Search

Drug development can be associated with slow timelines, particularly for rare or difficult-to-treat solid tumors such as glioblastoma. The use of external data in the design and analysis of trials has attracted significant interest because it has the potential to improve the efficiency and precision of drug development. A recurring challenge in the use of external data for clinical trials, however, is the difficulty in accessing high-quality patient-level data. Academic research groups generally do not have access to suitable datasets to effectively leverage external data for planning and analyses of new clinical trials. Given the need for resources to enable investigators to benefit from existing data assets, we have developed the Glioblastoma External (GBM-X) Data Platform which will allow investigators in neuro-oncology to leverage our data collection and obtain analyses. GBM-X strives to provide an uncomplicated process to use external data, contextualize single-arm trials, and improve inference on treatment effects early in drug development. The platform is designed to welcome interested collaborators and integrate new data into the platform, with the expectation that the data collection can continue to grow and remain updated. With such features, GBM-X is designed to help to accelerate evaluation of therapies, to grow with collaborations, and to serve as a model to improve drug discovery for rare and difficult-to-treat tumors in oncology.

Randomised controlled clinical trials are widely considered the preferred method for evaluating the efficacy or effectiveness of interventions in healthcare. Adaptive trials incorporate changes as the study proceeds, such as modifying allocation probabilities or eliminating treatment arms that are likely to be ineffective. These designs have been widely used in drug discovery studies but can also be useful in health services and implementation research and have been minimally used. In this article, we use an ongoing adaptive trial and two completed parallel group studies as motivating examples to highlight the potential advantages, disadvantages, and important considerations when using adaptive trial designs in health services and implementation research. We also investigate the impact on power and the study duration if the two completed parallel group trials had instead been conducted using adaptive principles. Compared with traditional trial designs, adaptive designs can often allow the evaluation of more interventions, adjust participant allocation probabilities (eg, to achieve covariate balance), and identify participants who are likely to agree to enrol. These features could reduce resources needed to conduct a trial. However, adaptive trials have potential disadvantages and practical aspects that need to be considered, most notably: outcomes that can be rapidly measured and extracted (eg, long term outcomes that take considerable time to measure from data sources can be challenging), minimal missing data, and time trends. In conclusion, adaptive designs are a promising approach to help identify how best to implement evidence based interventions into real world practice in health services and implementation research.

Sharing data from control groups across concurrent randomised clinical trials with identical enrolment criteria and the same control therapy can translate into efficiencies for the drug development process. We discuss potential benefits and risks of prospective data-sharing plans for concurrent randomised trials.

Pattern of HER2 amplification in the subset of HER2 heterogeneous tumors. Heatmap generated using data from central pathology evaluation of HER2 heterogeneity. The graphic includes patients classified as HER2 heterogeneous depicted in each row. The color code to the right of the heatmap refers to the percentage of HER2 amplified cells per core biopsy tumor area. Core biopsy tumor areas, A, B and C were obtained from core biopsy site 1 and D, E and F from core biopsy site 2.

FISH raw counts

Intratumor heterogeneity is postulated to cause therapeutic resistance. To prospectively assess the impact of HER2 (ERBB2) heterogeneity on response to HER2-targeted therapy, we treated 164 patients with centrally confirmed HER2-positive early-stage breast cancer with neoadjuvant trastuzumab emtansine plus pertuzumab. HER2 heterogeneity was assessed on pretreatment biopsies from two locations of each tumor. HER2 heterogeneity, defined as an area with ERBB2 amplification in >5% but P < 0.0001, adjusted for hormone receptor status). Single-cell ERBB2 FISH analysis of cellular heterogeneity identified the fraction of ERBB2 nonamplified cells as a driver of therapeutic resistance. These data suggest HER2 heterogeneity is associated with resistance to HER2-targeted therapy and should be considered in efforts to optimize treatment strategies.Significance:HER2-targeted therapies improve cure rates in HER2-positive breast cancer, suggesting chemotherapy can be avoided in a subset of patients. We show that HER2 heterogeneity, particularly the fraction of ERBB2 nonamplified cancer cells, is a strong predictor of resistance to HER2 therapies and could potentially be used to optimize treatment selection.See related commentary by Okines and Turner, p. 2369.This article is highlighted in the In This Issue feature, p. 2355

