Using artificial intelligence to reduce orthopedic surgical site infection surveillance workload: Algorithm design,...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Surgical Infection Prevention

Background:

• Surgical site infection (SSI) surveillance is resource-intensive.
• Accurate and efficient SSI detection is crucial for patient outcomes.
• Current surveillance methods require significant manual effort.

Purpose of the Study:

• To design and validate an algorithm for detecting SSI after hip replacement surgery.
• To report the successful implementation of this algorithm in a real-world clinical setting.
• To improve the efficiency and accuracy of orthopedic SSI surveillance.

Main Methods:

• Development of a multivariable algorithm, AI-HPRO, employing natural language processing (NLP) and extreme gradient boosting.
• Utilized data from 19,661 healthcare episodes across four public hospitals in Madrid, Spain.
• Validated the algorithm's performance using clinical data, including microbiological cultures and medication prescriptions.

Main Results:

• The AI-HPRO algorithm demonstrated high sensitivity (99.18%) and specificity (91.01%).
• Key markers for SSI included positive microbiological cultures, the term "infection", and clindamycin prescription.
• Achieved an AUC of 0.989 and a negative predictive value of 99.98%.

Conclusions:

• The AI-HPRO algorithm significantly reduces surveillance time (from 975 to 63.5 person/hours) and manual review volume (by 88.95%).
• This is the first report of an algorithm combining NLP and extreme gradient boosting for accurate, real-time orthopedic SSI surveillance.
• The model offers superior negative predictive value compared to NLP-only or NLP with logistic regression approaches.