Progression-Free and Overall Survival of First-Line Treatments for Advanced Renal Cell Carcinoma: Indirect Comparison of Six Combination Regimens

Simple Summary Recently, numerous treatments sharing similar mechanisms of action have been approved for advanced renal cell carcinoma. These combinations prolong survival compared to sunitinib, which was previously considered the standard of care in this context. Head-to-head comparisons between these innovative treatments are not available, but this information is needed to guide medical oncologists’ choices. To compare these combination therapies with one another and with sunitinib, our study used an innovative method (the Shiny method) that reconstructs individual patient data from published clinical trials. Using this approach, we demonstrated that pembrolizumab + lenvatinib is the most effective treatment in terms of progression-free survival (PFS) and overall survival (OS). Pembrolizumab + axitinib, nivolumab + cabozantinib and nivolumab + ipilimumab were similar in terms of PFS and superior to sunitinib, but pembrolizumab + axitinib also demonstrated a better OS. Our subgroup analysis showed that in favorable-risk patients, combination therapies showed no significant advantage over sunitinib, while in intermediate-poor risk patients, both pembrolizumab + axitinib and nivolumab + ipilimumab improved OS compared to sunitinib. Abstract Background: Recently, numerous combination therapies based on immune checkpoint inhibitors (ICI) and vascular endothelial growth factor (VEGF) inhibitors have been proposed as first-line treatments for advanced renal cell carcinoma (aRCC). Our study aimed to compare the efficacy of these combination regimens by the application of an innovative method that reconstructs individual patient data. Methods: Six phase III studies describing different combination regimens for aRCC were selected. Individual patient data were reconstructed from Kaplan–Meier (KM) curves through the “Shiny method”. Overall survival (OS) and progression-free survival (PFS) were compared among combination treatments and sunitinib. Results were summarized as multi-treatment KM curves. Standard statistical testing was used, including hazard ratio and likelihood ratio tests for heterogeneity. Results: In the overall population of aRCC patients, pembrolizumab + lenvatinib showed the longest median PFS and was expected to determine the longest OS. Pembrolizumab + axitinib, nivolumab + cabozantinib and nivolumab + ipilimumab were similar in terms of PFS, but pembrolizumab + axitinib also demonstrated a better OS. Our subgroup analysis showed that sunitinib is still a valuable option, whereas, in intermediate-poor risk patients, pembrolizumab + axitinib and nivolumab + ipilimumab significantly improve OS compared to sunitinib. Conclusion: The Shiny method allowed us to perform all head-to-head indirect comparisons between these agents in a context in which “real” comparative trials have not been performed.


Introduction
Over the past decades, systemic treatment for advanced renal cell carcinoma (aRCC) has considerably improved along with the development of new pharmacological targets. The main advancement in this setting has been the development of monoclonal antibodies and multitarget tyrosine-kinase inhibitors (TKI) that inhibit tumor growth and angiogenesis through the vascular endothelial growth factor (VEGF) pathway (e.g., bevacizumab, axitinib, lenvatinib, cabozantinib and sunitinib) [1].
In this ever-changing scenario, comparisons between these innovative treatments are essential to drivinge physician and regulatory authorities' decisions; however, randomized clinical trials (RCTs) have tested the efficacy of combination treatments only against sunitinib. On the other hand, in the field of methods for survival analysis, a new artificial intelligence technique (called the "Shiny method") has been developed and is increasingly used to reconstruct individual patient data from Kaplan-Meier (KM) curves and to generate cross-trial comparisons for which RCTs are lacking [6][7][8].
The Shiny method, also known as IPDfromKM, is an innovative tool of survival analysis in which software based on artificial intelligence reconstructs individual patient data. These reconstructed patients represent a new form of original clinical material; in particular, these reconstructed patients are suitable to perform indirect comparisons and consequently determine the place in the therapy of individual agents. In recent papers, we described the method and numerous experiences of application, especially in the field of oncology [9]. These analyses are helpful in providing an overview of new treatments, ranking their effectiveness based on indirect comparisons, and assessing equivalence from a regulatory viewpoint.
In this study, we applied the "Shiny method" to compare the efficacy of the main ICIbased combination regimens for aRCC, considering sunitinib as the standard of care (SOC).

