Predicting the Survival Status of Cancer Patients with Traditional Chinese Medicine Symptom Variation Using Logistic Regression Model

Background- In the study of rules in pathological changes, most of the traditional analyses are from the static perspective, which regard cross-sectional data as the input dataset to analyze the effect of non-time-varying factors. However, according to the clinical experiences, the changes of physical status can partly reflect the disease progression and the mortality which have been ignored in the existing studies. Thus, based on the dynamic perspective, by using the changing patterns of symptom as the model input, longitudinal data can be utilized to further explore the rules in pathological changes from the dynamic perspective which is a novel and effective solution. Method- The study proposed a dynamic pattern representation method; pretreated and transformed the original dataset, including the Traditional Chinese Medicine TCM and the western medicine clinical longitudinal data, into a 2-dimensional matrix composed of the symptom indexes and the changing patterns; and analyzed the influences between the changing patterns of symptom and III stage non-small cell lung cancer NSCLC patient's mortality by multivariate logistic regression. Result- The predicting accuracy using the transformed dataset by proposed representation method is 90.7%. Based on the enter stepwise regression method, the accuracy increased 26.3% and 14.5% than the baseline dataset and the last records respectively; based on the forward stepwise regression method, the accuracy increased 16.7% and 3% than the baseline dataset and the last records respectively. Conclusion- The experiment results indicated that the proposed data representation method is feasible and effective, meanwhile, the proposed novel dynamic perspective appears more appropriate for the TCM mainly III stage NSCLC patients' modeling than the traditional static method.