MCID

MCID predicts the probability that a patient undergoing total knee arthroplasty (TKA) will achieve the minimal clinically important difference (MCID) in the Knee Injury and Osteoarthritis Outcome Score–Physical Function Short Form (KOOS-PS) at one year using machine learning.

Key Features:

• Machine learning algorithms: Model development used five distinct machine learning algorithms, with elastic-net penalized logistic regression identified as the best-performing model based on discrimination (c-statistic), calibration, Brier score, and decision curve analysis.
• Predictive variables: Predictors include demographics, preoperative patient-reported outcome measures (PROMs), preoperative KOOS-PS scores, Visual Analog Scale (VAS) Pain scores, preoperative opioid use, and the PROMIS global mental health score.
• Outcome definition: The target outcome is achievement of MCID in KOOS-PS at one year post–TKA, where MCID was calculated using a distribution-based method.
• Dataset characteristics: Development and analysis used a retrospective cohort of 744 primary TKA patients treated between 2016 and 2018, with 72.8% achieving MCID at one year.

Scientific Applications:

• Preoperative risk stratification: Predicts individual probability of achieving KOOS-PS MCID to inform preoperative assessments.
• Expectation management: Provides probabilistic outcome information to aid in setting patient expectations regarding functional improvement after TKA.
• Resource optimization: Identifies patients more or less likely to attain clinically meaningful improvement to inform allocation of perioperative resources and intervention planning.

Methodology:

Retrospective review of primary TKA patients (2016–2018); MCID calculated by a distribution-based method; predictive models developed using five machine learning algorithms including elastic-net penalized logistic regression and evaluated with c-statistic, calibration, Brier score, and decision curve analysis.