[Skip to Navigation]
Sign In
Comment & Response
November 4, 2021

Analysis of Outcomes With Addition of Immunotherapy to Chemoradiation Therapy for Non–Small Cell Lung Cancer

Author Affiliations
  • 1Department of Biomedical Data Science, Stanford University, Stanford, California
  • 2Pfizer Inc, Groton, Connecticut
  • 3Department of Biostatistics, Harvard University, Boston, Massachusetts
JAMA Oncol. 2022;8(1):167-168. doi:10.1001/jamaoncol.2021.5605

To the Editor Jabbour and colleagues1 conducted an important, exploratory nonrandomized clinical trial to evaluate treatment outcomes and safety of pembrolizumab plus concurrent chemoradiation therapy for patients with non–small cell lung cancer in 2 cohorts: A (squamous/nonsquamous) and B (nonsquamous). One of the primary end points was the objective response. Their rates were 70.5% (95% CI, 61.2%-78.8%) in cohort A and 70.6% (95% CI, 60.7%-79.2%) in cohort B. The authors also conducted an analysis for the duration of response (DOR). The DOR is a clinically meaningful measure for efficacy beyond objective response rate and progression-free survival. However, the authors only considered DOR for responders. From the Kaplan-Meier curves for DOR reported in Figure 2,1 for cohort A, about 88.2% of responders had a duration of response lasting at least 6 months and 79.7% at least 12 months. The corresponding values for cohort B were 91.3% and 75.6%, respectively. With other colleagues in a recent publication in this journal,2 we discussed the issues of the conventional way to analyze DOR data, which was the procedure used for the present study. Specifically, it was pointed out that analysis of DOR data from the responders only violates the intention-to-treat principle and does not reflect the causal treatment effect. For example, in the present study,1 about 30% of patients did not respond during the study in either cohort. Therefore, the DORs were 0 for those nonresponders. It would be more informative to use the intention-to-treat analysis including those nonresponders for DOR assessment. Second, the Kaplan-Meier curves presented in Figure 21 for DOR are not valid as pointed out in the recent literature.2,3 This is because the DOR and its censoring time are not independent. A remedy to summarize the DOR data is to construct a current response rate curve over time, that is, at each time point, the proportion of the patients who responded and were still responding at this time point. The area under this curve is the mean DOR up to the study follow-up period.2,3 This mean value is quite informative about DOR. However, at this point, there are no statistical methods available to construct a valid estimate for the distribution of the DOR similar to the Kaplan-Meier curve in Figure 2.1 Thus, it is not clear if estimates of the proportion of patients reported in the article, whose DOR would be beyond 6 or 12 months, were valid. More research is needed along this line to draw attention to this important matter.

Add or change institution