Why ANCOVA Should Be Standard Practice in Rehabilitation Clinical Trials?

Abstract

Randomized Controlled Trials (RCTs) in rehabilitation often involve heterogeneous patient populations and small sample sizes, making precise estimation of treatment effects challenging. Analysis of covariance (ANCOVA) is gaining prominence as a superior statistical method in RCTs, especially in rehabilitation sciences, due to its ability to adjust for baseline differences and increase statistical power compared to unadjusted or change-score analyses.^1,2 This adjustment is important in rehabilitation trials where there is a huge variation in functional status at the baseline level.

Recent methodological studies reveal that ANCOVA provides unbiased, more precise and highly statistically enhanced estimation of treatment effects, even when model assumptions are moderately violated.³ For example, Wang et al. reported that ANCOVA ensures the integrity of inferential conclusions and enhances power in trials with continuous and binary outcomes. ⁴ Additionally, ANCOVA can decrease the required sample sizes by up to 25%, maximizing resource efficiency in clinical research.⁵

In rehabilitation sciences, where intervention protocols are complex and multi-component and the outcomes are multi-dimensional, ANCOVA’s ability to control for baseline heterogeneity enhances interpretability and trial rigour.⁶ Therefore, ANCOVA should be the standard analytic approach in rehabilitation RCTs to ensure accurate and efficient evaluation of therapeutic interventions.

Before using the ANCOVA, it must be noted that the dependent variable must be continuous and must be normally distributed, the covariate must be a continuous quantitative variable, the levels of the qualitative variable must be independent ⁷, the dependent variable and covariate should have a linear relationship⁸ the positive or negative sign and magnitude of the correlation coefficient at each level of the qualitative variable should be similar between the dependent variable and the covariate. ⁹ In other words, there should be equality of regression slopes⁷ there should be no relationship between the independent variable and the covariate variable. ¹⁰

How to correctly report the ANCOVA results:

• Revealing the correlation coefficient and statistically significant P-value of assessing the association between the covariate and dependent variable. ¹⁰
• Revealing the insignificant relationship between the independent variable and covariate variable, and thus the homogeneity of slope of regression lines.
• Tabular representation of means of the dependent variable before and after the adjustment of the impact of the covariate, as well as revealing the p-value of the means comparison separately.

References

1. Silva CN, et al. The effectiveness of progressions of difficulty during an exercise program in older individuals: a randomised clinical trial. Arch Gerontol Geriatr. 2025; 108:104975.
2. Johnson L, et al. Statistical and methodological considerations for randomised controlled trials in physical medicine and rehabilitation. PM&R. 2023;15(3):234-245.
3. Vickers AJ, Altman DG. Empirical comparison of baseline covariate adjustment methods in clinical trials. Trials. 2022;23(1):112.
4. Wang R, et al. Analysis of covariance (ANCOVA) in randomised trials: robustness and precision gain. Biometrics. 2023;79(2):345-356.
5. Yang S, Tsiatis AA. Robustness of ANCOVA in randomised trials under model misspecification. Biometrics. 2023;79(1):123-134.
6. Silva CN, et al. Application of ANCOVA in rehabilitation randomised clinical trials: a systematic review. J Rehabil Med. 2024;56(1):45-52
7. Field A. Discovering statistics using IBM SPSS Statistics. Sage Publications Limited; 2024 Feb 22.
8. Rutherford A. ANOVA and ANCOVA: a GLM approach. John Wiley & Sons; 2011 Oct 25.
9. Karpen SC. Misuses of regression and ANCOVA in educational research. American journal of pharmaceutical education. 2017 Oct 1;81(8).
10. Schneider BA, Avivi-Reich M, Mozuraitis M. A cautionary note on the use of the Analysis of Covariance (ANCOVA) in classification designs with and without within-subject factors. Frontiers in psychology. 2015 Apr 21; 6:474.