Purpose:We discuss designs and interpretable metrics of bias and statistical efficiency of “externally controlled” trials (ECT) and compare ECT performance to randomized and single-arm designs.Experimental Design:We specify an ECT design that leverages information from real-world data (RWD) and prior clinical trials to reduce bias associated with interstudy variations of the enrolled populations. We then used a collection of clinical studies in glioblastoma (GBM) and RWD from patients treated with the current standard of care to evaluate ECTs. Validation is based on a “leave one out” scheme, with iterative selection of a single-arm from one of the studies, for which we estimate treatment effects using the remaining studies as external control. This produces interpretable and robust estimates on ECT bias and type I errors.Results:We developed a model-free approach to evaluate ECTs based on collections of clinical trials and RWD. For GBM, we verified that inflated false positive error rates of standard single-arm trials can be considerably reduced (up to 30%) by using external control data.Conclusions:The use of ECT designs in GBM, with adjustments for the clinical profiles of the enrolled patients, should be preferred to single-arm studies with fixed efficacy thresholds extracted from published results on the current standard of care.

Verification of detection power calculation. Comparison between the predicted detection power of 18 replicates and the actual fraction of those replicates that were called MRD+. 18 independent samples from a 1:100k dilution were probed with a 488 SNV panel (see methods). Replicates had their duplex consensus reads downsampled in increments of 10% and detection power was recalculated. Reported are the median and range of predicted detection powers for all replicates at each downsampling.

Supplementary Figures and Figure Legends

Early stage breast cancer cohort. (A) Overview of subtype distribution of cohort as well as the timing of blood draws relative to surgery and treatment. (B) REMARK diagram of the early stage breast cancer cohort. (C) Distribution of the number of mutations tracked for each patient.

Purpose:Existing cell-free DNA (cfDNA) methods lack the sensitivity needed for detecting minimal residual disease (MRD) following therapy. We developed a test for tracking hundreds of patient-specific mutations to detect MRD with a 1,000-fold lower error rate than conventional sequencing.Experimental Design:We compared the sensitivity of our approach to digital droplet PCR (ddPCR) in a dilution series, then retrospectively identified two cohorts of patients who had undergone prospective plasma sampling and clinical data collection: 16 patients with ER+/HER2− metastatic breast cancer (MBC) sampled within 6 months following metastatic diagnosis and 142 patients with stage 0 to III breast cancer who received curative-intent treatment with most sampled at surgery and 1 year postoperative. We performed whole-exome sequencing of tumors and designed individualized MRD tests, which we applied to serial cfDNA samples.Results:Our approach was 100-fold more sensitive than ddPCR when tracking 488 mutations, but most patients had fewer identifiable tumor mutations to track in cfDNA (median = 57; range = 2–346). Clinical sensitivity was 81% (n = 13/16) in newly diagnosed MBC, 23% (n = 7/30) at postoperative and 19% (n = 6/32) at 1 year in early-stage disease, and highest in patients with the most tumor mutations available to track. MRD detection at 1 year was strongly associated with distant recurrence [HR = 20.8; 95% confidence interval, 7.3–58.9]. Median lead time from first positive sample to recurrence was 18.9 months (range = 3.4–39.2 months).Conclusions:Tracking large numbers of individualized tumor mutations in cfDNA can improve MRD detection, but its sensitivity is driven by the number of tumor mutations available to track.

Statistical Details and Supplementary Figures/Tables

Overview schematic of MRD assay. MRD assay workflow from characterizing a primary tumor to determining if a blood draw contains evidence of residual disease.

