Predictors of early death or hospice in curative inoperable lung cancer patients

Document Type


Publication Date


Publication Title

International Journal of Radiation Oncology, Biology, Physics


Purpose/Objective(s): The current treatment approach for inoperable stage II-III non-small cell lung cancer (NSCLC) involves aggressive chemo/radiotherapy (CRT). While outcomes have improved with immunotherapy, some patients transition to hospice or die early into their treatment. To help identify these patients and tailor treatment interventions, we developed a predictive model for early poor outcomes in this population. Materials/Methods: We included patients who received definitive CRT for stage II-III lung cancer from April 2012 - November 2019. Patient information was collected prospectively as part of a statewide consortium involving 27 sites. We defined an early poor outcome as termination of treatment due to hospice enrollment or death within 5 months of initiating radiation therapy. Potential predictors included patient and disease characteristics, patient reported outcomes (PROs), and treatment variables. Generalized linear models were used to describe the relationships between single predictors and outcomes of interest. Due to missing data, we imputed 25 datasets to fit multivariable models. We used Lasso regression on all 25 datasets to select variables of interest. Using these selected variables, we fit generalized linear models on all 25 imputed data sets and present pooled results. Results: A total of 2127 patients met criteria, of which 96 patients discontinued treatment early due to hospice enrollment or death. Of the 96 patients, 59% received concurrent chemotherapy and the mean age was 71 years old. When modeled individually, age, ECOG performance status, PTV volume, distance to critical structures, mean heart dose, insurance status, functional and physical well-being scale, and lung cancer symptom scale were all significant predictors of early hospice or death. Additionally, specific PRO questions, including “I have a lack of energy”, “I have been coughing”, and “I have been short of breath” were significant univariable predictors. Gender, marital status, baseline FEV1, treatment center, lung radiation dose (V5, V20, mean), and esophageal dose (D2cc) were not significant univariable predictors. Our Lasso procedure selected the following predictors for multivariable analysis: age, ECOG performance status, PTV volume, patient reported lack of energy, patient reported cough, mean heart dose, and patient insurance status. The pooled estimate of AUC for this multivariable model was 0.71. Conclusion: Our models identified a combination of initial patient, disease and treatment characteristics, and PROs that may help identify individuals undergoing curative CRT who are at high risk of early hospice enrollment or death. Further research to determine if interventions for this group (e.g. treatment modifications, supportive care) may help improve patients’ outcomes.




3 Supplement

First Page


Last Page