Reducing the Burden to Deliver Remote Postoperative Surveillance Using Machine Learning for Surgical-Site Infection (SSI)

Citation

McLean K, Sgrò A, Brown L, Buijs L, Potter M, Bouamrane MM & Harrison E (2024) Reducing the Burden to Deliver Remote Postoperative Surveillance Using Machine Learning for Surgical-Site Infection (SSI). British Journal of Surgery, 111 (Supplement_6), Art. No.: znae163.012. https://doi.org/10.1093/bjs/znae163.012

Abstract
Aim: To develop novel methods for automated assessment of patient-generated data from remote postoperative wound (RPW) surveillance to reduce the burden of clinical triage for in-person assessment. Method: This was a secondary analysis of an interventional study on RPW surveillance: “ImplementatioN of Remote Surgical wOund Assessment During the coviD-19 pandEmic” (INROADE). INROADE included adult patients undergoing gastrointestinal surgery, who could submit images of their surgical wound(s), and patient-reported symptoms of SSI for 30-days postoperatively. Separate models were developed to predict a clinical diagnosis of SSI within 48h based on patient-reported symptoms (logistic regression) and wound images (convolutional neural networks [CNN]). Model performance was evaluated using area under the curve (AUC), with 95% confidence intervals. The impact of implementation of automated assessment was simulated within the RPW surveillance pathway to “rule-out” responses needing clinician triage. Results: Of 200 patients enrolled, there were 1529 responses with 3.3% (n=51) submitted within 48h of an SSI. There was excellent discrimination using logistic regression with patient-reported symptoms (AUC: 0.830, 0.765-0.894). 2,125 images were used to develop the CNN, which showed moderate discrimination to identify 48h SSI (AUC: 0.615, 0.556-0.674). Usage to screen out “low-risk” responses prior to clinical triage was estimated to half the staff-time to deliver (12.2h vs 25.5h), while maintaining diagnostic accuracy (AUC: 0.760 [0.685-0.834] vs 0.789 [0.717-0.861]). Conclusions: Automated assessment can be successfully deployed within RPW surveillance pathways to reduce the burden on staff to deliver without compromising care and allow resources to be appropriately directed to those at greatest risk of SSI.

Keywords
adult; areas under curve; surgical wound infection; triage; diagnosis; surgical incisions; gastrointestinal surgical procedures; surveillance; medical; machine learning; clinical diagnosis; secondary data analysis; convolutional neural networks; coronavirus pandemic

Journal
British Journal of Surgery: Volume 111, Issue Supplement_6