The impact of a healthy lifestyle on Disability-Adjusted Life Years: a prospective cohort study

The EPIC-NL study consists of the two Dutch contributions to the European Prospective Investigation into Cancer and Nutrition (EPIC), which were set up simultaneously between 1993 and 1997. The design and rationale of the EPIC-NL study has been described in detail elsewhere [23]. In brief, the Prospect-EPIC study includes 17,357 women aged 49 to 70 years living in Utrecht and vicinity who participated in the nationwide Dutch breast cancer screening programme. The MORGEN-EPIC cohort consists of 22,654 men and women aged 20 to 65 years selected from random samples of the Dutch population in three different towns (Doetinchem, Amsterdam, and Maastricht). At baseline, a general questionnaire and a food-frequency questionnaire were administered and a physical examination was performed. This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects were approved by the institutional review board of the University Medical Center Utrecht (Prospect) and the Medical Ethical Committee of TNO Nutrition and Food Research (MORGEN). Written informed consent was obtained from all subjects.

From the total cohort (n = 40,011), subjects who did not give permission for linkage with disease registries were excluded (n = 2,910). Furthermore, men and women who suffered from any of the studied diseases (cancer, coronary heart disease (CHD), cerebrovascular accident (CVA), diabetes mellitus, chronic obstructive pulmonary disease (COPD), asthma, Parkinson’s disease, rheumatoid arthritis, osteoarthritis, and inflammatory bowel disease (IBD)) at baseline (n = 3,583) were excluded. Additionally, we excluded subjects without information on dietary intake (n = 142) or with implausible high or low scores for total energy intake (those in the top 0.5% and bottom 0.5% of the ratio of reported energy intake over estimated energy requirement based on basal metabolic rate) (n = 310). The final study population consisted of 33,066 men and women.

The general questionnaire included questions on demographics, presence of chronic diseases, and risk factors for chronic diseases, such as smoking behaviour and level of education (categorized as low (primary education up to those completing advanced elementary education), average (intermediate vocational education and higher general secondary education), or high (higher vocational education and university)). BMI was calculated from height and weight, which were measured during the physical examination. Daily dietary intakes were obtained from a self-administered validated food frequency questionnaire containing questions on the usual frequency of consumption of 79 main food items during the year preceding enrolment. This questionnaire allowed the estimation of the average daily consumption of 178 foods. Physical activity was assessed using the EPIC physical activity questionnaire and categorized according to the validated Cambridge Physical Activity Index (CPAI) [24]. This four-category index (inactive, moderately inactive, moderately active, or active) was derived by cross-classifying three questions referring to activities during the last year against classification of work activity. Because there was no information on physical activity (14%), smoking status (0.4%), educational level (0.8%), alcohol intake (categories) (3.3%), and/or BMI (0.1%) for some of the participants, missing data was imputed using single linear regression modelling (SPSS MVA procedure).

We investigated four important lifestyle factors combined into a previously defined pragmatic health behaviour score [25]. These lifestyle factors, all assessed at baseline, are smoking status, physical activity, BMI, and diet.

For smoking status, participants were grouped in two categories: (0) ever smokers (current and former smokers) and (1) never smokers. BMI was categorized in (1) BMI <25 (normal weight) and (0) BMI of 25 or higher (overweight and obese). Although being overweight or obese is not a lifestyle factor per se, it is a commonly included component in lifestyle indices beside physical activity, smoking, and dietary intake. For physical activity, the CPAI index was dichotomised into (0) inactive and (1) active (moderately inactive, moderately active, or active). The fourth lifestyle factor, diet, was categorized by adherence to the Mediterranean diet that has previously shown a beneficial impact on disease burden [26-29]. Adherence was measured using the modified Mediterranean Diet Score (mMDS) that consists of 8 components (fruit, nuts and seeds, vegetables, legumes, fish, cereals, unsaturated to saturated fat ratio, meat (products), and dairy) after excluding the alcohol component. For each component, 1 point was allocated depending on whether the intake was above or below the population median. Subsequently, these points were summed into one score that ranges from 0 (low adherence) to 8 (high adherence). For the health behaviour score, this mMDS score was further dichotomized into (0) low adherence (0 to 4 points) and (1) high adherence (5 to 8 points). We did not include alcohol in our health behaviour score. A moderate alcohol intake is associated to lower CVD risk but on the other hand related to a higher risk of breast cancer. Furthermore, excessive alcohol intake causes many health-related harms [30]. Therefore, we decided not to include moderate alcohol consumption as a potential beneficial behaviour and instead adjusted for alcohol consumption in the analyses.

We constructed the total health behaviour score by assigning 1 point per healthy behaviour (never smoking, normal weight, physically active, high adherence to Mediterranean diet), creating a total health behaviour score ranging from 0 to 4 points.

