Article Text

other Versions

PDF

Does physical activity moderate the association between alcohol drinking and all-cause, cancer and cardiovascular diseases mortality? A pooled analysis of eight British population cohorts
  1. K Perreault1,2,
  2. A Bauman2,3,
  3. N Johnson2,4,
  4. A Britton5,
  5. V Rangul3,
  6. E Stamatakis2,4,5
  1. 1Département de Médecine Sociale et Préventive, Université de Montréal, Institut de Recherche en Santé Publique, Montréal, Québec, Canada
  2. 2University of Sydney, Charles Perkins Centre, School of Public Health, Johns Hopkins Drive, Camperdown, New South Wales, Australia
  3. 3Faculty of Medicine, Department of Public Health and General Practice, HUNT Research Centre, NTNU, Norwegian University of Science and Technology, Levanger, Norway
  4. 4Faculty of Health Sciences, University of Sydney, Discipline of Exercise and Sport Sciences, Cumberland Campus, Lidcombe, New South Wales, Australia
  5. 5Department of Epidemiology and Public Health, University College London, London, UK
  1. Correspondence to Dr E Stamatakis, Charles Perkins Centre, University of Sydney, Johns Hopkins Drive, Sydney, NSW 2006, Australia; emmanuel.stamatakis{at}sydney.edu.au

Abstract

Objective To examine whether physical activity (PA) moderates the association between alcohol intake and all-cause mortality, cancer mortality and cardiovascular diseases (CVDs) mortality.

Design Prospective study using 8 British population-based surveys, each linked to cause-specific mortality: Health Survey for England (1994, 1998, 1999, 2003, 2004 and 2006) and Scottish Health Survey (1998 and 2003).

Participants 36 370 men and women aged 40 years and over were included with a corresponding 5735 deaths and a mean of 353 049 person-years of follow-up.

Exposures 6 sex-specific categories of alcohol intake (UK units/week) were defined: (1) never drunk; (2) ex-drinkers; (3) occasional drinkers; (4) within guidelines (<14 (women); <21 (men)); (5) hazardous (14–35 (women); 21–49 (men)) and (6) harmful (>35 (women) >49 (men)). PA was categorised as inactive (≤7 MET-hour/week), active at the lower (>7.5 MET-hour/week) and upper (>15 MET-hour/week) of recommended levels.

Main outcomes and measures Cox proportional-hazard models were used to examine associations between alcohol consumption and all-cause, cancer and CVD mortality risk after adjusting for several confounders. Stratified analyses were performed to evaluate mortality risks within each PA stratum.

Results We found a direct association between alcohol consumption and cancer mortality risk starting from drinking within guidelines (HR (95% CI) hazardous drinking: 1.40 (1.11 to 1.78)). Stratified analyses showed that the association between alcohol intake and mortality risk was attenuated (all-cause) or nearly nullified (cancer) among individuals who met the PA recommendations (HR (95% CI)).

Conclusions Meeting the current PA public health recommendations offsets some of the cancer and all-cause mortality risk associated with alcohol drinking.

Statistics from Altmetric.com

Introduction

Alcohol consumption is an integral part of western culture. In 2013, 88% of American adults reported having drunk alcohol at some point in their lifetime, 56% in the past month. Similarly, in England, 63% of the population aged 16+ years have reported drinking at least once a month,1 with 24% of men and 18% of women drinking more than the recommended amount.2 High alcohol consumption is linked to an increased risk of all-cause mortality,3 cancer mortality4–6 and cardiovascular diseases (CVDs) mortality.7 ,8 Public health strategies to reduce alcohol consumption have involved alcohol risk reducing campaigns and measures aimed at regulating sales, demand and supply.9 ,10 Despite these measures, alcohol consumption remains high and thus there is a need for strategies aimed at minimising the health risks associated with alcohol consumption.

There is compelling evidence showing that regular physical activity (PA) is associated with cardiovascular health,11 ,12 reduced all-cause mortality,13–17 cancer mortality18 ,19 and CVD mortality13 ,15 ,20 risk. However, the prevalence of physical inactivity is high worldwide,21 with estimates ranging from 34% in England22 to 50% in the USA.23 Mechanistic research suggests that alcohol consumption and PA may be linked to chronic disease through shared pathways but acting in the opposing direction. For example, the biological pathways through which alcohol is thought to induce carcinogenesis4 ,6 ,24 are similar to those by which PA may prevent cancer.25–27

Limited epidemiological evidence has suggested that a high level of PA, or high cardiorespiratory (CR) fitness, moderates the association between alcohol intake and all-cause and CVD mortality.28 ,29 These studies were limited in that they pooled non-drinkers and ex-drinkers, an approach that has recently been shown to distort the associations between alcohol and mortality by overestimating the protective effect of moderate drinking.30 No study has explicitly examined the hypothesis that regular PA may offset mortality risks associated with alcohol consumption across a detailed categorisation of weekly alcohol intake, a hypothesis that is difficult to subject to experimental testing due to ethical considerations.

