Seminars in Orthodontics | 2021
Use of Neural Network model to examine post-operative infections following orthognathic surgeries in the United States
Abstract
Abstract Objective The objective of this report is to identify predictors of systemic infections following orthognathic surgeries. We test the hypothesis Neural Network models are better at predicting infections compared to traditional analytical models while using a large number of predictor variables. Methods The Nationwide Inpatient Sample for the years 2006 to 2014 was used. All patients who had a jaw anomaly and underwent an orthognathic surgery were selected. Two models: Neural Network model and multivariable logistic regression model fit with the Maximum Likelihood method were used to predict systemic infections. Results 55,839 orthognathic surgeries for maxillary/mandibular hyperplasia/hypoplasia required hospitalization in the United States. The overall systemic infection rate was 1.1%. The success rate for accurately predicting the occurrence of an infection was 98.7% by the Neural Network model. The top six predictors included: age, co-morbid condition of fluid and electrolyte disorders, year of procedure, race, deficiency anemia, and mandibular osteoplasty/segmental or subapical osteotomy. The multivariable logistic regression model had poor goodness of fit. Conclusions Neural Network model is better at predicting systemic infections following orthognathic surgeries when compared to a traditional analytical model such as multivariable logistic regression approach while using a large set of predictor variables in a nationwide database.