Limitations of logistic regression in analyzing complex ambulatory blood pressure data: a call for non-parametric approaches

This commentary refers to ‘Ambulatory blood pressure monitoring, European guideline targets, and cardiovascular outcomes: an individual patient data meta-analysis’, by D.-Y. Zhang et al., https://doi.org/10.1093/eurheartj/ehaf220 and the discussion piece ‘Clarifying the use of multinomial logistic regression in analysis: a methodological response’, by D.-Y. Zhang et al., https://doi.org/10.1093/eurheartj/ehaf543. Zhang et al.1 on behalf of the International Database of Ambulatory Blood Pressure (ABP) in Relation to Cardiovascular Outcomes Investigators, performed an individual patient data meta-analysis to assess the relationship between office blood pressure (OBP) and ABP, both measured at baseline.1 Through multinomial logistic regression, they analyzed the probability of individuals belonging to each of four descending quartiles of the percentage of time that blood pressure remained within the target range, as established by the 2024 European Society of Cardiology guidelines. Their analysis encompassed a wide range of systolic and diastolic OBP values in both untreated and treated study participants. However, Zhang et al.1 do not appear to address the critical importance of verifying the underlying assumptions of their data analysis methods—a key step to ensure valid results. As outlined in the Edinburgh Medical School’s data analysis textbook and supported by numerous peer-reviewed articles,2–9 violating fundamental assumptions of logistic regression—such as linearity of the logit for continuous predictors, independence of observations, lack of multicollinearity, and appropriate distribution of residuals—can render statistical results, including P-values and confidence intervals, unreliable and potentially misleading against non-linear non-parametric biological data analysis. Because logistic regression is a parametric technique, it is particularly vulnerable to these violations; failure to assess and address these assumptions may compromise both the validity and interpretability of their study’s findings.