The aim of this study was to examine if health-enhancing PA moderates the association between alcohol intake and all-cause, cancer and CVD mortality risk in a large analysis of eight pooled British population-based cohorts.

Methods

Participants

The Health Survey for England (HSE)31 and the Scottish Health Survey (SHS)32 are continuous and annually repeated general population surveillance studies among independent samples of individuals living in private households in the two countries. Each sample is selected from a multistage, stratified probability design to give a sociodemographic nationally representative target population. Data collection was household-based and was carried out by interviewers. During the original survey interview, participants were asked to consent to their name, address and date of birth being sent to the Information Services Division (ISD) of National Health Service (NHS) for confidential linkage to their health records and mortality data. Ethical approval was obtained from the North Thames Multicentre Research Ethics Committee for England, the Local Research Ethics Council in England, the Research Ethics Committee for all Area Health Boards in Scotland and the Multicentre Research Ethics Committee for Scotland.

In this study, we used HSE years 1994, 1998, 1999, 2003, 2004 and 2006 and SHS years 1998 and 2003 that included information on both alcohol consumption and PA among individuals aged 40 years or older (n=50 198). We excluded those who did not consent to data linkage (n=5397) or had an incomplete covariates profile (n=8431), leaving 36 370 adult participants with a corresponding 5735 deaths and a mean of 353 049 person-years of follow-up (mean follow-up period of 9.7 years (SD 4.3)).

Outcomes

HSE and SHS were linked to NHS Central Register mortality data and participants were followed up for mortality until 31 December 2009 (SHS) or 31 March 2011 (HSE). Mortality was coded as a binary variable representing death or censoring. Diagnoses for primary causes of death were recorded according to the International Classification of Diseases, 9th Revision (ICD9) and 10th Revision (ICD10). Cancer and cardiovascular deaths were identified using the following codes: cancer: ICD9 140.0–239.9, ICD10 C00.0-D48.9; cardiovascular: ICD9 390.0–459.9, ICD10 I01.0-I99. These did not include injury, poisoning and other external causes.

Alcohol consumption

Adults were asked about whether or not they drink alcohol nowadays. Of those who reported no current alcohol consumption, a clarification was made of their status as never-drinkers or ex-drinkers. Supplementary questions were posed as ‘How often you consumed alcohol in the last 12 months?’; ‘Have you consumed alcohol in the last 7 days?’ and ‘How many and what size have you drunk in any 1 day?’.30 Total weekly UK units were calculated by summing the units of each type of beverage and multiplying by the reported frequency. In the UK, 1 unit is 8 g of alcohol, corresponding roughly with 25 mL measure of spirit, half a pint of beer, whereas a 175 mL glass of wine contains 2 units.33 A US standard drink (StdDrk) is any drink that contains about 14 g of pure alcohol.34 On that basis, one UK unit is equivalent to 0.57 US StdDrks.

Six sex-specific alcohol categories (UK units/week) were derived on the basis of the English Department of Health's suggested weekly limits35 and in accordance with recent findings emphasising the importance of carefully choosing the alcohol referent group:30 (1) never drunk; (2) ex-drinkers; (3) occasional drinkers (who declared not having drunk in the past 7 days); (4) within guidelines (<14 (women) (8 US StdDrks) and <21 (men) (12 US StdDrks)); (5) hazardous drinking (14–35 (women) (8–20 US StdDrks) and 21–49 (men) (12–28 US StdDrks)) and (6) harmful drinking (>35 (women) (20 US StdDrks) and >49 (men) (28 US StdDrks)). ‘Never drunk’ is synonymous to lifetime abstainers, that is, those who reported never having consumed alcohol. ‘Occasional drinkers’ declared being drinkers, but not having drunk in the past 7 days. We hypothesised their alcohol consumption to be less at risk than ‘within guidelines’.

