PREDICTIVE MODEL FOR INFECTION IN TOTAL KNEE ARTHROPLASTY SURGERIES IN A REHABILITATION HOSPITAL

With increased population longevity, total knee arthroplasty (TKA) procedures have increased. Although effective for pain relief and improvement of function and quality of life in severe cases, complications such as surgical site infection (SSI) remain relevant challenges. In 2024, the SSI rate for primary TKA was 2.1% in the Federal District and 2.0% in Brazil. Predictive models have been adopted to identify patient characteristics and surgical contexts at higher risk and to support SSI prevention in TKA surgeries. Thus, this study will compare the performance of two predictive risk models for SSI prediction in TKA surgeries performed in a Rehabilitation Hospital. This is a longitudinal, analytical, observational study, historical cohort type, based on hospital surveillance data. The population consists of all TKA surgical procedures in patients older than 18 years, performed in a rehabilitation hospital from 2008 to 2024. Two predictive models from the Centers for Disease Control and Prevention (CDC) were adopted and their results will be compared: the IRIC model – Surgical Infection Risk Index, which applies to any surgical procedure, and the Procedure-Specific Models (PSM). A total of 1,370 TKA surgeries were included in the study, with an incidence rate of 2.8% (n = 39) of SSI cases. The main demographic characteristics of the patients undergoing these surgeries were: 71% female, mean age 68 years (standard deviation: SD ± 9); 75% had ASA classification 2; mean Body Mass Index (BMI) 30.8 kg/m² (SD ± 4.7). Procedure characteristics were: mean duration 92 minutes (SD ± 28) and 93% were classified as clean surgeries. In both predictive models analyzed, the model variables age, sex, BMI, wound classification, and ASA showed no statistical association with SSI; procedure duration was the only variable associated with the outcome. Predictive performance of the two models was analyzed by the area under the ROC curve, yielding 0.64 and 0.61 for the IRIC and PSM models, respectively. Based on the obtained results, the PSM model did not show superior performance compared to the IRIC model for SSI prediction in TKA. In both models, mean procedure duration was an important predictive factor and deserves attention in the studied context.