Recursive partitioning analysis of 2105 patients treated in Radiation Therapy Oncology Group studies of head and neck cancer

Abstract

Background: The Radiation Therapy Oncology Group conducts large-scale prospective, randomized trials to test new concepts in cancer patient care and provide information about pretreatment and treatment factors that may influence outcome.

Methods: Recursive partitioning analysis (RPA) was used to examine the data derived from 2105 patients. RPA grouped patients according to the influence of tumor, of host, and of treatment variables on outcome.

Results: For survival, the most important factor was T classification. For lesions less than T3, the primary tumor was the next most important factor, whereas for T3 and T4 lesions the Karnofsky score was the next most predictive factor. Six distinct groups were formed by RPA, with median survivals ranging from 6.8 to 151.8 months. For local-regional control, the N classification was the most important factor. For patients with no adenopathy, T classification was the next most important factor, whereas for patients with adenopathy, the number of treatment fractions was the next most important factor. Such analysis created 5 distinct groups. In the most favorable, the median time to local-regional relapse has not yet been reached. In the least favorable group, fewer than 50% of the patients experienced complete response at any time following treatment.

Conclusions: RPA clarifies the relative importance and potential interactions of pretreatment and treatment variables and should permit more accurate stratification of patients in future trials.