Physical activity

The PA measures used here have validity against accelerometry in a large population-based validation study.36 Frequency and duration of leisure time PAs in the 4 weeks prior to the interview were assessed across three domains: (1) light/heavy manual work/gardening/do-it-yourself activity; (2) walking for any purpose and (3) light, moderate and vigorous sports/exercise. The questionnaire also included items on domestic activity but these were not included as part of the exposure variable in our analyses on the grounds of previous evidence highlighting the absence of links with all-cause mortality14 and CVD mortality.14 ,17 Metabolic equivalent task (MET)-hour/week was computed as MET for each specific activity36 multiplied by the number of hours the activity was performed per week. Extreme values of PA (on the basis of ≥5 SDs from the mean) were excluded. In line with the most recent PA public health guidance37 and recent evidence on the dose–response association between PA and mortality,38 we defined adherence to the lower PA recommendation as >7.5 MET-hour/week and adherence to the higher PA recommendation as >15 MET-hour/week.

Statistical analyses

Cox proportional-hazard models were used to examine the associations between alcohol consumption (reference category: never drunk) and all-cause, cancer and CVD mortality risks. Statistical interaction between Alcohol *PA was tested by adding an interaction term in both the unadjusted and the fully adjusted models. When interactions were significant, analyses by PA strata were performed.

Basic Cox models included adjustments for age (as a continuous variable) and gender. Further models were also adjusted for body mass index (continuous), cigarette smoking status (never regular smoker, ex-smoker, current smoker), psychological distress/depression (12-point General Health Questionnaire score), registrar general's social class (professional/managerial technical, skilled non-manual, skilled manual, semiskilled/unskilled manual, or other) and presence of long-standing illness. In all-cause and CVD mortality analyses, the middle models were also adjusted for doctor-diagnosed CVD (stroke or ischaemic heart disease, including angina) at baseline. Final models were additionally adjusted for PA (none, 0.1 to ≤7.5, 7.5 to 15, >15 MET-hour/week). All statistical analyses were carried out using SPSS V.21.0 for Mac (SPSS, Chicago, Illinois, USA).

In a sensitivity analysis, we performed stratified analyses excluding specific pre-existing conditions. For all-cause mortality, we excluded individuals with long-standing cancers or CVD at baseline. For cancer mortality, we excluded people with long-standing cancers, and for CVD mortality we excluded those with CVD.

Results

Sample characteristics

Characteristics of the population by groups of alcohol consumption (units per week) at baseline are presented in table 1. Age ranged from 40 to 102 years with a median of 56 years (IQR 47–67). In the final sample of 36 370 adults, 5307 (14.6%) reported no alcohol intake (never-drinkers and ex-drinkers together). In total, 4845 adults (13.3%) exceeded the recommended weekly limits for alcohol (women: 14 UK units (8 US StdDrks); men: 21 (12 US StdDrks)39). Those who reported alcohol consumption in the past 7 days consumed a median of 6.3 units/week (IQR 1.3–15.3). With regard to PA, 27.5% (9996/36 370) reported no non-domestic PA. The lower limit of the recommended level of PA (7.5 MET-hour/week) was met by 39.1% of the participants and the higher limit (15 MET-hour/week) by 23.3%. A median of 8.8 MET-hour/week (IQR 3.3–19.0) was found among those who reported any PA.

Table 1

Characteristics by groups of alcohol consumption (units per week) at baseline, Health Survey for England (1994; 1998; 1999; 2003; 2004 and 2006) and Scottish Health Survey (1998 and 2003) (n=36 370)

Associations between alcohol consumption and mortality

The HRs of death by alcohol categories for all-cause, cancer and CVD are provided in table 2, across three models with different levels of adjustments.

Table 2

HRs (95% CIs) for all-cause mortality, cancer mortality and CVD mortality by alcohol consumption, Health Survey for England (1994; 1998; 1999; 2003; 2004 and 2006) and Scottish Health Survey (1998 and 2003) (n=36 370)

In both the partially adjusted and fully adjusted models, we found a direct association between alcohol drinking and all-cause mortality, with ex-drinkers and drinkers at harmful level showing a clearly higher risk of all-cause mortality, compared with never-drinkers. We also observed a dose–response association between weekly alcohol consumption and cancer mortality, with increased risks from within guidelines consumption to harmful drinking. In contrast, we found weak evidence for an association between alcohol and CVD mortality, as only ex-drinkers were different to the referent group, and this also was persistent across all three models.

Ex-drinkers presented an increased risk of mortality for the three mortality outcomes, whereas occasional drinking was found to be protective against all-cause and CVD mortality. For example, the HR was 0.78 (0.63 to 0.97) for CVD mortality.

