Randomized clinical trials are often called the “gold standard” of research and have been the centerpiece of evidence-based medicine for nearly a century. In randomized clinical trials, researchers randomly assign patients to treatment or control groups. Researchers then determine if the treatment provides more benefit than the control. Those results then inform clinical guidelines and the care patients receive worldwide.
However, not all patients respond the same way in a clinical trial. Some subgroups of patients may benefit more, while others may experience poorer results than the average. Researchers refer to this variation from the average treatment effect as the heterogeneity of treatment effect (HTE).
“There is much more nuance to clinical trial results, but until now, we have not had the technology to systematically unlock this individualized information,” says Rohan Khera, MD, MS, assistant professor of medicine (cardiovascular medicine), assistant professor of biostatistics (health informatics), and director of the Cardiovascular Data Science (CarDS) Lab. “We now have the opportunity to individualize the interpretation of clinical trial results to provide much more personalized information for patients and the people who care for them.”
Khera aspires to completely innovate the evidence-generation process by applying machine learning to clinical trials. He hopes to learn more about how different types of patients respond to specific interventions. He aims to develop tools to identify individualized treatment effects in trial results and help physicians, health systems, and other stakeholders translate those results to a real-world population.
“We want to create tools that will help physicians decide if a treatment would benefit the patient in front of them and to help health systems and other stakeholders determine if their patient population would benefit from a specific intervention,” said Khera.
The National Institutes of Health (NIH) recently granted Khera a $3.8 million R01 award for his project, “Translating Personalized Inference from Randomized Clinical Trials to Real-World Cardiovascular Care.” This is the first systematic study to examine potential approaches to extract individualized information from randomized clinical trials.
“We know how important it is to move to personalized cardiovascular care so that we can improve individual patient outcomes and reduce disparities in care. Rohan’s exciting work will help get us closer to more individualized cardiovascular care,” said Eric Velazquez, MD, Robert W. Berliner Professor of Medicine and chief of Yale Cardiovascular Medicine. “This project will also help the medical community to get more out of our previous investments in clinical research to more effectively inform care of our patients, hopefully speeding the pace of our progress.”
The first aim of the study is to empirically evaluate several machine-learning approaches for detecting heterogeneous treatment effects. To evaluate models, Khera and his colleagues will use “digital twins” to compare participant-level data from five diverse NIH-funded randomized clinical trial datasets to populations in electronic health records (EHRs). The team is testing which method performs best, where they fail, thresholds of detection, and other performance metrics.
“We want to build trust in the machine-learning approaches so that health systems, clinical researchers, and other stakeholders trust the recommendations and understand how their patient populations will respond to certain treatments,” said Khera. “This trust is essential to translate these precision care approaches more seamlessly.”
The study will examine clinical trials in all aspects of cardiovascular care, including prevention, management within the health care system, and emergency care.
“That's the best part of what we are trying to do,” said Khera. “We are not restricting ourselves to just one cardiac condition. We want to create a framework that can be used for all clinical trials, not just those focusing on specific conditions.”
The study also aims to address technical barriers to achieving this vision. Khera and his colleagues are researching how to address missing data. They are investigating newer methods of making the models and tools resilient to the missingness of data. They also intend to better understand how to align data captured in different structural formats in medical records.
This research builds on foundational studies from Khera, Evangelos Oikonomou, MD, DPhil, clinical fellow (cardiovascular medicine), and other colleagues, with recent work published in The Lancet Digital Health, Diabetes Care and European Heart Journal.
Khera says this work is only possible at an institution such as Yale.
“The pool of people I get to work with is the most exciting thing ever,” said Khera. “Yale attracts the brightest individuals worldwide to participate in our programs, from undergraduates to graduate school, residency, and fellowship. I’ve found a lot of camaraderie working with everyone to try to figure out some of the most challenging problems.”
The Department of Internal Medicine at Yale School of Medicine is among the nation's premier departments, bringing together an elite cadre of clinicians, investigators, educators, and staff in one of the world's top medical schools. To learn more, visit Internal Medicine.