A spatial analysis of the determinants of pneumonia and influenza hospitalizations in Ontario (1992-2001)

Article (PDF Available)inSocial Science & Medicine 64(8):1636-50 · May 2007with147 Reads
DOI: 10.1016/j.socscimed.2006.12.001 · Source: PubMed
Abstract
Previous research on the determinants of pneumonia and influenza has focused primarily on the role of individual level biological and behavioural risk factors resulting in partial explanations and largely curative approaches to reducing the disease burden. This study examines the geographic patterns of pneumonia and influenza hospitalizations and the role that broad ecologic-level factors may have in determining them. We conducted a county level, retrospective, ecologic study of pneumonia and influenza hospitalizations in the province of Ontario, Canada, between 1992 and 2001 (N=241,803), controlling for spatial dependence in the data. Non-spatial and spatial regression models were estimated using a range of environmental, social, economic, behavioural, and health care predictors. Results revealed low education to be positively associated with hospitalization rates over all age groups and both genders. The Aboriginal population variable was also positively associated in most models except for the 65+-year age group. Behavioural factors (daily smoking and heavy drinking), environmental factors (passive smoking, poor housing, temperature), and health care factors (influenza vaccination) were all significantly associated in different age and gender-specific models. The use of spatial error regression models allowed for unbiased estimation of regression parameters and their significance levels. These findings demonstrate the importance of broad age and gender-specific population-level factors in determining pneumonia and influenza hospitalizations, and illustrate the need for place and population-specific policies that take these factors into consideration.
1 Figures
Social Science & Medicine 64 (2007) 16361650
A spatial analysis of the determinants of pneumonia and
influenza hospitalizations in Ontario (1992–2001)
Eric J. Crighton
a,
, Susan J. Elliott
b
, Rahim Moineddin
c,d
,
Pavlos Kanaroglou
b
, Ross Upshur
c,d,e
a
Department of Geography, University of Ottawa, 60 University Avenue, Simard Hall Room 06, Ottawa, Ont., Canada K1N 6N5
b
School of Geography and Geology, McMaster University, 1280 Main St. West., Hamilton, ON, Canada L8S 4L7
c
Department of Family and Community Medicine, University of Toronto, 256 McCaul Street, 2nd Floor, Toronto, ON, Canada M5T 2W5
d
Department of Public Health Sciences, University of Toronto, McMurrich Building, 12 Queen’s Park Crescent W.,
Toronto, ON Canada, M5S 1A8
e
Primary Care Research Unit, Sunnybrook and Women’s College Health Sciences Centre, 2075 Bayview Ave,
#E-349, Toronto, ON, Canada M4N 3M5
Available online 23 January 2007
Abstract
Previous research on the determinants of pneumonia and influenza has focused primarily on the role of individual level
biological and behavioural risk factors resulting in partial explanations and largely curative approaches to reducing the
disease burden. This study examines the geographic patterns of pneumonia and influenza hospitalizations and the role that
broad ecologic-level factors may have in determining them. We conducted a county level, retrospective, ecologic study of
pneumonia and influenza hospitalizations in the province of Ontario, Canada, between 1992 and 2001 (N ¼ 241,803),
controlling for spatial dependence in the data. Non-spatial and spatial regression models were estimated using a range of
environmental, social, economic, behavioural, and health care predictors. Results revealed low education to be positively
associated with hospitalization rates over all age groups and both genders. The Aboriginal population variable was also
positively associated in most models except for the 65+-year age group. Behavioural factors (daily smoking and heavy
drinking), environmental factors (passive smoking, poor housing, temperature), and health care factors (influenza
vaccination) were all significantly associated in different age and gender-specific models. The use of spatial error regression
models allowed for unbiased estimation of regression parameters and their significance levels. These findings demonstrate
the importance of broad age and gender-specific population-level factors in determining pneumonia and influenza
hospitalizations, and illustrate the need for place and population-specific policies that take these factors into consideration.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Canada; Pneumonia; Influenza; Hospitalizations; Determinants; Spatial analysis
Introduction
Pneumonia and influenza are the leading causes of
death from infectious disease and among the leading
causes of death overall in Canada (Health Canada,
2001). Accounting for over 60,000 hospitalizations
or approximately a third of all respiratory disease
ARTICLE IN PRESS
www.elsevier.com/locate/socscimed
0277-9536/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.socscimed.2006.12.001
Corresponding author. Tel.: +613 562 5800x1065;
fax: +613 562 5145.
E-mail addresses: eric.crighton@uottawa.ca (E.J. Crighton),
elliotts@mcmaster.ca (S.J. Elliott),
rahim.moineddin@utoronto.ca (R. Moineddin),
pavlos@mcmaster.ca (P. Kanaroglou),
ross.upshur@sunnybrook.ca (R. Upshur).
hospita lizat ion s (Health Canada, 2001), pneumonia
and influenza represent a significant public health and
health care system burden. While it is well established
that health and illness are influenced by bro ad
environmental, social, and economic factors (e.g.
Evans & Stoddart, 1990; Labonte, 1987), research
trying to explain variability in pneumonia and
influenza has typically focused on individual-level
biological and behavioural risk fact ors, resulting in
partial explanations and largely curative approaches
to reducing the disease burden (Loeb, 2003). In an
effort to address this shortcoming, we have conducted
an ecologic-level study of pneumonia and influenza
hospitalizations in Ontario using spatial and non-
spatial analytic techniques to examine patterns of
these illnesses, and the factors that determine them.
Furthering our knowledge in this area is critical for
the development of more population- an d place-
specific preventative strategies that take social, eco-
nomic, health care, and environmental factors into
consideration. This study adds to our previous
descriptive temporal (Crigh ton, Moineddin, Up shur,
& Mamdani, 2004)andspatial(Crighton, Elliott,
Moineddin, Kanaroglou, & Upshur, 2006)analysesof
pneumonia and influenza hospitalizations by showing
how hospitalization rates are associated with popula-
tion-level indicators measuring socioeconomic, beha-
vioural, health care, and environmental factors.
Background
There is a significant body of literature examining
various pneumonia and influenza outcomes, includ-
ing mortality (Fine et al., 1997), hospitalizations
(Morris & Munasinghe, 1994) and acquisition of
illness (Baik et al., 2000). Risk factors frequently
identified with pneumonia and influenza morbidity
and mortality include age, gender, chronic diseases,
recent viral infections and antibiotic use (e.g. Kaplan
et al., 2003). People at the extremes of age, in
particular the elderly, have been found to be among
those at the greatest risk of acquiring the illnesses
and experiencing the worst outcomes (Loeb, 2003).
Chronic illnesses associated with pneumonia include
asthma (Lange, Vestbo, & Nyboe, 1995), cardiovas-
cular disease (Kaplan et al., 2003) and HIV (Plouffe,
Breiman, & Facklam, 1996). Identified lifestyle risk
factors include smoking (Almirall et al., 2000),
obesity (Baik et al., 2000) and excessive drinking
(Nuorti et al., 2000).
While the role of these physiological and beha-
vioural factors in determining pneumonia and
influenza is significa nt, considerable variance re-
mains unexplained (Loeb, 2003). The above studies
take a largely biomedical approach to understand-
ing disease which is relevant primarily for clinical
decision making, but do not address the influence of
other determinants including social, economic,
cultural and physical environments (Evans &
Stoddart, 1990). There is, however, a well estab-
lished and growing body of evidence pointing to the
importance of these factors for numerous, typically
chronic, health conditions including, he art disease,
respiratory diseases, diabetes as well as all cause
mortality (e.g. Kawachi, Kennedy, Lochner, &
Prothrow-Stith, 1997; Ross et al., 2000; Wilkinson,
1996). Recent studies suggest that such factors are
similarly important in determining infectious dis-
eases such as pneumonia and influenza (Farr,
Bartlett, Wadsworth, & Miller, 2000; Wood, Sallar,
Schechter, & Hogg, 1999), resulting in calls for
further research in this area (Loeb, 2003
).
Several studies using individual level comparisons
have examined relationships between pneumonia
(various endpoints) and a range of socioeconomic
status (SES) measur es, including social class,
income and education (Farr et al., 2000; Stelianides,
Golmard, Carb on, & Fantin, 1999), although not
always coming to the same conclusions. A UK
study revealed that unemployed individuals were at
a significantly greater risk of being hospit alized with
pneumonia (Farr et al., 2000). Similarly, Wood et
al. (1999) found a strong negati ve association
between socioeconomic classes and avoidable mor-
tality from pneumonia. On the other hand, Stelia-
nides et al. (1999) found that low SES, defined as
long-term unemployment, homelessness or poor
living conditions, was not a significant risk factor
for pneumonia morbidity; and Vrbova, Mamdani,
Moineddin, Jaakimainen, & Upshur (2005) reported
no significant association between income and
mortality among the elderly.
