Cohort Profile: The Swedish National March Cohort

In the 1990s there was emerging evidence that physical activity was associated with a reduced risk of several cancers including colon,1 breast2 and prostate cancer.3 High physical activity and fitness had shown to be associated with a lower total risk of cancer in some,4–6 but not all, studies.7 There were still many questions:(i) could the findings be due to confounding? if not, (ii) was it mode, intensity, duration or frequency of physical activity that was important? and (iii) were certain time points in life more important than others to engage in physical activity in terms of prevention? This led to the initiation of the Swedish National March Cohort (SNMC) in 1997. The purpose of the SNMC was to create a large prospective cohort to investigate associations between lifestyle factors and chronic diseases, including cancer, in the Swedish population. In September 1997 the Swedish National March, a 4-day promotional and fund-raising event for the Swedish Cancer Society, took place in almost 3600 Swedish cities and villages. The event included fundraising galas and scientific programmes on television. It was promoted by the Swedish King and Queen and was heavily covered by local and national media. Participants were invited to fill out a questionnaire concerning lifestyle and medical history. To return it, a deal was made with one of the large chains of supermarkets across Sweden. Their 2300 stores were all equipped with a special mail box, making it possible for the participant to securely post the questionnaire at the local supermarket, no postage required. Next, the milk trucks exchanged milk for mailboxes and transported the questionnaires free of charge to Statistics Sweden for scanning. Participants could, if they wanted, send the questionnaires by regular mail, but due to lack of funding they had to cover the postage themselves. Compared with the general Swedish population in 1997, the cohort members were on average less educated (39% only had compulsory schooling compared with 25% of the Swedish population), they were more overweight or obese [43% had a body mass index (BMI) above 25 kg/m2 compared with 40% in the Swedish population] and they smoked less (9.6% compared with 19.2% of Swedish population aged 16–84) according to Statistics Sweden.8 Baseline characteristics of the cohort are shown in Table 1. Baseline characteristics of participants in Swedish National March Cohort (SNMC) MET, metabolic energy turnover; SD, standard deviation. Baseline characteristics of participants in Swedish National March Cohort (SNMC) MET, metabolic energy turnover; SD, standard deviation. The study was approved by the Regional Ethics Review Board at the Karolinska Institutet. All study participants provided informed consent. Due to the nature of the event, the number of participants could not be assessed in detail. In total, 43 880 participants completed the questionnaire and provided their individual and unique 10-digit national registration number.9 This allows for complete follow-up using record linkage to national registries. Subjects with an incorrect national registration number (inconsistent or non-existent) were excluded. The final number of subjects recruited into the cohort was 43 863. Thanks to the National Registration Number, used as identifier in a series of nationwide and essentially complete registers, all participants are monitored through linkage between the cohort file and these registers. There was no follow-up questionnaire, although only 20% denied future contact when asked if they could be contacted in the future. Each participant gave background data and exposure information in the 36-page questionnaire. The questionnaire was called ‘An hour for research’ since it was estimated to take 1 hour to complete, and participants were encouraged to do so at home. A list of the topics covered by the questionnaire can be found in Table 2. The English translation of the questionnaire is available on the web page of the SNMC [http://ki.se/en/meb/the-swedish-national-march-cohort-nmc], or can be requested from the principal investigators. Variables included in the questionnaire in the Swedish National March Cohort (SNMC) Variables included in the questionnaire in the Swedish National March Cohort (SNMC) Nearly bias-free, complete and cost-effective follow-up has been achieved through linkages to a broad range of national, well-validated registries.10 The Register of the Total Swedish Population and the Register of Population Changes provide information about dates of death, relocation and emigration.11 The Swedish population-based Patient Register covers all in-hospital care and provides information on medical data, anesthesiological procedures and up to eight discharge diagnoses coded according to the International Classification of Diseases (ICD). Since 1987 the register has had nationwide coverage and since 2001 it has also covered hospital-provided outpatient care. A validation study reported a generally high positive predictive value of 85–95% for diagnoses in the register.10 The Cause of Death Register holds information on date of death and underlying and contributory causes of death based on the ICD codes. This further allows calculation of expected numbers of deaths from specific causes by calendar year, gender and 5-year age group (see Table 3 for vital status of the participants). Vital status of participants included in the Swedish National March Cohort (SNMC) up to 31 December 2010 Vital status of participants included in the Swedish National March Cohort (SNMC) up to 31 December 2010 In the Swedish Cancer Register all cancers are recorded according to ICD-7, providing further information about medical data such as date of diagnosis, type of cancer, histology and pathology. The Swedish Cancer Register, the Patient Register and the Cause of Death Register, which are held by the National Board