Validation of tumor fraction confidence interval and effects of biological variability on tumor fraction inference. (A) Confidence intervals inferred from simulated data were aggregated into coverage probability for each condition and compared against theoretical coverage probability of 95% (see Methods). (B) Biological variability of clonal evolution loss was simulated by removing tumor evidence from a fraction of the fingerprint panel. Tumor fraction was calculated for simulated and original samples for comparison.

Figure S3: Operating characteristics, efficacy scenario 3 (an average of 3.2, 2.4 and 1.6 AEs per patients under toxicity scenarios 1, 2, 3) for RTOG 1016. Figure S4: Operating characteristics, efficacy scenario 1 (an average of 4.8, 3.6 and 2.4 AEs per patients under toxicity scenarios 1, 2, 3) for DeEscalate.

Patients who experienced local recurrence only after treatment for early stage breast cancer. cfDNA time points for all patients who experienced local recurrence only. End points represent time of last follow up or death.

In recent years several clinical studies have investigated deintensified treatments in human papillomavirus (HPV)-associated head and neck squamous cell carcinoma. Two large phase III trials, RTOG 1016 and De-ESCALaTE, which attempted to reduce toxicity by replacing radiotherapy in combination with cisplatin with the use of cetuximab in combination with radiotherapy, recently suggested that radiotherapy + cetuximab leads to inferior survival compared with standard therapy (observed HRs of 1.45 and 5 in RTOG 1016 and De-ESCALaTE), as well as increased rates of locoregional failure. These unexpected results should prompt a careful examination of deintensification trials, both in HPV-associated oropharyngeal cancer and in other contexts. Statistical designs for deintensification studies should be consistent with the study aims of reducing toxicities while maintaining survival nearly identical to the standard of care. We suggest criteria to design future deintensification trials and discuss important operating characteristics, including tradeoffs between power and stringent early stopping rules to reduce the number of patients exposed to inferior treatments. Using retrospective analyses of previous clinical studies, we compared designs with different operating characteristics. As an example, using outcomes data from RTOG 1016 and De-ESCALaTE, we conducted analyses to determine advantages of (i) stringent futility early-stopping rules and of (ii) study designs that leverage both toxicity and efficacy endpoints for interim analyses. We show that increasing the frequency of interim-futility analyses has little impact on power, but the average study duration and number of subjects enrolled before the trial is closed for inferiority can decrease substantially (from 57.8 to 18 months, and from 764 to 645 subjects). Moreover, the number of observed deaths during the study can be reduced by up to 68%.

Figure S5: Operating characteristics, efficacy scenario 2 (an average of 4.8, 3.6 and 2.4 AEs per patients under toxicity scenarios 1, 2, 3) for DeEscalate. Figure S6: Operating characteristics, efficacy scenario 2 (an average of 4.8, 3.6 and 2.4 AEs per patients under toxicity scenarios 1, 2, 3) for DeEscalate.

Supplementary text, figures and tables

Supplementary Methods

Online supplement

Purpose:While various studies have highlighted the prognostic significance of pathologic complete response (pCR) after neoadjuvant chemotherapy (NAT), the impact of additional adjuvant therapy after pCR is not known.Experimental Design:PubMed was searched for studies with NAT for breast cancer and individual patient-level data was extracted for analysis using plot digitizer software. HRs, with 95% probability intervals (PI), measuring the association between pCR and overall survival (OS) or event-free survival (EFS), were estimated using Bayesian piece-wise exponential proportional hazards hierarchical models including pCR as predictor.Results:Overall, 52 of 3,209 publications met inclusion criteria, totaling 27,895 patients. Patients with a pCR after NAT had significantly better EFS (HR = 0.31; 95% PI, 0.24–0.39), particularly for triple-negative (HR = 0.18; 95% PI, 0.10–0.31) and HER2+ (HR = 0.32; 95% PI, 0.21–0.47) disease. Similarly, pCR after NAT was also associated with improved survival (HR = 0.22; 95% PI, 0.15–0.30). The association of pCR with improved EFS was similar among patients who received subsequent adjuvant chemotherapy (HR = 0.36; 95% PI, 0.19–0.67) and those without adjuvant chemotherapy (HR = 0.36; 95% PI, 0.27–0.54), with no significant difference between the two groups (P = 0.60).Conclusions:Achieving pCR following NAT is associated with significantly better EFS and OS, particularly for triple-negative and HER2+ breast cancer. The similar outcomes with or without adjuvant chemotherapy in patients who attain pCR likely reflects tumor biology and systemic clearance of micrometastatic disease, highlighting the potential of escalation/deescalation strategies in the adjuvant setting based on neoadjuvant response.See related commentary by Esserman, p. 2771

