Determination of the most appropriate method for extrapolating overall survival data from a placebo-controlled clinical trial of lenvatinib for progressive, radioiodine-refractory differentiated thyroid cancer

A Flexible Ensemble Learning Method for Survival Extrapolation

OBJECTIVES

Survival extrapolation is an important statistical concept for estimating long-term survival from short-term clinical trial data. It is widely used in health technology assessment (HTA). Survival extrapolation is often performed by fitting one or two parametric models selected based on experience or selecting a model based on some goodness of fit statistics from a predefined collection of models. The main challenge in survival extrapolation is that the result is sensitive to model misspecification. In this study, we aim to propose a new approach that has a robust performance for survival extrapolation.

METHODS

We propose a new method called Ensemble Learning for Survival Extrapolation (ELSE). Instead of selecting one best model from a predefined collection, ELSE builds an ensemble model based on a collection of models from the model library. Under this framework, we construct a point estimate of the long-term survival with a weighted average of the estimates of all candidate models and derive confidence intervals using nonparametric bootstrap.

RESULTS

With our extensive numerical simulation studies, the proposed ELSE method shows better performance than the traditionally used model selection procedure based on Akaike Information Criterion (AIC). With a real data application to the Therapeutically Applicable Research to Generate Effective Treatment Wilms Tumor project (TARGET-WT) data, the ELSE method produces better survival extrapolation results in point estimate accuracy and confidence interval coverage.

CONCLUSIONS

We developed an ensemble learning method for survival extrapolation (ELSE) which is robust for the underline data model and has good real data performance.

Estimating Systolic Blood Pressure Intervention Trial Participant Posttrial Survival Using Pooled Epidemiologic Cohort Data

Background

Intensive systolic blood pressure treatment (<120 mm Hg) in SPRINT (Systolic Blood Pressure Intervention Trial) improved survival compared with standard treatment (<140 mm Hg) over a median follow‐up of 3.3 years. We projected life expectancy after observed follow‐up in SPRINT using SPRINT‐eligible participants in the NHLBI‐PCS (National Heart, Lung, and Blood Institute Pooled Cohorts Study).

Methods and Results

We used propensity scores to weight SPRINT‐eligible NHLBI‐PCS participants to resemble SPRINT participants. In SPRINT participants, we estimated in‐trial survival (<4 years) using a time‐based flexible parametric survival model. In SPRINT‐eligible NHLBI‐PCS participants, we estimated posttrial survival (≥4 years) using an age‐based flexible parametric survival model and applied the formula to SPRINT participants to predict posttrial survival. We projected overall life expectancy for each SPRINT participant and compared it to parametric regression (eg, Gompertz) projections based on SPRINT data alone. We included 8584 SPRINT and 10 593 SPRINT‐eligible NHLBI‐PCS participants. After propensity weighting, mean (SD) age was 67.9 (9.4) and 68.2 (8.8) years, and 35.5% and 37.6% were women in SPRINT and NHLBI‐PCS, respectively. Using the NHLBI‐PCS–based method, projected mean life expectancy from randomization was 21.0 (7.4) years with intensive and 19.1 (7.2) years with standard treatment. Using the Gompertz regression, life expectancy was 11.2 (2.3) years with intensive and 10.5 (2.2) years with standard treatment.

Conclusions

Combining SPRINT and NHLBI‐PCS observed data likely offers a more realistic estimate of life expectancy than parametrically extrapolating SPRINT data alone. These results offer insight into the potential long‐term effectiveness of intensive SBP goals.

Extrapolation of Survival Curves Using Standard Parametric Models and Flexible Parametric Spline Models: Comparisons in Large Registry Cohorts with Advanced Cancer

BACKGROUND

It is often important to extrapolate survival estimates beyond the limited follow-up times of clinical trials. Extrapolated survival estimates can be highly sensitive to model choice; thus, appropriate model selection is crucial. Flexible parametric spline models have been suggested as an alternative to standard parametric models; however, their ability to extrapolate is not well understood.

AIM

To determine how well standard parametric and flexible parametric spline models predict survival when fitted to registry cohorts with artificially right-censored follow-up times.

