Evaluating influencing factors in estimation of renal function by extending the Bland-Altman approach

Method comparison is frequently requested when newly developed methods are evaluated. In medical sciences in particular, numerous articles are published every year concerning new methods for improving diagnostics. Among several statistical approaches used to provide essential test characteristics, the Bland-Altman method is well established, but actually provides no evaluation of possible factors influencing test quality. We propose an analysis of residuals obtained from the Bland-Altman data alleviating the determination of possible factors influencing accuracy and precision compared with different methods and give a short tutorial on application of Bland-Altman analysis. We illustrate this extension of the classical Bland-Altman method with an actual clinical example: estimation of renal function applying the common formula of Cockcroft and Gault. The influence of diabetes, gender and age on accuracy and precision of this estimation is analysed and inulin clearance is used as the gold standard. The subsequent analysis of residuals allows detection of the variables significantly influencing accuracy and precision. In the given example, diabetes and age, but not gender, are revealed as significant factors in glomerular filtration rate estimation when applying the Cockcroft formula.