(1993) Elevated levels of insulin-like growth factor-binding protein-2 in the serum of prostate cancer patients. J Clin Endocrinol Metab 76: 1031-1035 Ho PJ and Baxter RC (1997) Insulin-like growth factor-binding protein-2 in patients with prostate carcinoma and benign prostatic hyperplasia. Clin Endocrinol 46: 333-342Mantzoros CS, Tzonou A, Signorello LB, Stampfer MJ, Trichopoulos D and Adami H-O (1997) Insulin-like growth factor 1 in relation to prostate cancer and benign prostatic hyperplasia. Br J Cancer 76: 1115-1118 Suboptimal analysis using 'optimal' cutpoints Sir, Oncology journals continue to publish many papers evaluating the prognostic importance of tumour markers in patients with various cancers. There is no agreed statistical methodology for handling such data, but some analyses are widely agreed to be misleading. One in particular, the so-called 'optimal' cutpoint method, is unsatisfactory for this purpose (Altman, 1991;Hilsenbeck et al, 1992;Hill, 1993;Altman et al, 1994 Altman et al, , 1995. Regrettably, this method, which is better referred to as the minimum P-value method, continues to appear in papers published in the British Journal of Cancer. Briefly, the minimum P-value method is as follows:(1) each distinct observed value of the marker is taken in turn as a cutpoint and two groups of patients created with values below or above this level (a variation is to use equi-spaced values across the observed range); (2) for each such grouping a log-rank test is performed and the P-value determined; (3) the cutpoint with the lowest P-value is called 'optimal', Kaplan-Meier curves are constructed for groups created with this cutpoint and the P-value reported; (4) in most cases the resulting binary variable is included with other variables in a Cox multiple regression analysis.The dangers of this approach have been outlined and include:(a) because of multiple testing the false-positive rate is around 40% rather than the nominal 5%; (b) the P-value is far too small (P = 0.002 corresponds to a genuine P = 0.05); (c) the value of the cutpoint has no clinical meaning; (d) the analysis gives no information about the shape of the relation between the level of the tumour marker and prognosis.In addition, when step (4) above is followed, the bias from the univariate analysis is transferred to the multivariate setting . It is not surprising, therefore, that such analyses often show that the tumour marker is apparently more important (i.e. has a smaller P-value) than other variables in univariate analyses, and that they usually retain significance after adjustment for standard risk factors. These problems arise from the search for the 'best' result. The consequence is a cutpoint that may be best in the narrow sense described, but which will not offer a true indication of the importance of the tumour marker.Looking quickly through a few recent issues of the journal I found three such studies. Dettmar et al (1997) studied the prognostic relevance of MIBl (Ki-67) and S-phase fraction (SPF) for disease-free sur...