of Health and Welfare, enable an accurate tracking of health and vital status of all cohort members. Starting from 2005 an additional register, the national Prescribed Drug Register, became available with information about drug utilization and expenditures of prescribed drugs in the entire Swedish population.12 Furthermore, a number of health care quality registries with clinical data most often lacking in central national health and administrative registers, are now available,13 whereof the SNMC so far has been linked to National Prostate Cancer Register. In the SNMC, exposure was assessed in much greater detail than in many other similar cohorts. Despite the absence of a well-defined population, the strong motivation of the participants to provide valid information has made the cohort a valuable source for research. Physical activity has been the primary focus so far, and its association with several outcomes has been investigated. The effects of physical activity, BMI, waist-to-hip ratio and waist circumference on total mortality risk were assessed. Results suggest that both inactivity and obesity increase total mortality risk. Further, no substantial multiplicative or additive interaction between different measures of adiposity and physical activity were detected.14 We also investigated associations of total and leisure time physical activity with the risk of heart failure; higher leisure time physical activity was associated with a lower risk of heart failure of any cause.15 Thanks to the cohort’s maturity, with follow-up since 1997, it has been possible to study the role of physical activity and anthropometric measures in the aetiology of several types of cancer. We found dose-dependent, homogeneous inverse associations between postmenopausal breast cancer incidence and physical activity across all strata of BMI. This association was further found between postmenopausal breast cancer incidence and BMI across all strata of physical activity. We found no evidence of effect modification between the two, suggesting independent effects.16 In another study investigating prostate cancer as an outcome, we found no association between total physical activity or BMI and localized or advanced prostate cancer.17 Because physical exercise has been associated with neuroprotective effects in the nigrostriatal dopaminergic system,18,19 physical activity and the risk of Parkinson’s disease was investigated. In male subjects, an inverse association with Parkinson’s disease for total physical activity was found; however, no association was observed for leisure time exercise or occupational physical activity.20 We have also compared self-estimates of physical activity obtained from an instrument developed for the study (the Energy Expenditure Questionnaire, EEQ) with the estimates obtained from questions typically asked in epidemiological investigations. Estimates of activity, based on questions focusing on leisure time exercise and some other selected activities, correlated poorly with total energy output from all physical activity and inactivity. We suggested that investigators should define explicitly which component of physical activity they investigate.21 Other studies, performed within a collaboration of a Swedish cohort consortium which includes the most important cohort studies in the country, have focused on the relation between the use of Swedish snus (sniff dipping) and the risk of cardiovascular outcomes. In summary, snus consumption is unlikely to confer any important increase in risk of atrial fibrillation, stroke or acute myocardial infarction.22–24 Previous, but limited, evidence suggested that the association between sleep duration and cardiovascular events is strongest in individuals who also report sleep disturbances.25–27 We therefore investigated sleep duration and insomnia symptoms in relation to incident cardiovascular events in the SNMC based on self-reported habitual sleeping pattern at baseline. We found that short-sleepers who reported frequent insomnia symptoms were at highest risk, suggesting that symptoms of sleep disturbance should be taken into account when assessing the association between short sleep and cardiovascular disease.28 We also compared selected percentiles of distributions of BMI between individuals with different habitual sleep patterns, performing quantile regression analyses of 40 197 Swedish adults who reported sleep duration and quality, weight, height and possible confounding factors at baseline. BMI was higher among short sleepers (≤ 5 h), relative to those sleeping 6–8 hours.29 Furthermore, we analysed the relationships between sleep patterns and risk of prostate cancer. The findings did not support any association.30 The cohort has been used to illustrate recent developments in mediation analysis where the main goal was to disentangle direct and indirect effects using the principal stratum approach.31 Bayesian semiparametric models based on the principal stratification approach were also proposed, with a parametric model for the potential outcome and a Bayesian non-parametric model for the potential intermediate outcomes. Both methodologies were applied to study the direct effect of physical activity on the incidence of cardiovascular disease, and its indirect effect through BMI.32 Moreover, the cohort has been used to demonstrate how to estimate attributable fractions to quantify the strength of an exposure-disease relationship, and their use in comparison with the traditional maximum likelihood estimator.33 Implementation of double robust estimators modelling both the association between the outcome and the exposure, and between the exposure and the confounders, to estimate a causal relationship between an exposure and an outcome has been shown.34 Pooling large-scale cohort studies to get sufficient power can be the only way to answer some research questions. The SNMC is currently part of several consortia. It