Biopsychosocial based machine learning models predict patient improvement after total knee arthroplasty
Total knee arthroplasty (TKA) is an effective treatment for end stage osteoarthritis. However, biopsychosocial features are not routinely considered in TKA clinical decision-making, despite increasing evidence to support their role in patient recovery. We have developed a more holistic model of patient care by using machine learning and Bayesian inference methods to build patient-centred predictive models, enhanced by a comprehensive battery of biopsychosocial features. Data from 863 patients with TKA (mean age 68 years (SD 8), 50% women), identified between 2019 and 2022 from four hospitals in NSW, Australia, was included in model development. Predictive models for improvement in patient quality-of-life and knee symptomology at three months post-TKA were developed, as measured by a change in the Short Form-12 Physical Composite Score (PCS) or Western Ontario and McMasters Universities Osteoarthritis Index (WOMAC), respectively. Retained predictive variables in the quality-of-life model included pre-surgery PCS, knee symptomology, nutrition, alcohol consumption, employment, committed action, pain improvement expectation, pain in other places, and hand grip strength. Retained variables for the knee symptomology model were comparable, but also included pre-surgery WOMAC, pain catastrophizing, and exhaustion. Bayesian machine learning methods generated predictive distributions, enabling outcomes and uncertainty to be determined on an individual basis to further inform decision-making.
Keywords: Bayesian inference; Biopsychosocial; Machine learning; Prediction algorithm; Total knee arthroplasty.
© 2025. The Author(s).
Full article available here.