Abstract
Machine learning (ML) algorithms have been developed to predict surgical site infection (SSI) in some populations, but not after emergent trauma laparotomy. We hypothesized that ML could identify patients at risk of SSI using only variables that are available perioperatively and that could be automatically extracted from the patient's chart.
Patients from the American College of Surgeons Trauma Quality Improvement Project database who received a laparotomy within 90 min of arrival were retrospectively reviewed. ML models were created to predict clinically meaningful in-hospital SSI (defined as either deep or organ space SSI), with subanalyses for both deep and organ space SSI individually. A game theoretical approach was used to estimate the relative significance of each variable toward the final prediction.
Of 5,481,046 patients in American College of Surgeons Trauma Quality Improvement Project from 2017 to 2021, 74,806 met the inclusion criteria. SSI incidence was 3.2%. The model for the composite SSI outcome achieved an area under the receiver-operator curve of 0.805 (95% confidence interval [CI] 0.787-0.824) with the organ space SSI alone model slightly outperforming the deep SSI alone model (area under the receiver-operator curve of 0.832 (95% CI 0.808-0.855) compared to 0.776 (95% CI 0.745-0.804). The most impactful variables were the facility SSI rate, colorectal injury, total number of injuries, and volume of packed red blood cells transfused.
ML can reliably identify emergency trauma laparotomy patients at anincreased risk for SSI. Such an approach can be integrated directly into electronic medical records to automatically identify high-risk patients on admission, allowing for personalized care plans tailored to each patient's risk profile.