Baseline imbalance and heterogeneity are present in meta-analyses of randomized clinical trials examining the effects of exercise and medicines for blood pressure management

Meta-analyses of randomized clinical trials (RCTs) inform clinical guidelines by synthesizing evidence. RCTs aim to create comparable groups at baseline so that post-intervention differences reflect treatment effects. However, baseline imbalances in prognostic factors (e.g., age, disease severity, blood pressure) can occur by chance or due to biases such as inadequate allocation concealment or deviations from intention-to-treat analyses. Prior research has shown that baseline imbalances can distort meta-analytic conclusions, as seen in meta-analyses of calcium for weight loss and oseltamivir for influenza. The authors previously observed notable baseline imbalances in systolic and diastolic blood pressure in an isometric exercise meta-analysis. This study examines the extent of baseline imbalance and heterogeneity across exercise and antihypertensive medicine RCTs and explores whether sample size, allocation concealment, and completeness of baseline reporting are associated with such imbalances.

Evidence from previous meta-epidemiological work indicates baseline imbalances can bias treatment effect estimates. For instance, in a calcium supplementation meta-analysis for weight loss, treatment groups had lower baseline body mass than controls, and adjusting for this removed the apparent treatment effect. Similarly, an oseltamivir meta-analysis found fewer influenza-positive participants in treatment groups than controls, suggesting allocation bias. The authors also reported substantial baseline SBP and DBP imbalances in a prior meta-analysis of isometric exercise versus nonexercise control, largely driven by a single large trial. These findings motivated a broader investigation of baseline imbalances across multiple exercise modalities and antihypertensive medicines.

Protocol preregistered on OSF (osf.io/dgu9b). Data were drawn from 391 RCTs (197 exercise and 194 antihypertensive medicine trials) included in a published network meta-analysis comparing exercise modes (endurance, resistance, isometric, combined) and antihypertensive medicine classes. For each included RCT, two authors independently extracted baseline mean and standard deviation (or converted equivalents) for systolic blood pressure (SBP), diastolic blood pressure (DBP), and age, prioritizing data reported for all randomized participants (typically Table 1) and secondarily data for analyzed participants. Values reported as medians with ranges/interquartile ranges were transformed to means and SDs using standard methods; some data were digitized from figures (WebPlotDigitizer). Where multiple arms evaluated the same exercise mode or medicine type, those arms were combined for analysis. Allocation concealment risk of bias was independently appraised (Cochrane RoB guidance). Studies not reporting baseline data for all randomized participants were flagged. Due to the large number of studies and the age of many medicine trials, authors were not contacted for missing data. Duplicate baseline datasets identified in four endurance vs control comparisons were removed, retaining the original publications. Statistical analyses focused on baseline imbalance across SBP, DBP, and age using fixed-effect meta-analyses. Heterogeneity and inconsistency were assessed with Cochran’s Q and I² statistics. Meta-regressions examined moderators: sample size, allocation concealment risk of bias, and completeness of baseline data reporting (all randomized vs not). Analyses were conducted separately by comparison (e.g., endurance vs control, resistance vs control, etc.).

- Exercise RCTs included: 190 (parallel-group designs); 7 low risk, 4 high risk, and 179 unclear risk of bias for allocation concealment; 94 reported baseline data for all randomized participants. - SBP at baseline (exercise comparisons): No baseline imbalances detected in any comparison. Heterogeneity observed: resistance vs control I²=33.0%; endurance vs control I²=14.4%; resistance vs combined I²=22.0%. No significant moderator effects for SBP. - DBP at baseline (exercise comparisons): No baseline imbalances detected. Notable heterogeneity: endurance vs control I²=30.3%; resistance vs control I²=41.0%; resistance vs combined I²=35.4%. Significant moderators: sample size in endurance vs control (β=0.00, 95% CI 0.00 to 0.01, p<0.01; larger samples associated with higher DBP in endurance groups) and resistance vs control (β=-0.06, 95% CI -0.11 to -0.01, p=0.01; larger samples associated with higher DBP in control groups). Completeness of baseline data reporting: endurance vs control difference between subgroups = -1.30 mmHg (95% CI -1.84 to -0.77), p<0.01; resistance vs combined difference between subgroups = 5.15 mmHg (95% CI 0.23 to 10.08), p=0.04. - Age at baseline (exercise comparisons): A small but statistically significant baseline imbalance in resistance vs control: mean difference -0.3 years (95% CI -0.6 to -0.1), indicating resistance groups were younger. Low-to-moderate inconsistency in some comparisons (e.g., endurance vs control I²=14.1%). - Medicines: Fewer data were available. No baseline imbalances detected and only a few instances of inconsistency were noted. - Overall: Baseline imbalances and substantial inconsistency were identified in some exercise comparisons; several moderator analyses (sample size and completeness of baseline reporting) were significant for DBP.

The study assessed whether baseline characteristics are balanced across RCT arms within meta-analyses of exercise and antihypertensive medicines. Findings showed generally balanced SBP and DBP at baseline across exercise comparisons, but a small age imbalance in resistance vs control suggests potential selection or reporting issues. Heterogeneity was more prominent in exercise comparisons than in medicines, indicating variability across trials that could influence interpretation of treatment effects. Significant moderator associations for DBP imply that larger sample sizes and whether studies reported baseline data for all randomized participants can systematically affect observed baseline differences, highlighting potential biases due to study conduct and reporting practices. The relative absence of baseline imbalances in medicine trials, albeit with limited data, may reflect differences in trial methodology, reporting standards, or eras of publication. Overall, the results reinforce that unrecognized baseline imbalances and heterogeneity can threaten causal inference in meta-analyses and that proactive evaluation of baseline prognostic variables is important for ensuring valid conclusions.

This meta-epidemiological analysis of RCTs comparing exercise modalities and antihypertensive medicines found largely balanced baseline SBP and DBP, but identified a small age imbalance in resistance exercise vs control and notable heterogeneity in several exercise comparisons. Medicines showed no baseline imbalances and minimal inconsistency, though fewer data were available. The study underscores the need for meta-analysts to routinely examine baseline prognostic variables to detect potential biases. Future research should standardize reporting of baseline data for all randomized participants, improve allocation concealment and risk-of-bias practices, explore sources of heterogeneity, and extend assessments to additional prognostic factors and more comprehensive datasets of medicine trials.

- Fewer data available for medicine trials limited precision and generalizability of findings for medicines. - Most trials had unclear risk of bias for allocation concealment; limited ability to detect associations with selection bias. - Baseline data were sometimes derived from medians/ranges or digitized from figures, introducing potential measurement error. - Authors were not contacted for missing data due to volume and age of studies, potentially biasing inclusion toward more completely reported trials. - Analyses focused on three baseline variables (SBP, DBP, age); other prognostic factors were not assessed. - Fixed-effect meta-analyses were used, which may not fully account for between-study variability. - Publication date disparities between exercise and medicine trials (mean difference ~14 years) may reflect differing reporting standards or methodologies across eras.