Cohen (1999) suggests that there are two general
mechanisms as to how SES might predispose to
pneumonia. The first is through increased exposure
to infectious agents from, for example, crowding or
institutional living. The second is through increased
susceptibility from a weakened immune system
related to factors including higher levels of stress,
poor or no housing, poor diet, or other life-style
factors associ ated with deprivation. Wilkinson
(1996), however, argues that in most Western
nations, absolute deprivation is somewhat rare,
and that it is in fact relative deprivation that is more
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1637
important because of its impact on social capital.
Typically measured in terms of social networks,
support, and cohesion, social capital has been found
to decrease as income inequality increases (Kawachi
et al., 1997). Social capital is then linked to health
through a range of mediary factors including
physiological stress responses, self-esteem, and
health behaviours (see Veens tra et al., 2005). While
numerous studies have linked social capital to
various health outcomes, it has not been studied
in the context of infectious diseases such as
pneumonia to our knowledge.
Health care system factors have also been
identified as being associated with pneumonia
and influenza. Although the ability of health care
to improve population health is questioned in
the public and population health literature (e.g.
Marmor, Barer, & Evans, 1994; McKeown, Record,
& Turner, 1975), there is evidence that access to
basic health services may reduce the burden of
diseases like pneumonia, and the risk of complica-
tions leading to hospitalization (Macinko, Starfield,
& Shi, 2003; Morris & Munasinghe, 1994). For
example, in a study examining the contributions of
Primary Care systems in OECD countries, Macinko
et al. (2003) found that the strength of the system,
measured using a variety of indicators including
service access, system financing, and comprehen-
siveness of care, was significantly negatively asso-
ciated with premature mortality from, among other
diseases, pneumonia and influenza (po0.01).
A further independent risk factor identified in
the literature is race. Being African American, for
example, has been found to be independently
associated with higher incidence of pneumococcal
infection in the USA (Harrison, Dwyer, Billman,
Kolczak, & Schuchat, 2000). In Alberta, hospitaliza-
tion rates for pneumonia among Aboriginal popula-
tions were found to be 5 times higher than among
non-Aboriginal populations (Marrie, Carriere, Jin,
& Johnson, 2004). While genetic endowment may be
a factor in these cases, it is expected that underlying
socioeconomic factors such as inadequate housing,
public sanitation and high unemployment, all of
which are chronic problems in many Canadian
Aboriginal populations, are at the root of this
inequity (Stavenhagen, 2005).
Missing from much of the work done on the
determinants of pneumonia and influenza is an
understanding of the geographical nature of these
illnesses and their determinants. A knowledge of the
geographical variation in infectious disease has, in
the past, significantly increased our understanding
of disease distribution and diffusion, as well as the
risk factors for infection or developing the disease
(e.g. Cliff, Haggett, & Ord, 1986; Cliff & Smallman-
Raynor, 1992; Morris & Munasinghe, 1994). An
historical study by Cliff et al. (1986), for example,
examined the introduction, spread and disappear-
ance of multiple epidemics of influenza in Iceland,
finding associations with various structural features
including road connections and population size.
Cliff and Smallman-Raynor (1992), using an ecolo-
gic study design, examined the spatial distribution
of AIDS in Uganda, revealing positive associations
with proximity to trucking routes, and army
recruitment rates.
There remains a paucity of studies examining the
geographical variation in pneumonia and influenza
and their determinants. A notable exception is an
ecologic-level study conducted by Morris and
Munasinghe (1994) looking at the geographic
variability and determinants of acute respiratory
illness hospitalizations (including pneu monia and
influenza) among the elderly in the US Findings
revealed marked regional elevations in rates that
were associated with socioeconomic and health care
system factors including education, household
crowding, income and physicians per capita. How-
ever, no control was provided in this analysis for
spatial dependence in model resi duals, thereby
potentially biasing the estimation of regression
parameters and their significance levels. More
recently, our group conducted a county-level spatial
analysis of pneumonia influenza hospitalizations in
Ontario (Crighton et al., 2006), revealing significant
geographic variability in hospitalization rates.
A moderate yet significant level of positive spatial
autocorrelation (Moran’s I ¼ 0.21; po0.05) was
found in the global data, with significant age-
specific ‘hot spots’ in several northern counties.
Determinants of these patterns were not examined.
These studies point to the importance of place in
health.
Conceptual framework
This study builds on past research by adopting a
population health perspective, which highlights the
importance of broad determinants of healt h and
disease. Evans and Stoddart’s (1990) population
health framework represents an important starting
point for this work. The framework emphasises the
relationship between health and genetic endowment,
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501638
the socioeconomic environment and the biophysical
environment. These broad categories are thought, in
turn, to condition an individual’s behavioural and
biological responses to external stimuli. While a
number of critiques of this framework have been
published (e.g. Poland, Coburn, Robertson, &
Eakin, 1998), a further significant criticism stems
from the fact that the framework does not explicitly
recognize the importance of the tempo ral and
spatial nature of health and illness.
According to Loytonen (1998), time and space
form an inseparable combination in all human
activity and yet until recently both have been
frequently ignored by health researchers. Acknowl-
edging this shortcoming, we propose a conceptual
framework that overtly addresses temporal and
spatial dimensions of health, while incorporating
the health de terminants outlined by Evans and
Stoddart (1990) as well as the risk factors identified
in the literature discussed above. We have grouped
potential health determinants into five broad con-
structs that concomitantly interact among them-
selves and over time and space: (1) the physical
environment including both natural and human
made factors such as air pollution, aeroallergens
and climate factors; (2) the social and economic
environment which is comprised of a broad range of
factors including occupation, education and in-
come, to levels of community involvement and
social support; (3) biological influences such as
age, gender, an d co-morbidities; (4) health beha-
viour including smoking and drinking; and finally,
(5) health and social services which comprises the
availability and utilization of health care services as
well as other social services, public health policy and
health education. Through a variety of either direct
or indirect processes these factors may affe ct both
the likelihood of exposure, infection, or transmis-
sion of pneumonia and influenza, and the biologic al
or behavioural response to infection. Pneumonia
and influenza morbidity also vary over time and
space, and conceptually, link back to the aforemen-
tioned constructs.
Informed by the conceptual framework, the
principal objective of this paper is to look beyond
the traditional biomedical focus of pneumonia and
influenza determinants, and to identify instead what
role social, economic, environmental, behavioural
and health service factors may have in determining
pneumonia and influenza rates over space, and how
this may vary by age and gend er. This study builds
on our previous work exploring temporal (Crighton
et al., 2004) and spatial patterns (Crighton et al.,
2006) of pneumonia and influenza hospitalizations
in Onta rio, and represents a further step towards
understanding the combined spatial and temporal
dimensions of infectious disease morbidity and
health service use presented in the conceptual
framework.
Methods and data
We conducted a retrospective, population-based,
ecological level study to assess spatial patterns of
pneumonia and influenza hospitalizations in Ontar-
io, and the factors that determine these patterns.
The geographical unit of analysis used is the census
division (N ¼ 49) (Fig. 1). Census divisions corre-
spond to the political regions, counties and districts
of the province. For convenience, census divisions
will be referred to as ‘counties’ . While the use of
ecological level data may limit one’s ability to make
firm policy recomm endations, ecological study de-
signs allow for nearly complete coverage of a
population in the study area, and are well suited
for generating hypotheses for future work. It is
further argued that many important determinants of
health (e.g. socioeconomic and environmental fac-
tors) are inherently contextual in nature and require
study designs that work at the ecological level
(Kawachi et al., 1997; MacIntyre & Ellaway, 2000;
Wilkinson, 1996).
There were approximately 12 million residents in
Ontario as of 2001 (Statistics Canada, 2004).
County populations range from approximately
13,000 in Manatoulin, in the North (Fig. 1), to
over 2.5 million in Toronto, in the South. Nort hern
Ontario is the most sparsely populated area in the
province, while Southern and Eastern Ontario is
made up of both sparsely populated rural agricul-
tural areas and the province’s major urban centres
(i.e. Tor onto, Ott awa, Hamilton and Windsor). For
residents of Ontario, access to health care services is
universal through the Ontario Health Insurance
Program (OHIP), although northern and rural
residents must frequent ly travel significantly longer
distances for health care services as compared to
their urban counterparts.
