do healthy and unhealthy behaviours cluster in new zealand?
TRANSCRIPT
2007 vol. 31 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 155© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
The role of health-related behaviours
as risk and protective factors in the
causation and prevention of chronic
disease is widely recognised.1 Yet few studies
internationally,2-9 and to our knowledge
none in New Zealand, have described the
population distribution of clustering of
health-related behaviours within individuals.
Such information about the distribution of
‘lifestyles’ within the population may be
helpful in designing and evaluating health
promotion policies and programs.
The relatively few published studies2-9 have
varied in their selection of behaviours, their
measure of clustering, and their analytical
approach. Some have reported only the
number of co-occurring behaviours rather
than specific behavioural patterns. Others
have privileged one behaviour over others
and defined clustering only in relation to this
‘central’ behaviour (most usually tobacco
use). Most studies have focused exclusively
on unhealthy behaviours to the neglect of
healthy behaviours.
The objective of the current study is
to describe the clustering of both healthy
and unhealthy behaviours in the New
Zealand population today using a symmetric
analytical approach. Since human behaviour
does not occur in isolation from its social
and cultural context, we examine how the
clustering of behaviours (co-occurrence of
Do healthy and unhealthy behaviours
cluster in New Zealand?
Martin TobiasPublic Health Intelligence, New Zealand Ministry of Health
Gary JacksonCounties Manukau District Health Board, New Zealand
Li-Chia Yeh, Ken HuangPublic Health Intelligence, New Zealand Ministry of Health
Abstract
Objective: To describe the co-occurrence
and clustering of healthy and unhealthy
behaviours in New Zealand.
Method: Data were sourced from the
2002/03 New Zealand Health Survey.
Behaviours selected for analysis were
tobacco use, quantity and pattern of alcohol
consumption, level of physical activity, and
intake of fruit and vegetables. Clustering
was defined as co-prevalence of behaviours
greater than that expected based on the
laws of probability. Co-occurrence was
examined using multiple logistic regression
modelling, while clustering was examined in
a stratified analysis using age and (where
appropriate) ethnic standardisation for
confounding control.
Results: Approximately 29% of adults
enjoyed a healthy lifestyle characterised
by non-use of tobacco, non- or safe use
of alcohol, sufficient physical activity and
adequate fruit and vegetable intake. This
is only slightly greater than the prevalence
expected if all four behaviours were
independently distributed through the
population i.e. little clustering of healthy
behaviours was found. By contrast, 1.5%
of adults exhibited all four unhealthy
behaviours and 13% exhibited any
combination of three of the four unhealthy
behaviours. Unhealthy behaviours were
more clustered than healthy behaviours,
yet Maori exhibited less clustering of
unhealthy behaviours than other ethnic
groups and no deprivation gradient was
seen in clustering.
Discussion: The relative lack of clustering
of healthy behaviours supports single issue
universal health promotion strategies at the
population level. Our results also support
targeted interventions at the clinical level
for the 15% with ‘unhealthy lifestyles’. Our
finding of only limited clustering of unhealthy
behaviours among Maori and no deprivation
gradient suggests that clustering does not
contribute to the greater burden of disease
experienced by these groups.
Key words: Behaviour, clustering, lifestyle,
health promotion.
(Aust N Z J Public Health. 2007; 31:155-63)
doi:10.1111/j.1753-6405.2007.00034.x
Submitted: September 2006 Revision requested: January 2007 Accepted: February 2007
Correspondence to: Dr Martin Tobias, Ministry of Health, PO Box 5013, Wellington, New Zealand. Fax: +64 4 495 4401; e-mail: [email protected]
behaviours in the same individual greater
than would be expected by chance) varies
across socio-demographic groups.
MethodsNew Zealand Health Survey 2002/03
The New Zealand Health Survey (NZHS)
was conducted from August 2002 to October
2003. The target population was the usually
resident, non-institutionalised civilian
adult population (aged 15 years and over)
living in permanent private dwellings. A
stratified multistage cluster sample design
was employed using an area-based sampling
frame. Details of the survey design and
analysis, including calculation of integrated
survey weights and standard errors for
estimates, are reported elsewhere.10
For the present analysis, survey participants
self-reporting any present or past cardio-
vascular disease or cancer (any type) were
excluded, since diagnosis of such diseases
may lead to behaviour change (e.g. smoking
cessation). This left 10,241 participants
for analysis, comprising 3,416 Maori, 821
Pacific and 6,005 European/Others.
Selection of behavioursHealth-related behaviours established as
major risk or protective factors for chronic
disease, in particular cardiovascular disease
Article Surveys
156 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2007 vol. 31 no. 2© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
and cancer, include tobacco use, alcohol use, diet and physical
activity.11
The healthy and unhealthy behaviours included in this study,
together with their definitions and abbreviations, are summarised
in Table 1. For the purposes of this paper we define a ‘healthy
lifestyle’ as one that incorporates all four healthy behaviours, and
an ‘unhealthy lifestyle’ as one that features any three of the four
unhealthy behaviours listed or all four of these behaviours.
