Xuxi Zhang

able help them realize that their body was still working well enough to perform the activity. 29 Regarding social frailty, an RCT in Spain found that a multicomponent exercise program was not only able to improve the physical aspects of frailty, but also to increase interaction with other people which could reduce the level of social frailty. 30, 31 However, studies to investigate the association between PA and psychological and social frailty among older adults are still scarce and more studies are needed. 27, 32 Finally, regarding psychological frailty, we found an interaction between 12 ‐ month change in moderate PA and country; in the Netherlands the results were different from the results from the other four countries (see Supplementary Table S4). More studies are needed to clarify this finding. Strengths and limitations One strength of our study is that we added longitudinal evidence on the association between PA and frailty among citizens aged 70 years or older from a diverse community ‐ based sample from five European countries. In addition, we used a validated instrument in order to consider frailty broadly from the physical, psychological and social perspectives, and to add to the current literature on the association between change in PA and the three domains of frailty. Social frailty in particular is a rarely explored domain and there is a dearth of studies on this subject. 33, 34 However, our study also has some limitations. First, PA was measured by one self ‐ reported question which is fairly crude and open to interpretation. This question does not differentiate between type of activity and does not take the duration of activity into account. Studies using a more comprehensive measurement of PA are needed to confirm our findings. However, some previous studies 35 ‐ 37 indicate that using a single question to measure PA is acceptable under certain conditions, e.g. when the sample size is large, when more complex methods would add to respondent burden, and when collecting data from a broad range of settings. Grill et al. (2012) also suggest that the reliability and validity of a single question to briefly classify PA levels is acceptable. 38 Therefore, taking into account the large sample size, the response burden and the aim of the study, we believe that using a single question to measure the frequency of PA is acceptable. Second, we transferred the ordinal variable of PA into a dichotomous one which might cause information loss. However, we conducted additional analyses on the association between PA and frailty with the ordinal variable of PA (Supplementary Figure S1), and the results were similar to our primary findings. Third, we found statistically significant differences in frailty scores between baseline and follow ‐ up. This finding was based on statistical methods rather than on clinical examinations. Hence, we cannot draw conclusions on the clinical meaning of the TFI scores. Future studies should explore whether this statistical difference corresponds to a clinically meaningful change in frailty level. Fourth, participants in both the intervention and control groups were included in the analyses. The intervention may have led to improvement in health which could result in the over ‐ estimation of the effect of PA on frailty. However, we controlled for the intervention 3 63 Longitudinal association between physical activity and frailty