Progression-Free Survival and Time to Progression as Potential Surrogate Endpoints for Overall Survival in Chemoradiotherapy Trials in Limited-Stage Small-Cell Lung Cancer: A Systematic Review and Meta-Analysis

Purpose To investigate whether progression-free survival (PFS) or time to progression (TTP) could be a valid surrogate endpoint for overall survival (OS) in patients with limited-stage small-cell lung cancer (LS-SCLC) receiving combined chemoradiotherapy. Methods Literature searching was performed in PubMed, Embase, and The Cochrane Library up to 2021. Prediction models were firstly established using data from phase III randomized controlled trials (RCTs) and then externally validated in phase II and retrospective studies. Correlation analysis was evaluated by a weighted linear regression model at both trial and arm levels. Cross-validation was performed to assess the consistency and robustness of the established models. Results 37 studies, including 15 phase III RCTs, 12 phase II studies, and 10 retrospective studies, were selected in the final analysis. In trial-level surrogacy, a very good correlation was observed between hazard ratios (HRs) of PFS/TTP and OS (R2 = 0.783, 95% CI 0.771–0.794). In arm-level surrogacy, very good correlations were also observed between 2-year (R2 = 0.823, 95% CI 0.814–0.832), 3-year (R2 = 0.843, 95% CI 0.833–0.850), 5-year (R2 = 0.852, 95% CI 0.843–0.859) PFS/TTP, and 5-year OS. An excellent correlation was observed between 4-year PFS/TTP and 5-year OS (R2 = 0.906, 95% CI 0.901–0.910). Cross-validation demonstrated reasonable overall consistency. External validation in phase II and retrospective studies showed good agreement (R2, 0.728–0.824). Conclusions PFS/TTP was a valid surrogate endpoint for OS in patients with LS-SCLC receiving combined chemoradiotherapy. The finding provides high-level evidence to support PFS/TTP as the primary endpoint in clinical trials so as to speed up introducing novel agents to the treatment of LS-SCLC.


INTRODUCTION
Small-cell lung cancer (SCLC) is the most aggressive subtype of lung cancer, with an estimated incidence of 4% and 250,000 cancer deaths worldwide (1,2). Limited disease accounts for one third of the total cases. Besides the patients with T1-2N0M0 disease (AJCC 8th) who may be surgical candidates, chemoradiotherapy is the standard of care for most of limitedstage small-cell lung cancer (LS-SCLC) (95%) (3) and results in a 5-year overall survival (OS) of 20%-30% (4,5).
OS is the gold-standard endpoint in randomized controlled trials (RCTs) as it is simple and unbiased. Especially, the 5-year OS rate is commonly used to assess the long-term benefits and toxicities of the treatment. However, using OS as the primary endpoint requires a large number of patients and long-term follow-up, leading to higher costs and delays in introducing novel drugs. Given these disadvantages, using an early surrogate endpoint in RCTs would shorten the time duration and save the research resources. Until now, The Food and Drug Administration has granted accelerated approval of many drugs based on surrogate endpoints of progression-free survival (PFS) or time to progression (TTP). For example, crizotinib was approved for anaplastic lymphoma kinasepositive non-small-cell lung cancer on the basis of PFS (6) and sunitinib for gastrointestinal stromal tumor and renal cell carcinoma on the basis of TTP (7). PFS and TTP have also been demonstrated to be valid surrogate endpoints for OS in some malignancies (8)(9)(10)(11)(12). However, an early valid surrogate endpoint has never been reported in LS-SCLC patients.
Reviewing various endpoints in clinical trials of LS-SCLC, PFS and TTP were potential surrogate endpoint for OS (4,13,14). Hereby, we investigated whether PFS/TTP could be used as an early efficient surrogate endpoint in LS-SCLC through literature-based analysis at trial and arm-level.

LITERATURE SEARCH AND STUDY SELECTION Search Strategy
Articles published before December 25, 2021 were identified via a systematic literature search of PubMed, Embase, and The Cochrane Library. The keywords were "Limited" and "Small Cell Lung Cancer" and "Chemoradiotherapy". The search strategy is shown in Supplementary Table 1. The database searches were carried out independently by two authors (YY and JY, W).
We excluded literatures without original data, phase I studies, inadequate survival data, systematic reviews, case reports, and other irrelevant publications.

