Study population
The data were from the Finnish Public Sector study, an ongoing prospective cohort study with identifiable questionnaire surveys. The eligible population of the original cohort included all employees who had been working for a minimum of six months in the target organizations, which included ten towns and six hospital districts, between 1991 and 2005 (n = 151,901) [11]. Nested survey cohorts included all those who were employed by the participating organizations at the time of surveys or had left the organizations after participating in an earlier survey, repeated by 4-year intervals. Surveys used in this study were collected in 2000-2002, 2004/2005, 2008/2009 and 2012/2013. Survey data for cohort members were successfully linked to employers’ records and comprehensive national health registers through unique personal identification codes, which are assigned to all citizens in Finland. The FPS study was approved by the Ethics Committee of the Hospital District of Helsinki and Uusimaa.
Of the cohort members, we identified persons who were at work and responded to at least one survey in 2000/2002, 2004/2005 or 2008/2009 (n = 81,587). Of those employees, 19,058 persons retired between 2000 and 2011. For this study we included persons who retired at statutory retirement age (statutory retirement, i.e., old age retirement), persons who entered to part-time retirement not due to health reasons (part-time retirement) and person who retired early on health grounds (from now on called as disability retirement). In case of entitlement to several pensions from different pension schemes at different times, the first one of awarded of above mentioned retirements types was selected.
For analytical purposes, the day of actual retirement was set as 0. From that point the duration to all previous and subsequent waves was calculated (-10, −6, −2, +2, +6, +10). In this study the pre-retirement period constitutes years −10 and −6, retirement transition years −2 and +2 and post-retirement period years +6 and +10. The difference between each wave is four years and years are averages across study population.
For this study, we included participants who had data on physical activity immediately before and after retirement (years −2 and +2) comprising a sample of 9,488 persons (80 % women). On average, participants provided physical activity data at 3.6 (range 2–4) of the possible four study waves.
Assessment of retirement
Data on retirement were obtained from the Finnish Centre for Pensions, which coordinates all earnings-related pensions for permanent residents in Finland [12]. All gainful employment is insured in a pension plan and accrues a pension; thus the pension data with successful linkage were available for all participants. The start dates for any pension were obtained for all participants from 2000 through 2011, irrespective of the participants’ employment status or workplace at follow-up.
According to the public sector Employees’ Pension Act, the statutory retirement age was generally from 63 to 65 years until 2005 and 63 to 68 years from 2005 onwards, although some individuals had kept their earlier retirement age from the previous pension act in which pension ages in some occupations were below 63 years (e.g., 60 years for primary school teachers, 58 for practical nurses). Part-time pension may be granted to a person who is at least 60 years old and who is transitioning from full-time work to part-time work. A disability pension may be granted if, due to an illness or injury, the employee cannot continue working even after attempts of rehabilitation, re-education, or assistance. In Finland employees may apply for disability pension when more than 300 reimbursed sickness absence days have accumulated during two consecutive years on the basis of the condition causing work disability. Around 70–80 % of all disability pension applications have been accepted [13].
Assessment of physical activity
Physical activity was assessed identically at each study wave. The respondents were asked to estimate their average weekly hours of leisure–time physical activity (including commuting) within the previous year in walking, brisk walking, jogging, and running, or their equivalent activities [14]. Each intensity grade had five response alternatives of which the class mid–points were used for the calculation of time spent in physical activity: no activity, less than 0.5 h (15 min used for calculation), ~1 h (45 min), 2–3 h (2.5 h), and ≥ 4 h/week (5 h). Cronbach’s alpha for the response categories was 0.68. The time spent on activity at each intensity level in hours per week was multiplied by the average energy expenditure of each activity, expressed in metabolic equivalent (MET). For example, walking, brisk walking, jogging and running corresponded to 3.5, 5, 8 and 11 METs, respectively [15]. In addition, the time spent (hours/week) in moderate intensity physical activity (walking, and brisk walking, or their equivalent activities) and vigorous physical activity (jogging, and running, or their equivalent activities) were calculated separately. Based on current physical activity recommendations the respondents were also classified as inactive if the weekly physical activity was less than 14 MET hours per week [16].
Assessment of covariates
Sex and occupational title were obtained from the employers’ registers. Based on occupational titles, occupational status was categorized to upper-grade non-manual workers (e.g. teachers, physicians), lower-grade non-manual workers (e.g. registered nurses, technicians), and manual workers (e.g. cleaners, maintenance workers) [17], and was determined by the last occupation preceding retirement.
Life-style related factors, that is body mass index (BMI), smoking and alcohol use, were based on the last questionnaire prior to retirement (wave -1). BMI was calculated from self-reported weight and height (kg/m2) and categorized into normal weight (BMI < 25.0 kg/m2), overweight (BMI 25–29.9 kg/m2) and obesity (BMI ≥ 30 kg/m2) [18]. Smoking status was categorized into never, former and current smokers. Alcohol use was categorized into none, moderate and heavy. The limit for heavy alcohol use was set to >16 drinks/week for women, >21 drinks/week for men which correspond with the medium risk levels of daily consumption set by the World Health Organization [19].
Disease status was constructed by taking into account chronic diseases in all pre-retirement waves available. Data on chronic illnesses was based on eligibility for special reimbursement based on the Social Insurance Institution of Finland’s Drug Reimbursement Register (asthma, diabetes, rheumatoid arthritis, coronary heart disease and depression), the Finnish Cancer Registry (cancer) and the questionnaires (osteoarthritis). For the analyses, the number of chronic diseases were modelled as a time-variant variable and participants were categorized as having no disease, one disease and two or more diseases.
Statistical analyses
Characteristics of the study population before retirement (year −2) for each retirement type are presented as mean values for continuous variables and as proportions for categorical variables. Trajectories of weekly MET hours (continuous outcome) were assessed using linear regression analyses with generalized estimation equations (GEE), and trajectories of physical inactivity (binary outcome) using log-binominal regression analysis with GEE. The GEE models control for the intra-individual correlation between repeated measurements using an exchangeable correlation structure and is not sensitive to measurements missing completely at random [20, 21].
To examine whether weekly average MET hours differed between pre-retirement period (years −10 to −6), during retirement transition (years −2 and +2) and post-retirement period (years +6 to +10), we tested period x time interaction effects separately for the different retirement types. In addition, interaction test of retirement type x period x time was conducted to examine the differences between retirement types. Adjusted mean estimates and their 95 % confidence intervals were calculated to represent an average of 4-year change of weekly total MET hours and change of hours of moderate-level and vigorous physical activity at different periods within retirement type. The analyses were adjusted for retirement age, sex, occupational status, life-style factors and number of chronic conditions before retirement.
We also examined whether retirement age, sex, occupational status and number of chronic conditions were associated with changes in weekly MET hours and physical inactivity during retirement transition and post-retirement by using contrast statements in GEE models. The models were adjusted for retirement age, sex, occupational status, life-style factors and number of chronic conditions before retirement.
Finally, to examine the role of selection (attrition bias), we conducted sensitivity analyses by repeating all analyses among those who had all four physical activity measurements available. The SAS 9.4 Statistical Package was used for all of the analyses (SAS Institute Inc., Cary, NC).
