders (ICD-10
F00-F99). Psychotic disorders, such as
schizophrenia were most common. The
next most frequent category was “injury,
poisoning and certain other consequences
of external causes” (ICD-10 codes F00
to T98) making up 17% (202/1,182) of
primary diagnosis codes. Within the injury
and poisoning category, benzodiazepine
poisoning was most frequently recorded.
The next category of primary diagnoses
was “Symptoms, signs and abnormal
clinical and laboratory findings, not else-
where classified” (ICD-10 codes R00 to R99),
comprising 6% (75/1,182) of all admissions.
Within this category, unspecified chest pain
was most frequently coded.
Cause of death information was available
in 87 of 126 deaths. The most frequent
causes included cardiovascular disease
(26.4%; ICD-10 IXX), cancer (22.0%; ICD-10
Figure 1: Scaled rectangle diagram, highlighting the relative
size and overlap of different characteristics of the cohort.
The figure has a small approximation; there are some low
frequency combinations that are not represented.
ARTICLE

88
ISSN 1175-8716 © NZMA
CXX), and respiratory diseases (17.2%;
ICD-10 JXX).
Mean length of stay in these index admis-
sions was right skewed with a median of
4 days, and interquartile range of 15 days.
The longest admission was 1,016 days (2.8
years) for a patient with a primary diag-
nosis of schizophrenia.
Table 2 does not show overlap between
these conditions, which are not mutually
exclusive, and many individuals in this
population have tobacco, cannabis and
alcohol addictions which are each severe
enough to be recorded as diagnoses (Figure
1). The figure shows a scaled rectangle
diagram7
which depicts selected features
of the cohort. In this figure, areas are
intended to be proportional to frequency.
The area of each rectangle that represents
an attribute (eg, ‘Alcohol’) is exactly propor-
tional to frequency, and areas of overlap are
approximately proportional to frequency
of co-occurrence of attributes. The outer
rectangle has an area that is proportional to
the total cohort (n=1,182). For example, the
figure shows substantial overlap between
Table 3: Crude and adjusted associations between sociodemographic and clinical factors and death during
follow-up (hazard ratio (HR); 95% confidence interval) (n=1,159; 112 deaths).
Risk factor Denominator risk Numerator risk Crude HR (95%
CI)
Adj. HR (95% CI)*
Gender
Female Male 1.45 (0.47 to 2.13) 1.59 (1.06 to 2.38)
Ethnicity
NZ European or Other Māori 2.20 (1.43 to 3.40) 1.80 (1.14 to 2.85)
Pacific 1.64 (0.98 to 2.73) 1.08 (0.60 to 1.94)
Indian 0.56 (0.08 to 4.08) 0.42 (0.06 to 3.13)
Chinese or Other
Asian
0.35 (0.05 to 2.52) 0.44 (0.06 to 3.22)
Hospital service
Mental Health Crisis Emergency Medicine 1.96 (0.99 to 3.84) 1.97 (0.99 to 3.91)
Geriatrics 1.73 (0.66 to 4.55) 1.58 (0.59 to 4.24)
Intensive care 1.47 (0.77 to 2.80) 1.58 (0.82 to 3.05)
Medicine paediatrics
and maternity
2.58 (1.54 to 4.32) 2.00 (1.14 to 3.49)
Other mental health 0.40 (0.05 to 2.95) 0.26 (0.03 to 1.93)
Surgery gynaecology
or dental
2.01 (0.82 to 4.92) 1.84 (0.74 to 4.57)
Clinical condition
Cannabis use No Yes 2.77 (1.44 to 5.31) 2.15 (1.10 to 4.22)
Tobacco use No Yes 1.87 (1.28 to 2.72) 1.56 (1.04 to 2.33)
Alcohol use No Yes 1.39 (0.85 to 2.27) 1.37 (0.83 to 2.28)
Other drug use No Yes 0.54 (0.24 to 1.24) 0.50 (0.22 to 1.16)
Diabetes No Yes 1.95 (1.22 to 3.13) 1.75 (1.05 to 2.93)
Obesity No Yes 1.68 (0.92 to 3.08) 1.44 (0.77 to 2.70)
Depression No Yes 0.74 (0.43 to 1.28) 1.03 (0.58 to 1.84)
Psychosis No Yes 0.80 (0.51 to 1.27) 0.74 (0.46 to 1.18)
CI: confidence interval. *Adjusted for sex, tobacco use, diabetes and ethnic group.
ARTICLE

