H. J. Gunn1, J. D. Cameron1, A. Mahajan2, P. D. Brown2, E. S. Yan2, S. A. Vora3, K. W. Merrell2, S. L. Stafford2, J. B. Ashman3, J. L. Peterson4, J. L. Leenstra2, Z. Wilson2, N. N. Laack II2, and W. Breen2; 1Department of Quantitative Health Sciences, Mayo Clinic, Rochester, MN, 2Department of Radiation Oncology, Mayo Clinic, Rochester, MN, 3Department of Radiation Oncology, Mayo Clinic, Phoenix, AZ, 4Department of Radiation Oncology, Mayo Clinic, Jacksonville, FL
Purpose/Objective(s): Patients with brain tumors often experience fatigue after treatment that impacts their quality of life (QOL). Patient-reported outcomes (PROs) including the Brief Fatigue Inventory (BFI) can quantitatively describe the effects of radiotherapy (RT) on patient QOL. Identifying which patient and treatment characteristics predict clinically significant changes in QOL using the minimal clinically important difference (MCID) can help guide treatment decisions and thus improve quality of life. It was previously reported that the MCID of the Brief Fatigue Inventory (BFI) for patients with brain tumors undergoing radiotherapy was 1.33, which was calculated using distribution-based and anchor-based approaches to assess change from baseline to end of treatment (EOT). The purpose of this study was to develop a machine learning variable selection algorithm to determine important patient and treatment predictors of this clinically significant change. Materials/
Methods: Patients with primary CNS tumors were enrolled on a multi-site prospective registry. Inclusion criteria included curative radiotherapy and completion of the BFI prior to treatment and at EOT. The outcome was a binary variable that measured if a patient experienced an MCID in the negative direction (worse fatigue). The LASSO was used to select important predictors of this outcome. Ten-fold cross-validation was used to increase generalizability of the model. Results: Overall, 356 patients had baseline and EOT BFI scores and non-missing values on all predictors included in the LASSO. The median age was 52.2 (range 18.1-90.9), 55% were males, and 94% were white. Patients received photons (52.8%) or protons (47.2%) with a median dose of 57 Gy (range 40-76 Gy) over a median of 28 fractions (range 10-40). The LASSO selected multiple predictors. Reaching the MCID threshold (i.e. clinically worse fatigue after RT) was associated with older age, female, non-white, living in an urban area, less fatigue at baseline, greater dose per fraction, and grade 2+ adverse events (AEs) during treatment. Number of fractions, dose, and modality were not predictive. Conclusion: Multiple patient and treatment characteristics were predictive of experiencing an MCID on the BFI. Baseline BFI scores were predictive of the MCID suggesting a ceiling effect such that patients with a high baseline BFI were not likely to get clinically significantly worse. Future research should further explore if other variables, particularly those that can be modified such as dose per fraction, predict clinically significant differences in fatigue and QOL for patients.