Data
Outcome variables
The Canadian Institute for Health Information
(CIHI) Discharge Abstract Database was used to
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1639
obtain information on hospitalizations for pneumo-
nia and influenza as the principal diagnosis, by
county of patients’ usual residence. This database
records discharges from all inpatient hospit al stays
in Ontario acute care hospitals using the Interna-
tional Classification of Diseases, Ninth Revision,
Clinical Modification (ICD-9-CM). Nine years of
aggregated CIHI hospitalization data were exam-
ined, covering the period between April 1, 1992 and
March 31, 2001. While the reliability of specific
pneumonia aetiologic information is somewhat low
(approximately 52%; Marrie, Durant, & Sealy,
1987), in aggregate form, pneumonia and influenza
have been found to be reliably coded (81%; Upshur,
1997).
All records with a princip al discharge diagnosis
of influenza or pneumonia (ICD-9 code: 480–487)
were selected (N ¼ 241,803). Average annual hospi-
talization rates were computed by dividing the
total number of hospitalizations over the study
period by the population at risk. Rates were
adjusted for age and sex by the direct method
(Breslow & Day, 1987). Comparative hospitaliza-
tion quotients (CHQ) were then calculated. CHQs
are defined here as a ratio between the observed
directly standardized hospitalization rate of a given
county, to the expected rate if the outcome had
occurred at the provincial rate (see Breslow &
Day, 1987 for a description of similar comparative
measures). Thus, CHQs below 1 indicate that rates
are below the provincial mean while CHQs above 1
indicate rates are above the mean.
Explanatory variables
The selection of explanatory variables was guided
by the conceptual framework. Biological factors
and co-morbidities have been examined in numer-
ous studies and are, therefore, not included here;
contributions and limitations of this literature
are outlined above. While the exclusion of these
variables represents a potential limitation given
the mediating role they may play in determining
pneumonia and influenza hospitalizations, it was
decided that the focus here would be on the less well
understood social, economic, environmental and
health care system factors. Explanatory variables
examined in this study are described in the following
paragraphs and in Table 1. While some variables are
self-explanatory, several warrant more explanation.
Explanatory variables were derived from two
major secondary sources. The first is the 1996/1997
Ontario Health Survey (OHS), which is comprised
ARTICLE IN PRESS
Fig. 1. Study area, Ontario counties, Canada.
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501640
of a sample of approximately 36,000 individuals,
and had a household response rate of over 78%
(Ontario Ministry of Health, 1997). OHS variables
were directly age and gender standardized to the
Ontario population. Variables obtained from the
OHS include percentages of daily smokers, passive
smokers, heavy drinkers, body mass index 427 and
influenza vaccination in past year. The variables,
social support and social involvement are indexes
based on the combination of several related ques-
tions. Low levels of social support and social
involvement have been found to affect a range of
health outcomes (e.g., Crighton, Elliott, van der
Meer, Small, & Upshur, 2003; Kawachi et al., 1997;
Veenstra et al., 2005), but have not, to our knowl-
edge, been studied in relat ion to pneumonia and
influenza.
Data on social and economic characteristics and
residential environment were taken from the 1996
census. Detailed descriptions of all census variables
used in this study can be fou nd in the 1996 Census
Data Dictionary (Statistics Canada, 1999). Instead
of using an absolute measure of income, ‘low
income’ is used as a general indicator of poverty
levels within the population. Statistics Canada
defines a low income person as being someone
who spends 20% more of total income on food and
shelter, than the amount spent by the average
person in the population, adjusting for size of
settlement to account for cost of living.
A variable for the proportion of family physicians
(FPs) and general practitioners (GPs) actively
practicing family medicine (i.e. billing for 450%
family medicine codes) per 100,000 population in
1996 was created using the Canadian Physician
Database (CPDB) and OHIP billing database. Also,
daily temperature data from Environment Canada’s
Daily Climate Database was used to calculate a 10-
year mean county temperature variable.
Analysis
Analysis was done to assess the degree of spatial
autocorrelation in the outcome variables using the
Moran’s I statistic. Significant spatial autocorrela-
tion indica tes a regula r pattern in the data over
space such that a value at a given location depends
on, and is similar to, a value of defined spatial
neighbours. Neighbour relationships are expressed
in a row standardized spatial weights matrix W
whose elements W
ij
represent the binary spatial
weights assigned to pairs of units i and j, where
ARTICLE IN PRESS
Table 1
Candidate in dependent variables
Variable Description Data Source
Physical environment
Temperature Average annual temperature over study period Environment Canada
Poor housing % of homes needing major repairs Census
Passive smoker % living in home with regular inside smoker OHS
Social and economic environment
Low income % low income households of total private households—household
spends 20% more of total income on food and shelter than average
Census
Unemployment % 415 years reporting to be unemployed Census
Low education % 15 years or older with less than high school education Census
Aboriginal % Aboriginal population Census
Marital status % 15 years or older not married Census
Social support % reporting low social support OHS
Social involvement % reporting low social involvement OHS
Behavioural
Daily smoker % 12 years and over who currently smoke daily OHS
Heavy drinker % 12 years and over drinking in excess 14 drinks/week (males) and 9
drinks/week (females)
OHS
Overweight % overweight (BMI greater than 27) OHS
Healthcare
Doctors Family medicine practitioners/100,000 population OHIP/CPDB
Flu shot % reporting influenza vaccination within 1 year OHS
OHS, 1996/1997 Ontario Health Survey; OHIP, Ontario Health Insurance Program; CPDB, Canadian Physician Database.
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1641
W
ij
¼ 1 for neighbours and 0 for non-neighbours.
For this analysis, neighbours were defined using
rook’s case adjacency, which considers all counties
with common borders as neighbours.
Initial variable screening included calculating
correlation coefficients, and checking for linearity
in the expected relationships, between explanatory
and outcome variables. Outcome variables were log
transformed (natural log) in order to yield approx-
imate normality in the data and to help stabilize the
variance. All explanatory variables underwent a
deviation from the mean transformation for easier
interpretation. In cases where there was a non-linear
relationship between the outcome and an explana-
tory variable, a log transformation was applied to
the latter.
For each outcome, the selected independent
variables were included in an ordinary least-squares
(OLS) regression. A backward stepwise procedure
was used with a significance level of po0.1 required
in the partial test for explanatory variables to
be retained in the final model. In cases where
significant correlation between explanatory vari-
ables was detected, each was entered into the
model selection procedure alternately to assess their
relative contributions. The fit of alternative models
were compared. Formal diagnostic tests for multi-
collinearity and heterosk edasticity were conducted.
Residuals were also tested for normal ity as well as
spatial dependence using the Moran’s I test. Further
analysis was done using a local indicator of spatial
autocorrelation or LISA (Anselin, 1995) to assess
the degree of localized clustering of model residuals
(for a good description of these techniques see
Bailey & Gatrell, 1995).
Spatial dependence in the model residuals repre-
sents a violation of OLS assumptions. If spatial
dependence was identified using the Moran’s I test,
spatial lag and spati al error models, two alternative
models that incorpora te spatial dependence, were
considered. These are discussed in detail by Anselin
(1992) and Bailey and Gatrell (1995). Based on the
results of Lagrange Multiplier tests (see Anselin,
1992), spatial error models were found to be the
appropriate alternative. Following Anselin’s (1992)
notation, the spatial error model takes the form:
Y ¼ X b þ with ¼ lW þ x,
where Y is a vector of observations on the
dependent variable, W is a spatial weights matrix,
e is a vector of error terms, We are spatially lagged
errors, l is an autoregressive coefficient, and x is a
vector of independent random errors. Models, using
the same explanatory variables as in the OLS
regression, were fit using maximum likelihood
estimation. Formal diagnostics tests were again
performed to assess the suitability of these models.
R
2
and log likelihood values are used to compare
the fit of the models.
To interpret the influence of significant explana-
tory variables in the models on the log CHQ, we
used the regression coefficient as the exponent of the
base of the natural log to assess a 1-unit increase in
x above its mean: e
b
1. All significant variables
examined in this study, other than the Aboriginal
variable, are interpreted in the same way. The
Aboriginal variable was natural log trans formed
and thus its coefficient is interpreted differently,
using the following formula:
x þ 1
x

b
1.
All data manipulation and statistical analyses
were done using S+ (v.6.2), SPSS (v.11.0.1) and
Geoda (v.0.9.5-i; Luc Anselin, University of Illinois,
USA).
Results
There were 241,803 pneumonia and influenza
hospitalizations in Ontario over the study period,
representing an overall rate of 242/100,000 popula-
tion. Rates were higher for males than females at
257/100,000 and 227/100,000, respectively. By age,
the rates are highest for the oldest (65+ years),
followed by the youngest (0–14 years) groups at
approximately 1,000 and 220 per 100,000, respec-
tively (data not shown).
