Statistical Innovation in Oncology Drug Development

JIANJUN GAN Chair
GlaxoSmithKline
 
Wednesday, Aug 7: 10:30 AM - 12:20 PM
5147 
Contributed Papers 
Oregon Convention Center 
Room: CC-E144 

Main Sponsor

Biopharmaceutical Section

Presentations

A New Scoring Method for Ordering Components within the Composite Endpoint in Oncology Trials

In clinical trials, patient-reported outcomes (PRO), encompassing multiple scales, have been widely used in measuring patients' quality of life, especially in oncology trials. The generalized pairwise comparison (GPC) combines the primary endpoint (e.g., overall survival) and PRO as the composite endpoint to assess the treatment effect. The prioritized GPC method requires a well-defined hierarchical order for each component within the composite endpoint. However, establishing a clear ranking order for each PRO scale is challenging. Non-prioritized GPC methods treat all components equally, thereby diminishing the importance of the primary endpoint. In response, we propose a semi-prioritized method that avoids the need for a strict order of PRO scales while simultaneously preserving the importance of the primary endpoint. We redefined the scoring algorithm, utilizing all endpoints observed in a trial to estimate the composite treatment effect, with particular emphasis on the primary endpoint. We conducted comprehensive simulation studies to assess the limitations of existing methods and applied all tests to data from an oncology trial. 

Keywords

Generalized Pairwise Comparison

Composite Endpoint

Patient-Reported Outcomes

Oncology Trials 

View Abstract 2182

Co-Author(s)

Ling Shi, UMass Boston, Department of Nursing
Edward Valachovic, University at Albany
Victoria Lazariu, University at Albany

First Author

Peiwen Yu, University at Albany

Presenting Author

Peiwen Yu, University at Albany

A Novel R Shiny Tool TVCurve for Survival Analysis with Time-Varying Covariate in Oncology Studies

The study of time-varying covariates (TVCs) gains attention in both statistical and medical fields. An example of a TVC is the receipt of hematopoietic cell transplantation (HCT) after CAR-T infusion, as patients may receive HCT after infusion, or not at all. The standard Cox model and Kaplan-Meier (KM) curve (Naïve method) may introduce "immortal time bias" since they assume TVC status known at baseline. Landmark analysis and time-dependent (TD) Cox model is two alternatives, but visualization of survival curves remains challenging. A novel visualization, Smith-Zee, based on TD Cox model, addresses this issue by mimicking new patients with TVC status change at different times, which overcomes drawbacks of the Naïve and Landmark methods. In this study, we developed a novel R Shiny tool called TVCurveTM to address these challenges and TVCurveTM incorporates various models: Naïve Cox, landmark Cox, and the TD Cox, along with multiple survival curves such as Naïve KM, Landmark KM, and Smith-Zee. Our tool TVCurve breaks collaboration barriers since it does not require data sharing between institutions but ensures standardized analyses across diverse datasets. 

Keywords

R Shiny

Time Varying Covariates

Time-dependent Cox model

Survival Curves

Visualization 

View Abstract 2209

Co-Author(s)

Yang Qiao, Iowa State University
Fei Gao, Fred Hutchinson Cancer Research Center
Jordan Gauthier, Fred Hutch
Qiang Zhang, Wills Eye Hospital
Jarcy Zee, University of Pennsylvania

First Author

Yimei Li, University of Pennsylvania

Presenting Author

Qian Wu, Fred Hutch

A Proof-of-Concept Clinical Trial Design for Evolutionary Guided Precision Medicine for Cancer

Current Precision Medicine (CPM) matches cancer therapies to consensus molecular characteristics at one or more timepoints. However, it is well-known that cancers contain extensive subclonal heterogeneity and that their subclonal compositions evolve dynamically in response to therapy. Mathematical modeling of this subclonal evolution has the potential to optimize the timing and sequencing of therapies in an even more effective and personalized manner than CPM. Clinical trial designs that test Evolutionary Guided Precision Medicine (EGPM) strategies that may prevent or delay relapse thereby improving outcomes, are needed. We evaluated Dynamic Precision Medicine (DPM), an EGPM, vs CPM in a stratified randomized design with two strata based on whether the patient was predicted to benefit from DPM, using an evolutionary classifier. We present this new proof-of-concept clinical trial design for this purpose and perform computer simulations which show high power, control of false positive rates, and robust performance in the face of anticipated challenges to clinical translation. The design is distinct from biomarker-driven designs of CPM, and can provide a robust evaluation of EGPM. 

Keywords

Precision Medicine

Clinical Trial Design

Tumour Evolution

Oncology 

View Abstract 2604

Co-Author(s)

Wei He, Georgetown University
Peter Mon, Purdue University
Matthew McCoy, Georgetown University
Chen-Hsiang Yeang, Academia Sinica, Teipei
Robert Beckman, Georgetown University

First Author

Deepak Parashar, University of Warwick

Presenting Author

Deepak Parashar, University of Warwick

Adjusting for Multiple Treatment Switching - Stratified Two-stage Method using G-estimation

In oncology clinical trials, estimates of OS is often confounded because some patients in the control group switch to the experimental treatment. Several switching adjustment methods have been developed to adjust for treatment switching, such as RPSFTM and two-stage method. However, many clinical trials now have patients in the control group switching to different treatment regimens other than the experimental treatments, and most of the existing methods still assume all switchers switch to the same treatment thus cannot handle multilevel switching. Stratified RPSFTM has been proposed to adjust for multilevel switching, however, the method does not change RPSFTM's assumption that treatment effect is the same for all participants regardless of when treatment is received. Two-stage method adjusts for switching that occurs after a specific disease-related time-point, which is a more practical assumption than RPSFTM. Thus, we extended the two-stage method to stratified two-stage model to adjust for multilevel treatment switching and outperforms other methods when the treatment effect is strong and when there is ineligible difference between the switching treatments. 