Participants were followed for mortality and morbidity through linkage with several registries. The selection of the diseases was based on their prevalence and disease burden in the Netherlands, but also on the availability of data information sources. Information on vital status and the date of death was obtained through linkage with municipal registries. The cause of death was obtained from Statistics Netherlands. Information on disease occurrence (cancer, CHD, CVA, diabetes mellitus, COPD, asthma, Parkinson’s disease, rheumatoid arthritis, osteoarthritis, and IBD) was obtained from the National Cancer Registry and the national hospital discharge diagnosis database from the Dutch National Medical Registry. The National Cancer Registry provided information on the type of cancer and the date of histological diagnosis. The national hospital discharge diagnosis database provided the date of diagnosis for CHD, CVA, diabetes mellitus, COPD, asthma, Parkinson’s disease, rheumatoid arthritis, osteoarthritis, and IBD. First diagnosis of the disease was assumed to be at the date of hospital discharge. The national hospital discharge diagnosis database was linked to the cohort with a validated probabilistic method using the following information: date of birth, gender, postal code, and name of the general practitioner [31]. Self-report and urinary glucose strip tests (Prospect-EPIC) provided additional information on diabetes mellitus at study entry. New diabetes mellitus cases ascertained during follow-up were verified against information of the general practitioner or pharmacist [32]. Follow-up was complete until 31 December 2007.

DALYs, i.e., the sum of the Years Lost due to Disability (YLD) and the Years of Life Lost due to premature mortality (YLL) [33], were calculated for each individual in the cohort. The YLL are computed as the number of years death occurred earlier than expected. The expected number of life years are the remaining years that a person of a certain age is expected to live on average defined at time of death, loss of follow-up, or end of follow-up [22]. The expected life expectancy was obtained from statistics Netherlands, which provides age, sex, and calendar year-specific life expectancies based on mortality rates [34]. The YLD are calculated by the number of years a person lives with a disability multiplied by a disability weight reflecting the severity of that disability. The disability weights were derived from the Dutch Disability Weight study by three panels of medical experts that evaluated a large number of disease stages using techniques such as person trade-off [35]. The disability weights can range between 0 (no burden) and 1 (death) (Additional file 1: Table S1). The years lived with a chronic disease are calculated from the disease onset until death or until the end of life expectancy. One DALY represents the loss of 1 year in full health. For example, a person who lived 5 years with diabetes (disability weight 0.20) and died 30 years before his life expectancy obtains 1 YLD and 30 YLL, which equals 31 healthy years of life lost (31 DALYs; Figure 1). The calculation of individual DALYs in a cohort and the advantages and disadvantage of different ways of doing that, have been described in more detail elsewhere [22].

Due to the distribution of the DALYs, i.e., a peak at 0 and a normal distribution in participants with DALYs >0, we used a two-part model to estimate the association of the individual lifestyle factors and the health behaviour score with DALYs as dependent variable [36]. This two-part model combines the estimation of the probability of having DALYs using logistic regression with the estimation of the number of DALYs among participants with DALYs >0 using linear regression. Confidence intervals were constructed with bootstrapping (500 samples). The individual health behaviours smoking status (never, former, and ever smokers), physical activity (CPAI index), BMI (normal weight (<25), overweight (25–30), obese (≥30)), and diet (mMDS score 0–2, 3–5, 6–8) were analyzed categorically. The health behaviour score was analyzed categorically using the health behaviour score 0 as the reference. The P for trend was estimated by applying a linear regression to the estimates of the categories. The analyses were adjusted for age, sex, education level, alcohol intake (ethanol in g/day and a categorical alcohol variable (never, quit, <1 drink/week, current drinker)), and energy intake. For the individual lifestyle factors, additional adjustment was done for smoking status and intensity (categorized as never, former (quit smoking >20 years ago, quit 10–20 years ago, quit ≤10 years ago), current smoker (1–15 cigarettes/day, 16–25 cigarettes/day, >25 cigarettes/day, pipe or cigar smoker), BMI (continuous), physical activity (CPAI index), and diet (continuous mMDS score) when the covariate was not the variable of interest. We investigated interaction with age and sex by including an interaction term in the logistic and linear model. We conducted additional analyses to investigate the effect of stratifying the analysis between the health behaviour score and DALYs for sex and age (>50/≤50 years), and restricting to several high-risk subgroups, persons with one of the four risk factors, hypertension or high cholesterol (total cholesterol/HDL ratio above 5)). Furthermore, we run analysis replacing BMI with waist circumference (1 point if waist circumference is below 94 cm for males or below 80 cm for females), and including a moderate alcohol intake (10–50 g of alcohol a day for men and 5–25 g of alcohol a day for women) in the mMDS score and without adjusting the analysis for alcohol intake.

Statistical analyses were conducted using SAS 9.2 (SAS Institute, Cary, US), except imputation which has been conducted using SPSS 14.0 (Chicago, IL, USA).