By county, rates ranged between 171 and 665 per
100,000 for all ages and both genders combined
(data not shown). This variability in hospitaliza-
tions is evident in Fig. 2, which shows county level
CHQs by gender. High, statistically significant
CHQs (po0.01) were generally seen in northern
counties including Kenora, Timiskaming and Man-
itoulin for both genders, wher e rates are approxi-
mately 2–3 times higher than the provincial mean.
In more southern areas, significantly high CHQs
were predominately found in rural counties along
the east coast of Lake Huron, Kawartha Lakes in
central Ontario, and Stormont-Dundas in the Far
East. The lowest CHQs are seen in urban and
southern areas in and around Toronto, Middlesex,
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501642
and Ottawa. Although there is somewhat more
heterogeneity in CHQs among females, the overall
spatial pattern of hospitalizations is consistent for
both sexes.
The Moran’s I statistic indicates that there is a
moderate, statistically significant degree of spatial
autocorrelation in the data for females (Moran’s
I ¼ 0.242; po0.015), indicating that counties with
similar hospitalization rates are clustered together.
Somewhat lower, but still significant autocorrela-
tion is seen for males (Moran’s I ¼ 0.173; po0.044).
Table 2 summarizes the results for the total
population and by gender models. There was no
evidence of heteroskedasticit y or multicollinearity in
these models. For the total population, the results
of the initial OLS model (data not shown)
were deemed unsuitable as the Moran’s I statistic
testing for spatial error dependence was significant
(Moran’s I ¼ 0.162; p ¼ 0.043). This indicates that
the model did not adequately account for spatial
dependence in the CHQs and the OLS assumption
of residual independence was not met. A spatial
ARTICLE IN PRESS
0.66 - 0.9
0.9 - 1.2
1.2 - 1.5
1.5 - 2
2 - 2.86
Male
Female
N
E
W
S
0
200
300 Kilometers
100
100
Moran’s I = 0.242
P-value = 0.015
Moran’s I = 0.173
P-value = 0.044
CHQs
Fig. 2. Comparative morbidity figures (CHQs) of pneumonia and influenza hospitalizations by gender in Ontario counties: 1992–2001.
Table 2
Regression coefficients and standard errors for the total and gender specific pneumonia and influenza models (1992–2001)
Explanatory
variables
Total
a
Females
a
Males
b
Coefficient SE p-value Coefficient SE p-value Coefficient SE p-value
Intercept 0.221 0.033 o0.001 0.231 0.032 o0.001 0.201 0.024 o0.001
Aboriginal 0.089 0.029 0.002 0.081 0.027 0.005
Low education 0.053 0.008 o0.001 0.037 0.009 o0.001 0.044 0.008 o0.001
Poor housing 0.037 0.014 0.009
Passive smoker 0.012 0.004 0.005
Heavy drinker 0.028 0.014 0.054 0.020 0.011 0.066
l 0.325 0.163 0.047 0.291 0.168 0.083
Model R
2
0.668 0.691 0.644
Model LL 19.671 20.19 19.18
LR test p ¼ 0.095 p ¼ 0.153
Note: SE, standard error; l, autoregressive coefficient (and standard error) of the spatial error model; Model LL, Model log likelihood; LR
test, likelihood ratio test.
a
Spatial error model.
b
Ordinary least-squares (OLS) model.
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1643
error model, which incorporates spatial dependence,
was therefore used. Three variables display signifi-
cant, positive relationships with the log CHQ
(natural log): Aboriginal population, low education
and drinking. The model explained approximately
67% of the variation (R
2
¼ 0.668). For the educa-
tion variable, the coefficient indicates that a 1-unit
(%) increase in low education from its mean is
associated with an increase of 0.0544% or 5.4%
in the CHQ on average . In the case of the heavy
drinking variable, a 1-unit (%) increase from its
mean is associated with an increase in the CHQ of
0.0283% or approximately 2.8%. For the same
increase in the aboriginal variable, the CHQ
increases by 0.0249% or 2.5%. The model log
likelihood, a measure of model fit, which can be
used to compare OLS and spatial error models, was
18.28 for the OLS models, and 19.67 for the spatial
error model. While this indicates that the fit is better
in the spatial versus the OLS model, the likelihood
ratio test (LR test) indicates that this difference is
not statistically significant.
For females, significant autocorrelation in the
OLS model residuals was again identified (Moran’s
I ¼ 0.176; p ¼ 0.027) and a spatial error model was,
therefore, employed. The model explained a sub-
stantial portion of the varia tion (R
2
¼ 0.67). CHQs
were found to increase with higher percentages of
low education, poor housing, and inside smokers.
The regression coeffici ent indicates that for every
1-unit (%) increase in the low education variable
from its mean, the CHQ increases by 0.0376% or
3.8% on average. For the same increase in either the
poor housing or passive smoking variables, the
CHQ increases by 3.8% or 1.2%, respectively. The
model log likelihood value was higher for the spatial
error model as compared to the OLS model (20.19
versus 18.85) indicating a better model fit. Again,
however, this difference was not significant (LR test
p ¼ 0.153).
For males, the OLS model was determined to be
appropriate as no significant error dependence was
found (Moran’s I ¼ 0.106; p ¼ 0.139). The model
explained a substantial portion of the variation in
the CHQs for pneumonia and influenza hospitaliza-
tions (R
2
¼ 0.64). CHQs were found to increase
with higher percentages of Aboriginal population,
low education and heavy drinker s. The regression
coefficient indicates that every 1% increase in the
Aboriginal population variable from the mean is
associated with an increase of the CHQ by 0.0226%
or approximately 2.3% on average. For every 1%
increase in the low education or heavy drinkers
variables, the CHQ increases by 4.5% and 2.0%,
respectively.
Table 3 summarize s the results of the female
models by age group. There was no evidence of
heteroskedasticity or multicollinear ity in this or any
of the age-specific female models. For the female
0–14-year age group OLS model (Table 3), no
significant error dependence was identified (Moran’s
I ¼ 0.081; p ¼ 0.218), and the model was deemed
appropriate. The model explains approximately
57% of the CHQ variation, with two variables
being retained in the model: Aboriginal populati on
and low education. With a 1% increase in the
Aboriginal population, the CHQ increases by
0.0383% or 3.8%. For the same increase in the
ARTICLE IN PRESS
Table 3
Regression coefficients and standard errors for the female pneumonia and influenza hospitalization models by age group (1992–2001)
Explanatory variables 0–14 years
a
15–64 years
b
65+ years
a
Coefficient SE p-value Coefficient SE p-value Coefficient SE p-value
Intercept 0.201 0.044 o0.001 0.356 0.054 o0.001 0.154 0.027 o0.001
Aboriginal 0.136 0.043 0.002 0.137 0.038 o0.001
Low education 0.078 0.014 o0.001 0.085 0.010 o0.001 0.033 0.009 o0.001
Daily smokers 0.014 0.007 0.029
Flu shot 0.005 0.003 0.078
Temperature 0.031 0.017 0.079
l 0.466 0.144 0.001
Model R
2
0.567 0. 733 0.431
Model LL 9.95 7.032 16.03
LR test p ¼ 0.009
a
Ordinary least-squares (OLS) model.
b
Spatial error model.
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501644
low education variable, the CHQ increases by
0.0811% or approximately 8%.
In the female 15–64-year age group OLS model,
significant error dependence was identified (Moran’s
I ¼ 0.259; p ¼ 0.002) making the spatial error
model more appropriate (Table 3). The spatial error
model accounted for approximately 73% of the
variation in CHQs (R
2
¼ 0.73). The LR test
indicates that the fit of the error model improved
significantly from the OLS (p ¼ 0.009). Again, only
the Aboriginal population and low education
variables were retained in the model. Here a 1%
increase in the Aboriginal population variable is
associated with an increase in the CHQ by 3.3%.
The same increase in the low education variable
accounts for an 8.1% increase in the CHQ.
For the 65+-year age group OLS model
(Table 3), no significant spatial error dependence
was identified (Moran’s I for model residuals ¼
0.121; p ¼ 0.081) . Approximately 43% of the
variation in the CHQs is explained, with four
variables being retained in the model: low education
(positive), daily smokers (positive), flu shot (nega-
tive), and temperature (negative). Increases of 1% in
proportions with low education or in proportions of
daily smokers are, respectively, associated with a
3.4% and 1.4% increase in hospitalizations. A 1%
increase in the flu shot is associated with a decrease
in the CHQ by 0.5% on average, and a 11 increa se
in the mean county temperature is associated with a
decrease in the CHQ by 3.1%.
