Data sources and searches
We registered the protocol for this systematic review and meta-analysis in the PROSPERO prospective register of systematic reviews (CRD42022327640). The conduct and reporting of this review adhered to PRISMA and MOOSE guidelines (25, 26) (Appendices 1-2). MEDLINE and Embase were searched from inception to 02 May 2022 with no language restrictions. The search strategy used a combination of MESH words or terms relating to the exposure (“physical activity”, “exercise”, “aerobic training”) and outcome (“chronic kidney disease”, “kidney failure”, “renal disease”). Details of the search strategy are presented in Appendix 3. One author (SKK) initially screened the titles and abstracts of the retrieved citations to assess their potential for inclusion. This was conducted using Rayyan (http://rayyan.qcri.org), an online bibliographic tool that helps to expedite the screening process using a process of semi-automation.(27) Full texts of the selected titles and abstracts were retrieved and detailed evaluation was done, which was independently conducted by three authors (SKK, MA and SS). To identify potential articles missed by the search of databases, manual scanning of reference lists of relevant studies and review articles was performed, and Web of Science was used to do a cited reference search.
Study selection
We included all population-based observational cohort (retrospective or prospective designs) studies that had evaluated the relationship between physical activity and risk of incident CKD in adult general populations that had reported at least one year follow-up duration for the ascertainment of outcomes. The following studies were excluded: (i) case-control and cross-sectional studies because of their lack of temporality; (ii) those involving elite athletes and/or evaluated competitive or endurance sports; and (iii) those evaluating the associations between measures of fitness (eg, cardiorespiratory fitness, physical fitness, exercise capacity) and risk of CKD; and (iv) those conducted in people with pre-existing diseases.
Data extraction and risk of bias assessment
Using a standardized data collection form which has been used for previous reviews of a similar nature,(12, 13, 15) one author (SKK) extracted relevant data from the eligible studies and two other authors (MA and SS) independently checked the data using the original articles. We extracted data on the following study characteristics: first author surname and year of publication, geographical location, year of recruitment/baseline data collection, specific study design, demographic characteristics (age and percentage of males), sample size, duration of follow-up, physical activity type and assessment method, definition of CKD, number of CKD events, risk comparisons, the most fully-adjusted risk ratios for CKD (and corresponding 95% confidence interval [CIs]), list of covariates adjusted for, and level of adjustment (‘+’ defined as minimally adjusted analysis, i.e. age and/or sex; ‘++’ as adjustment for conventional risk factors for CKD excluding inflammation, i.e. age and/or sex plus body mass index, socioeconomic status, alcohol consumption, smoking, and comorbidities ; and ‘+++” as adjustment for conventional risk factors plus inflammation). When there were multiple publications of the same cohort, we extracted data from the most recent study to avoid double counting the same cohort in the pooled analysis. The criterion for selection was the one with the most extended follow-up or analysis covering the largest number of participants and events). The risk of bias within individual observational studies was assessed using the Cochrane Risk of Bias in Non-randomised Studies – of Interventions (ROBINS-I) tool.(28) The risk of bias is assessed for the following domains: confounding, participant selection, classification of interventions, deviations from intended interventions, missing data, outcome measurements, and selective reporting. For each domain, the risk is quantified as low risk, moderate risk, serious risk, or critical risk and then an overall judgement of the risk of bias is provided for each study. The Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool was also used to assess the quality of the body of evidence, based on the following criteria: study limitations, inconsistency of effect, imprecision, indirectness and publication bias.(29)
Data analysis
Relative risks (RRs) with 95% confidence intervals (CIs) were used as the summary measures of association. All studies categorised physical activity exposure (e.g., leisure-time physical activity, total or any physical activity) into user-defined categories or quantiles. Due to the varied reporting of the RR comparisons, they could not be transformed to consistent comparisons (e.g., top versus bottom quantiles of the distribution of physical activity) using standard statistical methods previously described.(30, 31) However, to provide some consistency and enhance comparison and interpretation of the findings, the extreme groups (i.e., the top versus bottom or maximum versus the minimal amount of physical activity) reported for each study were used for the analyses. Several previous meta-analyses have utilised this approach (12, 13, 32, 33) and it is considered reliable as there is documented data that pooled estimates from transformed and untransformed data are qualitatively similar.(34) When a study reported specific types of physical activity in addition to any or total physical activity, we only used risk estimates for any or total physical activity in the pooled analysis as done for previous similar reviews.(12, 13, 32, 33) Relative risks were pooled using a random effects model to account for the effect of heterogeneity.(35) The extent of statistical heterogeneity across studies was quantified by standard chi-square tests and the I2 statistic.(36, 37) To determine the degree of heterogeneity, we also estimated 95% prediction intervals, which provide a region in which about 95% of the true effects of a new study are expected to be found.(38, 39) We explored for evidence of effect modification on the association (sources of heterogeneity) using pre-specified study-level characteristics such as geographical location, observational cohort design (prospective vs retrospective), the average age at baseline, the average duration of follow-up and number of CKD events, which was conducted using stratified analysis and random effects meta-regression.(40) To test the robustness of the observed association, we conducted a sensitivity analysis by investigating the influence of omitting each study in turn on the overall result (stata module metaninf). To explore for small study effects, we visually inspected constructed Begg’s funnel plots(41) and performed Egger’s regression symmetry test.(42) We employed Stata version MP 17 (Stata Corp, College Station, Texas) for all statistical analyses.