36 Design of an algorithm for the semi-automated detection of surgical site infections in elective colorectal surgery

Introduction and Objectives Surgical site infections (SSIs) represent a prevalent postoperative complication, classified into superficial, deep, and organ-space infections. Among surgical procedures, colorectal surgeries exhibit particularly elevated infection rates, ranging from 15% to 30%. These complications are associated with prolonged hospitalizations, increased antibiotic consumption, and substantial healthcare costs. Current surveillance methods for SSI detection heavily rely on labor-intensive manual reviews, thereby limiting their efficiency and scalability. This study seeks to design an algorithm to enable the semi-automated detection of SSIs in colorectal surgeries, with the objective of minimizing manual review requirements. Secondary goals include determining the most predictive indicators of infection risk, optimizing diagnostic accuracy, and evaluating the algorithm’s performance across distinct SSI categories. Methods A prospective, single-center study was conducted at HGG, with data collected by the Infection Control Team using pre-established surveillance protocols. The proposed algorithm was developed and tested retrospectively using data spanning from 2010 to 2023 to evaluate its capacity to detect SSIs. The study’s intervention entailed the development of a coding system designed to extract cases from the Minimum Basic Data Set (ICD-10) at discharge. Subsequently, an algorithm was implemented to filter cases based on a combination of variables, including extended hospital stays, 30-day readmissions, reoperations, postoperative mortality, antibiotic therapies, microbiological cultures, and computed tomography (CT) scans performed within 30 days. The study sample encompassed patients undergoing colorectal surgery recorded in the HGG SSI surveillance database. Given the retrospective nature of the study, no sample size calculation was required. Algorithm performance was assessed through the use of receiver operating characteristic (ROC) curves, enabling the identification of the optimal combination of variables for achieving maximum sensitivity and specificity. Results An analysis of 1213 cases demonstrated that the algorithm achieved an overall sensitivity of 86.77% and specificity of 73.17%, accurately predicting 118 events and 788 non-events. Excluding specific variables, such as antibiotic use or microbiological data, resulted in variations in sensitivity, which ranged from 78.68% to 86.03%. Importantly, organ-space infections were consistently detected with a high sensitivity of 97.6%. These results highlight the potential of semi-automated detection to substantially reduce the manual workload associated with SSI surveillance while maintaining robust diagnostic precision. Conclusions The semi-automated detection of SSIs through the proposed algorithm has demonstrated the capacity to reduce the workload of infection control personnel by up to 70%, thereby enhancing the efficiency of monitoring processes. Integrating this approach into clinical workflows may improve surveillance capabilities and contribute to better patient outcomes. Further research is warranted to validate the algorithm’s performance across diverse healthcare environments.