Table 4 summarizes the results of the male models
by age group. There was no evidence of hetero-
skedasticity or multicollinearity in any of the male
OLS models. Some differences between male age
groups can be seen in the fit of the models and the
variables that were retained in the models. For the
0–14-year age group model for males (Table 4),
OLS was again determined to be appropriate
(Moran’s I ¼ 0.005; p ¼ 0.065). The model explains
just over 50% of the variation of CHQs (R
2
¼ 0.50).
Again Aboriginal population and low education
variables were retained in the model. For every 1%
increase in the Aboriginal population variable from
the mean, the CHQ increases by 0.0332% or 3.3%,
and for the same increase in the low education
variable, the CHQ increases by 0.0790% or almost
7.9%.
For the 15–64-year age group OLS model
(Table 4), significant error dependence was identi-
fied (Moran’s I ¼ 0.261; p ¼ 0.009) and a spatial
error model was used instead. The model has high
explanatory power, accounting for approximately
76% of the variation in CHQs. The LR test
indicates that the fit of the error model improved
significantly from its OLS equivalent (p ¼ 0.039).
The model has three significant variables: Abori-
ginal population, low education and heavy drinkers.
The coefficients for these variables suggest that for
every 1% increase in either the Aboriginal popula-
tion, low education, or heavy drinkers variables, a
CHQ increases on average by 2.6%, 6.2% and
2.1%, respectively.
For the male 65+ age group model (Table 4),
OLS was determined to be appropriate, demon-
strating no spatial error dependence (Moran’s
I ¼ 0.081; p ¼ 0.119). The model has a relatively
weak explanatory power (R
2
¼ 0.40) with two
significant variables: low education and tempera-
ture. Here a 1% increase in the low education
ARTICLE IN PRESS
Table 4
Regression coefficients and standard errors for the male pneumonia and influenza hospitalization models by age group (1992–2001)
Explanatory variables 0–14 years
a
15–64 years
b
65+ years
a
Coefficient SE p-value Coefficient SE p-value Coefficient SE p-value
Intercept 0.200 0.047 o0.001 0.271 0.037 o0.001 0.138 0.022 o0.001
Aboriginal 0.118 0.046 0.013 0.099 0.031 0.001
Low education 0.076 0.015 o0.001 0.064 0.008 o0.001 0.026 0.007 o0.001
Heavy drinker 0.024 0.009 0.008
Temperature 0.039 0.014 0.006
l 0.372 0.158 0.018
Model R
2
0.502 0.757 0.402
Model LL 13.37 18.490 22.74
LR test p ¼ 0.039
a
Ordinary least-squares (OLS) model.
b
Spatial error model.
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1645
variable is associated with a 2.6% increase in the
CHQ, and a 11 increase in mean county temperature
is associated with a 3.8% decrease in the CHQ on
average.
Results from the LISA analyses (data not shown)
indicate that the four groups with significant global
error dependence in their OLS residuals (i.e. total
population; females all ages, females15–64 years;
and males 15–64 years) had similar patterns of local
error dependence. Clusters of large residuals centred
on Grey County in the southwest of the province
were identified, indicating that an underlying
process is not being accounted for by the explana-
tory variables in the models. Two significant clusters
of small residuals for these models are seen for in
Haldimand-Norfolk and Elgin counties in the south
indicating the strong predictive power of the models
in these areas. The spatial dependence in the OLS
model residuals was accounted for in the spatial
error models.
Discussion
The objectives of this research centred on develop-
ing a better understanding of the spatial character-
istics of pneumonia and influenza hospitalizations by
age and gender, and the broad contextual factors that
determine them. Several issues arise from the results
presented.
We found that pneumonia and influenza hospi-
talization ratios varied significantly across Ontario
counties, with similar ratios clustering together
(Fig. 2) suggesting that common spatial processes
may be at play in determining hospitalizations
in these areas. The highest rates are seen in the
sparsely populated northern and rural areas where
they are between 2 and 3 times the provincial
average. The lowest rates are seen in southern and
urban areas where rates are between 10% and 30%
below the provincial average. Efforts to reduce the
burden of pneumonia and influenza must be
directed accordingly.
Our results indicate that low education is strongly
associated with pneumonia and influenza hospitali-
zations over all age groups an d both genders.
Education is a good proxy for SES, particularly
among the elderly, because it remains mostly
unaffected over time as compared to employment
or income (Loeb, 2003). The association identified
here is supported by the results of an ecologic level
analysis of pneumonia hospitalizations in the USA
(Morris & Munasinghe , 1994). Similar associations
have been reported for other health conditions
(e.g. Chen, Dales, & Krewski, 2001; Joines, Hertz-
Picciotto, Carey, Gesler, & Suchindran, 2003;
Sundquist & Johansson, 1997). The process by
which edu cation affects hospitalizations is some-
what unclear in terms of whether or not there is a
direct relationship with incidence or severity of
symptoms, or whether the relationship is mediated
entirely through lifestyle, socioeconomic, environ-
mental or health care factors. In the case of health
care, for example, low education may affect help
seeking behaviour at the primary care level, as well
as adherence with medical regimes, thereby leading
to higher rates of hospitalizations. Structural
equation modelling represents a potential next step
towards better understanding such processes.
The Aboriginal population variable was posi-
tively associated with pneumonia and influenza
hospitalizations in most models except those for
the 65+-year age groups. This associ ation has not
been previously reported at this level of analys is,
however, it is consistent with studies showing that
Aboriginal populations exp erience higher rates of
pneumonia morbidity and mortality than non-
Aboriginal populations (Marrie et al., 2004).
Identified risk factors such as poor housing and
high rates of smoking and drinking, are found at
significantly higher rates among Aboriginal popula-
tions (Stavenhagen, 2005). Programs to improve
living conditions and access to education, and
public health measures to reduce smoking and
drinking rates could be expected to reduce pneu-
monia and influenza hospitalization rates in coun-
ties with large Aboriginal populations.
Smoking and drinking both ach ieved significance
in different models. Findings suggest that gender
may modulate the effect of these behavioural factors
on hospitalization rates. The association between
heavy drinking and hospitalizations among men,
but not women, could be anticipated, given that
heavy drinking rates are higher among men. The
significant association between daily smoking
among women, but not men, is somewhat surpris-
ing, however, as daily smoking rates among women
are lower.
We further found that influ enza vaccination was
weakly, negatively associated with hospitalizations
in the female 65+ mod el. While this finding
supports the continuation of influenza vaccina tion
campaigns targeted at the province’s elderly, the
development of better vaccination expo sure vari-
ables is required in order to accurately assess their
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501646
efficacy in reducing pneumonia and influenza
morbidity more widely.
The environmental factors examined in this study
were all found to be associated with pneumonia and
influenza hospitalization rates in one or more
models. Passive smoking was significant in the
female all age groups model. This finding contra-
dicts those of Farr et al. (2000) who found no
significant association between passive smoking and
pneumonia hospitalizations. Temperature was ne-
gatively associated in both 65+-year age group
models. Thi s result may be explained by the adverse
effects that cold can have on the immune system’s
resistance to respiratory infection (The Eurowinter
Group, 1997) particularly among the elderly who
may already have challenged immune systems. This
relationship is likely mediated through other factors
including poor housing conditions. Poor housing
was found to be independently associated with
pneumonia and influenza hospitalizations in the
female all age groups model. While this variable is
categorized in this research as a proxy for indoor
environmental conditions (e.g. indoor air quality,
inadequate heating, and so on), it also functions as
an indicator of low SES.
Also interesting are the variables that were not
found to be significant in the models. For example,
low income was not significantly associated with
pneumonia and influenza hospitalization rates, a
finding supported by a number of previous studies
(Stelianides et al., 1999; Vrbova et al., 2005 ), and
contradicted by others (e.g. Harrison et al., 2000;
Morris & Munasinghe, 1994; Wood et al., 1999).
The relationship must, therefore, not be discounted,
and further analysis examining alternative measures
should be considered that include absolute and
relative measures of income both at individual and
ecological levels, and at different spatial aggrega-
tions. The FP/GP availability variable was also not
significantly associated with pneumonia and influ-
enza hospitalizations. This suggests the following
possible explanations: the variable is not an
adequate measure of system access; there is a
mismatch between the scale of analysis and the
scale of the underlying process; or more simply, that
greater access to primary care does not equate
reduced risk of pneumonia and influenza hospitali-
zation. This last explanation seems unlikely given
that Macinko et al. (2003) found that, in a study
covering 18 OECD nations , strong primary care
systems were negatively associated with premature
mortality from pneumonia.
