Logistic regression as a relevant statistical tool for medical data investigation and evaluation

This paper presents the usage of logistic regression for predicting the classification of patients into one of the two groups. Our data come from patients who underwent Phadiatop test examinations and patients who underwent colectomy in the University Hospital of Ostrava. As the predictor variables were chosen personal and family anamneses for Phadiatop test and the physiological and operative scores for colectomy. For Phadiatop test, both of these anamneses were divided into four categories according to severity ranked by doctors. Scores for morbidity were based on the POSSUM system. The psychological score comprises 12 factors and the operative score comprises 6. The categorical dependent variable which we want to predict was Phadiatop test (respectively morbidity). The model for Phadiatop test was tested with the use of a medical database of 1027 clients and morbidity was tested upon a medical database of 364 clients. The developed models predict the right results with 75% probability for Phadiatop test and 70% probability for morbidity in surgery.