Data Extraction
The following information from included studies were extracted: publication year, design, treatments of groups, number of patients, median follow-up time, and endpoints. For phase III RCTs, the endpoints were HRs for OS and PFS/TTP, absolute PFS/TTP rates (1, 2, 3, 4, 5-year) and 5-year OS ( Table 1). The HRs or survival rates at different time point were obtained from the text or Kaplan-Meier curves, according to methods by Tierney et al. (26). For phase II trials and retrospective studies, the endpoints were absolute PFS/ TTP rates (year 1, 2, 3, 4, 5) and 5-year OS ( Table 2).

Endpoint Definition
OS was defined as the time from randomization, registration, diagnosis or the first day of treatment to death. PFS was defined as the time from randomization, registration, diagnosis or the first day of treatment to disease progression or death. TTP was defined as the time from randomization, registration, diagnosis or the first day of treatment to disease progression (Supplementary  Tables 2, 3). As surrogate endpoints were defined differently between the trials, two investigators (YY and JY,W) labelled an endpoint of a trial as PFS or TTP according to our established definitions. For the literature without detailed definition of PFS/ TTP, we tried to contact authors of original research, otherwise the definition from the text was adopted.

Quality Assessment
The quality of the candidate Phase II, III RCTs was evaluated on 7 domains according to the Cochrane Collaboration tool. The trials were excluded if high risk of bias in any domain was detected (Supplementary Table 4).
The quality of the candidate single-arm phase II, and retrospective studies was assessed in 3 domains with 9 items according to Newcastle-Ottawa Scale for cohort study. The studies were excluded if their scores were less than 6 points (Supplementary Table 5).

Correlation Evaluation
The correlations between surrogate endpoints and OS in phase III RCTs were performed at both trial-and arm-level. At trial level, the correlation of HRs for PFS/TTP and HRs for OS was qualified through a linear regression model, weighted by trial size. At arm-level, the linear correlation between the 1-, 2-, 3-, 4-, and 5-year PFS/TTP rates and 5-year OS rate was also evaluated by the linear regression model, with weight equal to each treatment-arm sample size. The coefficient of determination R 2 was calculated to assess the strength of correlation. R 2 values of 0-0.25, 0.25-0.5, 0.5-0.75, 0.75-0.9, 0.9-1 indicated poor, moderate, good, very good and excellent correlation. If the R 2 value was greater than 0.75, the following sensitivity analysis, leave-one-out cross-validation, and external validation were performed. If R 2 values showed great discrepancy between two adjacent time points, further subdivision of the time period and corresponding PFS/TTP rate extraction was performed to find a cut-off value.

Sensitivity Analysis
Phase III RCTs were classified into four subgroups depending on study designs (Supplementary Table 6). To assess the consistency and robustness of prediction models across different settings, sensitivity analyses were performed by leaving each subgroup of trials out at a time. The coefficient of determination R 2 value and its 95% CI were calculated by the weighted linear regression method mentioned above.

Leave-One-Out Cross-Validation
To assess the accuracy of prediction models, a leave-one-out cross-validation approach was performed. Each trial or treatment arm was left out once, and at each leave-one-out step a linear regression model was rebuilt on the other trials or  arms (n-1). This model was then applied to the left-out trial or arm and the corresponding 95% prediction interval was calculated to compare the predicted and actually observed treatment effect on OS.

External Validation of Phase III RCT Prediction Model
The arm-level predictive linear regression models established by phase III RCTs were applied to the phase II and retrospective studies for external validation. The predicted 5-year OS rate was calculated from the actual 1-5-year PFS/TTP rates in the phase II or retrospective studies using the established linear regression model from the phase III RCTs. More specifically, the equation

Trial-Level Correlation Between PFS/TTP on OS in Phase III RCTs
14 RCTs reported pairs of HRs for PFS/TTP and OS. A very good correlation was observed between 14 pairs of HRs for PFS/TTP and OS (R 2 = 0.783, 95% CI 0.771-0.794) (Figure 1). Sensitivity analysis showed very good correlations and robust consistency in most subgroups, except when leaving out 6 trials of different radiotherapy model (R 2 = 0.645, 95% CI 0.587-0.674) (Supplementary Figure 2A). This result was expected as the subgroup of different radiotherapy model close to half of the number of trials. Exclusion of these trials probably results in a lower correlation and wider confidence interval. The crossvalidation showed good consistency, as the observed HRs for OS were all in the 95% prediction intervals in 13 of 14 trials, and the HRs were very close to 95% prediction intervals in the remaining one trial (23) (Figure 2).    Figure 3).

