Using artificial intelligence to reduce orthopedic surgical site infection surveillance workload: Algorithm design, validation, and implementation in 4 Spanish hospitals

Abstract

Surgical site infection (SSI) surveillance is a labor-intensive endeavor. We present the design and validation of an algorithm for SSI detection after hip replacement surgery, and a report of its successful implementation in 4 public hospitals in Madrid, Spain. We designed a multivariable algorithm, AI-HPRO, using natural language processing (NLP) and extreme gradient boosting to screen for SSI in patients undergoing hip replacement surgery. The development and validation cohorts included data from 19,661 health care episodes from 4 hospitals in Madrid, Spain. Positive microbiological cultures, the text variable "infection", and prescription of clindamycin were strong markers of SSI. Statistical analysis of the final model indicated high sensitivity (99.18%) and specificity (91.01%) with an F1-score of 0.32, AUC of 0.989, accuracy of 91.27%, and negative predictive value of 99.98%. Implementation of the AI-HPRO algorithm reduced the surveillance time from 975person/hours to 63.5person/hours and permitted an 88.95% reduction in the total volume of clinical records to be reviewed manually. The model presents a higher negative predictive value (99.98%) than algorithms relying on NLP alone (94%) or NLP and logistic regression (97%). This is the first report of an algorithm combining NLP and extreme gradient-boosting to permit accurate, real-time orthopedic SSI surveillance.