The definition of ‘current smoking’ excludes non-daily smokers
because the self-reporting of occasional or non-daily use is
less robust than that of daily use.12 The Alcohol Use Disorders
Identification Test (AUDIT) is a widely used eight-item scale
identifying potentially hazardous drinking.13 A score of eight or
more is generally accepted as indicative of a harmful quantity or
pattern of drinking, although some investigators prefer a cut-point
of seven for females.13 The threshold for physical (in)activity of
150 minute per week of activity equivalised to moderate intensity
is widely accepted as indicating sufficient physical activity for
health.14 Five or more servings of fruit plus vegetables per day is
used in the New Zealand Health Survey as a proxy measure for
a healthy diet and correlates reasonably well with more compre-
hensive dietary assessments such as the Healthy Eating Index.15
Measurement of clusteringCorrelation between risk factors does not necessarily imply
clustering.3 Rather, clustering exists when the observed distribution
of the risk factors differs from that expected assuming the risk
factors to be independent of each other.3
The expected joint prevalence of any set of risk factors is
simply the product of the individual risk factor prevalences.3 For
example, if the prevalence of smoking is 20% and the prevalence
of drinking is 30%, then by the laws of probability the expected
joint prevalence of [smoking + drinking] is 6%. If the observed
prevalence exceeds the expected prevalence, clustering is said
to occur.
Analytical approachFirst, the observed prevalence of each singular behaviour and of
all possible behavioural patterns (co-occurrences of behaviours)
is briefly described.
Second, the association of each behavioural pattern with socio-
demographic variables is analysed by multiple logistic regression
modelling. The model adjusts for age (in four categories: 15-24,
25-44, 45-64, 65+), sex, ethnicity (three categories as indicated
Table 1: Selected health-related behaviours.
Healthy behaviours Definitions Unhealthy Definitions behavioursNon-smoking Never smoker, ex-smoker, Smoking One or more cigarettes per day or non-daily smoker
Healthy drinking Abstainer or AUDIT score 7 or less Unhealthy drinking AUDIT score 8 or above (both sexes)
Physically active 150 or more mins per week of activity Inactive <150 mins per week of activity equivalised equivalised to moderate intensity to moderate intensity (~3 METs)
Healthy diet 5 or more servings of fruit plus Unhealthy diet <5 servings of fruit plus vegetables per day vegetables per day
Healthy lifestyle All 4 of the above Unhealthy lifestyle All 4, or any combination of 3 of the above
above) and deprivation as a measure of socio-economic position
(five categories: quintiles of the NZDep2001 deprivation index,
1 being least deprived, 5 being most deprived). NZDep2001 is a
Census-based small area index of deprivation derived by principal
component analysis of nine socio-economic variables included in
the 2001 Census.16 The reference group for the regression models
is 25-44 years of age, male, European, deprivation quintile 5. While
age is adjusted for in all reported odds ratios, the effect of age
is not itself reported as this is biased by the exclusion of survey
respondents with diagnosed cardiovascular disease or cancer.
Finally, clustering (Observed/Expected prevalence ratio)
is analysed for each behavioural pattern separately by socio-
demographic subgroup (i.e. a stratified analysis). Confidence
intervals for the O/E or ‘clustering’ ratios were estimated by
standard parametric methods for Poisson random variables.
For all analyses, the prevalence or clustering of both healthy
and unhealthy behaviours is reported. These are not mirror images
of each other, because unhealthy behaviours are typically less
prevalent than their healthy counterparts – so higher degrees of
clustering are typically found for the former (since clustering
is defined as a ratio measure). While this could be corrected
by normalisation (re-scaling), this has not been done since it is
the actual degree of clustering that is important rather than the
comparison of clustering of healthy compared with clustering of
unhealthy behaviours.
Prevalence rates have been age standardised to the World Health
Organization (WHO) World Population17 by the direct method,
and ethnicity standardised to the New Zealand 2001 Census
population, where relevant. Ethnic standardisation is intended to
enable comparison of degrees of clustering by deprivation level
unconfounded by ethnicity. By contrast, comparison of clustering
by ethnicity does not require standardisation for deprivation
because deprivation is a mediator, not a confounder, of the
ethnicity/outcome relationship.
ResultsDescriptive epidemiologyHealthy behaviours
Table 2 shows the observed prevalence of single healthy
behaviours in the New Zealand adult (15+) population, as well
as the prevalence of a ‘healthy lifestyle’. Details of all healthy
behavioural patterns are shown in Table 6.
Tobias et al. Article
2007 vol. 31 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 157© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
Table 2: Healthy behaviour prevalence rates.
Numbera Prevalence (%) Non-smoking Healthy Physically Healthy Healthy drinking active diet lifestyleb
All (crude) 10,223 76.3 81.1 75.3 54.0 29.4
Age
15-24 1,517 73.2 67.1 76.1 44.9 20.5
25-44 4,664 72.4 80.5 74.5 51.9 26.2
45-64 2,918 80.0 87.9 78.6 60.0 36.8
65+ 1,124 90.2 94.7 66.8 66.7 41.9
Sex
Malec 3,961 75.1 70.9 80.4 44.3 22.4
Femalec 6,262 77.1 87.6 71.1 61.9 34.5
Ethnicity
Maoric 3,406 50.4 71.2 77.0 49.5 18.6
Pacificc 819 67.3 80.8 67.2 38.4 15.9
Otherc 5,998 79.8 80.5 75.9 54.7 30.6
SES
Dep Q1d 1,389 83.6 82.0 79.4 57.2 34.1
Dep Q2d 1,300 81.4 79.6 77.0 59.3 33.2
Dep Q3d 1,518 79.0 80.6 76.4 55.6 30.3
Dep Q4d 1,950 72.4 78.7 74.1 49.1 23.8
Dep Q5d 4,066 64.0 76.7 71.0 45.8 22.4Notes:(a) Number of respondents included in analysis.(b) Healthy lifestyle = all four behaviours reported.(c) Age standardised.(d) Age and ethnicity standardised.
Table 3: Unhealthy behaviour prevalence rates.