Treatment Arm-Level Correlation Between PFS/TTP and OS in Phase III RCTs
Sensitivity analysis showed very good correlations and robust consistency in most subgroups, except when leaving out subgroups of different radiotherapy model due to fewer trials ( Supplementary Figures 2B-E).
The prediction results of cross-validation analyses showed that the observed 5-year OS rate fell within the 95% prediction intervals in all arms based on 2-, 3-, and 4-year PFS/TTP. With respect to the 5-year PFS/TTP, the observed 5-year OS rates were FIGURE 1 | Trial-level correlation between hazard ratios for OS and PFS/TTP in phase III RCTs. Green circles represent trials with a size proportional to the number of patients, blue line for the estimated regression line and the light green zone for 95% confidence intervals. OS, overall survival; PFS/TTP, progression free survival/ time to progression; RCTs, randomized controlled trials.
FIGURE 2 | Leave-one-out cross-validation analysis of the prediction of HR for OS based on HR for PFS/TTP. Green circles represent predicted hazard ratio for OS, vertical lines for 95% prediction intervals, and blue squares for observed hazard ratios for OS. Red circles and lines indicate that the observed HR is beyond the 95% prediction intervals. HR, hazard ratio; OS, overall survival; PFS/TTP, progression-free survival/time to progression. all in the 95% prediction intervals in 22 of 24 arms, and the 5year OS rates of the remaining two trials (5,18) are very close to the 95% prediction intervals (Figure 4).
These findings indicated that improvements in 2-5-year PFS/ TTP are strongly associated with a higher 5-year OS.
External Validation of the Correlation Between PFS/TTP and OS 30 treatment arms from 12 phase II and 10 retrospective studies were used for external validation. Using the arm-level prediction models from the phase III RCTs, we calculated the predicted 5year OS rate for each phase II and retrospective studies using the actual 2-, 3-, 4-, or 5-year PFS/TTP rate. The actual and predicted 5-year OS rates were plotted in scatter plots, which indicated that the predicted 5-year OS was approximated to the actual 5-year OS. The predicted 5-year OS rate greatly correlated with the actual 5-year OS rate, with the R 2 ranging from 0.728 to 0.824 ( Figure 5). These results validated the hypothesis that PFS/ TTP is the efficient surrogate endpoint of OS.