Literature Search
First, we searched PubMed, Cochrane Library and the ClinicalTrials.gov databases to identify randomized controlled trials (RCTs) eligible for our analysis (last query on 30 October 2022 [10]. The main inclusion criteria were: (a) previously untreated adult patients with aRCC; (b) phase III trial; (c) PFS or OS endpoint; (d) results reported as a KM curve. For each curve of included trials, we collected the number of enrolled patients and the number of events (progression or death for PFS, death for OS). To avoid duplicate inclusion of patients of the same trial, we considered only the most recent publication.

Reconstruction of Individual Patient Data
We reconstructed patient-level data from the KM curves of treatment and control arms of each trial using the "Shiny method" [6][7][8]. For this purpose, the KM curves were digitized using Webplotdigitizer (version 4.5 online; https://apps.automeris.io/wpd/ (accessed on 15 February 2023)); then, the x-vs-y data points were input into the "Reconstruct individual patient data from Kaplan-Meier survival curve" function of the Shiny software (version: Cancers 2023, 15, 2029 3 of 12 1.2.2.0 online; last update: 1 April 2021); the total number of patients and events were input as well. In this way, we generated the reconstructed individual patient data for each arm of included RCTs. The reconstructed data of the patients are stored in archives containing the following information: (a) date of enrollment and last follow-up; the observation period of each patient results from the difference between these two dates. (b) patient outcome on the last follow-up date (alive, dead or censored).

Statistical Analysis
For each combination treatment, median PFS and OS were determined from reconstructed data and compared to sunitinib using Cox statistics for time-to-event end-points. We reported results as hazard ratio (HR) with a 95% confidence interval (95%CI). Indirect comparisons between treatments (in all head-to-head combinations) have been performed using the Cox model under the R-platform. Heterogeneity across control groups of different RCTs was quantified according to the likelihood ratio test and the concordance test. Statistical analyses were performed using the "survival" package under the R-platform (version 4.2.1) and SygmaPlot (version 13) software.

Included Trials and Application of the Shiny Method
Six trials met the criteria for inclusion in our analysis (see Figure 1 for the PRISMA flowchart and Table 1 for RCT characteristics).

Reconstruction of Individual Patient Data
We reconstructed patient-level data from the KM curves of treatment and control arms of each trial using the "Shiny method" [6][7][8]. For this purpose, the KM curves were digitized using Webplotdigitizer (version 4.5 online; https://apps.automeris.io/wpd/(accessed on 15 February 2023)); then, the x-vs-y data points were input into the "Reconstruct individual patient data from Kaplan-Meier survival curve" function of the Shiny software (version: 1.2.2.0 online; last update: 1 April 2021); the total number of patients and events were input as well. In this way, we generated the reconstructed individual patient data for each arm of included RCTs. The reconstructed data of the patients are stored in archives containing the following information: (a) date of enrollment and last follow-up; the observation period of each patient results from the difference between these two dates. (b) patient outcome on the last follow-up date (alive, dead or censored).

Statistical Analysis
For each combination treatment, median PFS and OS were determined from reconstructed data and compared to sunitinib using Cox statistics for time-to-event end-points. We reported results as hazard ratio (HR) with a 95% confidence interval (95%CI). Indirect comparisons between treatments (in all head-to-head combinations) have been performed using the Cox model under the R-platform. Heterogeneity across control groups of different RCTs was quantified according to the likelihood ratio test and the concordance test. Statistical analyses were performed using the "survival" package under the R-platform (version 4.2.1) and SygmaPlot (version 13) software.

Included Trials and Application of the Shiny Method
Six trials met the criteria for inclusion in our analysis (see Figure 1 for the PRISMA flowchart and Table 1 for RCT characteristics).  In our analysis of PFS based on these trials, 12 patient cohorts, along with their respective information on progressions, were reconstructed from the original Kaplan-Meier curves through the Shiny method. Then, the Kaplan-Meier PFS curves from these reconstructed patients were plotted individually and reported in a single multi-treatment graph (Figure 2A), which is the typical result generated by the Shiny method. In this graph, the 6 cohorts treated with sunitinib were pooled into a single cohort, thus generating a total of 7 curves. The same analysis was then performed for OS. The latter results are reported in Figure 2B. Detailed results of all head-to-head comparisons are reported in Table 2 and Figure S2. All values of event number were explicitly reported in the original trials or relative Supplementary Materials; events were then calculated as the difference of the total number of patients minus the total number of censored cases. § CLEAR included an arm treated with lenvatinib (18 mg orally once daily) + everolimus (5 mg orally once daily), which has not been included in our analysis.