METHODS

Adults with advanced breast, colorectal, small cell lung, non-small cell lung, or pancreatic cancer with a potential follow-up time of 10 y were selected from the SEER 1973-2015 registry data set. Patients were classified into 15 cohorts by cancer and age group at diagnosis (18-59, 60-69, 70+ y). Follow-up times for each cohort were right censored at 20%, 35%, and 50% survival. Standard parametric models (exponential, Weibull, Gompertz, log-logistic, log-normal, generalized gamma) and spline models (proportional hazards, proportional odds, normal/probit) were fitted to the 10-y data set and the 3 right-censored data sets. Predicted 10-y restricted mean survival time and percentage surviving at 10 y were compared with the observed values.

RESULTS

Across all data sets, the spline odds and spline normal models most frequently gave accurate predictions of 10-y survival outcomes. Visually, spline models tended to demonstrate better fit to the observed hazard functions than standard parametric models, both in the censored and 10-y data.

CONCLUSIONS

In these cohorts, where there was little uncertainty in the observed data, the spline models performed well when extrapolating beyond the observed data. Spline models should be routinely included in the set of models that are fitted when extrapolating cancer survival data.

Lenvatinib and sorafenib for differentiated thyroid cancer after radioactive iodine: a systematic review and economic evaluation

BACKGROUND

Thyroid cancer is a rare cancer, accounting for only 1% of all malignancies in England and Wales. Differentiated thyroid cancer (DTC) accounts for ≈94% of all thyroid cancers. Patients with DTC often require treatment with radioactive iodine. Treatment for DTC that is refractory to radioactive iodine [radioactive iodine-refractory DTC (RR-DTC)] is often limited to best supportive care (BSC).

OBJECTIVES

We aimed to assess the clinical effectiveness and cost-effectiveness of lenvatinib (Lenvima®; Eisai Ltd, Hertfordshire, UK) and sorafenib (Nexar®; Bayer HealthCare, Leverkusen, Germany) for the treatment of patients with RR-DTC.

DATA SOURCES

EMBASE, MEDLINE, PubMed, The Cochrane Library and EconLit were searched (date range 1999 to 10 January 2017; searched on 10 January 2017). The bibliographies of retrieved citations were also examined.

REVIEW METHODS

We searched for randomised controlled trials (RCTs), systematic reviews, prospective observational studies and economic evaluations of lenvatinib or sorafenib. In the absence of relevant economic evaluations, we constructed a de novo economic model to compare the cost-effectiveness of lenvatinib and sorafenib with that of BSC.

RESULTS

Two RCTs were identified: SELECT (Study of [E7080] LEnvatinib in 131I-refractory differentiated Cancer of the Thyroid) and DECISION (StuDy of sorafEnib in loCally advanced or metastatIc patientS with radioactive Iodine-refractory thyrOid caNcer). Lenvatinib and sorafenib were both reported to improve median progression-free survival (PFS) compared with placebo: 18.3 months (lenvatinib) vs. 3.6 months (placebo) and 10.8 months (sorafenib) vs. 5.8 months (placebo). Patient crossover was high (≥ 75%) in both trials, confounding estimates of overall survival (OS). Using OS data adjusted for crossover, trial authors reported a statistically significant improvement in OS for patients treated with lenvatinib compared with those given placebo (SELECT) but not for patients treated with sorafenib compared with those given placebo (DECISION). Both lenvatinib and sorafenib increased the incidence of adverse events (AEs), and dose reductions were required (for > 60% of patients). The results from nine prospective observational studies and 13 systematic reviews of lenvatinib or sorafenib were broadly comparable to those from the RCTs. Health-related quality-of-life (HRQoL) data were collected only in DECISION. We considered the feasibility of comparing lenvatinib with sorafenib via an indirect comparison but concluded that this would not be appropriate because of differences in trial and participant characteristics, risk profiles of the participants in the placebo arms and because the proportional hazard assumption was violated for five of the six survival outcomes available from the trials. In the base-case economic analysis, using list prices only, the cost-effectiveness comparison of lenvatinib versus BSC yields an incremental cost-effectiveness ratio (ICER) per quality-adjusted life-year (QALY) gained of £65,872, and the comparison of sorafenib versus BSC yields an ICER of £85,644 per QALY gained. The deterministic sensitivity analyses show that none of the variations lowered the base-case ICERs to < £50,000 per QALY gained.

LIMITATIONS

We consider that it is not possible to compare the clinical effectiveness or cost-effectiveness of lenvatinib and sorafenib.

CONCLUSIONS

Compared with placebo/BSC, treatment with lenvatinib or sorafenib results in an improvement in PFS, objective tumour response rate and possibly OS, but dose modifications were required to treat AEs. Both treatments exhibit estimated ICERs of > £50,000 per QALY gained. Further research should include examination of the effects of lenvatinib, sorafenib and BSC (including HRQoL) for both symptomatic and asymptomatic patients, and the positioning of treatments in the treatment pathway.

