Image-based algorithms for remote cesarean surgical site infection diagnoses in diverse populations

Project Summary/Abstract Post-cesarean surgical site infections (SSIs) contribute to maternal morbidity and mortality globally; as rates of cesarean delivery increase, so will the number of SSIs. Timely SSI diagnosis and treatment can improve maternal outcomes. However, in many settings, particularly in rural areas, postoperative wound monitoring is challenging due to physical and financial barriers. The overall goal of this proposal is to improve strategies for post-cesarean SSI monitoring by validating and updating two existing image-based diagnostic algorithms. The original algorithms were trained on image-SSI diagnosis dyads collected on women delivering via cesarean in rural Rwanda. One algorithm, using visible images (photographs), had a 83% sensitivity and 75% specificity. The second algorithm, using thermal images, had 95% sensitivity and 84% specificity. In this proposed research, we will prospectively follow 6,000 women in Rwanda, Ghana and Mexico (2,000 per site) and collect wound images and SSI diagnoses at postoperative day (POD) 10. These sites were chosen because of: a) high SSI rates; b) the potential to integrate an accurate SSI diagnostic algorithm into existing community health worker follow-up programs; and c) the diversity in skin tones across the study sites. Using this data, we will assess the generalizability of the existing visible image and thermal image algorithms, evaluating the sensitivity and specificity overall and by country (Aim 1). We will then retrain the algorithm to improve predictive properties across diverse populations, and we will explore adding clinical data to the algorithms to improve accuracy (Aim 2). Finally, for a subset of 1,200 women who are SSI negative at POD10, we will reevaluate at POD20 and POD30, and use these image-SSI diagnosis dyads to explore the need for later SSI monitoring and the ability to predict delayed SSIs with images captured at POD10 (Aim 3). The culmination of this research will provide strategies for home-based monitoring of cesarean-related SSIs that can accommodate a range of skin tones.