A new method for investigating the relation between change and initial value in longitudinal blood pressure data. II. Comparison with other methods.

Whether initial value affects blood pressure change or not is of interest in longitudinal blood pressure studies. The change/initial value relationship is, however, biased by the regression towards the mean phenomenon, as blood pressure measurements contain random errors (short-term intra-individual variation and measurement errors). Two methods (here called C and D) of avoiding this bias have earlier been proposed and used in population studies in South Wales and Framingham, with conflicting results. In the preceding paper a new method of avoiding the bias has been presented. In the present paper a comparison of the new method with the two earlier is made by applying them to the same data set. Method C indicates, as in South Wales, a highly significant positive relationship, while method D (as in Framingham) and the new method give a regression coefficient close to zero. The structure of the three methods is analysed. Method C is shown to be valid only under very restricted conditions and is best avoided. Method D gives a fairly accurate result, but underestimates the coefficient by approximately 30% in this case. The new method proposed seems to be the best method hitherto for this kind of analysis.