Artificial intelligence may help orthopedic surgeons better predict which patients will do well after joint replacement procedures and which are less likely to experience benefits.

In a new study, researchers at Hospital for Special Surgery (HSS) in New York City show that machine-learning algorithms can predict with reasonable accuracy which patients undergoing total knee or total hip replacement will report a minimally clinically important difference (MCID) in symptoms two years after the operation. “Machine learning has the potential to improve clinical decision-making and patient care by helping prioritize resources for postsurgical monitoring and informing presurgical discussions of likely outcomes” after total joint replacement, they reported in the June 2019 issue of Clinical Orthopaedics and Related Research.

“The least valuable health care is that which is not wanted or needed,” said Catherine MacLean, MD, PhD, HSS Chief Value Medical Officer and senior author of the study. “Accurate prediction of whether individual patients will achieve a meaningful improvement after a procedure will greatly assist patients and their physicians in determining the best course of therapy – and in avoiding those that are unlikely to work.”

Mark Fontana, PhD, Senior Director of Data Science at HSS and lead author of the new study, said this application of machine learning fits with the hospital’s dedication to patient-centered medicine. “We have an interest in understanding patient-reported outcomes; you can think of them as another vital sign. Pain and function are subjective, so asking patients themselves how they’re doing is necessary. For patients considering surgery, perhaps even more important, is understanding whether surgery is likely to improve their pain and function and get them back to the activities they love most – which is always our overarching goal at HSS,” Fontana said.

For the study, Fontana and his colleagues used data from 7,239 hip and 6,480 knee replacement cases at HSS conducted between 2007 and 2012. Using data about both physical and mental status of patients before and two years after the procedures, the investigators were able to calculate whether a patient achieved an MCID across four patient reported outcome measure (PROM) scores: one for general physical health score, one for general mental health, one for hip health, and one for knee health.

The goal of the study was to construct models that could accurately predict which patients were likely to experience meaningful gains in the various PROMs after surgery, taking into account a host of variables, including a person’s demographics (e.g., age, body mass index, where they lived), medical history (e.g., whether they had a previous joint surgery), and type of insurance.

“We’re predicting whether the improvement between the baseline and two-year score is big enough to be meaningful,” Fontana said.

In follow-up work, HSS investigators are creating printed and digital decision aids that surgeons can use to quickly ascertain the likelihood a patient would achieve an MCID based on their particular characteristics.

“In the future, surgeons will be able to say, ‘This is your baseline score, these are your other characteristics. An improvement of X points is one that patients think is meaningful, and we think the likelihood of you achieving that is Y percent,’” he said. “It’s that percentage that you can imagine being especially helpful to someone considering surgery.”

In addition to the article, the journal published a question-and-answer interview with Drs. MacLean and Fontana, as well as an editorial about the paper.