Artificial intelligence for surgical site infections risk prediction from routine hospital data

Abstract Surgical site infections (SSIs) are a major cause of postoperative morbidity and mortality. Predicting their occurrence could enable earlier identification of high-risk patients and improve the targeting of preventive strategies. The PRESID-IA project aimed to develop an artificial intelligence (AI) model for SSI risk prediction using data routinely collected through hospital information systems. The study cohort included patients who underwent surgery in the Cardiac Surgery and Breast Units at Fondazione IRCCS Policlinico San Matteo in Pavia, Italy, between 2018 and 2024. Data sources comprised the operating room management system, hospital discharge records, laboratory results, and outpatient visit documentation. SSIs were identified through manual chart review of outpatient reports at 30 or 90 days post-surgery, depending on the presence of prosthetic material. The study received ethical approval. A total of 5,382 patients were included (median age 62 years; 66.3% female). Relevant variables were extracted based on a systematic literature review and validated by a clinical expert panel to ensure consistency with both clinical practice and local data availability. Data management revealed several challenges, including the occurrence of infections within the same postoperative period across multiple procedures. Manual chart review also highlighted that many SSIs were not captured by administrative data alone. The development of reliable predictive models for SSI risk is heavily influenced by data generation and documentation practices, as well as the potential incompleteness of routinely collected infection-related data. Addressing these challenges in model development is essential to ensure that predictive AI tools can effectively support earlier clinical decisions and improve patient outcomes.