Using advanced artificial intelligence algorithms, a team from Harvard and Brigham Women’s pulled data from more than 7,000 hip arthroplasty patients and found that 3 factors, above all, are the highest predictors of early total hip arthroplasty (THA) revision surgery.
The resulting study, “The Utility of Machine Learning Algorithms for the Prediction of Early Revision Surgery After Primary Total Hip Arthroplasty,” appears in the June 1, 2022, edition of the Journal of the American Association of Orthopaedic Surgeons.
The researchers collected data from 7,397 consecutive primary total hip arthroplasty patients and found that 566 patients (6.6%) returned for early (less than two years from their primary hip surgery) revision surgery.
The team pulled patient demographics, implant characteristics, and other surgical variable data to feed into six machine learning algorithms and asked the machines to predict early THA revision risk.
Co-author Young-Min Kwon, M.D., Ph.D., vice chair of the Department of Orthopaedic Surgery at MGH and professor at Harvard Medical School explained what machine learning algorithms are and how they were used to predict which patients are most likely to return within two years for a revision surgery, “Machine learning algorithms are procedures used to predict output values from given data. In general, they can be categorized as either supervised or unsupervised. Supervised algorithms require input data and desired outcomes to be labeled, whereas unsupervised algorithms do not require labeling. The algorithms are termed machine ‘learning’ since they adjust their parameters over time (using mathematics and logic) as they are exposed to larger volumes of data.”
“The result of a trained machine learning algorithm is a machine learning model which consists of a prediction algorithm to predict the desired outcome variable,” added Dr. Kwon. “Their biggest advantage over traditional statistical techniques is their ability to accurately predict outcomes, even with incomplete or noisy data.”
When OTW asked for details on the model development process, Dr. Kwon explained: “We chose six models for development (artificial neural network, stochastic gradient boosting, support vector machine, k-nearest neighbor, random forest, and elastic-net penalized logistic regression) based on previous literature highlighting their utility in predicting clinical outcomes in patients undergoing total joint arthroplasty.”
“After cleaning to exclude missing data and outliers, we extracted all correlated features from the dataset. Subsequently, we performed recursive feature elimination using a random forest algorithm to select the final predictors for model development.”
“With over 1.2 million total hip arthroplasties (THA) expected to be performed annually by 2030, the rate of revision arthroplasties is also projected to increase. Of these, as high as 6% are expected to be early revision surgeries (within two years of index THA) with a multifactorial etiology. As revision THAs are more surgically challenging and associated with increased morbidity, mortality, and healthcare costs, preoperative identification of patients at risk for early revision THA may allow for reduction of future revision THA.”
What Were the Three Highest Predictors of Revision THA?
The team determined that the following were significantly associated with early revision surgery after primary THA:
- male sex,
- age <60 years,
- BMI >.35 kg/m2,
- depression,
- diabetes,
- renal failure,
- Charlson Comorbidity Index, and
- nonprivate (Medicare/Medicaid) insurance.
The strongest predictors for early revision after primary THA were Charlson Comorbidity Index, a body mass index of .35 kg/m2, and depression.
“Our study highlights the potential of six machine learning models to assist surgeons in a clinical setting through preoperative quantification of the patient-specific risk of early revision THA,” stated Dr. Kwon to OTW. “This, in turn, may allow surgeons to optimize at-risk patients better preoperatively and mitigate their risk of undergoing early revision THA.”