Figure S1: Operating characteristics, efficacy scenario 1 (an average of 3.2, 2.4 and 1.6 AEs per patients under toxicity scenarios 1, 2, 3) for RTOG 1016. Figure S2: Operating characteristics, efficacy scenario 2 (an average of 3.2, 2.4 and 1.6 AEs per patients under toxicity scenarios 1, 2, 3) for RTOG 1016.

Supplementary Tables

Application of MRD testing to healthy donor cfDNA samples. Patient specific panels designed for 14 different patients were pooled together and applied to cfDNA collected from four different healthy donors. Reported are the number of sites where ctDNA was detected. We required at least two sites to call a sample MRD+.

Purpose: Successful development of targeted therapy combinations for cancer patients depends on first discovering such combinations in predictive preclinical models. Stable cell lines and mouse xenograft models can have genetic and phenotypic drift and may take too long to generate to be useful as a personalized medicine tool.Experimental Design: To overcome these limitations, we have used a platform of ultra-high-throughput functional screening of primary biopsies preserving both cancer and stroma cell populations from melanoma patients to nominate such novel combinations from a library of thousands of drug combinations in a patient-specific manner within days of biopsy. In parallel, patient-derived xenograft (PDX) mouse models were created and novel combinations tested for their ability to shrink matched PDXs.Results: The screening method identifies specific drug combinations in tumor cells with patterns that are distinct from those obtained from stable cell lines. Screening results were highly specific to individual patients. For patients with matched PDX models, we confirmed that individualized novel targeted therapy combinations could inhibit tumor growth. In particular, a combination of multi-kinase and PI3K/Akt inhibitors was effective in some BRAF–wild-type melanomas, and the addition of cediranib to the BRAF inhibitor PLX4720 was effective in a PDX model with BRAF mutation.Conclusions: This proof-of-concept study demonstrates the feasibility of using primary biopsies directly for combinatorial drug discovery, complementing stable cell lines and xenografts, but with much greater speed and efficiency. This process could potentially be used in a clinical setting to rapidly identify therapeutic strategies for individual patients. Clin Cancer Res; 23(16); 4680–92. ©2017 AACR.

Screening results for drug combinations, expressed as percent of control (DMSO-treated) values.

Purpose:Deviations from proportional hazards (DPHs), which may be more prevalent in the era of precision medicine and immunotherapy, can lead to underpowered trials or misleading conclusions. We used a meta-analytic approach to estimate DPHs across cancer trials, investigate associated factors, and evaluate data-analysis approaches for future trials.Experimental Design: We searched PubMed for phase III trials in breast, lung, prostate, and colorectal cancer published in a preselected list of journals between 2014 and 2016 and extracted individual patient-level data (IPLD) from Kaplan–Meier curves. We re-analyzed IPLD to identify DPHs. Potential efficiency gains, when DPHs were present, of alternative statistical methods relative to standard log-rank based analysis were expressed as sample-size requirements for a fixed power level.Results:From 152 trials, we obtained IPLD on 129,401 patients. Among 304 Kaplan–Meier figures, 75 (24.7%) exhibited evidence of DPHs, including eight of 14 (57%) KM pairs from immunotherapy trials. Trial type [immunotherapy, odds ratio (OR), 4.29; 95% confidence interval (CI), 1.11–16.6], metastatic patient population (OR, 3.18; 95% CI, 1.26–8.05), and non-OS endpoints (OR, 3.23; 95% CI, 1.79–5.88) were associated with DPHs. In immunotherapy trials, alternative statistical approaches allowed for more efficient clinical trials with fewer patients (up to 74% reduction) relative to log-rank testing.Conclusions:DPHs were found in a notable proportion of time-to-event outcomes in published clinical trials in oncology and was more common for immunotherapy trials and non-OS endpoints. Alternative statistical methods, without proportional hazards assumptions, should be considered in the design and analysis of clinical trials when the likelihood of DPHs is high.

