Yale researchers have developed a machine learning tool that can guide personalized antihypertensive treatment
Courtesy of Evangelos Oikonomou
Researchers have developed a new tool that uses machine learning to help treat high blood pressure.
Hypertension is the most prominent modifiable risk factor for major cardiovascular events. According to the National Center of Health Statistics, part of the Centers for Disease and Control, it affects around 47 percent of adults in the United States. have high blood pressure or are taking an antihypertensive drug.
Evangelos K. Oikonomou, Yale Cardiology Fellow, Erica S. Spatz, Yale Associate Professor of Cardiology, and Rohan Khera, Yale Assistant Professor of Cardiology and Director of the Cardiovascular Data Science Lab, have developed a machine learning tool using data from two pioneering clinical trials can guide personalized antihypertensive treatment.
“At a time when multiple therapies are emerging, the ability to make personalized treatment decisions using data-driven approaches is essential to achieving accurate healthcare,” Khera said. “It is also important to fully leverage the contributions of participants in randomized clinical trials by learning information beyond population-level findings.”
Despite the availability of highly effective antihypertensive drugs, there has been debate in the medical field about the level at which measures should be taken to reduce systolic blood pressure.
To address this, two clinical trials, the Systolic Blood Pressure Intervention Trial, or SPRINT, and Action to Control Cardiovascular Risk in Diabetes Blood Pressure, or ACCORD, evaluated the cardiovascular benefit of having a target systolic blood pressure of <120 mmHg versus <140 mmHg in patients with and without type 2 diabetes mellitus.
The Yale team used the individualized participant data from these studies as the basis for their machine learning tool: Pressure Control in Hypertension, or PRECISION.
The team’s method uses computational representations to generate the results of the study for each individual in a test population, with patients with similar characteristics appearing closer together in the resulting simulated graph, or “space.” Then, to find the patient profiles that would benefit most from the intervention, the team conducts an analysis focused on each participant’s space.
“We can then use this graph or plot to compare outcomes in patients assigned to an intensive versus a standard blood pressure lowering strategy,” Oikonomou said.
This approach allows the team to monitor the effects of treatments at an individual and personalized level, as opposed to that of an entire population.
Khera emphasized the power of PRECISION’s framework and the value of machine learning. In addition, he stressed that artificial intelligence-assisted clinical trials would reduce trial costs as well as the time to access effective treatments.
“Our methods are innovative, but their greatest strength is the conceptual framework,” said Khera. “We hypothesized that the development of machine learning-based methods that can infer responses to therapies from different populations participating in studies. We have now proven the value of this approach in three different areas.”
The team hopes that PRECISION can help advance the development of safer clinical trials.
“First, we would like to emphasize that the proposed algorithm PRECISION is to be used for research purposes only until we prospectively validate its clinical usefulness,” said Oikonomou. “Second, we believe that our broader approach can be used to define a more personalized interpretation of clinical trials on diagnostic and therapeutic interventions. Finally, we are currently exploring the value of our technology to design smarter clinical trials that are more efficient and safer.”
Marc A. Suchard, a professor of biostatistics at the University of California at Los Angeles and a co-author of the study, agreed.
“We hope to be able to apply what we’ve learned about individualized treatment to a growing number of clinical areas,” added Suchard.
The study was presented in the journal Lancet Digital Health.