Numbera Prevalence (%) Smoking Unhealthy Insufficiently Unhealthy Unhealthy Unhealthy drinking active diet lifestyle – lifestyle – all 4 any 3 of 4 behaviours behavioursAll (crude) 10,223 23.7 18.9 24.7 46.0 1.4 13.6
Age
15-24 1,517 26.8 32.9 23.9 55.1 3.3 22.9
25-44 4,664 27.6 19.5 25.5 48.1 1.4 14.8
45-64 2,918 20.0 12.1 21.4 40.0 0.5 7.9
65+ 1,124 9.8 5.3 33.2 33.3
Sex
Maleb 3,961 24.9 29.1 19.6 55.7 1.8 17.2
Femaleb 6,262 22.9 12.4 28.9 38.1 1.4 11.8
Ethnicity
Maorib 3,406 49.6 28.8 23.0 50.5 2.4 26.3
Pacificb 819 32.7 19.2 32.8 61.6 4.0 28.1
Otherb 5,998 20.2 19.5 24.1 45.3 1.4 12.1
SES
DepQ1c 1,389 16.4 18.0 20.6 42.8 d d
DepQ2c 1,300 18.6 20.4 23.0 40.7 1.0 11.5
DepQ3c 1,518 21.0 19.4 23.6 44.4 0.7 10.0
DepQ4c 1,950 27.6 21.3 25.9 50.9 1.3 15.1
DepQ5c 4,066 36.0 23.3 29.0 54.2 3.3 25.9Notes:(a) Number of respondents included in analysis.(b) Age standardised.(c) Age and ethnicity standardised.(d) Count suppressed (<10).
Surveys Do healthy and unhealthy behaviours cluster?
158 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2007 vol. 31 no. 2© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
Overall, 76% of New Zealand adults are non-smokers, 81%
have a healthy drinking pattern, 75% are sufficiently physically
active for health, and 54% consume a healthy diet (as indexed
by sufficient fruit and vegetable intake). However, this overall
assessment disguises substantial variations between population
subgroups, with older people more likely to exhibit healthy
behaviours (except for physical activity) than youth; females
than males (again except for physical activity); European/Others
than Maori or Pacific people; and deprivation quintile 1 (the least
deprived) than quintile 5 (the most deprived).
Almost 30% of adults report a healthy lifestyle (as defined).
However, this again varies with age, from 20% in youth to more
than 40% in older people. It also varies with sex, being higher in
females (35%) than males (22%) after standardising for age. Maori
and Pacific ethnic groups have lower prevalence of the healthy
lifestyle (19% and 16% respectively) than European/Others (31%),
adjusting for age but not deprivation. A socio-economic gradient
was found, with the prevalence of the healthy lifestyle varying from
34% in deprivation quintile 1 (least deprived) to 22% in quintile
5 (most deprived), adjusting for age and ethnicity.
Unhealthy behaviours
Table 3 shows the observed prevalence of single unhealthy
behaviours in the New Zealand adult (15+) population, as well as
the prevalence of unhealthy lifestyle. Detailed prevalences of all
unhealthy behavioural patterns are shown in Table 7.
Overall, 24% of New Zealanders currently smoke cigarettes,
19% exhibit a hazardous drinking pattern, 25% are insufficiently
physically active for health, and 46% consume an unhealthy diet
(as indexed by inadequate intake of fruit and vegetables). As with
the corresponding healthy behaviours, differences are seen with
age, sex, ethnicity and deprivation as shown in Table 3.
Turning to the unhealthy lifestyle construct, approximately 15%
of the population may be characterised as having an unhealthy
lifestyle (defined as any three or all four unhealthy behaviours),
although only 1.5% of the population acknowledge all four
unhealthy behaviours. The prevalence of an unhealthy lifestyle
varies with age (26% in youth, 8% in middle-aged adults), sex
Table 4: Odds ratios and 95% CIs for healthy behaviour patterns.
S-A-P-D A-P-D S-P-D S-A-D S-A-PFemale 1.83 1.91 1.47 2.39 1.18 (1.60-2.09) (1.69-2.16) (1.30-1.67) (2.11-2.71) (1.05-1.34)Maori 0.57 0.89 0.62 0.47 0.47 (0.44-0.73) (0.72-1.09) (0.48-0.79) (0.37-0.59) (0.39-0.57)Pacific 0.52 0.65 0.49 0.50 0.86 (0.39-0.71) (0.48-0.88) (0.36-0.67) (0.38-0.67) (0.64-1.15)Q1 1.83 1.71 1.76 1.81 1.74 (1.41-2.38) (1.34-2.18) (1.35-2.29) (1.41-2.33) (1.39-2.17)Q2 1.84 1.75 1.80 1.84 1.50 (1.42-2.38) (1.37-2.24) (1.38-2.35) (1.42-2.38) (1.20-1.88)Q3 1.55 1.47 1.48 1.63 1.32 (1.21-1.99) (1.16-1.86) (1.16-1.90) (1.29-2.07) (1.05-1.67)Q4 1.10 1.10 1.09 1.17 1.11 (0.85-1.41) (0.88-1.38) (0.86-1.38) (0.91-1.50) (0.89-1.37)Notes:Reference group is male, European/Other, Deprivation Quintile 5.S=non-smoking; A=healthy drinking; P= sufficient physical activity; D=healthy diet.Two-behaviour patterns not shown (available from first author on request).
(19% in males, 13% in females, age adjusted), ethnicity (29%
in Maori, 32% in Pacific people, 13.5% in Others, adjusting for
age) and deprivation quintile (from 13% in quintile 2 to 29% in
quintile 5, adjusting for age and ethnicity).
Co-occurrence: regression analysisHealthy behaviours
Table 4 summarises the model output for all healthy behaviour
patterns (i.e. co-occurrence, not clustering); two-behaviour patterns
are not shown but are available from the first author on request.