Progression-Free Survival: Indirect Comparisons of the Six Combination Treatments Plus Sunitinib with One Another
In analyzing our results on PFS, the six combination treatments, along with sunitinib monotherapy, were ranked as follows: Forest plots of HRs with 95%CI for PFS of the six trials against pooled sunitinib controls are shown in Figure S1.

Overall Survival: Indirect Comparisons of the 6 Combination Treatments Plus Sunitinib with One Another
In analyzing our results on OS, the six combination treatments, along with sunitinib monotherapy, were ranked as follows: Forest plots of HRs with 95%CI for OS of the six trials against pooled sunitinib controls are shown in Figure S1. Detailed results of all head-to-head comparisons are reported in Table 3 and Figure S3.  Figure 3A,B shows the curves of PFS and OS obtained from the 6 control groups treated with sunitinib monotherapy that were compared for heterogeneity assessment. As controls should behave in a similar way, if heterogeneity is present, this may be due to differences in baseline patients' characteristics across the 6 trials. The results indicated that heterogeneity was present among the six trials both according to the likelihood ratio test and the concordance test applied to PFS (likelihood test = 50.06 with 5 df; p < 0.001; concordance, 0.552, SE = 0.007) and OS (likelihood test = 15.36 with 5 df; p = 0.009; concordance, 0.534, SE = 0.01. The presence of heterogeneity is driven mainly by the nivolumab + ipilimumab and lenvatinib + pembrolizumab RCT control arm, respectively, in the PFS and OS analysis.

Heterogeneity Analysis on Sunitinib Monotherapy Curves
In the case of PFS, there is no longer a significant heterogeneity when the nivolumab + ipilimumab control group is left out (likelihood ratio test = 8.86 on 4 df, p = 0.06); this demonstrates that in this case, the heterogeneity is driven mainly by this control arm.
In contrast, in the case of OS analysis, there is still significant heterogeneity after excluding the nivolumab + ipilimumab control group (likelihood ratio test = 14.6 on 4 df, p = 0.006). In this case, the lenvatinib + pembrolizumab control arm performed significantly better than the others, and so the heterogeneity was driven by this curve in the OS analysis. As expected, the likelihood ratio test after excluding the lenvatinib + pembrolizumab control group was not significant (likelihood ratio test = 3.56 on 4 df, p = 0.5).
heterogeneity was present among the six trials both according to the likelihood ratio test and the concordance test applied to PFS (likelihood test = 50.06 with 5 df; p < 0.001; concordance, 0.552, SE = 0.007) and OS (likelihood test = 15.36 with 5 df; p = 0.009; concordance, 0.534, SE = 0.01. The presence of heterogeneity is driven mainly by the nivolumab + ipilimumab and lenvatinib + pembrolizumab RCT control arm, respectively, in the PFS and OS analysis. In the case of PFS, there is no longer a significant heterogeneity when the nivolumab + ipilimumab control group is left out (likelihood ratio test = 8.86 on 4 df, p = 0.06); this demonstrates that in this case, the heterogeneity is driven mainly by this control arm.
In contrast, in the case of OS analysis, there is still significant heterogeneity after excluding the nivolumab + ipilimumab control group (likelihood ratio test = 14.6 on 4 df, p = 0.006). In this case, the lenvatinib + pembrolizumab control arm performed significantly better than the others, and so the heterogeneity was driven by this curve in the OS analysis. As expected, the likelihood ratio test after excluding the lenvatinib + pembrolizumab control group was not significant (likelihood ratio test = 3.56 on 4 df, p = 0.5).

Subgroup Analysis: Overall Survival in Favourable vs. Intermediate/Poor Risk Patients
In patients treated with nivolumab + ipilimumab and pembrolizumab + axitinib, OS's KM curves were available for patients with favourable versus intermediate/poor risk.
Multi-treatment KM curves are depicted in Figure 4A,B.

Subgroup Analysis: Overall Survival in Favourable vs. Intermediate/Poor Risk Patients
In patients treated with nivolumab + ipilimumab and pembrolizumab + axitinib, OS's KM curves were available for patients with favourable versus intermediate/poor risk.

