When Kathryn Doornbos finished her PhD and entered the biopharma world, she asked herself a question that stuck—why was it so hard for clinical sites to identify the right patients for the right trials at the right time? Her intuition was correct, and there is plenty of evidence to back it up. It’s the very reason why one out of every five oncology clinical trials will fail due to insufficient accrual.
“These new drugs that we all want to see make it to market are sometimes dependent on first-line clinical trials and a really specific selection of patients,” Doornbos told Inside Precision Medicine.
The party that suffers the most as a result of this shortcoming, according to Doornbos, isn’t the drug developer or the provider—it’s the patient.
“I believe that patients deserve clinical research as a care option,” Doornbos told Inside Precision Medicine. “I think oncology patients, in particular, get the best care when they’re part of a clinical trial, whether in a community or an academic medical center.”
So, when Doornbos came across Realyze Intelligence, a startup developing software to analyze patient data to identify and match patient populations to clinical trials, she hopped on board.
Assumptions, biases, and blind spots hindering research-as-care
According to Doornbos, many challenges stand in the way of optimal patient selection. These span from assumptions, biases, and blind spots to the cumbersome and time-consuming nature of working with patient data.
Some problems arise from the often-drawn imaginary line used to determine the extent to which a patient is willing to participate in a clinical trial. If a patient lives 10, 20, or 50 miles or more away from where they receive care, it may be assumed that they do not want to participate in a clinical trial due to the additional appointments and effort required.
Relatedly, there are sometimes assumptions about whether a patient will participate in a clinical trial based on limitations potentially imposed by socioeconomic status as well as their health insurance plans—just because someone is not on a premier health care plan or is uninsured doesn’t mean they don’t want to participate in clinical trials.
Doornbos stated, “We need to let patients decide that because a clinical trial affects their life and the trajectory of their treatment—I don’t think that anyone should be making that decision for a patient.”
Additionally, clinicians face the insurmountable challenge of matching each patient in need with the most appropriate clinical trial. The constant emergence of new clinical trials makes it impossible for any clinician to keep tabs on all the ones in their field. Doornbos added that it is not out of the question for doctors to favor recruiting patients for clinical trials headed by people they know rather than introducing them to an unfamiliar principal investigator.
Furthermore, doctors and other medical professionals do not have time to sift through patients’ medical records for hours on end.
“The burden of effort required to keep up with the level of precision information makes it impossible for a single human, or even teams of humans, to do all of the work. It’s unfathomable to think that you would go through 300 patient records for a patient. That seems like such a waste of human time.”
When it comes down to it, clinical trial matching is completely futile if the mountains of data stored in patients’ EHRs cannot be retrieved, processed, and understood to satisfy the rigorous inclusion/exclusion criteria of these studies.
Patient selection with human-in-the-loop machine learning
In the past few months, Realyze Intelligence’s platform has helped researchers at Memorial Sloan Kettering Cancer Center (MSKCC) augment the manual curation of cancer data elements and enabled a group of researchers from the UPMC Hillman Cancer Center, the Institute for Precision Medicine, and the University of Pittsburgh to generate real-world-data-led insights improve breast cancer patient care.
The startup’s software is based on “human-in-the-loop machine learning,” a collaborative approach that incorporates human input and expertise throughout the lifecycle of machine learning and artificial intelligence systems. Doornbos assures that this approach will not put anyone out of a job but rather supercharge their capabilities by drastically reducing the amount of tedious manual chart review, hunting, and pecking through clinical notes for some nuance.
“We’re not replacing human decision-making,” said Doornbos. “We can’t just turn everything over, both because we don’t know what’s really going to happen and also because that human element is so important—particularly with anything related to patients.”
With human efficiency in mind, Realyze Intelligence’s software dashboard is meant to be simple and easy to use. But, just as important, if not more, is that there is no “black box” regarding the software. There is complete transparency, and the system does not decide whether a patient should consent to or enroll in a trial.
The whole point of Realyze’s software is to surface the critical information so that clinical research specialists have it right at their fingertips, can make an assessment themselves, can interact with the system summarized very quickly, and then do what is a critical human task of informed consent and coordinating with the clinician to find all the human reasons to help that patient get on the proper trial.
“We respect the sacred space with a clinical research specialist, a PI, or a treating physician in that conversation with a patient that needs to happen about research as care,” said Doornbos. “We’re not trying to automate any of that. We’re just trying to empower everyone to have the most information quickly so that if that patient is a good match, they’re not overlooked, and they don’t get another cycle of therapy that disqualifies them or anything like that before they have the opportunity to know that that’s an option.”
The domino effect
Matching patients to clinical trials and providing them with the option to participate in clinical research is essential to not only enable care decisions, empower healthcare providers, and improve patient outcomes but also to address socioeconomic inequity, which has significant consequences for how beneficial a drug is across populations.
“You get a subset [of patients], so when you take that drug out into the wild, it does not perform as expected,” said Doornbos. “We’re providing a tool that helps sites take a look in the mirror and also provide future-proof research as care opportunities for their patients.”
At the end of the day, the greater the diversity captured by patient cohorts, the greater the potential reach of medicine.
Doornbos said, “When you start getting some patients to get treatment by being part of clinical trials, that helps other future patients and so forth. Then everything moves faster. It’s already moving fast, but the faster clinical trials enroll, and these great new therapies are proven, the faster all the other patients can benefit.”
Healthcare, in some ways, resembles a massive Rube-Goldberg chain reaction or an intricate domino run—when one piece is missing, everything else fails. The efforts of Doornbos and Realyze Intelligence have brought us one step closer to completing healthcare’s missing components and allowing precision medicine to reach its full potential.