PA as a moderator of the association between alcohol and mortality

All interaction terms of alcohol consumption by PA were found to be significant in the fully adjusted Cox models for the three mortality outcomes (all p<0.001).

A direct association between alcohol consumption and all-cause mortality was found among those who did not meet the lower PA recommendation (≤7.5 MET-hour/week) (figure 1A). HRs were attenuated up to a hazardous level of drinking among those who did meet the lower and higher recommendations (>7.5 and >15 MET-hour/week), but the pattern of the association remained the same across all PA strata (figure 1A–C). A protective effect of occasional drinking was observed among those who met the highest PA recommendation (HR (95% CI) 0.68 (0.46 to 0.99)). In sensitivity analyses, these findings broadly persisted when we excluded participants with CVD at baseline (see online supplementary figure S1A–C). When excluding those with neoplasms at baseline, the associations were similar in the inactive group but were attenuated in the >7.5 and >15 MET-hour/week PA groups (see online supplementary figure S1D–F).

Figure 1

HRs of all-cause mortality according to alcohol consumption categories and physical activity strata (MET-hour/week). Stratified analysis were performed on three physical activity strata: (A) ≤7.5 MET-hour/week; (B) >7.5 MET-hour/week and (C) >15 MET-hour/week. Five groups of alcohol consumption were defined (UK units/week (equivalent in US standard drinks)): never drunk and ex-drinkers: no drinking; occasional drinkers (drinkers, but have not drunk in the past 7 days); within guidelines: <14 (8) (women) and <21 (12) (men); hazardous: 14–35 (8–20) (women) and 21–49 (12–28) (men) and harmful: >35 (20) (women) and >49 (28) (men). Vertical bars represent the 95% CIs for the HRs. The reference group included participants who never drank. HRs of all-cause mortality were adjusted for age, gender, body mass index, social class, cigarette smoking status, psychological distress/depression score (12-point General Health Questionnaire score), long-standing illness, doctor-diagnosed CVD (stroke or ischaemic heart diseases, including angina) and physical activity level. CVD, cardiovascular disease; Drink., drinkers; Occ., occasional.

The risk of cancer mortality was increased in a dose–response fashion among inactive participants (figure 2A). This association was attenuated in those who met the lower recommendation (figure 2B), and substantially attenuated in the higher recommendation group (figure 2C). Results were broadly similar in the sensitivity analyses when individuals with neoplasms at baseline were excluded (see online supplementary figure S2A–C).

Figure 2

HRs of cancer mortality according to alcohol consumption categories and physical activity strata (MET-hour/week). Stratified analysis were performed on three physical activity strata: (A) ≤7.5 MET-hour/week; (B) >7.5 MET-hour/week and (C) >15 MET-hour/week. Five groups of alcohol consumption were defined (UK units/week (equivalent in US standard drinks)): never drunk and ex-drinkers: no drinking; occasional drinkers (drinkers, but have not drunk in the past 7 days); within guidelines: <14 (8) (women) and <21 (12) (men); hazardous: 14–35 (8–20) (women) and 21–49 (12–28) (men) and harmful: >35 (20) (women) and >49 (28) (men). Vertical bars represent the 95% CIs for the HRs. The reference group included participants who never drank. HRs of cancer mortality were adjusted for age, gender, body mass index, social class, cigarette smoking status, psychological distress/depression score (12-point General Health Questionnaire score), long-standing illness and physical activity level. Drink., drinkers; Occ., occasional.

No association between alcohol and CVD mortality risk was observed, although occasional drinking had a protective effect for CVD mortality in the lower and higher recommendations strata (HR (95% CI) for >7.5 MET-hour/week: 0.56 (0.34 to 0.93)|>15 MET-hour/week: 0.43 (0.19 to 0.97)) (figure 3A–C). Results were consistent in the sensitivity analyses when individuals with CVD at baseline were excluded from the sample (see online supplementary figure S3A–C).

