Surrogate endpoints for overall survival in randomized clinical trials testing immune checkpoint inhibitors: a systematic review and meta-analysis

Abstract

Introduction: There is debate on which are the best surrogate endpoint and metric to capture treatment effect on overall survival (OS) in RCTs testing immune-checkpoint inhibitors (ICIs).

Methods: We systematically searched for RCTs testing ICIs in patients with advanced solid tumors. Inclusion criteria were: RCTs i) assessing PD-(L)1 and CTLA-4 inhibitors either as monotherapy or in combination with another ICI, and/or targeted therapy, and/or chemotherapy, in patients with advanced solid tumors; ii) randomizing at least 100 patients. We performed a meta-analysis of RCTs to compare the surrogacy value of PFS and modified-PFS (mPFS) for OS in RCTs testing ICIs, when the treatment effect is measured by the hazard ratio (HR) for OS, and by the HR and the ratio of restricted mean survival time (rRMST) for PFS and mPFS.

Results: 61 RCTs (67 treatment comparisons and 36,034 patients) were included in the analysis. In comparisons testing ICI plus chemotherapy, HR_PFS and HR_mPFS both had a strong surrogacy value (R² = 0.74 and R² = 0.81, respectively). In comparisons testing ICI as monotherapy, HR_PFS was the best surrogate, although having a moderate correlation (R² = 0.58). In comparisons testing ICI plus other treatment(s), the associations were very weak for all the surrogate endpoints and treatment effect measures, with R² ranging from 0.01 to 0.22.

Conclusion: In RCTs testing ICIs, the value of potential surrogates for HR_OS was strongly affected by the type of treatment(s) tested. The evidence available supports HR_PFS as the best surrogate, and disproves the use of alternative endpoints, such as the mPFS, or treatment effect measures, such as the RMST.

Keywords: immune check inhibitor (ICI); immunotherapy; methodology; randomized clinical trial; surrogate.

Figure 1

Forest plots showing meta-analytic pooled estimate (with 95% CI) of treatment effects on OS and potential surrogate endpoints, meta-analytic pooled estimate (with 95% CI) of the ratio between surrogate endpoint and HR_OS, R² coefficient (with 95% CI) from the weighted linear regression and surrogate threshold effect (STE), by type of treatment administered in the experimental arm. The figure shows in the left panel the meta-analytic pooled estimate (circles) of treatment effects on potential surrogate endpoints, by type of treatment administered in the experimental arm. Horizontal lines indicate the 95% CI and the solid vertical line indicates a HR or rRMST of 1, which is the null-hypothesis value. Values <1 indicate a treatment effect in favor of the experimental arm, while values >1 indicate treatment effects in favor of the control. The meta-analytic pooled estimates of HR_OS are also displayed. The central panel shows the meta-analytic pooled estimate (circles) of the ratio between surrogate endpoint and HR_OS, by type of treatment administered in the experimental arm. Horizontal lines indicate the 95% CI, and the solid vertical line indicates a ratio of 1, which is the null-hypothesis value. Values <1 indicate a surrogate endpoint that overestimates the protective treatment effect size observed with HR_OS, while values >1 indicate a surrogate endpoint that underestimates it. The right panel shows the R² coefficient (with 95% CI) estimated from the weighted linear regression model, by type of treatment administered in the experimental arm. Surrogate threshold effect (STE) values are also reported on the right.

Figure 2

Correlations between effects of ICIs on OS and the potential surrogate endpoints, PFS (A, C) and mPFS (B, D) by type of treatment administered in the experimental arm. The figure shows the correlations between effects of ICIs on OS and the potential surrogate endpoints, PFS (A, C) and mPFS (B, D) according to the type of treatment administered in the experimental arm [i.e., ICI alone, ICI plus chemotherapy, and ICI plus ICI or other treatment(s)]. The treatment effects are measured by the HR for OS, and by the HR and the rRMST for the two surrogate endpoints. Each circle represents a comparison, and the surface area of the circle is proportional to the number of patients in the corresponding comparison. Red circles represent comparisons with ICI alone as experimental arm, blue circles represent comparisons with ICI plus chemotherapy as experimental arm, and green circles represent comparisons with ICI plus ICI or other treatment(s) as experimental arm. Dashed lines represent weighted regression lines. The R² coefficients, with their 95% CI (displayed in square brackets), were reported in the legend.

Figure 3

The correlations between effects of ICI alone on OS and the potential surrogate endpoints, PFS (A, C) and mPFS (B, D). The figure shows the correlations between effects of ICI alone on OS and the potential surrogate endpoints, PFS (A, C) and mPFS (B, D). The treatment effects are measured by HR for OS, and by the HR and the rRMST for the two surrogate endpoints. Each circle represents a comparison, and the surface area of the circle is proportional to the number of patients in the corresponding comparison. Straight line represents weighted regression line. Dashed lines represent 95% prediction bands based on the values predicted by the weighted regression model. The surrogate threshold effect (STE) is represented by the intersection point between the horizonal line y=1 and the upper 95% prediction band. Black diamond indicates the meta-analytic pooled estimate. The diamond’s width represents the 95% CI of the surrogate pooled estimate, and height represents the 95% CI of the HR_OS pooled estimate. The surrogacy equation between the log-transformed treatment effects and the ln-HR_OS estimated from the weighted linear regression, the R² coefficient, and the pooled ratio between surrogate endpoint and HR_OS were also reported with their 95% CI (displayed in square brackets).