A network meta-analysis evaluates a new decision algorithm for fractional polynomial selection in the first-line treatment of RCC

A clinical algorithm-based approach that was developed to improve fractional polynomial model selection used additional criteria based on face validity, predictive accuracy, and expert opinion, which improved the plausibility of the results of survival in patients with renal cell carcinoma.

A clinical algorithm-based approach that was developed to improve fractional polynomial (FP) model selection used additional criteria based on face validity, predictive accuracy, and expert opinion, which improved the plausibility of survival outcomes in patients with renal cell carcinoma (RCC) who were receiving first-line therapy compared with models based on statistical adjustment alone, according to data from a network meta-analysis ( NMA).1

The results of the NMA, which were presented during the Colloquium on genitourinary cancers 2022also suggested that use of the algorithm in other indications beyond RCC in the first-line setting will require customization based on prior knowledge of treatments, their mechanism of action and criteria of interest for this specific indication.

“Wider application of this decision algorithm may also improve the accuracy and relevance of indirect comparisons of oncology treatments by strengthening the predictive accuracy of selected model estimates and better aligning with clinical expectations,” wrote L. Study lead author Bradley McGregor, MD, clinical director of the Lank Center for Genitourinary Oncology, senior physician at the Dana-Farber Cancer Institute and instructor in medicine at Harvard Medical School, and colleagues, in the poster.

Determining the cost-effectiveness (CE) of cancer treatments requires estimating long-term survival beyond what has been assessed so far, but no trial has compared all relevant treatment options to assess this EC in the first-line treatment of RCC. Differences in the mechanism of action of TKIs in monotherapy versus immunotherapy/TKI combinations result in differences in survival trends and produce disproportionate risks over time.

“This variability over time necessitates indirect treatment comparisons that take into account time-varying hazards, such as NMA using FP,” the study authors wrote. “FP NMA involves estimating 1- or 2-parameter models, called first-order and second-order models. Each of these parameters is assigned its own polynomial with a power selected from a range of predefined possible polynomial powers. This results in a wide variety of power combinations, leading to a large pool of candidate models. A single ideal model is then selected from this pool of available models.

Clinical plausibility and validation against available information are important drivers of FP model selection, but the statistical fit criteria used for model selection, such as deviance information criteria, do not hold. disregard clinical data. As such, they only illustrate the overall statistical fit of the model to the test results that have been reported; this can lead to models that demonstrate unlikely trends in risk over time and lack face validity.

For this analysis, the researchers developed an algorithm with the aim of improving the FP NMA model selection process by considering not only the statistical goodness of fit, but also the predictive accuracy and clinical plausibility. The algorithm was applied for analysis and to make indirect comparisons of overall survival (OS) and progression-free survival (PFS) of first-line RCC treatment options.

The landscape of first-line RCC treatment has evolved from the use of single-agent TKIs to new combinations of immunotherapies. Data from pivotal Phase 3 trials such as the JAVELIN study (NCT02684006), CLEAR trial (NCT02811861), CheckMate-9ER trial (NCT03141177) and KEYNOTE-426 trial (NCT02853331) supported the approvals for avelumab (Bavencio) plus axitinib (Inlyta), pembrolizumab (Keytruda) plus lenvatinib (Lenvima), nivolumab (Opdivo) plus cabozantinib (Cabometyx), and pembrolizumab plus axitinib, respectively.

The data from these combinations was used in the algorithm to examine the current process of selecting the fractional polynomial NMA model. Sunitinib (Sutent) was the common comparison arm across all trials considered in the analysis, and all of these data were used to form a connected web of evidence for OS and PFS outcomes.

Investigators reconstructed synthetic PFS and OS results from JAVELIN, CLEAR, and KEYNOTE-426, and reviewed individual patient data from CheckMate-9ER from the September 2020 database lockout. Forty-four models were considered for data review, and they represented several combinations of parameters.

For the OS network, researchers calculated absolute modeled outcomes using the sunitinib arm of CheckMate-9ER as the anchor treatment to which relative treatment effects were applied; this was done because the trial was conducted recently and was noted to reflect current clinical practice regarding subsequent treatments. For the PFS network, the investigators pooled all data collected across the sunitinib arms of all trials to balance the significant differences observed among the survival trends of the individual sunitinib data.

