Methods
Information on variables used in this study was collected during a field phase including a health examination, interviews, and self-administered questionnaires. Moreover, information was drawn from national health registers.
IWL was defined by combining questions concerning dieting attempts during the year prior to baseline (no/yes) and weight loss during the year prior to baseline (no/yes) from a self-administrative questionnaire. The questionnaire included a question concerning the amount of weight loss during the year prior to baseline (among those with weight loss: range 1–38 kg, mean 5.2, SD 4.0), but, in line with previous studies on self-report IWL, individuals who had attempted to lose weight and had lost any amount of weight were considered as satisfying the IWL criteria, irrespective of the amount of weight lost during the year prior to baseline.17–19 30
Data on sex and age were obtained from the sampling frame. Educational attainment and smoking habits were asked about during an interview. Education was divided into a three-class variable including categories: low (did not graduate from upper secondary school or vocational school), intermediate (graduated from upper secondary school or vocational school) and high (graduated from university or university of applied sciences). Individuals were categorized according to their smoking status as never-smokers, former smokers and current smokers.
A self-administered questionnaire was used to measure leisure-time PA, alcohol consumption (g ethanol/week) and habitual sleep duration during 24 hours. PA was categorized in three levels: not physically active (‘low’), regularly engaging in light PA such as walking or cycling (‘moderate’) and exercising for 3 hours or more per week or training for competitive sports (‘regular vigorous training’). Individuals were categorized according to their alcohol consumption (g ethanol/week) as non-users, moderate users (1–199 for male or 1–99 for female) and heavy users (200 or over for male or 100 or over for female). Sleep duration was divided into a three-class variable including the categories: ‘≤6 hours’, ‘7–8 hours’ and ‘≥9 hours’.
A self-administered Food Frequency Questionnaire assessing habitual food intake during the last 12 months31 32 was used to measure energy intake and quality of diet. Average daily intakes of food groups, energy and nutrients were calculated using The National Food Composition Database (Fineli) and in-house software (Finessi).33 Quality of diet was measured with The Alternate Healthy Eating Index (AHEI).34 In this study, the AHEI was constructed to suit the Finnish food culture while imitating the original AHEI as closely as possible.35
Data for body mass index (BMI) and metabolic factors was collected during a health examination. Height and weight were measured by trained study nurses, with the participants only wearing light clothing and no shoes, and BMI was calculated. Normal weight was defined as BMI <25 kg/m2, overweight as 25 ≤BMI <30 kg/m2 and obesity as BMI ≥30 kg/m2. As the proportion of individuals with underweight was small (n=29), they were included in the group with normal weight. Waist circumference was measured and abdominal obesity was, in accordance with the International Diabetes Federations (IDF) metabolic syndrome (MetS) criteria, defined as a waist circumference of ≥80 cm for women and ≥94 cm for men.36
Blood pressure was measured twice using a standard mercury manometer, with 2 min intervals (Mercuro 300; Speidel & Keller, Jungingen, Germany). The mean of the two measurements was used. The use of antihypertensive medication was asked about during the interview. The IDF’s definition of elevated blood pressure was used: systolic pressure ≥130 mm Hg or diastolic pressure ≥85 mm Hg, or use of antihypertensive medication.36
Concentrations of serum triglycerides (automated enzymatic method, Olympus system reagent, Germany), serum HDL cholesterol (enzymatic method, Roche Diagnostics, Mannheim, Germany) and serum fasting glucose (hexokinase, Olympus System Reagent, Germany) were determined from frozen (−70C) serum samples. Categorization of these variables was conducted according to threshold values for the MetS: serum triglycerides (mmol/L) <1.7 and ≥1.7, serum HDL cholesterol (mmol/L) ≥1.03 in men or ≥1.29 in women and <1.03 in men or <1.29 in women and fasting serum glucose (mmol/L) <5.6 and ≥5.60.36 MetS was defined as having a waist circumference of ≥80 cm in women or ≥94 cm in men and meeting two or more of the aforementioned unfavorable values of serum triglycerides, serum HDL cholesterol, fasting serum glucose and blood pressure.36
Four variables representing different indicators of poor health were formed. Severe MetS was defined as an unfavorable value in each MetS component. Mental health status was determined with a self-administered questionnaire, including the General Health Questionnaire (GHQ).37 Individuals with a GHQ score >2 were categorized as having poor mental health. Self-perceived health was determined during an interview with a five-category question including the options: good, quite good, mediocre, quite poor, poor. Additionally, a two class-variable was formed including categories: 1) good, quite good and mediocre and 2) quite poor and poor. Specially trained physicians diagnosed osteoarthritis in the knee and hip joints during the health examination on the basis of physical status, symptoms and medical history, according to detailed written instructions with uniform diagnostic criteria.29
The study was conducted using a cohort study design with T2D incidence as the outcome. The T2D cases occurring during a 15-year follow-up were identified from nationwide registers covering information on medication use, hospitalization and cause of death, with the presence of any of the International Classification of Diseases, Tenth Revision codes E10–E14 (see online supplementary file S1). In Finland, under the Health Insurance Act, the costs of diabetes medication are reimbursed for patients with diabetes with a diagnosis from an attending physician.38 In order to receive the medication allowance, the physician must provide a certificate describing the diagnostic criteria applied for T2D diagnosis and the certificate must be checked and accepted by special advisers at the Social Insurance Institution of Finland (Kela). The nationwide register of patients receiving diabetes medication reimbursement is maintained by Kela. Moreover, information from the Finnish Hospital Discharge Register39 and the National Causes of Deaths Register were used. Study participants were linked to these registers with a unique social security number identifying each Finnish citizen. During the 15-year follow-up, 417 individuals (241 men and 176 women) developed T2D.
Statistical methods
Cox’s proportional hazards model40 was used to estimate the HR and its 95% CI of T2D in relation to the different predictors considered. The follow-up time was defined as the number of days from the baseline examination to the date of T2D occurrence, death or end of follow-up, whichever came first. Statistical significance was tested using the likelihood ratio test. Potential confounding factors were first selected based on the literature, and the variables which satisfied criteria for confounding in this data were included in the models.41 Since it is not easy to draw the line between confounding factors and mediators, four main effects models and one interaction model were defined. The first model included age, sex and an exposure variable in question. The second model included age, sex, waist circumference and IWL. The third model included age, sex, IWL, education (low, intermediate, high), alcohol consumption (none, moderate, heavy), leisure time PA (low, moderate, regular vigorous training), smoking status (never, past, current), AHEI (quintiles), energy intake (quintiles), BMI (continuous) and sleep duration (<6, 7–8, ≥9 hours/day). The fourth model included the variables of the third model and the variables of the MetS, that is, waist circumference (continuous), blood pressure (raised, normal), serum glucose (continuous), serum triglycerides (continuous) and serum HDL cholesterol (continuous). Finally, possible modification by sex, age, leisure time PA, body mass index, energy intake, AHEI, sleep duration and MetS on the prediction of the IWL on T2D risk was studied by including an interaction term between IWL and the potential effect modifying factor considered in the fourth model.
The calculations were performed using SAS (V.9.3, SAS Institute, Cary, North Carolina, USA).