Females are almost twice as likely as males to enjoy a healthy
lifestyle (odds ratio 1.8). Females also have signif icantly
higher prevalences of all three and two-behaviour patterns (co-
occurrences) except for the [non-smoking + sufficient activity]
pattern that occurs 20% less often than in males, a difference that
is also statistically significant at the 95% level.
Maori are less likely than European/Others to exhibit healthy
lifestyles and all three- and two-behaviour patterns except for
[physically active + healthy diet], for which there is no difference.
The results for Pacific people are essentially the same, although
not all differences are statistically significant – perhaps due to the
relatively smaller numbers in the sample.
A clear socio-economic gradient in the prevalence of healthy
lifestyles and all three- and two-behaviour patterns is evident from
the deprivation analysis, adjusting for age, sex and ethnicity.
Unhealthy behaviours
Table 5 summarises the model output for all unhealthy behaviour
patterns (i.e. co-occurrence, not clustering); two-behaviour
patterns are not shown but are available from the first author on
request.
Females are less likely than males to show an unhealthy lifestyle,
although not all differences are statistically significant. Females are
significantly less likely than males to show all of the two-behaviour
patterns except [smoking + insufficient physical activity] (which
they are significantly more likely to exhibit) and [unhealthy diet
+ insufficient physical activity] for which the gender difference
is not statistically significant.
Tobias et al. Article
2007 vol. 31 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 159© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
Table 5: Odds ratios and 95% CIs for unhealthy behaviour patterns.
S-A-P-D A-P-D S-P-D S-A-D S-A-PFemale 0.67 0.42 1.18 0.42 0.71 (0.42-1.08) (0.29-0.62) (0.86-1.62) (0.32-0.55) (0.48-1.05)Maori 1.48 1.27 1.80 2.44 1.68 (0.67-3.29) (0.69-2.33) (1.17-2.78) (1.68-3.55) (0.84-3.36)Pacific 1.91 1.56 1.53 1.65 2.10 (0.83-4.38) (0.83-2.94) (0.90-2.61) (0.87-3.12) (0.97-4.54)Q1 0.55 0.60 0.31 0.50 0.47 (0.21-1.42) (0.31-1.20) (0.16-0.60) (0.27-0.92) (0.18-1.23)Q2 0.39 0.70 0.31 0.54 0.48 (0.15-1.03) (0.36-1.36) (0.18-0.54) (0.31-0.96) (0.21-1.08)Q3 0.26 0.34 0.31 0.67 0.30 (0.09-0.76) (0.15-0.75) (0.19-0.52) (0.39-1.17) (0.12-0.74)Q4 0.46 0.71 0.56 0.68 0.70 (0.21-1.02) (0.40-1.26) (0.36-0.86) (0.43-1.09) (0.34-1.45)Notes:Reference group is male, European/Other, Deprivation Quintile 5. S=current daily smoking; A=unhealthy drinking; P=insufficient activity for health; D=unhealthy diet (inadequate fruit and vegetable intake).Two behaviour patterns not shown (available from first author on request).
Confidence intervals for Maori and Pacific ethnic groups are
wide, making it difficult to demonstrate significant differences.
However, the results are suggestive of higher prevalences for all
unhealthy behaviour patterns. Unlike the situation for healthy
behaviours, there is less of a clear gradient in unhealthy behaviour
patterns across the deprivation quintiles after adjustment for age,
sex and ethnicity. There is a tendency, however, for quintiles 1-4, and
especially quintiles 1-3, to have lower (sometimes significantly lower)
prevalences of all unhealthy behaviour patterns than quintile 5.
Clustering: stratified analysisHealthy behaviours
Table 6 shows the observed and expected prevalences for all
healthy behaviour patterns (except for the two-behaviour patterns
– available from the first author on request), and the corresponding
O/E ratios (‘clustering ratios’). Confidence intervals for the
ratios are not shown for clarity (available from the first author on
request), but those with p values <0.05 are identified.
In the total New Zealand population, a small but statistically
significant degree of clustering is seen for most healthy behaviour
patterns (i.e. the prevalence of co-occurrence is typically greater
than would be expected if the component behaviours were
independent). The healthy lifestyle pattern shows the highest
degree of clustering (a ratio of 1.17, 95% confidence interval
1.13-1.21). That is, the proportion of individuals exhibiting all
four behaviours was 29%, which is 17% higher than the 25%
that would have been expected had the four component healthy
behaviours been independently distributed across the population.
All three-behaviour patterns show clustering ratios ranging from
1.05 to 1.15. Less clustering is seen for the two-behaviour patterns,
with [non-smoking + sufficient activity] and [healthy drinking +
sufficient activity] showing no clustering at all.
Clustering varies with age, with younger age groups showing
more clustering than older age groups. Indeed, older people
(65+) show no clustering at all except for overall healthy lifestyle
(clustering ratio 1.10, 95% CI 1.01-1.20). Few significant gender
differences were found after standardising for age.
Maori demonstrate higher degrees of clustering of healthy
behaviours than do European/Others. Healthy lifestyle has an O/E
ratio of 1.36 (CI 1.26-1.46) in Maori compared with only 1.15 (CI
1.09-1.20) in European/Others. Other major differences include
[non-smoking + healthy drinking + healthy diet]: 1.27 versus 1.14;
[non-smoking + sufficiently active + healthy diet]: 1.21 versus 1.08;
[non-smoking + healthy drinking]: 1.10 versus 1.05; and [non-
smoking + healthy diet]: 1.11 versus 1.05 respectively.