STUDY REGISTRATION

This study is registered as PROSPERO CRD42017055516.

FUNDING

The National Institute for Health Research Health Technology Assessment programme.

Matching-Adjusted Indirect Comparison of the Efficacy of Apalutamide and Enzalutamide with ADT in the Treatment of Non-Metastatic Castration-Resistant Prostate Cancer

INTRODUCTION

Apalutamide and enzalutamide are next-generation androgen receptor inhibitors that demonstrated efficacy in placebo-controlled studies (SPARTAN for apalutamide; PROSPER for enzalutamide) when used in combination with androgen deprivation therapy (ADT) for treatment of non-metastatic castration-resistant prostate cancer (nmCRPC). In the absence of comparative studies between these agents, the present study sought to indirectly compare metastasis-free survival (MFS) and overall survival (OS) in patients with nmCRPC who received these therapies.

METHODS

Individual patient-level data from SPARTAN (apalutamide plus ADT) and published data from PROSPER (enzalutamide plus ADT) were utilized. An anchored matching-adjusted indirect comparison (MAIC) was conducted by weighting the patients from the SPARTAN study to match baseline characteristics reported for PROSPER. Hazard ratios (HRs) for MFS and OS were re-estimated for SPARTAN using weighted Cox proportional hazards models and indirectly compared with those of PROSPER using a Bayesian network meta-analysis.

RESULTS

From the SPARTAN population (N = 1207), a total of 1171 patients were matched to the PROSPER population (N = 1401). The recalculated HRs (95% confidence interval) for apalutamide versus ADT based on the reweighted SPARTAN data to mimic the PROSPER patient population were 0.26 (0.21; 0.33) for MFS and 0.62 (0.41; 0.94) for OS. MAIC-based HRs (95% credible interval) for apalutamide versus enzalutamide were 0.91 (0.68; 1.22) for MFS and 0.77 (0.46; 1.30) for OS. The Bayesian probabilities of apalutamide being more effective than enzalutamide were 73.6% for MFS and 83.5% for OS.

CONCLUSIONS

MAIC results suggest that nmCRPC patients treated with apalutamide have a higher probability of a more favorable MFS and OS compared with those treated with enzalutamide.

Survival Extrapolation in Cancer Immunotherapy: A Validation-Based Case Study

BACKGROUND

Immune-checkpoint inhibitors may provide long-term survival benefits via a cured proportion, yet data are usually insufficient to prove this upon submission to health technology assessment bodies.

OBJECTIVE

We revisited the National Institute for Health and Care Excellence assessment of ipilimumab in melanoma (TA319). We used updated data from the pivotal trial to assess the accuracy of the extrapolation methods used and compared these to previously unused techniques to establish whether an alternative extrapolation may have provided more accurate survival projections.

METHODS

We compared projections from the piecewise survival model used in TA319 and those produced by alternative models (fit to trial data with minimum follow-up of 3 years) to a longer-term data cut (5-year follow-up). We also compared projections to external data to help assess validity. Alternative approaches considered were parametric, spline-based, mixture, and mixture-cure models.

RESULTS

Only the survival model used in TA319 and a mixture-cure model provided 5-year survival predictions close to those observed in the 5-year follow-up data set. Standard parametric, spline, and non-curative-mixture models substantially underestimated 5-year survival. Survival estimates from the TA319 model and the mixture-cure model diverge considerably after 5 years and remain unvalidated.

CONCLUSIONS

In our case study, only models that incorporated an element of external information (through a cure fraction combined with background mortality rates or using registry data) provided accurate estimates of 5-year survival. Flexible models that were able to capture the complex hazard functions observed during the trial, but which did not incorporate external information, extrapolated poorly.

Long-term Survival and Cost-effectiveness Associated With Axicabtagene Ciloleucel vs Chemotherapy for Treatment of B-Cell Lymphoma

Importance

Axicabtagene ciloleucel, a chimeric antigen receptor T-cell therapy, represents a new and potentially curative treatment option for B-cell lymphoma. It is expected to have long-term survival benefits; however, long-term survival data are limited.

Objective

To estimate the long-term survival and cost-effectiveness of axicabtagene ciloleucel for treatment of relapsed or refractory B-cell lymphoma.

