Comparison of robustness against missing values of alternative decision tree and multiple logistic regression for predicting clinical data in primary breast cancer

Nomogram based on multiple logistic regression (MLR) is a standard technique for predicting diagnostic and treatment outcomes in medical fields. However, the applicability of MLR to data mining of clinical information is limited. To overcome these issues, we have developed prediction models using ensembles of alternative decision trees (ADTree). Here, we compare the performance of MLR and ADTree models in terms of robustness against missing values. As a case study, we employ datasets including pathological complete response (pCR) of neoadjuvant therapy, one of the most important decision-making factors in the diagnosis and treatment of primary breast cancer. Ensembled ADTree models are more robust against missing values than MLR. Sufficient robustness is attained at low boosting and ensemble number, and is compromised as these numbers increase.