Figure 3

HRs of CVD mortality according to alcohol consumption categories and physical activity strata (MET-hour/week). Stratified analysis were performed on three physical activity strata: (A) ≤7.5 MET-hour/week; (B) >7.5 MET-hour/week and (C) >15 MET-hour/week. Five groups of alcohol consumption were defined (UK units/week (equivalent in US standard drinks)): never drunk and ex-drinkers: no drinking; occasional drinkers (drinkers, but have not drunk in the past 7 days); within guidelines: <14 (8) (women) and <21 (12) (men); hazardous: 14–35 (8–20) (women) and 21–49 (12–28) (men) and harmful: >35 (20) (women) and >49 (28) (men). Vertical bars represent the 95% CIs for the HRs. The reference group included participants who never drank. HRs of CVD mortality were adjusted for age, gender, body mass index, social class, cigarette smoking status, psychological distress/depression score (12-point General Health Questionnaire score), long-standing illness, doctor-diagnosed CVD (stroke or ischaemic heart diseases, including angina) and physical activity level. CVD, cardiovascular disease; Drink., drinkers; Occ., occasional.

Discussion

In this large British general population cohort, we found that the association between alcohol intake and mortality risk was moderated by PA.

In stratified analyses, alcohol-related mortality risk varied by PA levels. Participants who did not meet the lower recommendation of 7.5 MET-hour/week showed a clear dose–response risk for cancer mortality from drinking within guidelines and up to a harmful level. The risk of cancer mortality was attenuated or nullified in participants who met the lower and higher PA recommendations (>7.5 and >15 MET-hour/week), as none of the drinking categories were found to be substantially different to the referent group (never-drinkers). The association between alcohol and all-cause mortality was similar across PA strata, although HRs were slightly attenuated up to a hazardous level of drinking among active participants (>7.5 and >15 MET-hour/week). With regard to CVD mortality, no association was found except for a protective effect of occasional drinking in active participants only.

Regardless of PA levels, we found a nearly J-shaped association between alcohol consumption and all-cause and CVD mortality (with the beneficial associations being evident in the occasional drinkers only). Similar to the conclusions of another recent analysis from a similar pool of British cohorts, our results confirmed that the inclusion of former drinkers in the referent group (forming a broad category of ‘non-drinkers’) has the potential to overestimate the protective effect of drinking at the recommended level.30 As for cancer mortality, we found no protective effect of occasional drinking, and an increased risk from within guidelines consumption. The public health relevance of our results is further emphasised by the recently updated alcohol consumption guidelines review of the UK Chief Medical Officers that found that cancer mortality risk starts from a relatively low level of alcohol consumption.40

Previous studies in the field

Very few studies until now have specifically examined the effects of alcohol use jointly with PA on mortality. In a prospective study including 29 402 men, Shuval et al28 found that moderate and high CR fıtness (a direct indicator of habitual PA41) was protective against all-cause mortality at all levels of drinking. For example, the heavy drinkers in the highest category of CR fitness presented an all-cause mortality risk reduction of 38% compared with the heavy drinkers in the lowest CR fitness category. This study did not observe such associations with CVD mortality and the only protective effect found for CVD mortality was in the high-fitness/moderate drinking category (3–14 US StdDrks/week or 5–25 UK units/week). Our results indicate similar findings, as lower and higher PA recommendations slightly attenuated the risk of all-cause mortality up to a hazardous level of drinking; and occasional drinkers presented a reduced risk of all-cause and CVD mortality among active participants.

Potential biological mechanisms

Alcohol has been recognised as a major contributor to cancer mortality.5 ,42 The study of the links between PA and cancer is a recent field and current evidence shows that there is a robust decreased risk with increased PA for breast and colon cancers and perhaps several other cancer sites (eg, ovarian, lung and prostate cancers).25 ,43 ,44 In a very large recent study from 12 US and European cohorts, high levels of leisure-time PA were associated with lower risks for 13 of 26 site-specific cancers,45 including several types of cancer that are specifically associated with increased alcohol consumption (eg, oesophageal, gastric cardia, liver, colon, rectum, breast).40 Our results on the presence of a nearly linear association between alcohol and cancer mortality in the inactive group strengthen further current knowledge on the links between alcohol and health risks. The associations we observed were substantially attenuated among participants who met both the lower (>7.5 MET-hour/week) and higher (>15 MET-hour/week) PA guidelines. There are several biological mechanisms by which alcohol is believed to contribute to carcinogenesis,4 ,6 ,24 which appear to be similar to those by which PA may prevent cancer,25–27 but acting in the opposite direction. These pathways include (alcohol vs PA): increase46–49 versus decrease50–53 of oxidative stress and inflammatory markers, elevation54 ,55 versus diminution56–58 of sex steroid hormone levels, and reduced59 ,60 versus enhanced61–63 immune function. A recent study based on an animal model demonstrated that exercise decreases tumour incidence and growth by over 60% through enhanced immune function.64 These results suggest that PA may be a powerful preventive strategy against cancer onset and also mediate cancer regression. Such exercise-activated antitumour immunity pathways are yet to be demonstrated in humans but they are receiving increased attention.65

