External Validation of the Mayo Prosthetic Joint Infection Risk Scoring System for Total Ankle Arthroplasty

Research Type: Level 3 - Retrospective cohort study, Case-control study, Meta-analysis of Level 3 studies Introduction/Purpose: The Mayo Prosthetic Joint Infection (PJI) risk scoring system has been used to predict the risk of PJI after arthroplasty surgery on various joints. While there have been many reports on the accuracy of this algorithm, its applicability and accuracy pertaining to total ankle arthroplasty (TAA) and its patient demographic remains unclear. In this study, we aimed to assess the accuracy of the Mayo PJI risk scoring system on patients who underwent TAA surgery in three Boston referral hospitals. Methods: We conducted a retrospective analysis of 66 patients who underwent TAA at any of three main referral hospitals in Boston. PJI was detected by reviewing patients’ profiles, which were confirmed by an expert clinician and based on CDC criteria. The baseline Mayo PJI risk score was calculated using predefined criteria including BMI, prior ankle surgery, prior TAA, immunosuppression, ASA score, and procedure time. The performance of the scoring system was depicted via area under the ROC curve (AUROC), Youden’s index, accuracy, sensitivity, specificity, and positive and negative predictive values. Results: From a cohort of 66 patients, 6 developed PJI after TAA based on CDC criteria. The mean age of our patient population was 68.3±9.7 with 64% males and 36% females. We used Youden’s J index as the main criterion to select the highest accuracy for the threshold of the baseline risk score and selected “Mayo score 0” as the most accurate cut-off value. AUROC showed 0.90, and the accuracy was 0.78. Sensitivity was calculated at 0.83, specificity of 0.78, a PPV of 0.28, an NPV of 0.98, and Youden’s J index of 0.71 (Table 1). Conclusion: Despite the low sample size, our preliminary outcomes showed that the Mayo PJI Risk method had an acceptable accuracy for predicting PJI after TAA. Our cut-off value in this study, zero, differed from previous studies of different populations, suggesting the need for assessing the performance of this method on different, larger, and more granular populations. Further studies might adjust the cut-off for every population or reach consensus on a single cut-off value based on a larger sample size. Researchers should also pay attention to potentially modifying factors included in the prediction model that might vary based on inherent population differences.