Weight-Bearing Physical Activity, Lower-Limb Muscle Mass, and Risk of Knee Osteoarthritis

Key Points Question Are weight-bearing recreational physical activities associated with increased risk of knee osteoarthritis? Findings In this cohort study of 5003 participants, weight-bearing recreational physical activity was significantly associated with increased odds of incident knee osteoarthritis among participants with low levels of lower-limb muscle mass. Meaning This study provides evidence for future tailored physical activity recommendations based on a person’s muscle mass and osteoarthritis risk, which can help optimize the benefits of physical activity while minimizing the potential risk of developing osteoarthritis.


Study design and population
This study was embedded in the Rotterdam Study (RS), a large population-based prospective cohort study started in 1990.The design of the Rotterdam Study has been previously described in detail 1 (Figure 1).The Rotterdam Study has been approved by the Medical Ethics Committee of the Erasmus MC (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG).As a longitudinal population-based cohort Study, the Rotterdam Study has no pre-specified health exclusion criteria, meaning that all persons older than 55 years of age living in the area were invited to participate.The current analysis included 5003 participants from the three RS sub-cohorts (RS-I, RS-II, and RS-III) who had complete data of baseline recreational physical activity, baseline knee pain, and knee radiographs from baseline and at least one time from follow-up visits Participants with x-ray-defined osteoarthritis for one or both knees at baseline were excluded (Figure 2).
The observation periods are 1997 to 2002 for RS-I, 2000 to 2011 for RS-II, and 2006 to 2012 for RS-III, with an average of 6.3 years follow-up after baseline.Participants of RS-I and RS-II underwent two follow-up evaluations, while those in RS-III while those in RS-III had one follow-up assessment post-baseline.

Recreational physical activity
We assessed physical activity as an exposure variable using two validated questionnaires: an adapted version of the Zutphen Physical Activity Questionnaire for RS-I and RS-II and the Longitudinal Aging Study Amsterdam (LASA) physical activity questionnaire for RS-III 2,3 .Both questionnaires asked participants about the frequency and duration of various types of physical activity.For RS-I and RS-II participants, physical activity information was gathered using the Zutphen questionnaire, covering the two weeks preceding the interview.In contrast, participants in RS-III, interviewed with the LASA questionnaire, had their physical activity data collected separately for winter and summer periods and expressed as averages.The metabolic equivalent of task (MET) 4 , defined as the ratio of the rate of energy expended during an activity to the rate of energy expended at rest, was used to weigh the intensity of physical activity and all activities were expressed in MET*hours/week.
Based on data from a previous biomechanical study, physical activity levels were divided and summarized into weight-bearing and non-weight-bearing physical activity levels 5 .Non-weight-bearing activities were defined as those in which the knee joint did not bear the body weight, while weight-bearing activities were defined as those in which the knee joint did bear the body weight.All activities in the questionnaires were classified into weightbearing or non-weight-bearing activities, except for household work.Because household work is an umbrella term that covers multiple activities, both weight-bearing and non-weight-bearing, we could not distinguish.The detailed classification of physical activities is listed in Supplementary eTable 1.Total physical activity, weight-bearing physical activity and non-weight-bearing physical activity, all expressed in MET*hours/week, of each participant were the three exposures of this study.Total physical activity was defined as the combination of weight-bearing and non-weight-bearing activities.Since physical activity was assessed using different questionnaires across the RS cohorts.total physical activity, weight-bearing physical activity and non-weight-bearing physical activity were standardized into Z-scores per cohort.The original MET*hours/week physical activity data each unit represents after standardization are summarized in supplementary eTable 2.

Co-variates
We have used the following co-variates: Age, sex, BMI, baseline KLG, RS sub-cohorts, education level, alcohol intake, smoking, systolic blood pressure, HDL/total cholesterol ratio, and diabetes mellitus prevalence.Co-variates data were collected at baseline: age (years), education level (Primary education, Lower/intermediate general or lower vocational education, Intermediate vocational or higher general education, and Higher vocational education or university), alcohol intake (g/day), and smoking (Current smoker.former smoker, never smoker) were measured in-home interviews.BMI (kg/m 2 ), systolic blood pressure (mmHg), and HDL/total cholesterol ratio were measured at the research center.Diabetes mellitus prevalence data were collected by consulting medical records.The missing rate of all co-variates was below 3.4% except for alcohol intake (8.6%), and missing values were imputed by using the multiple imputation function from the MICE package in R 6 .Parameter settings (datasets=10, iteration=5) 7 .

Osteoarthritis
We assessed two outcomes: incident radiographic knee osteoarthritis based on x-ray radiographs and incident symptomatic knee osteoarthritis based on a self-reported pain questionnaire and x-ray radiographs.Radiographic knee osteoarthritis is the primary outcome and symptomatic knee osteoarthritis is secondary outcome.

