Machine Learning-Driven Probability Scoring Enhances Diagnostic Certainty and Reduces Costs in Suspected Periprosthetic Joint Infection

Background: Accurate diagnosis of periprosthetic joint infection (PJI) remains challenging, particularly in culture-negative and borderline cases where current practices lead to high diagnostic uncertainty. SynTuition™, a machine-learning-based probability score integrating preoperative biomarkers, was developed to support clinical decision-making. This study compared its diagnostic performance and economic impact with standard physician practice. Methods: A total of 12 physicians provided diagnoses of 274 clinical vignettes representing suspected PJI cases. SynTuition probabilities were converted to binary diagnostic classifications using a validated threshold. Diagnostic accuracy, agreement, indecision rates, decision curve analysis, and misdiagnosis-related costs were evaluated. Results: SynTuition achieved an overall percent agreement of 96.0% when compared against the expert adjudicated clinical reference, outperforming the pooled physician group at 90.8%. Physicians showed high indecision (38–48%) in inconclusive 2018 ICM cases, whereas SynTuition generated a definitive diagnosis with an 86.7% agreement against expert adjudication. Decision curve analysis demonstrated a higher net benefit for SynTuition across a broad range of thresholds, reducing projected unnecessary revision by up to 5.8%. Economic modeling showed a reduction in misdiagnosis-related costs from $6.9 million to $2.9 million per 1000 suspected PJI cases, yielding estimated savings of $4000 per suspected case. Conclusions: SynTuition demonstrated high diagnostic accuracy, lower uncertainty, and significant clinical and economic advantages over routine physician practice, supporting its integration into clinical decision-making for suspected PJI, particularly in diagnostically ambiguous cases.