B-ENT

Decision tree and prediction model for hypopharyngeal cancer survival

1.

Department of Pathology and Laboratory Medicine, Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan

2.

Department of Nursing, Meiho University, Pingtung, Taiwan

3.

Department of Otolaryngology, Head and Neck Surgery, Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan

4.

School of Medicine, National Defense Medical Center, Taipei, Taiwan

5.

Department of Otolaryngology, Head and Neck Surgery, Tri-Service General Hospital, Taipei, Taiwan

6.

Division of Hematology and Oncology, Department of Internal Medicine, Kaohsiung Medical University Hospital, Kaohsiung, Taiwan.

B-ENT 2019; 15: 93-100
Read: 687 Downloads: 403 Published: 28 January 2020

Objective: Data mining methods have been used to build prediction models from cancer databases. However, the hierarchy and relative contribution of the prognostic factors affecting hypopharyngeal cancer survival have not been reported and were analyzed here using a decision tree and Cox regression models.

Methods: A total of 923 patients with American Joint Committee on Cancer (AJCC) stage III - IV hypopharyngeal cancer were identified from 2007 to 2013 in the Surveillance, Epidemiology, and End Results database. Classification and Regression Tree (CART) and Cox regression models were used to analyze race, staging, insurance status, treatment modality, sex, tumor differentiation, and marital status. The endpoint was the 5-year disease-specific survival.

Results: Univariate and multivariate analyses using Cox regression models revealed the 5-year disease-specific survival rates were higher in (1) non-black patients vs. black patients; (2) AJCC stage III vs. IV patients; (3) the “insured” group vs. the “other” insurance groups; (4) the “operation (OP)+radiotherapy (RT)” group vs. the “RT alone” group and the “RT alone” group vs. the “OP alone” group. The 5-year disease-specific survival curves based on treatment also showed “OP+RT” was superior to “RT alone” and “RT alone” was superior to “OP alone”. The relative importance of variables influencing the survival of patients with hypopharyngeal cancer was generated using the CART algorithm, with race being the most important, followed by staging, insurance status, and treatment modality.

Conclusion: A decision tree algorithm easily elucidates the hierarchies and relative contribution of variables for predicting survival.

Files
EISSN 2684-4907