Statistical analysis
The relationship between recreational physical activity at baseline (total, weight-bearing, and non-weight-bearing) and incidence of radiographic osteoarthritis and symptomatic osteoarthritis was assessed using logistic regression, adjusted for RS sub-cohorts, baseline KLG, follow-up time, age, sex, and BMI (model 1).When analyzing incident radiographic knee osteoarthritis, both knees from the same individual were included.A generalized estimating equation (GEE) was used to account for the correlation of the knees from the same participant.We also adjusted for education level, alcohol intake, smoking, systolic blood pressure, HDL/total cholesterol ratio, and diabetes mellitus prevalence (model 2).To study the influence of LMI on the relation between recreational physical activity and osteoarthritis, a pre-specified stratification analysis based on tertiles of LMI was conducted in the subgroup of participants with LMI data available (n=1881).To reduce the chance of type I error, Benjamin-Hochberg multiple testing correction was applied to all analysis results with false discovery rate of 0.05 10 .All analyses were done separately in participants with baseline pain and without.All statistical analyses were performed using R (version 4.2.1, R Foundation for Statistical Computing, Vienna, Austria).

Equity, diversity and inclusion statement
The authors include women and men in biology and clinical specialties from Europe and Asia.The study population included a spectrum of demographics.In discussing the generalizability and limitations of the findings, we acknowledge that RS is an ethnically non-diverse population cohort.

eTable 1. Classification of physical activities included Non-weight bearing activities Weight-bearing activities
© 2024 Wu Y et al.JAMA Netw Open.

eTable 2. Physical activities level of the Rotterdam Study sub-cohorts
Data are presented in mean and Standard deviations.Unit is MET*Hour per week.MET= Metabolic equivalent of task.eTable 3.

weight-bearing physical activity, mean (SD), MET*Hour/week
BMI= Body mass index, HDL= high-density lipoprotein.RS= Rotterdam study.KLG= the Kellgren and Lawrence Grade (KLG).MET*Hour/week = Hours of Metabolic equivalent of task per week.Continuous data are shown in mean with Standard deviations, and category variables are shown in the number of cases and percentage.

eTable 4. Baseline characteristics of the study population No (%) Total study populatio Study Population without baseline knee pai
BMI= Body mass index, HDL= high-density lipoprotein.RS= Rotterdam study.KLG= the Kellgren and Lawrence Grade (KLG).MET*Hour/week = Hours of Metabolic equivalent of task per week.Continuous data are shown in mean with Standard deviations, and category variables are shown in the number of cases and percentage.

eTable 5: Association of physical activities and incident symptomatic knee osteoarthritis in the pain-free population
* Indicates p-value remain significant after multiple testing corrections using Benjamini and Hochberg method.© 2024 Wu Y et al.JAMA Netw Open.eTable 6.

Association between physical activities and incident radiographic knee osteoarthritis in population with baseline knee pain
The statistical model used is a multivariate logistic regression model-Model1 adjusted for age, sex, Rotterdam study sub-cohorts, BMI, follow up time, and baseline KLG.Model 2 additionally adjusted for education level, alcohol intake, smoking, systolic blood pressure, HDL/total cholesterol ratio, and diabetes mellitus.OR= Odds ratio, CI= confidence interval.* Indicates p-value remain significant after multiple testing corrections using Benjamini and Hochberg method.eTable 7.

bearing physical activity, mean (SD), % of total physical activity
DXA= dual X-ray Absorptiometry, BMI= Body mass index, HDL= high-density lipoprotein.RS= Rotterdam study.KLG= the Kellgren and Lawrence Grade (KLG).MET*Hour/week = Hours of Metabolic equivalent of task per week.Continuous data are shown in mean with Standard deviations, and category variables are shown in the number of cases and percentage.

eTable 9. Stratification analysis on tertiles of lower-limb muscle mass index (LMI) for association between physical activities and incident radiographic knee osteoarthritis in population without baseline knee pain
© 2024 Wu Y et al.JAMA Netw Open.

eTable 10. Stratification analysis on tertiles of lower-limb muscle mass index (LMI) for association between physical activities and incident radiographic knee osteoarthritis in population with baseline knee pain
* Indicates pvalue remain statistically significant after multiple testing corrections using Benjamini and Hochberg method.© 2024 Wu Y et al.JAMA Netw Open.eTable 11.

Association between physical activities and incident radiographic knee osteoarthritis in populations without baseline knee pain and excluding baseline KLG>=1
* The statistical model used is a multivariate logistic regression model-Model1 adjusted for age, sex, Rotterdam study sub-cohorts, BMI, follow up time, and baseline KLG.Model 2 additionally adjusted for education level, alcohol intake, smoking, systolic blood pressure, HDL/total cholesterol ratio, and diabetes mellitus.* Indicates pvalue remain statistically significant after multiple testing corrections using Benjamini and Hochberg method.© 2024 Wu Y et al.JAMA Netw Open.eTable 12.

Association between physical activities and incident radiographic knee osteoarthritis in populations excluding baseline KLG>=1
The statistical model used is a multivariate logistic regression model-Model1 adjusted for age, sex, Rotterdam study sub-cohorts, BMI, follow up time, and baseline KLG.Model 2 additionally adjusted for education level, alcohol intake, smoking, systolic blood pressure, HDL/total cholesterol ratio, and diabetes mellitus.
* Indicates p-value remain statistically significant after multiple testing corrections using Benjamini and Hochberg method.