Adaptive trial design integrating Bayesian and frequentist approaches

Background: Clinical trials are fundamental to advancing public health, as they generate robust evidence on the effectiveness, risk profile, and health outcomes of interventions. Statistical techniques are essential for accurately analyzing and interpreting the results of these trials. Classical frequentist methods are based on rigid hypotheses and sample sizes which could limit their power if applied to small trials or under moderate effect sizes. Bayesian adaptive methods, provide a flexible framework that permits to update evidence, make decisions in interim, and incorporate prior information.
Methodology: The present research aims to juxtapose frequentist and Bayesian adaptive approaches in the assessment of Sequential Yoga Poses (SYP) and Fascial Manipulation (FM) for Mechanical Neck Pain (MNP). One hundred patients suffering from MNP were randomly allocated to the experimental (n = 51) receiving FM and home-based SYP, or the control (n =48) subjects receiving conventional treatment. The study involved recording pain intensity (NPRS), Elbow Extension Range of Motion (EEROM) during ULNT1, Patient-Specific Functional Status (PSFS), and Fear-Avoidance Behaviors (FABQ). The frequentist analysis employed repeated measures ANOVA, while the Bayesian adaptive analysis utilized Bayes Factors (BF??) and posterior probabilities to facilitate interim evidence evaluation.
Results: The frequentist approach yielded small effect sizes (d = 0.211) and low statistical power (0.18), which in turn led to non-significant p-values even though there were clinically relevant improvements. Bayesian adaptive analysis presented strong support for the intervention effects being the case, as it had high posterior probabilities and Bayes Factors, pinpointing the gains in pain reductions, mobility, function, and psychosocial outcomes that were less visible during the frequentist analysis.
Interpretation: Besides, Bayesian adaptive approaches, like in the case of the current study, are advantageous in clinical trials especially where the sample size is small, effect sizes are modest, or when interim assessment is needed. They empower researchers to observe the gradual effect of the treatment and at the same time to modify their plans accordingly.