DISCUSSION
This is the first study combining data from high-quality Phase III RCTs, Phase II studies, and retrospective studies to explore the efficacy of PFS or TTP as a surrogate endpoint of OS in patients with limited-stage SCLC who underwent chemoradiotherapy. Previous meta-analyses have assessed surrogate endpoints in other types or stages of lung cancer. Mauguen et al. (11) indicated that PFS and DFS are valid surrogate endpoints in locally advanced non-small-cell lung cancer. As for extensivestage SCLC, Foster et al. (12) firstly reported that PFS was a potential surrogate endpoint using individual data from 6 singlearm and 3 RCTs in 2011 and then validated the results by seven new phase II/III trials in 2015 (54). However, analysis of surrogate endpoints in limited-stage SCLC has never been tested. Our study showed that there were strong correlations between PFS/TTP and OS at the trial level, and 2-5-year PFS/ TTP and 5-year OS at the treatment arm level. The coefficient R 2 ranged from 0.783 to 0.907, which indicated that nearly 78.3%-90.7% of the variation on OS can be indicated by PFS or TTP. The sensitivity analysis showed good consistency across different settings, and the cross-validation also showed good accuracy of the prediction models. The external validation with phase II and retrospective studies showed excellent agreement between the actual and predicted 5-year OS rates derived from the established linear regression models. These findings confirmed the feasibility of taking PFS/TTP as the primary endpoint for clinical trials of LS-SCLC.
However, the predictive value of 1-year PFS/TTP for 5-year OS was quite lower compared with that of the 2-year PFS/TTP (correlation R 2 0.379 vs. 0.823), which indicated that 1-year PFS/ TTP was not an appropriate surrogate endpoint. PFS/TTP data of phase III RCT ( Table 1) showed that around 50% of patients without progression relapsed during the second year after upfront treatment. From the third year, the PFS/TTP did not reduce significantly. Table 2 also showed that the PFS/TTP was relatively stable after 2 years of chemoradiotherapy in phase II and retrospective studies. This decreasing trend of PFS/TTP in LS-SCLC was consistent with our clinical experience. Thus, longer follow-up time such as 2-5-year PFS/TTP was necessary.
There has been debate on how a surrogate endpoint should be considered as valid. We employed the correlation approach which has been used to assess the possibility of PFS or TTP as a surrogate endpoint for OS in locally advanced NSCLC (11), nasopharyngeal carcinoma (8), and diffuse large B-cell lymphoma (10). Candidate surrogate endpoints could be valid only if the correlation coefficient was greater than 0.75 (55).
Exploring effective treatment or drugs for SCLC is urgent as its prognosis is still poor compared with other malignancy. There was no improvement of outcome for extensive-stage SCLC in the past more than three decades until atezolizumab was added in the classic regimen of etoposide and cisplatin as first-line chemotherapy (56). However, the immunotherapy plus chemotherapy only prolonged the median OS by 2 months compared with chemotherapy alone after more than a 2-year study period. For locally advanced lung cancer, new treatments have significantly increased the OS in NSCLC based on results of the PACIFIC trial (57), and the mature OS was achieved after 6 years from the beginning of the first patients enrolled (58). For l i m i t e d -s t a g e S C L C u n d e r s t a n d a r d c o n c u r r e n t chemoradiotherapy followed by prophylactic cranial irradiation, 5-year OS was still low ranging from 26% to 34% and did not change in the past two decades (4,25). There are several ongoing phase II and III trials investigating the PD-1/PD-L1 consolidation immunotherapy (59)(60)(61)(62). High-level evidence for immunotherapy as concurrent of consolidation treatment has not been reported until now. However, parts of these ongoing trials (59,60) have already defined PFS as the primary endpoint, OS as the second endpoint. Given that no study has reported valid surrogate endpoints for limited-stage SCLC, our analysis is of great importance to provide a rationale to define PFS or TTP as the primary endpoint in clinical trials, so as to speed up introducing novel effective agents to improve outcome of LS-SCLC. Another advantage of this study is comprehensively enrolled published literatures with high quality and proper sample size. In addition to the strong correlations demonstrated by phase III RCTs, the positive relationships between 2-5-year PFS/TTP and 5-year OS rates were externally validated by independent data from phase II and retrospective studies. The validation method was unique and firstly used in diffuse large B-cell lymphoma (10) by our department, which showed good efficacy to find early surrogate endpoints. This time, the validation method was used again in limited-stage SCLC to improve the reliability of the conclusions. Moreover, our study found multiple time points between 1-year and 2-year PFS/TTP which were not suitable, indicating that the 2-year PFS/TTP rate or more was really valid when used as a primary endpoint.
One major statistical challenge is inconsistencies and absences of the definition of endpoints across the trials in the current study. The starting time for endpoints was defined from randomization, registration, diagnosis, or the first day of treatment, differently. As SCLC is one of the most aggressive cancers, a difference of 1 or 2 months caused by definition of starting time may result in bias from different arms. Second, this is a literature-based systematic review and meta-analysis without individual patient data; therefore, a potential publication bias cannot be excluded. Third, the prediction models were based on data of patients who received first-line combined chemoradiotherapy, so the extrapolation to other treatments was cautious, especially when more effective second or more line treatments were developed in the future. Nevertheless, LS-SCLC patients usually died after disease progression because there are still no effective second-line treatments nowadays (25).
In conclusion, the current study provides first literature-based evidence to evaluate the correlation of PFS/TTP with OS in patients with limited-stage SCLC. The finding supports PFS/TTP as a valid surrogate endpoint for OS in LS-SCLC patients who underwent combined upfront chemoradiotherapy.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.