Xuxi Zhang

Covariates Some covariates were assessed at baseline, including age (in years), sex, country, educational level, living situation, smoking, alcohol risk and multi ‐ morbidity. Educational level concerned the highest level of education completed by the participant and was categorized according to the 2011 International Standard Classification of Education (ISCED) into primary or less (ISCED 0 ‐ 1), secondary or equivalent (ISCED 2 ‐ 5), and tertiary or higher (ISCED 6 ‐ 8). 23 Living situation was categorized as ‘not living with others’ or ‘living with others’ (a partner, child(ren) and/or others). Smoking was measured with one item that assessed whether a person currently smoked. Alcohol risk was measured with the Alcohol Use Disorder Identification Test (AUDIT ‐ C), 24 which is a 3 ‐ item screener to grade high ‐ risk alcohol use on a scale from 0 (lowest risk) to 12 (highest risk). A score of at least 4 for men and 3 for women was regarded as hazardous drinking or active alcohol use disorder. 24 Multi ‐ morbidity was defined as having at least two of the following 14 chronic conditions 25 : heart attack, hypertension, diabetes, stroke, high blood cholesterol, asthma, arthritis, osteoporosis, chronic lung disease, cancer or malignant tumor, stomach or duodenal ulcer, Parkinson’s disease, cataract and hip fracture or femoral fracture. 26 Statistical analysis The longitudinal association between frequency of moderate PA and frailty was estimated with multivariate linear regression models. Four separate regression models were built for overall, physical, psychological or social frailty at follow ‐ up as dependent variable, and frequency of moderate PA at baseline as independent variable. The first set of models were adjusted for country and for frailty at baseline ( crude model ). The second set of models were additionally adjusted for age, sex, educational level, living situation, smoking, alcohol risk and multi ‐ morbidity ( adjusted model ). Since the UHCE project was an intervention study and participants were divided over an intervention and a control group, intervention condition was also added to the adjusted mode as a covariate. The association between the 12 ‐ month change in frequency of moderate PA and overall, physical, psychological or social frailty was assessed using the same crude and adjusted multivariate linear regression models as described above, taking change in frequency of moderate PA as the independent variable. Furthermore, interactions between baseline frequency of moderate PA or 12 ‐ month change in frequency of moderate PA and age, sex, country, educational level, living situation and intervention on the frailty scores were assessed with UNIANOVA. Bonferroni correction was applied for multiple testing ( P = 0.05/48 = 0.001). Apart from an interaction between country and change in frequency of moderate PA regarding psychological frailty, no statistically significant interaction was found. All P ‐ values of the interaction analyses are presented in Supplementary Table S1. 3 55 Longitudinal association between physical activity and frailty