907. Technology and Medicine: Prediction of Surgical Site Infection in Clean Surgeries using Artificial Neural Networks

Background In Belo Horizonte, a city with 3,000,000 inhabitants, a survey was performed in six hospitals, between July 2016 and June 2018, about surgical site infection (SSI) in patients undergoing clean surgery procedures. The main objective is to statistically evaluate such incidences and enable an analysis of the SSI predictive power, through MLP (Multilayer Perceptron) pattern recognition algorithms. Methods Through the Hospital Infection Control Committees (CCIH) of the hospitals, a data collection on SSI was carried out through the software SACIH - Automated System for Hospital Infection Control. So, three procedures were performed: a treatment of the collected database for use of intact samples; a statistical analysis on the profile of the collected hospitals; an evaluation of the predictive power of five types of MLPs (Backpropagation Standard, Momentum, Resilient Propagation, Weight Decay and Quick Propagation) for SSI prediction. The MLPs were tested with 3, 5, 7 and 10 neurons in the hidden layer and with a division of the database for the resampling process (65% or 75% for testing, 35% or 25% for validation). They were compared by measuring the AUC (Area Under the Curve - ranging from 0 to 1) presented for each of the configurations. Results From 45,990 records, 12,811 were able for analysis. The statistical analysis results were: the average age is 49 years old (predominantly between 30 and 50); the surgeries had an average time of 134.13 minutes; the average hospital stay is 4 days (from 0 to 200 days), the death rate reached 1% and the SSI 1.49%. A maximum prediction power of 0.742 was found. Conclusion There was a loss of 60% of the database samples due to the presence of noise. However, it was possible to have a relevant sample to assess the profile of these six hospitals. The predictive process, presented some configurations with results that reached 0.742, what promises the use of the structure for the monitoring of automated SSI for patients submitted to surgeries considered clean. To optimize data collection, enable other hospitals to use the prediction tool and minimize noise from the database, two mobile application were developed: one for monitoring the patient in the hospital and other for monitoring after hospital discharge. The SSI prediction analysis tool is available at www.nois.org.br. Disclosures All Authors: No reported disclosures