901. Risk Prediction for Surgical Site Infection in Patients Subject to Knee Arthroplasty Surgery

Abstract

BackgroundThis research represents an experiment on surgical site infection (SSI) in patients undergoing knee arthroplasty surgery procedures in hospitals in Belo Horizonte, between July 2016 and June 2018. The objective is to statistically evaluate such incidences and enable a study of the prediction power of SSI of pattern recognition algorithms, in this case the Multilayer Perceptron (MLP).MethodsData were collected on SSI in five hospitals. The Hospital Infection Control Committees (CCIH) of the hospitals involved collected all data used in the analysis during their routine SSI surveillance procedures and sent the information to the Nosocomial Infection Study Project (NOIS). Three procedures were performed: a treatment of the database collected for use of intact samples; a statistical analysis on the profile of the hospitals collected and; an assessment of the predictive power of five types of MLP (Backpropagation Standard, Momentum, Resilient Propagation, Weight Decay, and Quick Propagation) for SSI prediction. MLPs were tested with 3, 5, 7, and 10 hidden layer neurons and a database split for the resampling process (65% and 75% for testing, 35% and 25% for validation). They were compared by measuring AUC (Area Under the Curve - ranging from 0 to 1) presented for each of the configurations.ResultsFrom the 1438 data collected, 390 records were usable and it was verified: the average age of the patients who underwent this surgical procedure was 70 (ranging from 29 to 92), average surgery time was 171 minutes (between 50 and 480), 47% presented a hospital contamination, 1% SSI and no deaths. During the MLP experiments, due to the low number of SSI cases, the prediction rate for this specific surgery was 0.5.ConclusionDespite the large noise index of the database, it was possible to have a relevant sampling to evaluate the profile of hospitals in Belo Horizonte. However, for the predictive process, despite some results equal to 0.5, the database demands more samples of SSI cases, as only 1% of positive samples generated an unbalance of the database. To optimize data collection and enable other hospitals to use the SSI prediction tool (available in www.sacihweb.com), two mobile application were developed: one for monitoring the patient in the hospital and the other for monitoring after hospital discharge.Disclosures All Authors: No reported disclosures