Physical activity (PA) is recognised to be a major protective factor against the development of several chronic diseases associated with ageing, such as type 2 diabetes, cardiovascular disease, weight gain and some cancers, as well as cognitive decline 1 2 3 . In older people, PA may also help to decrease falls and disability, and improve independence 4 5 6 . In addition to PA, there is increasing interest in the relationships between sedentary behaviour and health outcomes 7 8 . Sedentary behaviour refers to activities that do not substantially increase energy expenditure above the resting level; they include sleeping, sitting, lying down and watching television, along with other forms of screen-based entertainment 9 . Sedentary behaviour, typically assessed by time spent viewing television (TV), along with low PA levels, represent complementary aspects of human movement and are independent risk factors for major chronic diseases 2 10 .

Retirement is a major transition in the life course and concerns a growing segment of the population. The number of persons who reach retirement age is rapidly increasing in developed countries. According to the World Health Organisation, in the year 2000, 600 million people were aged 60 or above throughout the world; this figure will reach 1.2 billion by 2025 and 2 billion by 2050 11 . Retired persons generally have more time for leisure activities, potentially including leisure-time physical activity (LTPA, including structured activities such as sports, and unstructured activities such as walking for pleasure). Retired individuals may also have more time to spend on sedentary occupations such as TV viewing. However, the manner in which retirement affects PA and sedentary behaviour is poorly understood.

Cross-sectional surveys from the US and Europe suggested that PA patterns change at retirement 12 13 14 15 ; however, few longitudinal data exist which describe these changes in the same individuals 16 17 18 19 20 . Discrepancies exist in the findings of these studies; one study found no relationship between retirement and sports activities or non-sports LTPA 18 , while others observed that retirement was associated with an increase in sports activities and exercise 17 or increased overall PA 19 20 . Except for one study 17 , detailed information on changes in specific leisure-time activities is lacking. Moreover, to our knowledge, only one study examined the effect of retirement on sedentary behaviour 17 ; it concluded that retirement was associated with increased TV viewing. However, whether or not LTPA and sedentary behaviour changes are interrelated remains unknown.

The objective of the present study, which used a longitudinal design in a sample of middle-aged French adults, was to investigate the relationship between retirement and 3-year changes in LTPA and sedentary behaviour, as assessed by time spent watching TV.

Our study assessed the influence of retirement on LTPA and sedentary behaviour in French subjects. In both genders, retirement was associated with an increase in LTPA, especially with increased duration of moderate LTPA and walking. Retirement did not modify the ranking of the most frequently performed LTPAs, but the number of participants and the duration increased at retirement. In subjects who retired from less physically demanding jobs, retirement was associated with an increase in time spent watching TV. In women retirees, a high increase in time spent walking was associated with a decrease in TV viewing.

In our study, the retirement-related increase in LTPA was substantial, as it represented about 2 h/week in both genders. This increase in LTPA is consistent with previous findings 17 19 20 . However, unlike previous studies that showed modifications only in population subgroups 16 , we observed a positive impact of retirement on LTPA whether or not subjects initially met recommended PA levels and whatever the physical demands at work prior to retirement. An increase in LTPA of 2 h/week on average may be substantial in terms of potential health-related benefits. A recent trial examined in inactive overweight women the effect of exercising at 50%, 100%, and 150% of the currently recommended physical activity dose (150 min/week of moderate-intensity physical activity 1 2 ) on cardiorespiratory fitness 34 . Subjects experienced a graded dose-response change in fitness across levels of exercise training, with statistically significant improvement observed even for the lowest intervention level 34 . Increases in cardiorespiratory fitness are associated with decreased mortality risk 1 2 .

There are several potential explanations for an increase in LTPA with retirement, including reduced time-related barriers, long-term perspective on health, concern about health and independence and changes in social networks, support systems and daily routines 17 35 . However, in our study, the increase in LTPA associated with retirement did not compensate for the loss of occupational PA.

The effect of retirement assessed in our study may be a short-term effect, corresponding to what happened 1-to-3 years after retirement. Caution should be used when extrapolating these findings to the entire post-retirement period. Indeed, cross-sectional data suggested that at retirement, there existed a tendency in improving physical activity patterns, but this positive effect decreased with time after retirement, to disapeare during the final period of life 15 . In the GLOBE study 18 , including a 13-year interval between baseline and final PA measures, no significant relationship was observed between retirement and LTPA. One can hypothesize that there may have been an increase in LTPA in the first years following retirement, but this increase was no longer visible 10-13 years after retirement.

Our study offered the opportunity of investigating in detail specific LTPAs performed in each gender and potential modifications of such LTPA patterns during retirement. One major finding was that retirement quantitatively modified the LTPA pattern (an increase in the number of participants and in the duration), but not qualitatively (there was no change in gender-specific ranking of the most frequent LTPAs). In our study, walking was the most frequently performed LTPA and was the main contributor to the retirement-related LTPA increase. In the ARIC study, walking was the most frequently cited activity by subjects who adopted a LTPA or a sports activity at retirement 17 . Walking is easily performed, requires no specific equipment, does not incur any substantial cost and is the basis for public health recommendations concerning PA 2 .