Discussion
The present study investigated the main first-line combination treatments for aRCC by application of an innovative method of indirect comparison of survival data, the "Shiny method". Our choice to employ this approach to conduct these indirect comparisons, not a standard network meta-analysis [17][18][19], was because this method offers some advantages, such as considering the length of the follow-up of the trials and managing in a more precise way the variance of the model. The fact that the Shiny method evaluates not only the number of events but also the time of their occurrence is particularly important in a context like the one studied in here, where the follow-up lengths were considerably different across the agents investigated, and some differences were substantial. For example, some treatments (such as nivolumab + ipilimumab or sunitinib monotherapy and, to a lesser extent, lenvatinib + pembrolizumab and nivolumab + cabozantinib) had a particularly long follow-up as opposed to atezolizumab + bevacizumab, the follow-up of which was short. Consequently, some event rates were influenced by the length of the follow-up; in this framework, the rankings estimated by the Shiny method have an advantage in that they accounted for this important factor.
In recent years, the Shiny method has increasingly been used not only in oncology but also in other areas of of therapeutics, such as medical devices [20], surgery [21,22] and also in patients with COVID-19 [23]. However, oncology and oncohematology remain the two main areas of application, where the Shiny method represents a quite simple alternative to network meta-analysis [8,9,24,25].
Although combination regimens for aRCC target similar pathways, our results indicate that their efficacy is not equal. Median PFS and OS obtained with atezolizumab + bevacizumab did not differ significantly from that obtained with sunitinib, and so this result does not yet justify its use. In contrast, several combinations of TKI + ICI demonstrated a significant advantage, in terms of both PFS and OS, when compared to sunitinib monotherapy, in particular pembrolizumab + lenvatinib. Long-term follow-up will tell whether this therapy will confirm its superiority compared to the other combinations.
In subgroup analysis, patients with favourable risk obtained excellent results in terms of OS also with sunitinib monotherapy, and so combination therapy in these patients seems to add little OS benefit. In contrast, the intermediate/poor risk group of patients seems to benefit mostly from combination therapy, though we found no difference between nivolumab + ipilimumab vs. pembrolizumab + axitinib. In intermediate/poor risk patients, no separate KM curves were reported for patients with intermediate/low risk treated with pembrolizumab + lenvatinib, and this prevented us from applying the Shiny method to test the superiority observed by Bosma et al. in their meta-analysis [17].
Although the criteria for patient selection were similar among the six studies, our heterogeneity analysis of control arms demonstrated a significantly better PFS and OS in the controls of the CheckMate214 trial (the one testing nivolumab + ipilimumab) compared with the other control arms. Hence, this raises the possibility that the remarkable survival found in the active arm of CheckMate214 depends on the favorable characteristics of the patients and is therefore an overestimate. To further assess this issue, we compared in more detail the inclusion criteria and the characteristics of patients in the selected RCTs, but we found no remarkable difference. Of note, CheckMate214 did not report the Karnofsky performance status score (PS) of patients, while the CLEAR trial presented a slightly higher number of patients with favorable MSKCC prognostic score and with better PS.
In the light of our results, at least three combination therapies seem to have the best therapeutic profile. In the clinical use, the selection of the most suitable treatment should be guided by the patient's characteristics, particularly, the fitness of the subject to tolerate the adverse drug reactions (ADR), combined with a foresight on the best sequence of treatments if progression occurs [26]. Nivolumab + ipilimumab is burdened by more intense immunerelated ADRs, while lenvatinib + pembrolizumab is characterized by a higher incidence of grade >3 ADR (82.4% of patients) than other TKI + ICI combinations [27]. Although ADR may impact on health-related quality of life (HRQOL), a recent study observed that patients treated with pembrolizumab + lenvatinib have similar HRQOL score than patients receiving sunitinib [28]. It would be worthwhile to extend the comparison of HRQOL scores to the combination regimens, evaluated in the present study in terms of OS and PFS.
The use of these combination treatments in real practice may generate further evidence to guide treatment selection. For example, a recent study investigating the efficacy of nivolumab + ipilimumab vs. pembrolizumab + axitinib in real practice showed a reduced PFS compared with that reported in the RCTs [29].

Conclusions
In conclusion, the "Shiny method" permitted to generate valuable clinical results on new treatments for aRCC and to discuss their efficacy on the basis of original indirect comparisons. In addition, the "one to many approach", recently developed by our group, will allow us to quickly update the results of the present analysis as new treatments or longer follow-up will become available [30].
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/cancers15072029/s1, Table S1: Comparison of PFS and OS between combination treatments and sunitinib; Figure S1: Forest plots of HRs with 95%CI for PFS and OS of the six trials against pooled sunitinib controls; Figure