89
ISSN 1175-8716 © NZMA
the alcohol, cannabis and tobacco-related
conditions. It also reveals that alcohol and
cannabis-related disorders are relatively
rarely documented in hospital records after
the age of 50 years in this population.
The results of Cox modelling, without
adjustment, and after adjustment for
confounders are shown (Table 3).
The crude and adjusted associations
(Table 3) show an increase in risk of death
if subjects had diabetes or were discharged
from medicine, paediatrics or maternity
services (compared to discharge from
mental health services). Cannabis and tobac-
co-related diagnoses and complications, or
a diagnosis of diabetes were most strongly
associated with premature death, after
adjustment for confounders. Men with insuf-
ficient housing were 1.6 times (95% CI: 1.1
to 2.4) more likely to die during follow-up,
compared to corresponding women. Māori
who were inadequately housed were 1.8
times (95% CI: 1.1 to 1.9) more likely to die
during follow-up than NZ European and
Other. People with cannabis-related condi-
tions were twice as likely as those without to
die during follow-up (adjusted hazard ratio
[aHR] 2.15; 95% confidence interval: 1.10 to
Figure 3: Nomogram that illustrates the relationships between
selected risk factors and predicted survival at 50 years of age
in the study population (n = 1,159; 112 deaths).
Figure 2: Kaplan-Meier plots comparing the survival of people with diagnoses associ-
ated with substance use or diabetes (dashed line), with those who did not have such a
diagnosis (solid line).
ARTICLE

90
ISSN 1175-8716 © NZMA
4.22). People diagnosed with diabetes (aHR
1.75; 1.07 to 2.93), and tobacco smokers (aHR
1.56; 1.04 to 2.33) were also at increased risk
of death.
Kaplan-Meier survival plots (Figure 2)
show a steeper reduction in survival for
those with diabetes and substance abuse;
most notably for diabetes, with median
survival 39.4 years in diabetes and 65.3
without. Also, median survival was only 53.5
years in people with cannabis diagnoses,
and about 64.5 years in subjects without.
Median survival, here, is comparable to life
expectancy, calculated in official statistics.
The median survival contrasts with the
average life expectancy of the New Zealand
population being about 83 years in women
and 80 years in men, calculated from death
rates between 2012 and 2014.8
The life span
of this population is, therefore, about 15
years shorter than the New Zealand average,
and considerably lower than for the Māori
population (77 years females, compared to
73 years males).
Figure 2 also shows a steep mortality
increase in people with diabetes, in the late
30s to 40s, compared to those without the
diagnosis. Mortality in people with alco-
hol-related diagnoses deviates most steeply
from those without the diagnosis in the
50-to-60 year age range. A global test for
non-proportionality on an adjusted Cox
model (cox.zph) was not significant
(P=0.512).
After adjustment for confounding factors,
people with inadequate housing who iden-
tified as Māori were associated with a 68%
increase (95% CI: 7% to 166% increase) risk
of death, compared to ‘NZ European and
Other’ ethnic groups. No other convincing
mortality differences were observed by
ethnic group.
A regression nomogram,9
Figure 3, illus-
trates the relative contribution of different
factors to survival probability (taken at
the age of 50) in this cohort. The predicted
survival for an individual is calculated from
the nomogram by summing the individual
points per variable for each risk factor,
marking this point on the “total points”
line and marking a vertical line down to
calculate survival at the age of 50 years.
For example, a Māori (100 points) man (20
points) who does not smoke tobacco or
cannabis (0 points), but has diabetes (40
points) has a total of 100+20+40=160 points,
corresponding to a survival probability with
these characteristics to the age of 50 years
of 70%. This compares to a ‘NZ European or
Other’ man with the same characteristics
of the previous example who will have
about 120 points, so that survival until 50
years is about 10% higher than his Māori
counterpart.
People in the cohort who were discharged
from Adult Medical services (as well as
Paediatrics and Maternity) were twice as
likely to die during follow-up (aHR 2.00;
95% CI: 1.14 to 3.49), compared to those
discharged from the Crisis Mental Health
team, after adjustment for confounders.
Discharge from emergency and geriatric
services showed higher crude mortality
risks than discharge from the mental health
crisis team, but these differences were not
significant after adjustment.
When population attributable risks were
calculated for the potentially modifiable
exposures examined, tobacco accounted
for the greatest proportion of premature
death, followed by cannabis-related disease,
then diabetes, with alcohol-related disease
accounting for the lowest proportion of
disease (Table 4). The population attrib-
utable risk can be interpreted as the
percentage reduction in the death rate
which would occur if the exposure were
completely removed from the population,
assuming that exposures are causal.
Table 4: Population attributable risks for substance use and diabetes, among
people coded with insufficient housing during an admission to Counties Manukau
inpatient services, 2002 to 2014.
Risk factor Hazard ratio Prevalence
(%)
Population
attributable risk (%)
Tobacco smoking 1.56 46.8 20.8
Cannabis related disease 2.15 8.9 9.3
Diabetes 1.75 9.1 6.4
Alcohol-related disease 1.37 13.9 4.9
ARTICLE

