The use of logistic and ordinal regression for the prediction of the phadiatop test results

This paper presents the use of logistic and ordinal regressions for predicting the classification of patients into the Phadiatop test groups. Our data come from patients who underwent Phadiatop test examinations in the University Hospital of Ostrava, the Clinic of Occupation and Preventive Medicine. Two databases were available for the research giving the records about family and personal anamnesis of each patient. These anamnesis were selected and appropriately rated as predictive variables for discriminant analysis. These data were obtained by an evaluation of the atopy score. Moreover, both of these anamnesis were divided into four categories according to the severity ranked by the doctors (asthma, allergic rhinitis, eczema and others). Logistic regression with independent variables was used for the personal anamnesis in the older database. The model was tested using a medical database of 1027 clients in the first case and 1132 clients in the second case. In the newer database, complete family and personal anamnesis were used. The developed model predicts the results of the Phadiatop test with 75% probability dividing the patients into one of two groups and with 64% probability dividing the patients into five groups, which may represent significant financial savings for the diagnosis of atopy.