Much of the spatial error dependence identified in
the hospitalization data was explained by the
covariates in our OLS models. Possible explana-
tions for the remaining spatial error dependence
include the omission of a relev ant variable from the
models, or perhaps, again, a mismatch between the
scale of analysis and the scale of the underlying
process. In models where spatial dependence re-
mained, spatial error models provided control, and
confirmed the independent contribution of identi-
fied social, economic and behavioural covariates.
While the spatial effect on the models was not large
in any given case, it is important to recognize that
spatial effects are frequently found to be significant
(e.g., Green, Hoppa, Young, & Blanchard, 2003;
Joines et al., 2003), and as such wider use of spatial
modelling techniques should be encouraged.
This study has several limitations. Firstly, it is an
ecologic study. While this approach lends itself to
the examination of contextual determinants of
health, complete population coverage, and the
generation of hypotheses, it limits our ability to
draw conclusions concerning factors that are
responsible for variations in hospitalization rates
at the individual level. The modifiable areal unit
problem (MAUP) (Openshaw, 1984) repres ents a
second potential limitation in that the patterns
identified here may dep end on the areal aggrega-
tions used (i.e. counties). Unfortunately, testing for
this is difficult due to population size constraints at
larger scales. Thirdly, the use of hospitalizations is
not necessarily reflective of morbidity in the
population, and does not account for differential
access to services (Eyles, Birch, Chambers, Hurley,
& Hutchison , 1991). However, given that health
insurance is universal in Ontario, hospitalizations
are believed to represent a good estimate of severe
morbidity. Findings from a British study identifying
a strong correlation (r ¼ 0.69; po0.01) between
hospitalizations and morbidity for respiratory ill-
ness (Payne, Coy, Patterson, & Milner, 1994)
reinforce this point, although this has not been
confirmed in a Canadian context. Fourthly, access
(as defined by proximity) to hospitals was not
controlled for in this analysis despite it being a
potential determinant of use. Our results, however,
indicate that rural areas typically had the highest
hospitalization rates, despite having the least access.
Thus, the absence of this variable is not believed to
have confounded our findings. Finally, despite
aetiological differences, influenza and the pneumo-
nias represented by the ICD-9 diagnostic codes
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1647
480–487, were combined for this analysis. Although
the aggregation is commonly seen (e.g. Morris &
Munasinghe, 1994; Saynajakangas, Keistinen, &
Tuuponen, 2001), due to the low reliability in
specific aetiologic information , it is expected that
age and gender-specific differences in our results at
least partly reflect illness-specific prevalence rates.
This research makes a number of theoretical,
methodological and substantive contributions to the
study of pneumonia and influenza, and more
generally, infectious disease. Theoretically, our
results point to the usefulness of the conceptual
framework in guiding the investigation of the
underlying causes of spatial variability of pneumo-
nia and influenza outcomes. Its utility is borne out
by the range of explanatory variables, which
emerged as significant in the models, and the large
proportion of geographic variation in pneumonia
and influenza hospitalizations that it has helped
explain. The framework further illustrates the
complexity of the relationships that exist and the
need for more sophisticated analytic techniques that
can integrate both spatial and temporal dimensions.
Methodologically, this research demonstrates how
spatial analytic techniques can be applied in study-
ing infectious disease. The use of spatial error
regression models allowed for unbiased estimation
of regression parameters and their significance
levels. Wider use of spatial modelling techniques
should be encouraged. Substantively, a better
understanding of the geographic variability in
pneumonia and influenza hospitalizations in Ontar-
io, and of the importance of the broad range of age
and gender-specific ecologic-level factors associated
with this outcome, wi ll help inform the development
of more region and population-specific social,
economic, public health and health care programs.
The significance of these findings further extends to
various (re)emerging infectious diseases including
HIV/AIDS, tuberculosis, and the more recent
concern, Avian Influenza, where similar factors
play a role in determining disease exposure,
susceptibility, and health care system use (Cohen,
1999; Lede rberg, Shope, & Oaks, 1992). Efforts
made to reduce pneumonia and influenza rates
through, for example, improved social welfare
programs, could, therefore, be expected to reduce
rates of other infectious diseases.
As the populations of Canada and most Western
nations age, and the burden of pneumonia and
influenza increases, developing a better understand-
ing of these illnesses and related healt h services use
is crucial. Further work is needed that includes the
examination of different geographical aggregations
and alternative pneumonia and influenza outcomes.
Multilevel modelling to understand the relative roles
of individual and contextual level factors in
determining pneumonia and influenza outcomes is
also necessary. This paper compliments work
previously done examining temporal patterns
(Crighton et al., 2004) and spatial patterns (Cright-
on et al., 2006) of pneumonia and influenza
hospitalizations, and sets the stage for analyses
that incorporates spatial and temporal dimensions
together.
References
Almirall, J., Bolibar, I., Vidal, J., Sauca, G., Coll, P., Niklasson,
B., et al. (2000). Epidemiology of community-acquired
pneumonia in adults: A population-based study. European
Respiratory Journal, 15, 757–763.
Anselin, L. (1992). SpaceStat tutorial. A workbook for using
spacestat in the analysis of spatial data. Urbana, IL:
University of Illinois, Urbana-Champaign.
Anselin, L. (1995). Local indicators of spatial association—LISA.
Geographical Analysis, 27, 93–115.
Baik, I., Curhan, G. C., Rimm, E. B., Bendich, A., Willett, W. C.,
& Fawzi, W. W. (2000). A prospective study of age and
lifestyle factors in relation to community-acquired pneumonia
in US men and women. Archives of Internal Medicine, 160,
3082–3088.
Bailey, T. C., & Gatrell, A. C. (1995). Interactive spatial data
analysis. Essex: Prentice-Hall.
Breslow, N. E. & Day, N. E. (1987). Statistical methods in cancer
research, vol. II: The design and analysis of cohort studies.
IARC Scientific Publications No. 82. Lyon: International
Agency for Research on Cancer.
Chen, Y., Dales, R., & Krewski, D. (2001). Asthma and the risk
of hospitalization in Canada: The role of socioeconomic and
demographic factors. Chest, 119(3), 708–713.
Cliff, A., Haggett, P., & Ord, J. (1986). Spatial aspects of influenza
epidemics. London: Pion.
Cliff, A. D., & Smallman-Raynor, M. (1992). The AIDS
pandemic: Global geographical patterns and local spatial
processes. Geographical Journal, 158, 182–198.
Cohen, S. (1999). Social status and susceptibility to respiratory
infections. Annals of the New York Academy of Science, 896,
246–253.
Crighton, E. J., Elliott, S. J., van der Meer, J., Small, I., &
Upshur, R. (2003). Impacts of an environmental disaster on
psychosocial health and well-being in Karakalpakstan
(Uzbekistan). Social Science & Medicine, 56(3), 557–573.
Crighton, E. J., Moineddin, R., Upshur, R., & Mamdani,
M. (2004). Pneumonia and influenza hospitalizations:
A population level time series analysis. Epidemiology and
Infection, 132, 1167–1174.
Crighton, E. J., Elliott, S. J., Moineddin, R., Kanaroglou, P., &
Upshur, R. (2006). An exploratory spatial analysis of
pneumonia and influenza hospitalizations in Ontario by age
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501648
and gender. Epidemiology and Infection July 7; 1–9 (Epub
ahead of print).
Evans, R. G., & Stoddart, G. L. (1990). Producing health,
consuming health care. Social Science & Medicine, 31(12),
1347–1363.
Eyles, J., Birch, S., Chambers, S., Hurley, J., & Hutchison, B.
(1991). A needs-based methodology for allocating health care
resources in Ontario, Canada: Development and an applica-
tion. Social Science & Medicine, 33, 489–500.
Farr, B. M., Bartlett, C. L. R., Wadsworth, J., & Miller, D. L.
(2000). Risk factors for community-acquired pneumonia
diagnosed upon hospital admission. Respiratory Medicine,
94, 954–963.
Fine, M. J., Auble, T. E., Yealy, D. M., Hanusa, B. H.,
Weissfeld, L. A., Singer, D. E., et al. (1997). A prediction rule
to identify low-risk patients with community-acquired pneu-
monia. New England Journal of Medicine, 336, 243–250.
Green, C., Hoppa, R. D., Young, T. K., & Blanchard, J. F.
(2003). Geographic analysis of diabetes prevalence in an
urban area. Social Science & Medicine, 57(3), 551–560.