contributes to projects in the National Cancer Institute Cohort Consortium [http://epi.grants.cancer.gov/Consortia/members/nmc.html] and it is an associated cohort in the European Cohort Consortium. At a national level, the SNMC is part of the SNUS collaboration and a member of the Swedish Cohort Consortium. The SNMC is a prospective cohort with almost complete follow-up ascertained through linkages to nationwide, continuously updated registers. Due to the large sample size and long follow-up duration, the cohort provides adequate statistical power to investigate a range of different outcomes. The cohort study further provides detailed exposure information about a broad range of potential risk and confounding factors. We used validated questions to assess exposures. The information was collected among motivated people who wished to support cancer research by taking part in a fund-raising event, and who then actively chose to take the effort to answer the questionnaire. The proportion of missing or inconsistent data is unusually low. For example, less than 1% missing for age at menarche, treatment for childlessness and number of births, less than 2% for education and less than 5% for vitamin use and BMI. Weaknesses of this cohort study are related to the fact that all information collected in the questionnaire is self-reported. Because exposure was measured only at baseline, it is possible that some subjects have changed their exposure status over the years. In addition, in the assembly of the cohort, biological samples were not included. Since subjects participated on a voluntary basis during a fund-raising event, the SNMC may be prone to a potential healthy volunteer bias. Further, this may unintentionally affect the representativeness of the cohort and reduce generalizability of study findings.35 Representativeness and validity of population-based and selected cohorts have been actively discussed in the field of epidemiology. Population-based cohort studies are often confronted with the problems of poor response rates at baseline and incomplete follow-up.36 Further, initial non-response of study individuals limits the external validity of the cohort sample, since non-response is hardly randomly distributed across segments of the population.37 The approach to selecting study subjects from a restricted source population according to specified criteria has been frequently used in cohort studies.38 The purpose of such approaches is to enhance feasibility of a study and to increase prevalence of exposure and completeness of follow-up, which further increases validity and precision of the study.38 It has been argued that knowledge of specific conditions such as effect measure modification, skilful control for confounding factors and the understanding of mechanisms of effect is more important than representativeness of the study subjects in scientific studies.39 Additionally, potential bias due to lack of representativeness needs to be balanced against the probable bias from incomplete follow-up in a more representative sample.35 It has further been commented that the more representative a sample is, the more difficult it may be to obtain repeated data.36 One important criticism, which has been discussed and investigated, is that if the exposure of interest is associated with the probability of being selected, the estimated exposure-outcome association may be biased in a non-representative fashion.35 Pizzi et al. investigated validity of exposure-disease association estimates in a restricted source population using simulation analyses for scenarios with more extreme values than typically seen in epidemiology. Under a range of scenarios they reported that the use of a restricted source population produces weak bias in the estimates of exposure-disease associations. To introduce bias, selection has to be associated with one or more unmeasured or unknown disease risk factors. This, however, can likewise happen in a general population.38 Nevertheless, on some occasions, representativeness is inevitable and appropriate, such as in the case of sampling for purely descriptive purposes.36,39 Hence, whether representativeness is harmful or beneficial depends on the context of the particular study.35 Representativeness should be evaluated in each individual study setting, rather than just avoiding or uncritically adopting it.40 The data of the SNMC are held and managed by the Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden. Further information about the study, the English translation of the questionnaire and existing publications are available on the study webpage [http://ki.se/en/meb/the-swedish-national-march-cohort-nmc]. The SNMC is a large prospective cohort recruited to investigate the association between lifestyle, cancer and cardiovascular disease in the Swedish population. In September 1997 the Swedish National March, a fund-raising event for cancer, took place in almost 3600 Swedish cities and villages. A total of 43 880 participants aged 8–94 completed the baseline questionnaire and entered into the study. No follow-up questionnaire was planned, but thanks to the National Registration Number all participants are monitored through linkages to a broad range of national well-validated registries. All participants filled in a detailed questionnaire on physical activity, anthropometrical measures, occupation, education, smoking, vaccination, sun and UV exposure, sleeping, mobile phone use, medical and pharmacological history, an 85-item food frequency questionnaire and an only-for-women section. Potential joint collaborations can be discussed with the principal investigators, Karolinska Institutet, Stockholm, Sweden. Further information can be found on the SNMC webpage [http://ki.se/en/meb/the-swedish-national-march-cohort-nmc]. This work was supported by the Swedish Cancer Society, ICA AB and Telefonaktiebolaget LM Ericsson. Conflict of interest: None declared.