Clinical Trials
SBRT vs Sublobar Resection for Stage IA NSCLC: Causal Inference Analysis in SEER-Medicare
Linderman et alJAMA surgery

In patients with stage IA non-small cell lung cancer (NSCLC), determining whether stereotactic body radiation therapy (SBRT) matches the oncologic outcomes of surgery in operable candidates has been hindered by early trial closures and confounding in observational data. To address hidden confounding stemming from patient frailty and comorbidity, investigators performed a retrospective cohort study using the Surveillance, Epidemiology, and End Results (SEER)-Medicare database spanning 2010 to 2019, with follow-up through December 21, 2021.
The cohort included 4,251 patients with stage IA NSCLC (mean age, 75.6 years; 60.7% female, 39.3% male), of whom 2,558 received SBRT and 1,693 underwent sublobar resection. The primary outcome was lung cancer-specific mortality, analyzed across three models: unadjusted, standard regression-adjusted for measured covariates, and proximal causal inference utilizing negative control exposures derived from billing codes and negative control outcomes based on other-cause mortality.
In standard analyses, SBRT was associated with higher lung cancer cause-specific mortality, showing an unadjusted hazard difference (HD) of 0.047 person-years (95% CI, 0.038–0.055) and a covariate-adjusted HD of 0.024 person-years (95% CI, 0.014–0.034). However, strong associations between treatment selection and other-cause mortality highlighted substantial confounding. When proximal causal inference was applied to indirectly control for unmeasured frailty and comorbidity, the excess cancer-specific mortality associated with SBRT narrowed to an HD of 0.012 person-years (95% CI, 0.0013–0.027).
The authors concluded that observational comparisons between SBRT and surgery are biased by unmeasured confounding. While proximal causal adjustment revealed a smaller but significant advantage favoring surgery, potential residual bias related to differences between clinical and pathologic staging could not be ruled out.
Key Takeaways
- In the proximal causal inference analysis adjusting for unmeasured frailty and comorbidity, SBRT was associated with a higher lung cancer cause-specific mortality compared with sublobar resection (hazard difference, 0.012 person-years; 95% CI, 0.0013–0.027).
- Accounting for hidden confounding reduced the estimated mortality difference associated with SBRT relative to surgery compared to unadjusted (hazard difference, 0.047 person-years; 95% CI, 0.038–0.055) and standard covariate-adjusted models (hazard difference, 0.024 person-years; 95% CI, 0.014–0.034).
Read the original sourceJAMA surgery