Harrison, L. H., Dwyer, D. M., Billman, L., Kolczak, M. S., &
Schuchat, A. (2000). Invasive pneumococcal infection in
Baltimore, MD. Implications for immunization policy.
Archives of Internal Medicine, 160, 89–94.
Health Canada. (2001). Respiratory disease in Canada. Ottawa,
Canada: Health Canada.
Joines, J. D., Hertz-Picciotto, I., Carey, T. S., Gesler, W., &
Suchindran, C. (2003). A spatial analysis of county-level
variation in hospitalization rates for low back problems in
North Carolina. Social Science & Medicine, 56(12),
2541–2553.
Kaplan, V., Clermont, G., Griffin, M. F., Kasal, J., Watson, R.
S., Linde-Zwirble, W. T., et al. (2003). Pneumonia: Still
the old man’s friend? Archives of Internal Medicine, 163,
317–323.
Kawachi, I., Kennedy, B. P., Lochner, K., & Prothrow-Stith,
D. (1997). Social capital, income inequality and mortality.
American Journal of Public Health, 50, 245–251.
Labonte, R. (1987). Community health promotion strategies.
Health Promotion, 5, 32.
Lange, P., Vestbo, J., & Nyboe, J. (1995). Risk factors for death
and hospitalization from pneumonia. A prospective study
of a general population. European Respiratory Journal, 8,
1694–1698.
Lederberg, J., Shope, R. E., & Oaks, S. C. (1992). Emerging
infections: Microbial threats to health in the United States.
Washington, DC: National Academy Press.
Loeb, M. (2003). Community-acquired pneumonia in older
people: The need for a broader perspective. Journal of the
American Geriatrics Society, 51, 539–543.
Loytonen, M. (1998). GIS, time, geography and health. In
A. Gatrell, & M. Loytonen (Eds.), GIS and health
(pp. 97–110). London: Taylor & Francis (Chapter 7).
Macinko, J., Starfield, B., & Shi, L. (2003). The contribution
of primary care systems to health outcomes within orga-
nization for economic cooperation and development
(OECD) countries, 1970–1998. Health Services Research, 38
,
831–865.
Macintyre, S., & Ellaway, A. (2000). Chapter 14: Ecological
approaches: Rediscovering the role of the physical and social
environment. In L. Berkman, & I. Kawachi (Eds.), Social
epidemiology. Oxford, England: Oxford University Press.
Marmor, T. R., Barer, M. L., & Evans, R. G. (1994). The
determinants of a population’s health: What can be done to
improve a democratic nation’s health status? In R. G. Evans,
M. L. Barer, & T. R. Marmor (Eds.), Why are some people
healthy and others not (pp. 217–230). Hawthorne, NY: Aldine
de Gruyter.
Marrie, T. J., Durant, H., & Sealy, E. (1987). Pneumonia—
The quality of medical records data. Medical Care, 25(1),
20–24.
Marrie, T. J., Carriere, K. C., Jin, Y., & Johnson, D. H. (2004).
Hospitalization for community-acquired pneumonia in Al-
berta First Nations Aboriginals compared with non-First
Nations Albertans. Canadian Respiratory Journal, 11(5),
336–342.
McKeown, T., Record, R. G., & Turner, R. D. (1975). An
interpretation of the decline in mortality in England and
Wales during the twentieth century. Population Studies
(Cambridge), 29(3), 391–422.
Morris, R. D., & Munasinghe, R. L. (1994). Geographical
variability in hospital admission rates for respiratory disease
among the elderly in the United States. Chest, 106(4),
1172–1181.
Nuorti, J. P., Butler, J. C., Farley, M. M., Harrison, L. H.,
McGeer, A., Kolczak, M. S., et al. (2000). Cigarette smoking
and invasive pneumococcal disease. Active bacterial core
surveillance team. New England Journal of Medicine, 342(10),
681–689.
Ontario Ministry of Health. (1997). Ontario health survey (OHS)
1996/1997. Toronto: Ontario Ministry of Health & Long
Term Care.
Openshaw, S. (1984). The modifiable areal unit problem. Concepts
and techniques in modern geography, No. 38. Norwich: Geo
Books.
Payne, J. N., Coy, J., Patterson, S., & Milner, P. C. (1994). Is use
of hospital services a proxy for morbidity? A small area
comparison of the prevalence of arthritis, depression,
dyspepsia, obesity, and respiratory disease with inpatient
admission rates for these disorders in England. Journal of
Epidemiology and Community Health, 48, 74–78.
Plouffe, J. F., Breiman, R. F., & Facklam, R. R. (1996).
Bacteremia with Streptococcus pneumoniae. Implications for
therapy and prevention. Franklin County Pneumonia Study
Group. Journal of the American Medical Association, 275(3),
194–198.
Poland, B., Coburn, D., Robertson, A., & Eakin, J. (1998).
Wealth, equity and health care: A critique of a ‘‘population
health’’ perspective on the determinants of health. Social
Science & Medicine, 46(7), 785–798.
Ross, N. A., Wolfson, M. C., Dunn, J. R., Berthelot, J., Kaplan,
G. A., & Lynch, J. W. (2000). Relation between income
inequality and mortality in Canada and in the United States:
Cross sectional assessment using census data and vital
statistics. British Medical Journal, 320, 898–902.
Saynajakangas, P., Keistinen, T., & Tuuponen, T. (2001).
Seasonal fluctuations in hospitalisation for pneumonia in
Finland. International Journal of Circumpolar Health, 60(1),
34–40.
Stelianides, S., Golmard, J. L., Carbon, C., & Fantin, B. (1999).
Influence of socioeconomic status on features and out-
come of community-acquired pneumonia. European Journal
of Clinical Microbiology & Infectious Diseases, 18(10),
704–708.
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–1650 1649
Statistics Canada. (1999). 1996 Census Dictionary (final ed).
Ottawa, Canada: Statistics Canada.
Statistics Canada (2004). Annual demographic statistics 2003.
Catalogue no. 91-213-XPB. Ottawa, Canada: Statistics
Canada.
Stavenhagen, R. (2005). Report of the special rapporteur on the
situation of human rights and fundamental freedoms of
indigenous people, Rodolfo Stavenhagen, on His Mission to
Canada (21 May–4 June 2004). United Nations Commission
On Human Rights.
Sundquist, J., & Johansson, S. E. (1997). Self reported poor
health and low educational level predictors for mortality: A
population based follow up study of 39,156 people in Sweden.
Journal of Epidemiology and Community Health, 51(1), 35–40.
The Eurowinter Group. (1997). Cold exposure and winter
mortality from ischaemic heart disease, cerebrovascular
disease, respiratory disease, and all causes in warm and cold
regions of Europe. Lancet, 349, 1341–1345.
Upshur, R. (1997). Measuring the impact of influenza on
hospitalizations of the elderly in Ontario. M.Sc. thesis.
Toronto, Canada: University of Toronto.
Veenstra, G., Luginaah, I., Wakefield, S., Birch, S., Eyles, J., &
Elliott, S. (2005). Who you know, where you live: Social
capital, neighbourhood and health. Social Science & Medi-
cine, 60(12), 2799–2818.
Vrbova, L., Mamdani, M., Moineddin, R., Jaakimainen, L., &
Upshur, R. E. G. (2005). Does Socioeconomic status affect
mortality subsequent to hospital admission for community
acquired pneumonia among older persons? Journal of
Negative Results in BioMedicine, 4,4.
Wilkinson, R. G. (1996). Unhealthy societies: The afflictions of
inequality. London: Routledge.
Wood, E., Sallar, A. M., Schechter, M. T., & Hogg, R. S. (1999).
Social inequalities in male mortality amenable to medical
intervention in British Columbia. Social Science & Medicine,
48, 1751–1758.
ARTICLE IN PRESS
E.J. Crighton et al. / Social Science & Medicine 64 (2007) 1636–16501650
    • Comorbidities such as type 2 diabetes, obesity, chronic obstructive pulmonary disease (COPD), and asthma have been associated with severe influenza illness, resulting in increased rates of hospitalization and premature death [49][50][51][52] . Likewise, individual behaviors such as smoking and alcohol abuse can be detrimental to health and are associated with influenzarelated hospitalization [53] . To combat the higher prevalence of comorbidities and detrimental health behaviors observed in disadvantaged neighborhoods, efforts should aim to increase vaccination coverage, improve health behaviors, and address specific underlying comorbidities (eg, type 2 diabetes, obesity, and COPD) to help decrease susceptibility to and severity of influenza illness in disadvantaged neighborhoods [54][55][56].