91
ISSN 1175-8716 © NZMA
Discussion
This study shows that among the adult
population of people with insufficient
housing and who present to hospital,
about 10.6% (126/1182) will die within 6
years. Māori experience a higher mortality
compared to NZ European and other ethnic
groups, after adjustment for confounding
factors. Substance and tobacco use is
relatively common, and tobacco use was
associated with the greatest burden of
premature death. Diagnoses linked to
cannabis use were associated with a
two-fold increase in the risk of death.
This study may have some limitations.
The variable for selection of the popu-
lation—that is, the Z codes for insufficient
housing—may be subjective and unreliable.
The degree of homelessness indicated by
this diagnosis is uncertain, as discussed in
the introduction. The selection based on
hospital treatment and ICD code is likely
to bias the people in this study toward
more severe forms of homelessness.
Other limitations were that subjects were
assumed to be in the country and able to
be recorded with a death event during
follow-up. It is possible that some subjects
may have moved overseas and died there.
Clinical information is likely to have some
measurement error that may bias measures
of association. For example, many more
subjects in the cohort may use cannabis,
as only those with complications of use are
likely to be recorded in hospital diagnoses.
In the mortality analysis, the comparisons
made are within a small population, so that
some results—particularly those relating
to low risk of mortality among patients
with mental health conditions—should be
treated with caution. Studies conducted
in general populations, for example, show
increased risk of mortality in patients with
these conditions.10
The mortality among this
group is also likely to be higher than a more
general population-based sample of people
affected by homelessness, due to the criteria
for selection into the study. Further popu-
lation-based research is needed to more
accurately define the burden of mortality
among homeless people that do not present
to hospital.
Despite the uncertainty about the meaning
of the homelessness diagnoses in this study,
results from research overseas are rela-
tively concordant with ours, with regard
to prevalence of disorders and mortality.
In a meta-analysis of 29 eligible surveys,
from 7 countries, the most common mental
disorders reported were alcohol dependence
(prevalence: 8.1 to 58.5%), drug dependence
(prevalence: 2.8 to 42.3%), with psychiatric
disorders such as major depression and
psychosis also common.11
The most recent study (2013), describes a
cohort of homeless people in Boston, from
healthcare records taken from 1 January
2003 to 31 December 2008, which were then
linked to death records from the state of
Massachusetts. The mean age of the 28,033
people in the cohort was 41 years (standard
deviation 12.4), with 66% male. During the
average 3.2 years of follow-up, 1,302 deaths
occurred (4.6%). The major causes of death
included drug overdose (n=219), cancer
(n=206) and heart disease (n=203).2
The
overdose deaths involved opioids in 81% of
cases, with alcohol use mentioned in 32%.
The mean age at death was 51.2 years, with
81% (1,055/1,302) occurring in men. The
study authors described a high prevalence of
chronic pain and opioid use in the homeless
group and described obviating diversion
of opioids as a priority for improving the
health of this population. Our study found
a similar mortality rate (10.6% in 5.7 years
compared to 4.6% after mean follow-up of
3.2 years [US study]). Opiates are, however,
less available in New Zealand than in the
US, and no patients were coded with a death
due to overdose in the mortality information
that was available in our cohort.
In a Swedish study4
of 2,283 people regis-
tered with the Office for the Homeless in
Stockholm between 1995 and 1996, the
cohort was followed up until 2005 and
mortality rates then compared with the
general population. The study found that
homeless people had an excess mortality,
between 2.5 to 3.0 times that of the general
population. Homeless people—compared to
people who were not homeless, but had been
hospitalised for drug-related treatment—had
a similar risk of mortality (relative risk 1.13;
95% confidence interval: 0.93 to 1.38). With
the high rate of alcoholism in this study, the
authors concluded that alcohol was almost
entirely the cause of the excess mortality in
their study population. Our study shows that
ARTICLE