Keywords

Treatment switching

Two-stage method

Multi-level switching

Randomized Clinical Trial

RPSFTM 

View Abstract 2547

Co-Author

Debarghya Nandi, University of Illinois At Chicago

First Author

Luoying Yang

Presenting Author

Luoying Yang

Bias in the IPCW estimator for censored pairwise comparisons: the importance of explicit estimands

Net benefit and win ratio are gaining interest in oncology and cardiovascular research for assessing multifaceted clinical outcomes. They compare multiple outcomes hierarchically by using all possible pairs from treatment and control groups. For time-to-event outcomes suffering from censoring, the inverse probability of censoring weighting (IPCW) is available. However, Dong's (2021) original IPCW formulation does not distinguish (a) "ties" in uncensored prioritized outcomes and (b) "uninformative" comparison due to censoring when considering lower-priority outcomes. The resulting censoring-dependent bias has been overlooked because of the lack of clear estimands in pairwise comparisons. In this talk, we introduce explicitly defined estimands for net benefit/win ratio with censored outcomes in terms of the "probabilities that would be observed if we had removed censoring". We show the bias by the series of simulations by varying dependencies in censoring, outcome correlations, and treatment effects for separate outcomes. We propose the modified IPCW estimator that reduces the bias but sacrifices efficiency by excluding uninformative censored pairs. 

Keywords

generalized pairwise comparisons

inverse probability of censoring weighting

net benefit

prioritized outcomes 

View Abstract 2179

Co-Author(s)

Musashi Fukuda
Tomohiro Shinozaki, Tokyo University of Science

First Author

Taku Chikamochi, Tokyo University of Science

Presenting Author

Taku Chikamochi, Tokyo University of Science

Impact of informative censoring in randomized trials with delayed treatment effect

Time-to-event endpoints like progression-free survival and overall survival in oncology randomized trials sometimes show differential censoring patterns between study arms which can be indicative of informative censoring, depending on censoring reasons. Informative censoring can bias treatment effect estimates but few simulation studies characterized the magnitude of its impact, particularly in the context of delayed treatment effects. We use copula methods to model dependent censoring data and assess the impact of informative censoring. To improve the understanding of copula models in this context, we propose a new measure of the strength of informative censoring, the probability of events being informatively censored. To guide the choice of copula, we further propose a visual tool for examining the underlying correlation pattern. We conduct simulation studies to assess the impact of informative censoring on estimation bias for hazard ratios, as well as on empirical power of unweighted, weighted log-rank tests, and the MaxCombo test. 

Keywords

Informative Censoring

Copula

Randomized clinical trials

Generalized log-rank tests 

View Abstract 2002

Co-Author(s)

Yujie Zhao, Merck & Co., Inc.
Gregory Chen, Merck & Co., Inc.
Margarita DONICA, Merck & Co., Inc.
Larry León, Merck & Co., Inc.
Ludovic Trinquart, Tufts Medical Center
Sabrina Wan, Merck & Co., Inc.

First Author

Jingyi Lin, Merck & Co., Inc

Presenting Author

Jingyi Lin, Merck & Co., Inc

Predict PFS and OS using ORR for PD1/PDL1 combination therapy development

In solid tumor therapy development, with early phase clinical efficacy objective response rate (ORR), deriving predictive models to median progression free survival (mPFS) and median overall survival (mOS) is important as it could optimize late phase trial design. In literature, there are only some ORR/mPFS/mOS association research with limited number of included trials. In this paper, we attempt to predict mPFS and mOS by ORR and optimize late phase trial design of PD1/PDL1 combination therapy. To include adequate eligible clinical trials, we derive comprehensive quantitative clinical trial landscape database by combining information from various sources. A tree-based machine learning regression model is derived to borrow strength and account for ORR/mPFS (ORR/mOS) relationship heterogeneity to ensure the predictive model is robust and has clear structure for interpretation. Over 1000 times cross validation, predictive MSE of proposed model is competitive to random forest, extreme gradient boosting, and superior to common additive and interaction regression models. Example application of proposed model on PD1/PDL1 combination therapy late phase trial POS evaluation is illustrated. 

Keywords

PD1 / PDL1 combination therapy development

Predict late phase clinical endpoint by early phase clinical endpoint

Quantitative clinical trial landscape data base

Tree based machine learning regression model

Borrow strength and account for heterogeneity

Clear structure for clinical interpretation and robustness 

View Abstract 1930

First Author

Lei Yang, Sanofi US

Presenting Author

Lei Yang, Sanofi US