A rank-based robust change-point regression approach in human intervention studies of functional foods

Heterogeneity of treatment effects is common in functional-food studies, particularly when baseline characteristics influence the magnitude of response. Change-point regression models (CPRM) are useful for identifying the baseline level at which an intervention effect begins to appear. However, conventional CPRM relies on ordinary least squares (OLS-CPRM) and is sensitive to outliers, which can lead to unstable estimation of the change point (τ). To address this issue, I developed a rank-based robust CPRM (RR-CPRM) and evaluated its utility using two clinical datasets. RR-CPRM produced estimates of τ consistent with those of the conventional model while providing narrower bootstrap confidence intervals, indicating improved stability and reduced sensitivity to outliers. In datasets with little to no influence from outliers, RR-CPRM yielded results comparable to OLS-CPRM, demonstrating that robustness is achieved without loss of performance. These findings suggest that RR-CPRM offers a reliable and flexible framework for evaluating baseline-dependent intervention effects in human functional-food trials and may contribute to more accurate interpretation of clinical outcomes.