92
ISSN 1175-8716 © NZMA
a range of substance use disorders, as well
as diabetes, were associated with high risk of
mortality.
This report draws attention to the poor
health status of people affected by insuffi-
cient housing in Counties Manukau district.
This population comprises a disproportion-
ately high number of Māori (about 120%
higher than the level expected in the popu-
lation) and Pacific people (about 60% higher
than the level expected in the population),
raising issues of equity.
With high mortality associated with
substance use, and mental health issues rela-
tively common, these findings highlight some
possibilities for intervention. The co-occur-
rence of mental health and addiction issues
suggests that mental health services, particu-
larly crisis services, are an important referral
point for addiction treatment.
These findings also raise the issue of
whether addiction and mental health issues
are likely to be exacerbated by a lack of
secure housing. While the lack of housing
may be partly caused by mental health and
addiction issues, it is likely that lack of secure
and safe shelter will also hamper recovery
and treatment. There are a number of case-
studies from Canada12
and the US,13
reporting
that state investment in long-term housing of
the homeless results in overall cost savings,
because homeless people are frequent users
of expensive psychiatric, medical, corrections
and emergency housing services.14
This study highlights the excess mortality
and spectrum of disease among people
with insufficient housing. Māori are
disproportionately represented and have
higher adjusted mortality rates. Lack of
secure housing is likely to hamper efforts
of this population to recover from prev-
alent addiction and mental health related
conditions. Overseas studies indicate that
improved health, and cost savings result
from programmes to provide secure shelter
to populations such as these.
Competing interests:
Nil.
Acknowledgments:
The author thanks Dean Papa, analyst and statistician for help with data linking, and Dr
Doone Winnard, Dr Wing Cheuk Chan and Ms Jude Woolston for helpful comments on
drafts.
Author information:
Simon Thornley, Public Health Physician, Counties Manukau District Health Board, Manu-
kau, New Zealand; Roger Marshall, Associate-Professor of Biostatistics, Section of Epidemiol-
ogy and Biostatistics, University of Auckland, New Zealand.
Corresponding author:
Simon Thornley, Public Health Physician, Māori Health Development, Private Bag 94052,
South Auckland Mail Centre, Auckland, New Zealand.
sithor@gmail.com
URL:
https://www.nzma.org.nz/journal/read-the-journal/all-issues/2010-2019/2016/vol-129-
no-1440/6975
ARTICLE

93
ISSN 1175-8716 © NZMA
1. Baker MG, Howden-Chap-
man P. Time to invest in
better housing for New
Zealand children. N Z
Med J. 2012; 125:6–10.
2. Baggett TP, Hwang
SW, O’Connell JJ, et al.
Mortality among home-
less adults in Boston:
shifts in causes of death
over a 15-year period.
JAMA internal medicine.
2013; 173:189–95.
3. Amore K, Viggers H, Baker
M, Howden-Chapman P.
Severe Housing Depri-
vation: The problem and
its measurement. Official
Statistics Research Series,
2013. Wellington: Statistics
New Zealand, 2013.
4. Beijer U, Andreasson S,
Ågren G, Fugelstad A.
Mortality and causes of
death among homeless
women and men in
Stockholm. Scand J Public
Health. 2011; 39:121–7.
5. Thiébaut ACM, Bénichou
J. Choice of time-scale in
Cox’s model analysis of
epidemiologic cohort data:
a simulation study. Stat
Med. 2004; 23:3803–20.
6. Ministry of Health.
Population of Counties
Manukau DHB. Edition.:
Ministry of Health,
cited 31 March 2016].
Available from: http://
www.health.govt.nz/
new-zealand-health-sys-
tem/my-dhb/
counties-manukau-dhb/
population-coun-
ties-manukau-dhb
7. Marshall RJ. Scaled
rectangle diagrams can be
used to visualize clinical
and epidemiological
data. J Clin Epidemiol.
2005; 58:974–81.
8. Statistics New Zealand.
Life expectancy. Edition.:
Statistics New Zealand,
cited 15/12/2015 2015].
Available from: http://
www.stats.govt.nz/
browse_for_stats/health/
life_expectancy.aspx
9. Harrell F. Regression
modeling strategies:
with applications to
linear models, logistic
and ordinal regression,
and survival analysis:
Springer, 2015.
10. Healy D, Le Noury J,
Harris M, et al. Mortality
in schizophrenia and
related psychoses:
data from two cohorts,
1875–1924 and 1994–2010.
BMJ Open. 2012; 2.
11. Fazel S, Khosla V, Doll H,
Geddes J. The prevalence
of mental disorders
among the homeless
in western countries:
systematic review and
meta-regression analysis.
PLoS Med. 2008; 5:e225.
12. Medicine Hat Community
Housing Society. At home
in Medicine Hat: our plan
to end homelessness.
Medicine Hat, 2014.
13. National Alliance to End
Homelessness. The state of
homelessness in America
2015. Washington DC:
National Alliance to End
Homelessness, 2015.
14. Pomeroy S. The cost of
homelessness: analysis
of alternate responses
in four Canadian cities.:
National Secretariat on
Homelessness 2005.
REFERENCES:
ARTICLE

Thornley-Final

More Related Content

What's hot (20)

Similar to Thornley-Final (20)

More from Simon Thornley (20)

Thornley-Final