Treatment Allocation Modeling for Clinical Stage I Non–small Cell Lung Cancer Patients to Facilitate the Shared Decision Making Process for Stereotactic Body Radiation Therapy Versus Surgical Resection

Abstract

We sought to create a predictive model of treatment allocation to surgery or stereotactic body radiation therapy (SBRT) for clinical Stage I non-small cell lung cancer (NSCLC). A model tailored to patient physiology and comorbidities could add pertinent data to shared decision-making in the clinical setting. Retrospective databases of clinical Stage I NSCLC from the departments of radiation oncology and surgery from 2000 to 2014 at a single high-volume institution were merged after IRB approval. Multivariate logistic regression modeling identified patient and tumor variables associated with treatment type, which was used to calculate propensity scores for the likelihood of receiving SBRT. A propensity score close to 0 indicated a low likelihood of receiving SBRT while a score close to 1 represented a high likelihood of receiving SBRT. Scores for all clinical Stage I NSCLC patients were ranked into deciles to determine patterns of treatment allocation and three-year overall survival (OS). Patient data was then divided into training (70%) and validation (30%) cohorts to create and test the regression model. For patients in the validation cohort, their predicted treatment type was compared to their actual treatment delivered to assess accuracy. From 2000 to 2014, 830/1,195 (69%) clinical Stage I NSCLC patients received surgery, while 365/1,195 (31%) had SBRT. Surgical patients were younger (66.3 years vs 73.9), had a higher FEV1-percent-predicted value (79.1% vs 55.1%), had less coronary artery disease (21.4% versus 33.7%), and less congestive heart failure (CHF) (2.2% versus 16.4%), all p<0.001. Variables associated with SBRT included increasing age (per year OR 1.1) and CHF (OR 8.6), while increasing FEV1-percent-predicted had a decreased likelihood of SBRT (per FEV1% increase: OR 0.94), all p<0.001. The model predicted the actual treatment received in 83% of the validation cohort. Patients with a tenth decile propensity score had high rates of SBRT use (87%) and similar 3-year OS regardless of treatment type (32% for SBRT and 36% for surgery). By comparison, 99% of the first decile patients received surgery, with a 3-year OS of 88%. Of the surgical patients, 92/107 (86%) of those in the first decile received a lobectomy. In the tenth decile, of the few patients that did receive surgical resection, 7/14 (50%) had a sublobar resection. A tailored model accounting for specific comorbidities could offer guidance to Stage I NSCLC patients regarding treatment selection. For patients with the highest propensity values indicating a high likelihood of receiving SBRT, overall 3-year survival is low, and similar among patients that received SBRT or surgical resection. Clinical use of this model could assist in shared-decision making discussions among the radiation oncologist, surgeon and patient.