Hashimoto T, Ohtaki M, Ueoka H, Munaka M, Kuramoto A, Kamada N
Institute of Statistical Mathematics, Tokyo, Japan.
Int J Hematol. 1994 Jul;60(1):59-69.
Chronic phase duration was analyzed in 148 Japanese patients with Ph-positive chronic myelocytic leukemia (CML) treated only with busulfan, and multivariate proportional hazard models were constructed to infer the prognosis of a given patient, and to evaluate the prognostic importance of the clinical findings. To make an accurate estimation of an individual patient's prognosis, we introduced a piecewise linear model for describing a hazard function in each prognostic variable. Of the 11 prognostic variables initially analyzed, the absolute count of peripheral basophils was revealed as the most important prognostic factor. A high basophil count was clearly a factor in cases with poor prognosis. The statistical model selection based on the criteria of model approximation finally suggested a model with three covariates; absolute count of peripheral basophils, platelet count, and presence or absence of additional chromosome abnormalities superimposed on the translocation of 9;22. By analyses based on the above model, we estimated that the median of the chronic phase duration for the patients with poor and good prognosis was 12 and 75 months, respectively. The piecewise linear model was well suited for an accurate estimation of an individual patient's prognosis. Thus, the model constructed here would become a reliable control to evaluate investigational approaches, such as hydroxyurea, interferons, and/or bone marrow transplantation.