Digital droplet PCR (ddPCR) testing for single mutation in dilution series. (A) The observed VAFs of HapMap serial dilutions measured by using a ddPCR assay for a single mutation (with three replicates) vs. the expected VAFs; (B) The representative ddPCR raw data and fluorescence intensity thresholds for determining positive or negative droplets.

Supplementary Material Table S1: Operating characteristics. The Bayesian design conducts interim analyses for inferiority (using the probability of HR>1.2) every 1, 3, 6 or 12 months, starting from months 1, 3, 6 or 12. The far-right column reports operating characteristics for the RTOG 1016 protocol design. Table S2: Operating characteristics for a sample size N=860, which is larger than in RTOG 1016. Table S3: Operating characteristics (N=800). Table S4: Operating characteristics. The Bayesian design conducts interim analyses for inferiority every 1, 3, 6 or 12 months using the posterior probability of HR>1 for inferiority stopping. Results are based on 10,000 simulations using either a prior variance of 1/2, 1, 5 or 10. Outcome data of the control arm have been generated from reported Kaplan-Meier curves from RTOG 1016.

Technical performance of the MRD Assay. All samples from early stage breast cancer cohort. Includes plasma cfDNA and buffy coat gDNA samples. Samples from the same patient and probed with the same panel are represented independently. (A) On target fraction calculated from Picard CollectHsMetrics and using pct_selected_bases. (B) Coefficient of variation of duplex fragment depth across fingerprint sites. (C) Fold 80 base penalty score for duplex fragment depth across fingerprint sites. (D, E) Error rate comparison between conventional sequencing (requiring Q20 base quality and removing duplicate fragments) and our duplex sequencing with added filters (see methods). (F) Fraction of total duplexes after downsampling raw reads. We considered a sample to be saturated if it still contained at least 95% of its duplexes after downsampling to 80% of its raw reads.

Mutational signatures shed insight into the range of mutational processes giving rise to tumors and allow a better understanding of cancer origin. They are typically identified from high-throughput sequencing data of cancer genomes using non-negative matrix factorization (NMF), and many such techniques have been developed towards this aim. However, it is often of particular interest to compare mutational signatures across multiple conditions, e.g. to understand which signatures are present across different treatments, or to identify signatures that are shared or specific across cancer types. Existing techniques within the NMF context only allow decomposition within a single dataset, so that integrating results across multiple conditions requires running separate analyses on each dataset, followed by subjective and manual comparisons of the identified signatures. To address this issue, we propose a Bayesian multi-study NMF method that jointly decomposes multiple studies or conditions to identify signatures that are common, specific, or partially shared by any subset. We propose two models: a “discovery-only” model that estimates de novo signatures in a completely unsupervised manner, and a “recovery-discovery” model that builds informative priors from previously known signatures to both update the estimates of these signatures and identify any novel signatures. We then further extend these models to estimate the effects of sample-level covariates on the exposures to each signature, enforcing sparsity through a non-local spike-and-slab prior. We demonstrate our approach on a range of simulations, and apply our method to colorectal cancer samples to show its utility.

