LABORATORY-BASED MACHINE LEARNING ALGORITHM REDUCES ECONOMIC BURDEN BY ENHANCING DIAGNOSTIC ACCURACY AND CLARITY IN PERIPROSTHETIC JOINT INFECTION
Author(s)
Van Thai-Paquette, MS1, James Parr, PhD2, Amy Worden, BS, MPH3, Krista Toler, MS4.
1Research Principal Engineer II, Zimmer Biomet, Claymont, DE, USA, 2Machine Learning Lead Engineer, Zimmer Biomet, Swindon, United Kingdom, 3Global Health Economics and Market Access Manager, Zimmer Biomet, Centennial, CO, USA, 4Research & Development Assoc Director, Zimmer Biomet, Warsaw, IN, USA.
1Research Principal Engineer II, Zimmer Biomet, Claymont, DE, USA, 2Machine Learning Lead Engineer, Zimmer Biomet, Swindon, United Kingdom, 3Global Health Economics and Market Access Manager, Zimmer Biomet, Centennial, CO, USA, 4Research & Development Assoc Director, Zimmer Biomet, Warsaw, IN, USA.
OBJECTIVES: Periprosthetic joint infection (PJI) diagnosis remains complex, with erroneous application plus implementation challenges of current standard of care (SOC) methods. Diagnostic uncertainty often defaults to unnecessary two-stage revision surgeries and antibiotic overuse. This analysis assessed the potential economic impact of integrating a PJI Score Model (“Model”), which utilizes machine learning (ML) to generate a PJI probability score from synovial fluid, into the diagnostic pathway from a U.S. payer perspective.
METHODS: A decision-analytic model simulated 1,000 suspected PJI evaluations. Potential misdiagnosis costs were compared between a physician panel and the Model using diagnostic performance against SOC 2018 International Consensus Meeting (ICM) criteria, with inconclusive cases adjudicated by expert physicians. Based on previous studies, inconclusive rate and diagnostic accuracy was 23% and 92% by physicians and 0.4% and 97% by the Model. Disease prevalence was 17%. Cost of a false-negative result was estimated at $55,000 (blended rate incorporating physical therapy and repeat diagnostic testing (56%), or aseptic revision followed by a 2-stage revision for infection (44%)). Cost of a false positive result was estimated at $75,000 for 2-stage revision for infection.
RESULTS: The Model demonstrated greater ability to accurately differentiate diagnostically ambiguous results compared to physicians. Physicians predominantly defaulted to a false-positive diagnosis in inconclusive cases, leading to lower overall accuracy. In a 1,000-patient suspected PJI cohort, compared to physicians using SOC, the Model reduced potential errors from 82 to 36 cases, thus reducing net costs by 46%, from over $6M to approximately $3M.
CONCLUSIONS: In this study, implementing an ML model to aid in the diagnosis of suspected PJI reduced the rate of false results, particularly false-positive results, which can lead to unnecessary two-stage revision surgery and high-dose systemic antibiotics. Overall costs were reduced by nearly half, amounting to an estimated net savings of $2,793 per suspected PJI.
METHODS: A decision-analytic model simulated 1,000 suspected PJI evaluations. Potential misdiagnosis costs were compared between a physician panel and the Model using diagnostic performance against SOC 2018 International Consensus Meeting (ICM) criteria, with inconclusive cases adjudicated by expert physicians. Based on previous studies, inconclusive rate and diagnostic accuracy was 23% and 92% by physicians and 0.4% and 97% by the Model. Disease prevalence was 17%. Cost of a false-negative result was estimated at $55,000 (blended rate incorporating physical therapy and repeat diagnostic testing (56%), or aseptic revision followed by a 2-stage revision for infection (44%)). Cost of a false positive result was estimated at $75,000 for 2-stage revision for infection.
RESULTS: The Model demonstrated greater ability to accurately differentiate diagnostically ambiguous results compared to physicians. Physicians predominantly defaulted to a false-positive diagnosis in inconclusive cases, leading to lower overall accuracy. In a 1,000-patient suspected PJI cohort, compared to physicians using SOC, the Model reduced potential errors from 82 to 36 cases, thus reducing net costs by 46%, from over $6M to approximately $3M.
CONCLUSIONS: In this study, implementing an ML model to aid in the diagnosis of suspected PJI reduced the rate of false results, particularly false-positive results, which can lead to unnecessary two-stage revision surgery and high-dose systemic antibiotics. Overall costs were reduced by nearly half, amounting to an estimated net savings of $2,793 per suspected PJI.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EE18
Topic
Economic Evaluation
Topic Subcategory
Budget Impact Analysis, Value of Information
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Infectious Disease (non-vaccine), SDC: Injury & Trauma