    Full-text · Article · Sep 2016
    • A case–control study on child ALRI mortality in Gambia did not find a significant association with SES [9], and, similarly, a prospective cohort study of children under 5 years in Matlab did not find an association with incidence of respiratory infections and various sociodemographic measures [6]. However, other studies have found various measures of social status including income, home ownership, and education to be predictive of respiratory infections in various age groups and country set- tings [53, 54, 66]. SES may affect respiratory infections by increasing exposure to pathogens in crowded living quarters or by decreasing an individual's immune status due to stress or poor nutrition [58].
    [Show abstract] [Hide abstract] ABSTRACT: Background Respiratory infections continue to be a public health threat, particularly to young children in developing countries. Understanding the geographic patterns of diseases and the role of potential risk factors can help improve future mitigation efforts. Toward this goal, this paper applies a spatial scan statistic combined with a zero-inflated negative-binomial regression to re-examine the impacts of a community-based treatment program on the geographic patterns of acute lower respiratory infection (ALRI) mortality in an area of rural Bangladesh. Exposure to arsenic-contaminated drinking water is also a serious threat to the health of children in this area, and the variation in exposure to arsenic must be considered when evaluating the health interventions. MethodsALRI mortality data were obtained for children under 2 years old from 1989 to 1996 in the Matlab Health and Demographic Surveillance System. This study period covers the years immediately following the implementation of an ALRI control program. A zero-inflated negative binomial (ZINB) regression model was first used to simultaneously estimate mortality rates and the likelihood of no deaths in groups of related households while controlling for socioeconomic status, potential arsenic exposure, and access to care. Next a spatial scan statistic was used to assess the location and magnitude of clusters of ALRI mortality. The ZINB model was used to adjust the scan statistic for multiple social and environmental risk factors. ResultsThe results of the ZINB models and spatial scan statistic suggest that the ALRI control program was successful in reducing child mortality in the study area. Exposure to arsenic-contaminated drinking water was not associated with increased mortality. Higher socioeconomic status also significantly reduced mortality rates, even among households who were in the treatment program area. Conclusion Community-based ALRI interventions can be effective at reducing child mortality, though socioeconomic factors may continue to influence mortality patterns. The combination of spatial and non-spatial methods used in this paper has not been applied previously in the literature, and this study demonstrates the importance of such approaches for evaluating and improving public health intervention programs.
    Full-text · Article · Sep 2016
    • Previous studies have used statistical methods to describe empirical relationships between the burden of influenza and social determinants of health (Charland et al., 2011; Crighton et al., 2007). We used a simulation model to further explain one such
    [Show abstract] [Hide abstract] ABSTRACT: Factors associated with the burden of influenza among vulnerable populations have mainly been identified using statistical methodologies. Complex simulation models provide mechanistic explanations, in terms of spatial heterogeneity and contact rates, while controlling other factors and may be used to better understand statistical patterns and, ultimately, design optimal population-level interventions. We extended a sophisticated simulation model, which was applied to forecast epidemics and validated for predictive ability, to identify mechanisms for the empirical relationship between social deprivation and the burden of influenza. Our modeled scenarios and associated epidemic metrics systematically assessed whether neighborhood composition and/or spatial arrangement could qualitatively replicate this empirical relationship. We further used the model to determine consequences of local-scale heterogeneities on larger scale disease spread. Our findings indicated that both neighborhood composition and spatial arrangement were critical to qualitatively match the empirical relationship of interest. Also, when social deprivation was fully included in the model, we observed lower age-based attack rates and greater delay in epidemic peak week in the most socially deprived neighborhoods. Insights from simulation models complement current understandings from statistical-based association studies. Additional insights from our study are: (1) heterogeneous spatial arrangement of neighborhoods is a necessary condition for simulating observed disparities in the burden of influenza and (2) unmeasured factors may lead to a better quantitative match between simulated and observed rate ratio in the burden of influenza between the most and least socially deprived populations. Copyright © 2015 The Authors. Published by Elsevier B.V. All rights reserved.
    Full-text · Article · Mar 2015
    • CHWs/MCHBs are also well-placed to identify emerging needs among communities and to aid the health system in preventing families from falling through service gaps. Undervaluing CHWs/MCHBs' work, however, can result in reduced uptake of services , compelling marginalized populations to delay seeking care or to present at emergency departments (Khandor et al. 2011), resulting in higher costs for the healthcare system (Crighton et al. 2007). By implication, federal, provincial and territorial health ministries should consider formal recognition of CHWs/MCHBs as part of the health workforce.
    [Show abstract] [Hide abstract] ABSTRACT: This article reports findings from an applied case study of collaboration between a community-based organization staffed by community health workers/multicultural health brokers (CHWs/MCHBs) serving immigrants and refugees and a local public health unit in Alberta, Canada. In this study, we explored the challenges, successes and unrealized potential of CHWs/MCHBs in facilitating culturally responsive access to healthcare and other social services for new immigrants and refugees. We suggest that health equity for marginalized populations such as new immigrants and refugees could be improved by increasing the role of CHWs in population health programs in Canada. Furthermore, we propose that recognition by health and social care agencies and institutions of CHWs/MCHBs, and the role they play in such programs, has the potential to transform the way we deliver healthcare services and address health equity challenges. Such recognition would also benefit CHWs and the populations they serve. Copyright © 2014 Longwoods Publishing.
    Full-text · Article · Aug 2014
    • Crighton (35) investigated the relationship between mean annual temperature and pneumonia and influenza hospitalizations in Ontario, Canada, 1992–2001. An increased temperature was associated with a decreased number of pneumonia cases among both men and women over 65 years.
    [Show abstract] [Hide abstract] ABSTRACT: Background: The Arctic and subarctic area are likely to be highly affected by climate change, with possible impacts on human health due to effects on food security and infectious diseases. Objectives: To investigate the evidence for an association between climatic factors and infectious diseases, and to identify the most climate-sensitive diseases and vulnerable populations in the Arctic and subarctic region. Methods: A systematic review was conducted. A search was made in PubMed, with the last update in May 2013. Inclusion criteria included human cases of infectious disease as outcome, climate or weather factor as exposure, and Arctic or subarctic areas as study origin. Narrative reviews, case reports, and projection studies were excluded. Abstracts and selected full texts were read and evaluated by two independent readers. A data collection sheet and an adjusted version of the SIGN methodology checklist were used to assess the quality grade of each article. Results: In total, 1953 abstracts were initially found, of which finally 29 articles were included. Almost half of the studies were carried out in Canada (n=14), the rest from Sweden (n=6), Finland (n=4), Norway (n=2), Russia (n=2), and Alaska, US (n=1). Articles were analyzed by disease group: food- and waterborne diseases, vector-borne diseases, airborne viral- and airborne bacterial diseases. Strong evidence was found in our review for an association between climatic factors and food- and waterborne diseases. The scientific evidence for a link between climate and specific vector- and rodent-borne diseases was weak due to that only a few diseases being addressed in more than one publication, although several articles were of very high quality. Air temperature and humidity seem to be important climatic factors to investigate further for viral- and bacterial airborne diseases, but from our results no conclusion about a causal relationship could be drawn. Conclusions: More studies of high quality are needed to investigate the adverse health impacts of weather and climatic factors in the Arctic and subarctic region. No studies from Greenland or Iceland were found, and only a few from Siberia and Alaska. Disease and syndromic surveillance should be part of climate change adaptation measures in the Arctic and subarctic regions, with monitoring of extreme weather events known to pose a risk for certain infectious diseases implemented at the community level.
    Full-text · Article · Jul 2014
    • CHWs/MCHBs are also well-placed to identify emerging needs among communities and to aid the health system in preventing families from falling through service gaps. Undervaluing CHWs/MCHBs' work, however, can result in reduced uptake of services , compelling marginalized populations to delay seeking care or to present at emergency departments (Khandor et al. 2011), resulting in higher costs for the healthcare system (Crighton et al. 2007). By implication, federal, provincial and territorial health ministries should consider formal recognition of CHWs/MCHBs as part of the health workforce.
    [Show abstract] [Hide abstract] ABSTRACT: This article provides results from an empirical case study that showcases a community health worker practice targeting immigrants and refugees in Canada. The study focuses on the Multicultural Health Brokers practice, which offers an innovative approach to health promotion outreach and community development addressing broad social determinants of health. This article offers new evidence of both the role of community health worker interventions in Canada and community health workers as an invisible health and human services workforce. It also discusses the Multicultural Health Brokers contribution both to the "new public health" vision in Canada and to a practice that fosters feminist urban citizenship.
    Full-text · Article · Jan 2014
Show more