For almost half of Americans with hypertension, high blood pressure is a potential death sentence, and will cause nearly 700,000 deaths in 2021, according to the US Centers for Disease Control and Prevention. It also increases the risk of stroke and chronic heart failure. However, prevention and early detection, such as eating well, exercising more, and drinking less alcohol, are relatively easy to relieve, but can be difficult to treat. Physicians have many potential treatments for high blood pressure to choose from, each with their own strengths and weaknesses, making prescribing the most effective drug a challenge. Beta-blockers slow the heart, but they can cause asthma. ACE inhibitors relax blood vessels, but can cause severe coughing. Now, new artificial intelligence programs could help doctors match the right drugs to the right patients.
Developed jointly by data scientists and physicians at Boston University, this data-driven model gives clinicians real-time hypertension treatment recommendations based on patient-specific characteristics such as demographics, vital signs, past medical history, and laboratory records. intended to provide. This model, described in a recent study published in BMC Medical Informatics and Decision Making, is more effective than the current standard of care for systolic blood pressure (measured when the heart is beating rather than at rest). ) may help reduce The program’s approach to transparency could also help doctors increase confidence in artificial intelligence-generated results, the researchers said.
“This is a new machine-learning algorithm that leverages information from electronic medical records and demonstrates the power of AI in medicine,” said Yoannis Ioannis, Distinguished Professor of Engineering at BU and Director of the Institute for Computational Sciences, Rafik B. Hariri. Pascalidis said. engineering. “Our data-driven model not only predicts outcomes, but also suggests the most appropriate drug to use for each patient.”
Today, when choosing which drug to prescribe for a patient, physicians consider the patient’s medical history, treatment goals, and the benefits and risks associated with a particular drug. Often, when there are multiple options, and when none of the options are better or worse than the others, choosing which drug to prescribe is the , can be a bit like a coin toss.
In contrast, a model developed by BU uses individual patient profiles to generate custom hypertension prescriptions, providing physicians with a list of recommended drugs with associated probabilities of success. The researchers’ aim was to highlight the therapy that best controlled systolic blood pressure for each patient based on its efficacy in similar patient groups.
“Our goal is to facilitate a personalized approach to hypertension treatment based on machine learning algorithms, aiming to maximize the effectiveness of hypertension medications at the individual level,” said Paschalidis. says.
The model was developed using anonymized data of 42,752 hypertensive patients from Boston Medical Center (BMC), the BU’s primary teaching hospital, collected between 2012 and 2020. Patients were classified into similarity groups based on similarities in clinically relevant features such as past demographics. such as blood pressure records and past medical history. During the study, the model’s efficacy was compared to the current standard of care and his three other algorithms designed to predict appropriate treatment plans. The researchers found that this model achieved a 70.3% greater reduction in systolic blood pressure than standard care, and he performed 7.08% better than the second best model. The algorithm has been clinically validated, with researchers manually reviewing his 350 random his samples.
The model also showed benefits of reducing or stopping prescriptions for some patients on multiple medications. The algorithm suggests to doctors multiple optimal treatments, according to the researchers, in a situation known as clinical equilibrium, a situation in which the medical community is divided over the efficacy of one drug versus another. , which may provide valuable insight.
“These advanced predictive analytics have the ability to enhance clinician decision-making and positively impact the quality of care we provide and, in turn, patient outcomes,” said a former professor at BU. Rebecca Mishlis, who recently became the director of Mass, said: General Brigham’s Chief Medical Information Officer. “This is an important first step to show that these models actually outperform standard care and could help us become better physicians.”
While many recognize that machine learning’s ability to process large amounts of data and uncover patterns and correlations can be useful in medicine, the results can be difficult to interpret and often challenge artificial intelligence. Its adoption has been limited due to reasons such as low trust levels. Until now, machine learning in healthcare has been hampered by incomplete or inaccurate data and sparse patient histories that can skew predictive results. An important aspect of this study was to ensure data transparency so that clinicians, especially those without technical expertise, could understand how the algorithm works and how the model makes specific therapeutic recommendations. It was to give me a clear understanding of how and why I propose.
“Using data from a diverse patient population at Boston Medical Center, this model provides tailored care to an underrepresented patient population and provides individualized recommendations to improve outcomes for these patients.” It provides an opportunity to do so,” said Nicholas J. Cordera, BU, Chovanian Avedisian School of Medicine. Assistant Professor and BMC Medical Director in charge of Quality and Patient Safety. “Such personalized medicine and models offer opportunities to better serve populations that are not necessarily well represented in national studies or considered when developing guidelines.”
This study was funded by the National Science Foundation and was previously published with William G. Adams, Professor of Pediatrics, BU Chovanian & Avedisian School of Medicine and Director of the Clinical Research Informatics Program at the BU Institute of Clinical and Translational Sciences. Based on collaborative research.