Little if any difference in the nature or extent of clustering of
healthy behaviours is seen across deprivation quintiles 1-4, but
quintile 5 displays higher degrees of clustering for some patterns,
including the healthy lifestyle pattern itself (1.25 versus 1.11 for
quintile 1). Compared with quintile 1, quintile 5 also exhibits
substantial clustering in regard to [non-smoking + sufficiently
active + healthy diet]: 1.17 versus 1.05; and [non-smoking +
healthy drinking]: 1.10 versus none (1.00).
Unhealthy behaviours
Table 7 shows the observed and expected prevalences for
all unhealthy behaviour patterns (other than the two-behaviour
patterns, which are available from the first author on request), and
the corresponding O/E (‘clustering’) ratios. Confidence intervals
for the ratios are not shown for clarity (available from the first
author on request), but those with p values <0.05 are identified.
For the total New Zealand population, moderate or high
degrees of clustering are seen for some patterns (especially the
four- and three-behaviour patterns), with slight or no clustering
for most of the two-behaviour patterns. The ‘four behaviour
unhealthy lifestyle’ pattern is highly clustered (O/E ratio 2.75,
CI 2.29-3.21). Almost as much clustering is found for one of the
three-behaviour unhealthy lifestyle patterns, namely [smoking +
unhealthy drinking + unhealthy diet] (clustering ratio 2.43, CI
2.19-2.66). The remaining three-behaviour unhealthy lifestyle
patterns show moderate to slight clustering. One two-behaviour
pattern – [smoking + unhealthy drinking] – also shows a high
degree of clustering (ratio 1.79, CI 1.65-1.92).
Males and females show the same overall pattern, but the degree
of clustering of unhealthy behaviours (when present) is always
Surveys Do healthy and unhealthy behaviours cluster?
160 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2007 vol. 31 no. 2© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
greater for females than males. For example, the four-behaviour
unhealthy lifestyle pattern has a clustering ratio of 4.48 in females
compared with 2.28 in males, after adjusting for age. Other major
gender differences are seen with the three-behaviour pattern
[smoking + unhealthy drinking + unhealthy diet]: 3.24 versus 1.81;
and the two-behaviour pattern [smoking + unhealthy drinking]:
2.11 versus 1.57 respectively.
There are few differences in the extent or nature of clustering
with age, at least to 64 years. Not all behaviour patterns could be
assessed in the 65+ age group because of small numbers in some
Table 6: Clustering of healthy behaviours.
S-A-P-D S-A-P S-A-D S-P-D A-P-DAll (crude) Observed 29.4 48.7 38.5 34.0 35.0 Expected 25.2 46.6 33.4 31.0 33.0 Ratio 1.17a 1.05a 1.15a 1.10a 1.06a
15-24 years Observed 20.5 40.3 25.7 28.0 24.7 Expected 16.8 37.4 22.1 25.0 22.9 Ratio 1.22a 1.08a 1.17a 1.12a 1.0825-44 years Observed 26.2 45.4 35.5 30.2 32.6 Expected 22.5 43.4 30.2 28.0 31.1 Ratio 1.16a 1.05a 1.17a 1.08a 1.05a
45-64 years Observed 36.8 57.1 45.9 41.1 42.9 Expected 33.2 55.3 42.2 37.7 41.5 Ratio 1.11a 1.03 1.09a 1.09a 1.0365+ years Observed 41.9 57.9 59.0 43.6 45.4 Expected 38.1 57.1 57.0 40.2 42.2 Ratio 1.10a 1.01 1.04 1.08 1.08Maleb Observed 22.4 45.8 27.4 29.0 26.7 Expected 19.0 42.8 23.6 26.7 25.3 Ratio 1.18a 1.07a 1.16a 1.08a 1.06Femaleb Observed 34.5 50.0 46.6 37.9 40.9 Expected 29.7 48.0 41.8 33.9 38.6 Ratio 1.16a 1.04a 1.11a 1.12a 1.06a
Maorib Observed 18.6 30.6 22.6 23.2 30.2 Expected 13.7 27.6 17.8 19.2 27.1 Ratio 1.36a 1.11a 1.27a 1.21a 1.11a
Pacificb Observed 15.9 42.1 22.9 17.9 22.0 Expected 14.0 36.5 20.9 17.4 20.9 Ratio 1.13 1.15a 1.10 1.03 1.06Otherb Observed 30.6 50.4 39.9 35.8 35.2 Expected 26.7 48.8 35.1 33.1 33.4 Ratio 1.15a 1.03a 1.14a 1.08a 1.05a
Quintile 1c Observed 34.1 55.1 42.4 39.1 38.5 Expected 29.4 52.2 37.2 36.5 35.9 Ratio 1.15a 1.03 1.11a 1.07 1.06Quintile 2c Observed 33.2 51.0 42.6 38.1 38.7 Expected 28.5 49.9 37.3 35.8 35.5 Ratio 1.15a 1.04 1.13a 1.06 1.09Quintile 3c Observed 30.3 48.4 39.9 35.4 35.6 Expected 26.5 47.9 34.7 33.1 33.8 Ratio 1.14a 1.01 1.14a 1.07 1.05Quintile 4c Observed 23.8 44.3 32.3 28.9 29.0 Expected 20.8 42.3 28.0 26.5 28.6 Ratio 1.14a 1.04 1.15a 1.09a 1.01Quintile 5c Observed 22.4 40.5 29.3 27.4 28.0 Expected 17.1 35.5 23.9 22.7 26.0 Ratio 1.31a 1.14a 1.22a 1.21a 1.08a
Notes:(a) p value <0.05. (b) Age standardised.(c) Age and ethnicity standardised.S=non smoking; A=healthy drinking; P=sufficient physical activity; D=healthy diet.
cells. However, it does appear that clustering of the two-behaviour
pattern [smoking + unhealthy drinking] does increase with age,
from a clustering ratio of 1.58 in youth to 1.93 in older people.
