Clinical Data Visualizations at BMS Merge Trial and Real-World Data

October 26, 2020

By Deborah Borfitz

October 26, 2020 | The probability of success is low in early drug discovery and gets progressively more expensive on the journey to marketing approval by the Food and Drug Administration (FDA). To “fail early and fail often” has thus become the mantra of medicine developers needing to direct resources to the candidates most likely to triumph, says Philip Ross, director of translational bioinformatics data science at Bristol-Myers Squibb (BMS). Only 9.6% of drugs that make it to phase I trials ever get FDA approved.

Ross spoke on the topic of clinical data visualizations to drive clinical and biomarker exploration using both clinical trial and real-world data (RWD) at the recent Bio-IT World Conference & Expo Virtual. He is developer of a Spotfire application supporting near-real-time visualizations of clinical data for all ongoing clinical trials at BMS.

To maximize the impact of biomarker data from clinical trials, BMS accelerates the availability of biomarker assay results and clinical data to translational researchers, Ross says. The company also leverages “standardized programs that merge, transform, and derive data for key analysis endpoints.”

Exploratory visualizations get refreshed automatically when data arrives, Ross continues, and all steps from receipt of data to visualizations are automated. The delivery time from data receipt to exploratory analysis and review in Spotfire is typically less than a day.

With the recent acquisition of Celgene, BMS is now in the midst of growing a new IPS ecosystem from two different pharma systems with disparate workflows, databases and S3 buckets by “choosing the most effective components of each,” says Ross. That ecosystem will include a Gen3 Data Commons layer to “democratize” all the data—from omics, biomarkers, CAR T, reporting outputs and other RWD—and make it available to search and provide a way to effectively leverage the information in a variety of analysis and visualization environments.

Ross included external collaborations, chemical structure optimization, predictive drug substance design, reproducible research, and translational epidemiological environments. The idea is to optimize the environments for the particular type of work being done but also to enable better communication across those environments.

His Knowledge Science Research (KSR) team can pull clinical trial data from a variety of points in its life cycle, notes Ross, be it stable views from a clinical data warehouse or statistically analyzed data from various interim and final locks. Extracted clinical variables can be leveraged for formal or exploratory analysis. In the latter case, biomarker data from a variety of sources can also be used, potentially from database lock or source statistical systems as well as the Sage database where BMS houses its immunohistochemistry, flow cytometry, and cytokine data. Manifest and result files get extracted in computer-readable format and merged with clinical variables for further analysis and to create visualizations for nearby teams.

“This is a very mature effort at this point,” says Ross. The work includes manual provisioning of formal analysis data, automated clinical data workflow, automated clinical biomarker workflow, automated worldwide safety workflows, and manual integrated study workflows—all having different levels of accessibility, depending on the appropriate access to that data.


Impactful Analyses

In concert with the clinical organization at BMS, the KSR team automated review for 192 studies “to allow one to look at a variety of aspects of the blinded clinical data.” For a view of efficacy, the alternative approach would be an exploratory Spotfire build “so we don’t have to worry about the timing of the statistical variation of efficacy,” says Ross. His cited example showed a treatment timeline, percent change in baseline for tumors, size of tumors over time, and the emergence of non-target, non-measurable lesions.

Automated Spotfire builds also provide customized views of biomarkers, “just for a particular type of visual exploration that is most useful for the biomarker data and indication being studied with it in a given trial,” says Ross. “This is being used for variety of clinical and biomarker analyses questions and to prepare for future biomarker study design.”

For purposes of worldwide safety, his team has analyzed data from 150 clinical trials involving 50,000 treated patients, which has proven “very useful” for evaluating safety broadly across therapeutic programs based on characteristics such as therapeutic area, education, and treatment type. 

So far in 2020, the KSR team has delivered analysis-ready data for more than 200 clinical trials, 107 clinical trial database locks, and updated 529 Spotfire files, reports Ross. Of those, 192 visualizations incorporate blinded data and are visible to the broader clinical team; 117 look at tabular biomarker data combined with safety and efficacy data for translational scientists; 128 contain unblinded safety and efficacy data for specific clinical teams; and 92 contain clinical safety program data for worldwide safety. All told, 236 users have accessed and explored the Spotfire builds.

“We know we are developing things that are actually having impact,” says Ross. Accelerating the availability of clinical, biomarker and RWD, and reusing analysis data, maximizes the effects.

The Gen3 Data Commons layer now being utilized was released by the University of Chicago in 2019, notes Ross. In addition, BMS is tapping a variety of modern software approaches coming to maturity. Among those specifically mentioned were the “microservices revolution,” accelerated software delivery through the adoption of agile, lean practices, and adoption of open-source tools.

Within BMS, a research and early development (R&ED) project called Helix seeks to integrate all systems used in the R&ED data and analysis environment, he says. This is in addition to the IPS ecosystem that aligns around the “best and brightest” from BMS and Celgene legacy systems for maximizing work efficiency. “We expect this will facilitate our ability to bring clinical and biomarker data together with real-world data.”

Ross recapped findings of a published study on how BMS brought together in-house clinical trials and RWD around tumor mutational burden (TMB). Some tumors have a higher rate of mutations, which may have a higher neoantigen load and possibly a greater response to certain therapeutic efforts. A literature review found that the number of mutations in various tumors correlated with the overall response rate of those same tumors across multiple clinical trials.

The expectation is that there is a relationship between TMB and treatment, and prior studies have shown pretty poor survival over time in patients with high TMB receiving chemotherapy. BMS’s Nivolumab (targeting tumors expressing PD-1 and PD-L1) was subsequently shown to increase the number of patients who survive for a longer period of time despite having a higher TMB, Ross says. Using RWD, BMS was able to show similar results among high TMB patients in the real world who were receiving PD-1 and PD-L1 targeting therapy.

Moving forward, says Ross, BMS intends to enhance interactions currently happening between data wranglers, bioinformaticians and RWD experts. The company also plans to “improve the ability of scientists to locate and analyze exploratory data and visualizations, analysis results and reports, and clinical, nonclinical and real-world data.”