In our study, the increase in LTPA associated with retirement occurred concurrently with an increase in time spent watching TV. Changes in time spent watching TV had been assessed in only one previous study on retirement, in which an increase was observed, in agreement with our results 17 . The relationship between PA and sedentary behaviour is complex; it has been pointed out that 'sedentary' is not simply the opposite of 'active' 9 10 36 . Several studies have reported that sedentary behaviour and PA are related to risk of major chronic disease independently of one another 27 37 38 39 . Some of our findings contrast with this notion of independence of PA and sedentary behaviour. Indeed, we found that subjects who had held a less physically demanding profession were more inclined to adopt sedentary behaviour during their leisure time once retired, independently of age or education level. This suggests that this group should be targeted for measures aimed at limiting increases in sedentary behaviour at retirement. Moreover, overall, the retirement related 2 h/week increase in LTPA did not compensate for the 30-40 min/day increase in TV viewing. However, in stratified analyses, only women who declared the highest increase in walking duration after retirement simultaneously decreased TV viewing. This suggests that certain groups of population exist for whom the benefit of retirement is enhanced, since these subjects compensate for the decrease in sedentary occupations by increasing their LTPAs.

In contrast to our highly-detailed PA questionnaire, our proxy for sedentary behaviour was limited to time spent watching TV, which may have led to an underestimation of time devoted to sedentary occupations. To better assess attributes of a sedentary lifestyle, an evaluation of time spent sitting down while traveling and at work, along with other screen-based sedentary occupations such as computer use during leisure time, would be of interest. Indeed, sedentary lifestyle represents a complex set of behaviours which may each have separate effects on health outcome 28 40 and may be modified differentially by retirement. It should be noted, however, that in women, TV viewing has been suggested as a robust marker of overall sedentary lifestyle 41 .

Several limitations to our study should be pointed out. First, despite a mean retirement age in our population (57-60 y) consistent with retirement age in France in the early 2000's (55 y for some professions such as teachers, and 60 years for the majority of other professions, mean age: 58.1 y 42 ), our subjects were volunteers participating in a nutritional intervention study 22 who generally had a higher education level and occupational status, along with a healthier lifestyle than the general population 21 . Thus, our study sample cannot be considered as nationally representative of individuals aged 45-64, and caution is needed when extrapolating these findings to the national level. Our analyses focused on a part of the SU.VI.MAX cohort; however, sociodemographic differences observed between those subjects who were and those who were not included in the analyses were unlikely to have caused additional bias in the studied associations. Second, measurements of LTPA and TV viewing were derived from self-reporting, which may be a source of potential misclassification bias (especially due to over-reporting of PA 43 ). However, there is no reason to expect that misclassification would differ according to retirement status. Third, our study did not explore the reasons for retirement, which may have differential effects on PA and other types of health behaviour. A possible selection bias could have occurred if participants in poor health had retired earlier for that reason, and if poor health rather than retirement per se negatively influenced PA. However, in our study, this type of bias would not explain the observed increase in LTPA and would have minimised the true effect. Moreover, a previous study showed that tobacco use and alcohol intake, but not PA, were differentially affected by voluntary or involuntary retirement 19 . Validity of using the MAQ in older subjects and in a longitudinal design may be discussed. For reliability, in the first description of the MAQ by Kriska et al. 24 , 1-3 week test-retest correlations for past-year LTPA were found of about the same magnitude in the older compared to the younger adult subjects (37-59 y, rho = 0.88 and 21-36 y, rho = 0.92, respectively). For validity (i.e. comparison against a gold standard method), we are not aware of study data that have assessed the performance of the MAQ according to age groups. Regarding the appropriateness of using the MAQ in a longitudinal design, repeated use of this questionnaire allowed to detect annual evolution of PA during follow-up (median: 2.8 y) in the Diabetes Prevention Program 25 . In a subsample of DPP participants (n = 274, 50.6 ± 11.3 y), LTPA (MET-h/week) measured by the MAQ was correlated with self-reported minutes of physical activity at both 1 year (r = 0.41, P < 0.0001) and 2 years (r = 0.51, P < 0.0001) 44 .

One of the strengths of our study lay in its prospective design and the use of a detailed PA questionnaire, which specified the types of LTPA performed during the preceding year and the duration of each reported activity. In addition, to our knowledge, only one previous study analysed the influence of retirement on both PA and sedentary behaviour 17 . Our data enabled adjustment for confounding factors and considered individual heterogeneity by stratified analyses. Moreover, in previous studies 12 17 18 , definitions of work and retirement were not mutually exclusive, since some subjects were considered as retired, but were still working part-time or full-time, and retirement status was reversible. In our study, the definition of retirement was less ambiguous: we considered retired persons to be subjects who had definitively stopped all professional activity. Indeed, in France, in 1998-2001, people were considered "retired" and started to receive their retirement allocations only when they stopped any professional activity. Thus, a new profession would not have been allowed. As for potential volunteer implication in associations or charities, corresponding physical activity was counted as LTPA.

Our study provides important information about retirement-related changes in physical activity and sedentary behaviour. Few data exist on this topic, despite the increasing number of ageing individuals concerned 11 and the major health consequences related to insufficient physical activity and excessive sedentary behaviour 1 2 3 . Our results suggest that retirement is a key period of change in PA and sedentary behaviour. The substantial 2 h/week increase in LTPA through retirement observed in our study did not compensate for loss of occupational PA. This emphasises the need to reinforce public health measures aimed at increasing regular PA and decreasing sedentary behaviour during the critical retirement period, as well as during other lifetime transition periods.

The authors declare that they have no competing interests.

MT performed statistical analysis, interpreted results and drafted the manuscript. SB contributed to designing the study, performing analyses and interpreting results. HC and ACV contributed to interpreting results and editing the manuscript. SH is coordinator of the SU.VI.MAX study and contributed to interpreting results and editing the manuscript. JMO supervised the design, analysis and interpretation of data as well as writing of the manuscript. All authors critically reviewed the manuscript and approved the final version.