Pacific people show much the same pattern and extent of
clustering as European/Others. However, Maori consistently
show less clustering of unhealthy behaviours than the other ethnic
groups (despite showing more clustering of healthy behaviours).
For example, the four-behaviour unhealthy lifestyle is clustered to
an O/E ratio of only 1.45 in Maori versus 3.15 in Pacific people
and 3.26 in European/Others. As another example, the [smoking
Tobias et al. Article
2007 vol. 31 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 161© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
+ unhealthy drinking] pattern has a clustering ratio of only 1.26
in Maori (although this is still statistically significant) compared
with 1.85 in both of the other ethnic groups.
There is very little if any evidence for a gradient in clustering
across deprivation quintiles. If anything, for several unhealthy
behaviour patterns the degree of clustering tends to be slightly
greater in quintiles 1-2 than in the other quintiles. For example, the
important two-behaviour pattern [smoking + unhealthy drinking]
has a clustering ratio of 2.26 in Q1, 1.79 in Q2, 1.79 in Q3, 1.67
in Q4 and 1.59 in Q5.
Table 7: Clustering of unhealthy behaviours.
S-A-P-D S-A-P S-A-D S-P-D A-P-DAll (crude) Observed 1.4 2.0 5.0 4.0 2.6 Expected 0.5 1.1 2.1 2.7 2.1 Ratio 2.75a 1.81a 2.43a 1.49a 1.21a
15-24 years Observed 3.3 4.2 8.8 5.5 4.4 Expected 1.2 2.1 4.9 3.5 4.3 Ratio 2.84a 1.99a 1.81a 1.56a 1.0225-44 years Observed 1.4 2.0 5.5 4.4 2.9 Expected 0.7 1.4 2.6 3.4 2.4 Ratio 2.12a 1.46a 2.12a 1.30a 1.21a
45-64 years Observed 0.5 0.9 2.9 2.9 1.2 Expected 0.2 0.5 1.0 1.7 1.0 Ratio 2.41 1.74a 3.00a 1.69a 1.1665+ years Observed – – – 1.7 1.3 Expected – – – 1.1 0.6 Ratio – – – 1.57 2.22a
Maleb Observed 1.8 2.4 7.3 3.7 3.8 Expected 0.8 1.4 4.0 2.7 3.2 Ratio 2.28a 1.69a 1.81a 1.36a 1.20a
Femaleb Observed 1.4 2.0 3.5 4.5 1.8 Expected 0.3 0.8 1.1 2.5 1.4 Ratio 4.48a 2.44a 3.24a 1.78a 1.32a
Maorib Observed 2.4 3.8 10.9 8.0 3.6 Expected 1.7 3.3 7.2 5.8 3.3 Ratio 1.45a 1.16 1.51a 1.39a 1.08Pacificb Observed 4.0 5.2 8.4 8.7 5.4 Expected 1.3 2.1 3.9 6.9 3.9 Ratio 3.15a 2.53a 2.17a 1.32a 1.39a
Otherb Observed 1.4 1.8 4.5 3.3 2.5 Expected 0.4 0.9 1.8 2.2 2.1 Ratio 3.26a 1.90a 2.52a 1.50a 1.17a
Quintile 1c Observed – 1.8 4.8 2.8 2.9 Expected – 0.7 1.5 1.6 1.8 Ratio – 2.57a 3.20a 1.75a 1.64a
Quintile 2c Observed 1.2 1.7 4.2 2.8 2.8 Expected 0.4 0.9 1.7 1.9 2.0 Ratio 3.10a 1.83a 2.46a 1.41a 1.41a
Quintile 3c Observed 0.7 1.0 5.2 2.4 1.4 Expected 0.5 1.0 1.9 2.2 2.1 Ratio 1.40 1.04 2.76a 1.09 0.70Quintile 4c Observed 1.3 2.3 5.3 4.5 3.0 Expected 0.8 1.5 3.0 3.6 2.8 Ratio 1.68 1.51a 1.80a 1.24 1.03Quintile 5c Observed 3.3 4.3 8.5 8.3 4.8 Expected 1.2 2.4 4.4 5.0 3.6 Ratio 2.75a 1.77a 1.94a 1.66a 1.33a
Notes:(a) p value <0.05.(b) Age standardised.(c) Age and ethnicity standardised.S=smoking; A=unhealthy drinking; P=insufficient physical activity; D=unhealthy diet.Rates were not calculated when counts less than 10.
DiscussionConcept of clustering
The concept of clustering is not clearly defined in the literature,
and different authors have theorised and operationalised this
construct in different ways. An important contribution of this study
has been to clarify the definition of clustering as co-occurrence
greater than that expected by chance. Clustering thus implies that
the distribution of the risk factors concerned is not independent of
each other, but instead reflects a common, more distal, determinant.
Surveys Do healthy and unhealthy behaviours cluster?
162 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2007 vol. 31 no. 2© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
Thus finding clustering – or not – has both policy and research
implications, some of which are elaborated below.
