Determining the Efficacy of Pembrolizumab in Patients With Previously Treated Advanced Hepatocellular Carcinoma

To the Editor:

Qin et al¹ conducted an important study to evaluate the efficacy of pembrolizumab compared with placebo among patients in Asia with advanced hepatocellular carcinoma who progressed or were intolerant of sorafenib or oxaliplatin-based chemotherapy. The primary end point was overall survival (OS), with a one-sided significance threshold of 0.0193. The median OS was 14.6 months (95% CI, 12.6 to 18.0) for pembrolizumab compared with 13.0 months (95% CI, 10.5 to 15.1) for placebo (hazard ratio [HR], 0.79; 95% CI, 0.63 to 0.99; P = .0180), suggesting an OS benefit to pembrolizumab.

The validity of the HR relies on the proportional hazards (PH) assumption.² In this study, the protocol states that in the event of a PH violation, an alternate survival assessment, such as the difference/ratio of restricted mean survival times, should be performed. Unfortunately, the manuscript does not indicate whether the PH assumption was assessed. Visually, the Kaplan-Meier curves for OS in Figure 2A¹ appear to overlap early in the follow-up period before separating around 12 months then reconverging around 44 months, an indication that the PH assumption may not hold. Using reconstructed individual patient-level data³ from the OS curves in Figure 2A,¹ the Schoenfeld residual test provides evidence that the PH assumption was not met (P = .042).

Furthermore, HRs are difficult to interpret in a clinical context without knowledge of the baseline hazard profile. Median survival time is often reported alongside the HR to provide a clinically understandable summary of the treatment effect, but median survival is only a local summary of the survival profile, which does not capture either short- or long-term differences. Alternatively, difference of mean survival times is a clinically meaningful and assumption-free summary of the global treatment effect.⁴ With reconstructed data from Figure 2A, the 3-year restricted mean survival times were 17.9 (95% CI, 16.5 to 19.4) months for pembrolizumab versus 15.5 months (95% CI, 14.3 to 16.8) months for placebo. The difference was 2.41 months (95% CI, 0.49 to 4.33; P = .014) in favor of pembrolizumab, meaning that across 3 years, patients receiving pembrolizumab survived 2.41 months longer, on average, than those receiving placebo. By comparison, the reconstructed difference in median survival was 1.54 months (95% CI, –0.013 to 4.89; P = .40). The statistical nonsignificance of the difference in median survival is at odds with the statistically significant HR. Not only does the restricted mean survival time difference suggest a statistically significant treatment effect, which the median difference lacks power to detect, but it provides a clear interpretation in terms of the expected survival benefit, unlike the HR. Moreover, the difference in 3-year restricted mean survival time difference has a narrower CI than the difference in median survival times, indicating that it is more precise.

We commend the authors on completing this impactful study but wish to emphasize the importance of assessing the PH assumption, particularly when the prespecified study protocol requires it. We also encourage greater consideration of alternate survival measures, such as the mean survival time, which (1) remain statistically valid even in the presence of non-PH, (2) provide a more clinically comprehensible measure of the treatment effect than the HR, and (3) provide a more precise assessment of the treatment difference as compared with median survival.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Determining the Efficacy of Pembrolizumab in Patients With Previously Treated Advanced Hepatocellular Carcinoma

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).