2000 年 20 巻 5 号 p. 405-412
Cox’s proportional hazard model is now widely used to evaluate prognostic factors of diseases. But it has been often pointed out that the Cox’s method is sometimes not sufficient because it assumes temporally constant effects of prognostic factors on the disease process. In this paper we developed a new evaluation method of prognostic factors, based on the more detailed modeling of the course of the disease. This method describes the disease process by longitudinal Markov model (temporal succession of simple Markov model) and the prognostic factors are evaluated by the multiple logistic regression analysis.
Stage II breast cancer is chosen as the subject in this study. Data are retrospective ones, collected in National Cancer Center Central Hospital.
As first step, a longitudinal Markov model was constructed to describe the longitudinal disease state transition of a breast cancer. Then based on the detailed investigation of the longitudinal change of transition probability between each disease state, we discriminated the breast cancer patients by two groups, good and poor prognosis one, by using a nonparametric test. We extracted a patient group who showed a recurrence for the first two and a half years as the poor prognosis group (p < 0.05), which corresponds to clinical accepted view.
As second step, a multiple logistic regression analysis was applied to evaluate prognostic factors that contribute the discrimination between good and poor prognostic group, and the results were obtained that three factors (n-classification of pathological diagnosis, ductal spread, and existence of estrogen receptor) were considered as the major prognostic factors for the early death in Stage II breast cancer.
These results well correspond to the clinically accepted view concerning the prognostic factors of breast cancer, so that this method would be an efficient quantification method to transform clinical impressions to objective evidences in prognosis of diseases.
