Development and validation of an Artificial Intelligence and machine learning model to predict 30 day mortality in patients with secondary peritonitis

Survival rates are lower following emergency abdominal surgery, with 30-day mortality rates ranging from 4% to 8%. Early identification of risk factors and proactive risk mitigation strategies could potentially avert some of these fatalities. Although numerous tools exist to assist clinicians in assessing the risk of mortality or serious complications post-surgery (6), most are designed for preoperative risk evaluation and rely solely on variables available before surgery (7). However, the patient’s intraoperative course provides additional valuable information, offering the potential for early detection of modifiable deterioration. Numerous risk stratification models are currently available, however these models were primarily derived from and for elective surgery patients, raising questions about their accuracy and applicability to emergency surgery patients.Therefore, we aim to construct a deep-learning algorithm, an AI model, that will provide dynamic predictions of postoperative mortality based on pre- and intraoperative electronic medical record data. We hypothesize that our model will accurately predict 30-day mortality and morbidity in patients undergoing EL in the Indian population. Unlike existing risk assessment models that include only preoperative data or both pre- and intraoperative data to develop a model, we intend to incorporate both preoperative, intraoperative, and postoperative electronic medical record data of the patients. By implementing deep-learning algorithms, we aim to analyze complex interactions among patient characteristics, intraoperative events, and postoperative outcomes. It offers the potential for personalized risk assessment and modified intervention strategies, ultimately aiming to improve outcomes and reduce mortality rates in this high-risk patient population. A total of 500 patients will be included, divided into three groups at a ratio of 7:1:2. For model development, 350 patients will be included, while 50 patients will be allocated for validation, and 100 patients for testing the model