Dose–response effects of PA

Notably, we did not observe appreciable differences in mortality across alcohol intake categories between the upper and lower PA cutoffs (>7.5 and >15 MET-hour/week). We hypothesised that a higher level of PA would confer an extra reduction in mortality risk, based on the recent large pooled study that found a dose–response risk reduction in all-cause mortality and cancer mortality with several multiples of the minimal dose of PA recommendations.38 This might be due to the reduced statistical power in our most active category (the >15 MET-hour/week strata included n=8123 corresponding to 713 all-cause mortality events and 288 cancer mortality events), or a ceiling effect often observed in smaller epidemiological studies studying the dose–response of PA and mortality risk.66

Strengths and limitations

This study has some strengths and some limitations. To the best of our knowledge, this is the first set of population cohorts to examine the influence of PA as a moderator of all-cause, cancer and CVD mortality risks across a detailed categorisation of weekly alcohol intake. Moreover, all previous studies in the area pooled all non-drinkers as the reference group (combining lifelong abstainers and former drinkers who may have quit due to health issues, increasing the chance for bias30 ,67). On the other hand, the alcohol measure used in this study assessed the quantity of alcohol intake (UK units/week), but not the drinking pattern (eg, binge drinking), which have important health implications.67 ,68 Also, the possibility of unmeasured confounding (including dietary6970) factors remains. Misclassification because of under-reporting of heavy drinkers71 ,72 may explain the limited number of participants in the harmful drinking group, which requires cautious interpretation. Since alcohol drinking73 and PA74 had been shown to be dynamic behaviours, the baseline measurements we used may not reflect the actual alcohol intake/PA level over the follow-up period.

Conclusion

We found evidence of a dose–response association between alcohol intake and cancer mortality in inactive participants but not in physically active participants. PA slightly attenuates the risk of all-cause mortality up to a hazardous level of drinking. The protective effects of PA were evident from a level of meeting the minimal public health recommendations of PA. Our results provide an additional argument for the role of PA as a means to promote the health of the population even in the presence of other less healthy behaviours.

What are the findings?

  • These are the first population set of cohorts to study the influence of physical activity (PA) as a moderator of all-cause, cancer, and cardiovascular risks death risk across a detailed categorisation of weekly alcohol intake, using those who never drunk as the reference group.

  • PA moderated the association between alcohol consumption and all-cause and cancer mortality risk. This effect is evident from the lower PA recommendations (7.5 MET-hour/week).

  • This study brings novel information about the potential of PA to promote health and reduce alcohol consumption-related risk of all-cause and cancer mortality.

How might it impact on clinical practice in the future?

  • Our results suggest that being physically active at the lower limit of public health recommendations (7.5 MET-hour/week) mitigates some mortality risks associated with alcohol drinking.

  • From a public health angle, our conclusions are oriented towards a behavioural strategy to reduce mortality risk associated with alcohol consumption.

References

View Abstract

Footnotes

  • Contributors KP contributed to the study concept and design; data analysis and interpretation; and to the manuscript drafting and revisions. ABa, NJ, ABr and VR participated in the critical revision of the manuscript for important intellectual content; and in the results interpretation. ES contributed to the conception of study idea and study design; data set acquisition and harmonisation; study supervision; redrafting of the manuscript; critical revision of the manuscript for important intellectual content; and to the results interpretation. All the authors approved the final version of the manuscript before its submission and publication. ES had full access to the data and KP had partial controlled access in a secure data environment; they both take joint responsibility for the integrity of the data and the accuracy of the data analyses.

  • Funding The Health Survey for England and the Scottish Health Survey were commissioned by the Department of Health and conducted by the Joint Health Surveys Unit of the National Centre for Social Research (NatCen) and the Department of Epidemiology and Public Health, University College Medical School, London. ES is funded by the National Health and Medical Research Council through a Senior Research Fellowship. KP did this work as part of a research internship at Charles Perkins Centre, University of Sydney.

  • Competing interests None declared.

  • Patient consent Obtained.

  • Ethics approval The North Thames Multicentre Research Ethics Committee for England, the Local Research Ethics Council in England, the Research Ethics Committee for all Area Health Boards in Scotland, and the Multicentre Research Ethics Committee for Scotland.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.