PURPOSE Adaptive clinical trials use algorithms to predict, during the study, patient outcomes and final study results. These predictions trigger interim decisions, such as early discontinuation of the trial, and can change the course of the study. Poor selection of the Prediction Analyses and Interim Decisions (PAID) plan in an adaptive clinical trial can have negative consequences, including the risk of exposing patients to ineffective or toxic treatments. METHODS We present an approach that leverages data sets from completed trials to evaluate and compare candidate PAIDs using interpretable validation metrics. The goal is to determine whether and how to incorporate predictions into major interim decisions in a clinical trial. Candidate PAIDs can differ in several aspects, such as the prediction models used, timing of interim analyses, and potential use of external data sets. To illustrate our approach, we considered a randomized clinical trial in glioblastoma. The study design includes interim futility analyses on the basis of the predictive probability that the final analysis, at the completion of the study, will provide significant evidence of treatment effects. We examined various PAIDs with different levels of complexity to investigate if the use of biomarkers, external data, or novel algorithms improved interim decisions in the glioblastoma clinical trial. RESULTS Validation analyses on the basis of completed trials and electronic health records support the selection of algorithms, predictive models, and other aspects of PAIDs for use in adaptive clinical trials. By contrast, PAID evaluations on the basis of arbitrarily defined ad hoc simulation scenarios, which are not tailored to previous clinical data and experience, tend to overvalue complex prediction procedures and produce poor estimates of trial operating characteristics such as power and the number of enrolled patients. CONCLUSION Validation analyses on the basis of completed trials and real world data support the selection of predictive models, interim analysis rules, and other aspects of PAIDs in future clinical trials.

BACKGROUND: Short-term fine particulate matter (PM(2.5)) exposure is positively associated with acute cardiovascular and respiratory events. Understanding whether this association varies across specific cardiovascular and respiratory conditions has important biologic, clinical, and public health implications. METHODS: We conducted a time-stratified case–crossover study of hospitalizations from 2000 through 2014 among United States Medicare beneficiaries aged 65+. The outcomes were hospitalizations with any of 57 cardiovascular and 32 respiratory discharge diagnoses. We estimated associations with two-day moving average PM(2.5) as a piecewise linear term with a knot at PM(2.5) =25 μg/m(3). We used Multi-Outcome Regression with Tree-structured Shrinkage (MOReTreeS) to identify de novo groups of related diseases such that PM(2.5) associations are: 1) similar within outcome groups; but 2) different between outcome groups. We adjusted for temperature, humidity, and individual-level characteristics. We introduce an R package, moretrees. RESULTS: Our dataset included 16,007,293 cardiovascular and 8,690,837 respiratory hospitalizations. 51 of 57 cardiovascular diseases were grouped and positively associated with PM(2.5). We observed a stronger positive association for heart failure, which formed a separate group. We observed negative associations for groups containing the outcomes other aneurysm and intracranial hemorrhage. 31 of 32 respiratory outcomes were grouped and were positively associated with PM(2.5). Influenza formed a separate group with a negative association. CONCLUSIONS: We used a new statistical approach, MOReTreeS, to uncover variation in the association between short-term PM(2.5) exposure and hospitalizations for cardiovascular and respiratory causes controlling for patient characteristics, time trends, and environmental confounders.

A common assumption of data analysis in clinical trials is that the patient population, as well as treatment effects, do not vary during the course of the study. However, when trials enroll patients over several years, this hypothesis may be violated. Ignoring variations of the outcome distributions over time, under the control and experimental treatments, can lead to biased treatment effect estimates and poor control of false positive results. We propose and compare two procedures that account for possible variations of the outcome distributions over time, to correct treatment effect estimates, and to control type-I error rates. The first procedure models trends of patient outcomes with splines. The second leverages conditional inference principles, which have been introduced to analyze randomized trials when patient prognostic profiles are unbalanced across arms. These two procedures are applicable in response-adaptive clinical trials. We illustrate the consequences of trends in the outcome distributions in response-adaptive designs and in platform trials, and investigate the proposed methods in the analysis of a glioblastoma study.