Design, Setting, and Participants

Economic evaluation study using a survival analysis that digitized and extrapolated survival curves published in the ZUMA-1 trial (Safety and Efficacy of KTE-C19 in Adults With Refractory Aggressive Non-Hodgkin Lymphoma), which enrolled patients between November 2015 and September 2016 and had a maximum follow-up of 24 months. Five different survival models (standard parametric, flexible parametric, 2 mixture cure models, and a flexible parametric mixture model) were used to extrapolate the survival curves to a lifetime horizon from January through June 2018. A cost-effectiveness analysis, from both a trial-based and lifetime horizon, was also conducted to inform the value of this novel therapy. The model was based on data from 111 patients with B-cell lymphoma who were enrolled in the ZUMA-1 trial.

Interventions

One-time administration of axicabtagene ciloleucel compared with chemotherapy.

Main Outcomes and Measures

Undiscounted and discounted life-years (LYs) and quality-adjusted life-years (QALYs), total costs, and incremental costs per LY and QALY gained.

Results

The modeled cohort of 111 patients started at 58 years of age. At the end of the trial, treatment with axicabtagene ciloleucel resulted in 0.48 more LYs and 0.34 more QALYs than chemotherapy, producing a cost-effectiveness estimate of $896 600 per QALY for public payers and $1 615 000 per QALY for commercial payers. Extrapolated long-term survival for patients treated with axicabtagene ciloleucel ranged from 2.83 to 9.19 discounted LYs and from 2.07 to 7.62 discounted QALYs. Incrementally, treatment with axicabtagene ciloleucel was associated with 1.89 to 5.82 discounted LYs and 1.52 to 4.90 discounted QALYs vs chemotherapy. With the use of these incremental estimates of survival, cost-effectiveness estimates ranged from $82 400 to $230 900 per QALY gained for public payers and from $100 400 to $289 000 per QALY gained for commercial payers.

Conclusions and Relevance

Treatment with axicabtagene ciloleucel appears to be associated with incremental gains in survival over chemotherapy. The range in projected long-term survival was wide and reflected uncertainty owing to limited follow-up data. Cost-effectiveness is associated with long-term survival, with further evidence needed to reduce uncertainty.

Matching-adjusted indirect treatment comparison of ribociclib and palbociclib in HR+, HER2- advanced breast cancer

Background

Ribociclib (RIBO) and palbociclib (PALBO), combined with letrozole (LET), have been evaluated as treatments for hormone receptor-positive, human epidermal growth factor receptor 2-negative advanced breast cancer in separate Phase III randomized controlled trials (RCTs), but not head-to-head. Population differences can lead to biased results by classical indirect treatment comparison (ITC). Matching-adjusted indirect comparison (MAIC) aims to correct these differences. We compared RIBO and PALBO in hormone receptor-positive/human epidermal growth factor receptor 2-negative advanced breast cancer using MAIC.

Methods

Patient-level data were available for RIBO (MONALEESA-2), while only published summary data were available for PALBO (PALOMA-2). Weights were assigned to MONALEESA-2 patient data such that mean baseline characteristics matched those reported for PALOMA-2; the resulting matched cohort was used in comparisons. Limited by the results reported in PALOMA-2, progression-free survival (PFS) was the primary comparison. Cox regression models were used to calculate adjusted hazard ratios (HRs) for PFS, before indirect treatment comparison (ITC) was performed with 95% confidence intervals. An exploratory analysis was performed similarly for overall survival using earlier PALBO data (PALOMA-1). Grade 3/4 adverse events were also compared.

Results

Racial characteristics, prior chemotherapy setting, and the extent of metastasis were the most imbalanced baseline characteristics. The unadjusted PFS HRs were 0.556 (0.429, 0.721) for RIBO+LET versus LET alone and 0.580 (0.460, 0.720) for PALBO+LET versus LET alone. MAIC adjustment resulted in an HR of 0.524 (0.406, 0.676) for RIBO+LET versus LET. PFS ITC using unadjusted trial data produced an HR of 0.959 (0.681, 1.350) for RIBO versus PALBO, or 0.904 (0.644, 1.268) with MAIC. Unadjusted overall survival HR of RIBO versus PALBO was 0.918 (0.492, 1.710); while exploratory MAIC was 0.839 (0.440, 1.598). ITC of grade 3/4 adverse events yielded a risk ratio of 0.806 (0.604, 1.076).

Conclusion

MAIC was performed for RIBO and PALBO in the absence of a head-to-head trial: though not statistically significant, the results favored RIBO.