A proof-of-concept study associating artificial intelligence surveillance of surgical site infections with antibiotic prophylaxis from 765,962 surgeries

Surveillance of superficial surgical site infections (SSSIs) is currently either a manual process, or relies on adminstrative coding. We hypothesized that an artificial intelligence (AI) based Natural Language Processing (NLP) model used on Electronic Health Record (EHR) data could be used for SSSI surveillance, and associated with breaks in relevant procedural events, using prophylactic antiobiotics as an example. Retrospective cohort study with EHR data from 18 Danish hospitals (May 2016–December 2021). All inpatient surgical procedures were included ( n = 765,962), with subgroup analyses of hip and knee arthroplasties ( n = 36,378) and exploratory laparotomies ( n = 5,250). SSSIs within 30 days postoperatively were identified using a validated NLP model. The primary exposure was prophylactic antibiotic administration within 120 min before incision. Multivariable logistic regression estimated odds ratios (ORs) controlling for relevant confounders. Among 765,962 surgeries, 14,018 SSSIs (1.8%) were identified, including 6 months where SSSI rates exceed 1 standard deviation (SD) and 3 months exceeding 2 SD of the mean rate. The exposure occurred in 37.3% of procedures overall, 94.4% of arthroplasties, and 54.6% of laparotomies. This was associated with reduced infection risk overall (OR 0.90; p <.001) and in laparotomies (OR 0.65; p <.001), but not significantly in arthroplasties (OR 0.90; p =.58). AI based SSSI surveillance can identify SSSI incidences at scale and be associated with potential breaks in procedural events.