Expert Discusses How Data Informatics Is Driving Drug Development

Subha Madhavan, PhD

Subha Madhavan, PhD

Adaptive clinical trial designs are likely to become a staple of clinical research in the future, reducing both the time and cost of traditional trial development, explained Subha Madhavan, PhD, a chief data scientist at Georgetown University Medical Center.

“The single-drug, single-disease era is gone,” said Madhavan. “We as data scientists have to figure out how to utilize big data to better inform clinical studies that can help patients much more quickly.”

Reliant on real-world repositories of data, adaptive clinical trial designs are building on preexisting knowledge to further clinical development. I-SPY 2 (NCT01042379), led by Laura Esserman, MD, MBA, of the University of California, San Francisco (UCSF), is one such trial that is currently evaluating approximately 20 neoadjuvant treatments in patients with breast cancer. Unlike traditional models, adaptive trial designs enable patients to stop or switch to another arm of therapy, based on their response.

“We're still in early stages of applying novel techniques, such as machine learning and artificial intelligence to understand these data sets better,” explained Madhavan. “Once we do, we can improve how efficient our clinical trials are.”

OncLive: What is the goal of the Innovation Center for Biomedical Informatics?

In an interview with OncLive, Madhavan, who is also the director of the Innovation Center for Biomedical Informatics (ICBI) and an associate professor of oncology at Georgetown University Medical Center, discussed the role of data science and informatics in bringing novel clinical study designs to the field of oncology.Madhaven: We worked on a collaborative team with Massachusetts Institute of Technology, Amgen, and UCSF, to think about innovative ways to design clinical studies. There are many challenges to designing clinical trials. We tend to design clinical studies from first principles, be it a single drug, indication, or phase. We flipped that [approach around] and asked whether we could combine these different models of clinical trials to make it more efficient and reduce the time and cost [of traditional clinical trials]. We recently published a paper titled Pipelines on that.

Does the I-SPY trial fall into this category?

There are basket trials and there are umbrella trials. The basket trials test a certain drug across various tumor types. Umbrella trials test multiple drugs in a given tumor type across various gene mutations. A hypothesis-generating phase I study [is generally] a basket trial, which then identifies the populations that we can pass on to a phase II umbrella trial. Then, we can move on to a pragmatic phase III clinical trial. By assimilating this assembly of different study designs, we found that we could drastically reduce the amount of development time for clinical studies.Yes. I-SPY was designed and launched by Dr Laura Esserman at UCSF. It's an adaptive trial in which patients with breast cancer are randomized to different treatment arms as they come in, based on genetic testing and protein testing. The notion is that every patient who walks in the door benefits from our experience with prior patients on the study. Additionally, if patients develop comorbidities, relapse, or fail to respond to a certain [arm], we’re able to shut down the treatment arm or move them into a different treatment arm.

What are the key points to remember about these types of studies?

How do you contribute towards this goal?

[At the 2018 Ruesch Symposium on Gastrointestinal Cancers], we also talked about the LUNG-MAP study, which is an umbrella study, as well as the NCI-MATCH study. We looked at the positives and negatives for those study designs as well as how the pipeline model can help enhance these study designs.One of the key points is the use of real-world evidence. The notion with pipelines is that there is always a repository of knowledge from each one of the phases. Then, all that knowledge is built into designing the next phase for the study. You can imagine a real-world evidence repository that's constantly collecting data as we proceed. As every patient is registered in the study, there is information that's collected and registered in the real-world study. Also, once a drug is approved, there are a ton of data that have been collected through observational studies via electronic medical records. As patients are seen in hospitals and their data are entered in electronic health records (EHRs), it can be fed back into new clinical designs. One of the big concepts [in the space] is how to use a real-world evidence repository to inform these study designs.I founded the Innovation Center for Biomedical Informatics at Georgetown University with a 3-part mission in mind. Our first mission is to conduct investigator-initiated research in data science and informatics. We're constantly developing new hypotheses and developing new informatic data science approaches and testing them. The second part of the mission is education. How do we educate the next generation of data scientists to work with these complex data sets to develop novel methodologies, technologies, and apply them to real-world problems? The third part of the mission is partnerships. This symposium is a model for how multidisciplinary teams, from private to academic organizations, can come together to achieve a common goal. We do a lot of partnerships with industry and with federal agencies to apply data insights to health problems.

Our center has been operational since 2012. We work very closely with the Ruesch Center, specifically with John Marshall, MD, and his team. We've become data stewards for a larger data ecosystem. We're not just looking at the research data or clinical study data; we’re also working very closely with our health partners, such as MedStar Health System. We also work with Hackensack University Hospital in New Jersey, which is part of the Lombardi Cancer Consortium.

However, how do we conduct observational studies utilizing electronic medical record data? There is a treasure trove of information that's buried in EHRs as information gets collected during routine patient care. How do we apply that to better patient care and also improve research and novel hypothesis generation?