Model selection was performed either solely on the basis of statistical fit criteria (DIC-based approach) or on the basis of a selection algorithm using a priori criteria (predictive accuracy with respect to experimental data). trial), face validity and clinical plausibility of survival beyond the observed trial period, and statistical adequacy criteria. Due to the lack of relevant long-term survival data in the studies, clinical plausibility of adjustments and long-term survival extrapolations from clinical algorithm-based approach models were shared with 2 treatment oncologists patients with RCC.

The DIC-based model selected the second-order model with p1 = -2; p2 = -2 for PFS, and the second-order model with p1 = 1; p2 = -2 for the operating system. In the long term, both models resulted in clinically implausible survival extrapolations, the researchers said. Notably, the results showed that the PFS with pembrolizumab plus axitinib lacked face validity, as the results of the immunotherapy/TKI combinations were expected to be comparable; however, the model showed favorable PFS outcomes for pembrolizumab plus axitinib compared to avelumab plus axitinib over 20 months.

With respect to OS, it was noted that the trial data were heterogeneous, and the Kaplan-Meir curves demonstrated that the curves for pembrolizumab plus lenvatinib and sunitinib crossed twice, as other trials had a single crossover or no crossover. Although a single functional formulary was fitted to all included trials, the formulary poorly reflected CLEAR results and the modeled OS for pembrolizumab/lenvatinib did not adequately reflect trial results. Investigators postulated that this was due to preferential adjustment to other trials or excessive adjustment to risk trends in the results.

Additionally, the DIC-based approach selected an OS model that estimated tail risks for pembrolizumab plus lenvatinib and avelumab plus axitinib over the first few months, resulting in a rapid decline in survival curves. Long-term OS with sunitinib was also found to lack face validity, showing a plateau.

Six viable models for PFS and 2 for OS were identified using the clinical algorithm-based approach, considering predictive accuracy and clinical plausibility for survival extrapolations. The selected model predicted the highest median PFS with pembrolizumab plus lenvatinib, at 23.5 months, as well as with nivolumab plus cabozantinib, at 19.2 months. The lowest predicted median PFS was for sunitinib, at 10.2 months.

Additionally, PFS models were found to overestimate outcomes for most of the treatments examined, and this was thought to be partly due to anchoring absolute outcomes to a pooled sunitinib arm; this led to overestimates for treatments with lower performing comparator sunitinib arms. The OS model was selected in accordance with the opinion of clinical experts that the use of immunotherapy after sunitinib treatment would lead to favorable results when compared to trial data, and would likely cause the curves to cross survival of immunotherapy plus TKI compared to sunitinib over time. Notably, OS results for immunotherapy/TKI combinations were similar.

The researchers noted that while DIC-based fitting is still an effective measure for assessing the fit of FP models to observed test data, it can still generate clinically implausible results. The use of additional model selection criteria improved the clinical plausibility of the survival results compared to statistical adjustment alone.

The study authors noted that the decision to use a pooled sunitinib arm from individual sunitinib arms from all available trials resulted in over- and under-predictions of PFS models by both selection approaches due to because the relative efficacy of the treatments which was estimated based on the sunitinib arm of each trial, and the absolute survival results were estimated based on pooled data from the sunitinib arm spanning the trials.

“In our study, the FP models generally performed imperfectly against some of the trial data,” the study authors wrote. “This likely reflects heterogeneity between trial populations, differing mechanisms of action and subsequent treatment regimens and underscores the need for further research on incorporating adjustment for intra-assay heterogeneity and inter-trial in fitting FP models to survival data.”

Reference

  1. McGregor B, Petersohn S, Klijn SL, and all. Network meta-analysis (NMA) of first-line treatments for advanced renal cell carcinoma (1L aRCC): development of a decision algorithm for the selection of fractional polynomial (FP) models. J Clin Oncol. 2022;40(supplement 6):391. doi:10.1200/JCO.2022.40.6_suppl.391

Martin E. Berry