Healthy behavioursOur study finds that, in 2002/03, approximately 29% of adult
New Zealanders enjoyed a healthy lifestyle characterised by non-
use of tobacco, safe or non-use of alcohol, sufficient physical
activity for health, and a healthy diet as indexed by adequate
fruit and vegetable consumption. This is very similar to rates
reported for Australia,5 the United Kingdom3 and Finland,2 despite
differences between studies in the definitions and thresholds for
these health-related behaviours. It is higher than rates reported
for the United States,6-8 with better physical activity and fruit and
vegetable consumption rates being reported for New Zealand, but
again differing definitions make the comparison inexact. While
in line with at least some other developed countries, and perhaps
higher than many public health workers might have predicted,
our results imply that more than two-thirds of adults are still
not meeting even modest thresholds for healthy living. Or more
positively, this finding reveals the great scope still available for
health gain via improvements in lifestyle.
Of even more interest, we found that healthy behaviours show
little clustering within individuals. For example, independent
distribution of all four health-related behaviours selected for
study would have yielded a joint prevalence of 25% rather than
the 29% actually observed – indicating only 17% clustering
(i.e. a clustering or observed/expected ratio of 1.17). Even less
clustering was found for all the three-behaviour and two-behaviour
patterns. Such low levels of clustering of healthy behaviours has
also been found in other studies.6,8 This finding has implications,
for example, for research into the health benefits of fruit and
vegetable consumption. Such research has often been criticised
on the grounds that fruit and vegetable consumption is merely
a marker of a healthy lifestyle, so any association between such
consumption and health outcomes may be subject to residual
confounding by other dimensions of a healthy lifestyle (e.g. non-
smoking, healthy drinking, sufficient physical activity) rather
than reflecting a true dietary impact. Our analysis, finding little
or no clustering of healthy behaviours within individuals, refutes
this hypothesis.
Unhealthy behavioursOur study found that, in 2002/03, less than 1.5% of adult New
Zealanders exhibited all four of the selected unhealthy behaviours
(i.e. daily tobacco use, potentially hazardous alcohol consumption
in terms of volume or drinking pattern, insufficient physical
activity for health, and inadequate intake of fruit and vegetables
as an indicator of wider dietary pattern). Again, this is similar to
the rate found in some other developed countries,2-5 especially
if allowance is made for differences in variable definition,
behavioural thresholds, and study design. Using a wider definition
of unhealthy lifestyle to include any three as well as all four
unhealthy behaviours, then 15% of the adult population would
be so categorised. Youth (23%), males (17%), people living in
the most deprived 20% of small areas (26%), Maori (26%) and
Pacific people (28%) exhibited higher-than-average prevalences
of these behaviour patterns. For the latter three socio-demographic
groups the proportion with an unhealthy lifestyle was higher than
the corresponding proportion with a healthy lifestyle.
Turning to clustering, even the low prevalence of the ‘four-
behaviour unhealthy lifestyle’ observed (approximately 1.5%)
is almost three times higher than the 0.5% prevalence expected
given independent risk factor distributions. The three-behaviour
unhealthy lifestyles show lesser but still substantial degrees of
clustering. The two-behaviour patterns show progressively less
clustering than the three-behaviour patterns with the exception of
the co-occurrence of tobacco and potentially hazardous alcohol
consumption, which co-occurs almost twice as often as expected,
a finding that has been reported previously in other countries.2,4,5,7,9
Interestingly, insufficient physical activity and unhealthy diet
(insufficient fruit and vegetable intake) were not found to cluster
to more than a very minor degree – contrary to what might have
been hypothesised if physical inactivity contributed substantively
to the obesity epidemic.
Unlike the situation with healthy behaviours, Maori exhibited
less clustering than other ethnic groups for most unhealthy
behaviour patterns. While Maori experience higher prevalences of
most of the individual risk factors,10 recent research suggests that
unhealthy behaviours – including tobacco use – make a smaller
contribution to the ethnic disparity in health than previously
estimated.18 Our study suggests that clustering of unhealthy
behaviours makes no contribution to the Maori/non-Maori health
disparity at all.
Analogously, it is generally the case that more disadvantaged
or lower status groups have higher risk factor prevalences than
their more powerful and privileged counterparts, yet differences
in the distribution of singular unhealthy behaviours explain
relatively little of the socio-economic gradient in ischaemic heart
disease.19 It has been hypothesised that greater clustering of risk
factors might explain more of the gradient,3 yet our results do not
support the notion of a social gradient in clustering of unhealthy
behaviours.
Health system Our findings on clustering (or lack thereof) of both healthy
and unhealthy behaviours are new (for New Zealand) and have
implications for health promotion policy and practice both
in New Zealand and abroad. Conventional population-based
health promotion programs have tended to focus on single
issues, while individual-based (clinical) services tend towards
the case management approach of dealing with multiple factors
simultaneously (e.g. absolute cardiovascular risk). The current
work supports both these approaches. The relative lack of
clustering of healthy behaviours shows that single issue initiatives
(e.g. separately promoting fruit and vegetable consumption and
walking) are generally appropriate for most universal strategies
(because the lack of substantial clustering means that little spin-
off can be expected from one dimension of health to another).
Conversely, the clustering of multiple unhealthy behaviours in
Tobias et al. Article
2007 vol. 31 no. 2 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 163© 2007 The Authors. Journal Compilation © 2007 Public Health Association of Australia
a relatively small number of people justifies a case management
approach to ‘high-risk’ strategies. Targeted chronic care
management and integrated care initiatives emanating from the
personal health sector will thus meet – and reinforce – universal
public health initiatives coming the other way.
Yet three caveats should be borne in mind in using our results
for planning either clinical or population-based preventive
interventions.
First, exclusion of survey participants with a history of
cardiovascular disease or cancer may have introduced bias,
especially for older males. It is unlikely, however, that this could
have been sufficient to affect our conclusions regarding the extent
of clustering.
