A new prognosis factor analysis based on nonhomogeneous Markov description

To evaluate prognosis factors, Cox's proportional hazard model has been used. But it was found that the analytical ability was not sufficient. So we propose a new evaluation method combining Markov chain model and multiple logistic regression analysis to estimate the prognosis factors. Stage II breast cancer was chosen as the subject. The data was retrospective data gathered in National Cancer Center Central Hospital. As first step, a simple Markov chain model was constructed to describe the state transition of a breast cancer. Then the multiple property of each state transition was investigated in detail. And the patients who had gotten a recurrence for the first two and a half years were discriminated as the poor prognosis group by a nonparametric test (p<0.05). And the result proved to corresponding with the clinical experience. As second step, three factors (n classification of pathological diagnosis, ductal spread, and estrogen receptor) were selected as the prognosis factors for the early death in Stage II breast cancer by a multiple logistic regression analysis. This new prognosis factor analysis could find out some scientific evidences. Especially, it was found to be remarkable efficient in proving clinically experienced observation.