Second, different results may be obtained if different
behaviours are selected for study, if those behaviours are defined
and operationalised differently, and if different thresholds
for categorising the behaviours are chosen. Indeed, our
dichotomisation of behaviours represents a major simplification,
and a more nuanced approach that recognised varying degrees of
behavioural expression (e.g. light, moderate and heavy smoking)
might be preferable. Yet contrasts between groups and trends over
time in the extent of clustering may still be validly estimated,
provided the method is consistently applied.
Some analyses have included overweight/obesity (defined via
body mass index or waist circumference) in behaviour clustering
studies. Given that obesity is not a behaviour per se, we prefer to
use measures of physical activity and nutrition directly.
Finally, a policy focus on behavioural clustering risks being
counterproductive if it decontextualises human behaviour and
neglects the social forces that shape behavioural repertoires
and their expression in different settings. The lifestyle construct
as applied in this study does not imply free and unconstrained
behavioural choices by individuals or families. Provided this is
understood, we conclude that measurement and monitoring of
behavioural clustering can be a useful tool for health promotion,
helping to identify intervention points (‘over-represented’
behavioural patterns within socio-demographic and geographic
subgroups) and evaluate outcomes relating to such intervention
points.
We have shown that 29% of adults in New Zealand live
a relatively healthy lifestyle, being non-smokers, eating 5+
vegetables and fruit a day, being moderately active, and indulging
in non-harmful use (or non-use) of alcohol. Conversely, 15%
live a relatively unhealthy lifestyle, with any three or all four
unhealthy behaviours. There is clearly still much scope for
health gain via health promotion. Clustering tendencies were not
strong, especially for healthy behaviours, and did not display any
social gradient or ethnic disparity. Hence clustering of unhealthy
behaviours cannot explain any of the well-established ethnic
or socio-economic disparities in the burden of chronic disease.
Moreover, while a clinical case management approach seems
appropriate for the 15% of the population with unhealthy lifestyles,
promotion of healthy lifestyles via population-based programs will
need to focus on single issues and little spin-off from one issue to
another can be expected.
AcknowledgementsWe thank the 13,000 New Zealanders who freely gave of their
time to participate in the 2002/03 New Zealand Health Survey.
This report is published with the approval of the Deputy Director-
General (Public Health). However, opinions expressed are those of
the authors and do not necessarily reflect the views of the Ministry
of Health or the Counties Manukau District Health Board.
References1. Oppenheimer GM. Profiling risk: the emergence of coronary heart disease
epidemiology in the United States 1947-70. Int J Epidemiol. 2006;35:720-30.
2. Laaksonen M, Prattala R, Karisto A. Patterns of unhealthy behaviour in Finland. Eur J Public Health. 2001;11:294-300.
3. Ebrahim S, Montaner D, Lawler D. Clustering of risk factors and social class in childhood and adulthood. Br Med J. 2004;328:861-4.
4. Klein-Getlink J, Choi B, Fry R. Multiple exposures to smoking, alcohol, physical inactivity and overweight: Prevalences according to the Canadian Community Health Survey Cycle 1.1. Chronic Dis Canada. 2006;27:25-33.
5. Australian Institute of Health and Welfare. Living Dangerously: Australians with Multiple Risk Factors for Cardiovascular Disease. Canberra (AUST): AIHW; 2005.
6. Newsom J, McFarland B, Kaplan M, et al. The health consciousness myth: implications of the near independence of major health behaviours in the North American population. Soc Sci Med. 2004;60:433-7.
7. Fine L, Philogene G, Gramling R. Prevalence of multiple chronic disease risk factors: 2001 National Health Interview Survey. Am J Prev Med. 2004;27:18-24.
8. Reeves M, Rafferty A. Healthy lifestyle characteristics among adults in the United States, 2000. Arch Intern Med. 2005;165:854-7.
9. Riyami A, Afifi M. Clustering of cardiovascular risk factors among Omani adults. East Mediterr Health J. 2003;9:893-903.
10. Ministry of Health. A Portrait of Health. Key Results of the 2002/03 New Zealand Health Survey. Wellington (NZ): The Ministry; 2004.
11. Tobias M, Turley M, Paul S, et al. Debunking the ‘only 50%’ myth: prevalence of established risk factors in New Zealanders with self-reported ischaemic heart disease. Aust N Z J Public Health. 2005;29:405-11.
12. World Health Organization. Guidelines for Controlling and Monitoring the Tobacco Epidemic. Geneva (CHE): WHO; 1998.
13. Saunders JB, Aasland OG, Babor TF, et al. Development of the Alcohol Use Disorders Screening Test (AUDIT). WHO collaborative project on early detection of persons with harmful alcohol consumption. Addiction. 1993;88:791-804.
14. Department of Health and Human Services. 1996. The Report of the Surgeon General on Physical Activity and Health. Washington (DC): United States DHHS.
15. Weinstein S, Vogt TM, Gerrior SA. Healthy eating index scores are associated with blood nutrient concentration in the Third National Health and Nutrition Examination Survey. J Am Diet Assoc. 2004;104:576-84.
16. Crampton P, Salmond C, Kirkpatrick B. Degrees of Deprivation. 2nd ed. Wellington (NZ): David Bateman; 2004.
17. World Health Organization. Age Standardisation of Rates: A New WHO Standard. Geneva (CHE): WHO; 2000. GPE Discussion Paper No.: 31.
18. Blakely T, Fawcett J, Hunt D, et al. What is the contribution of smoking and socio-economic position to ethnic inequalities in mortality in New Zealand? Lancet. 2006;368(9529):4-6.
19. Marmot M, Wilkinson RG. Social Determinants of Health. Oxford (UK): Oxford University Press; 2005.
Surveys Do healthy and unhealthy behaviours cluster?