MACHINE LEARNING IN THE PERIOPERATIVE SETTING: UNCOVERING THE VALUE OF DATA SCIENCE AND LARGE INSTITUTIONAL DATASETS AMONG PATIENTS UNDERGOING SURGERY

Abstract

4.2 million people worldwide die within 30 days of surgery yearly, and about 17% develop at least one complication. Many of these are preventable. Hence, there is a crucial need to improve perioperative outcomes, which benefit individual patients and the health systems. A key enabler to such efforts would be a curated registry comprising high-quality, real-world data of the entire pre-operative, intra-operative, and post-operative segments of the hospitalization. This registry would allow studies of factors influencing patient outcomes and evaluation of clinical interventions. Currently, no large, high-resolution database integrates all patient care areas available globally. My Ph.D. project is to describe the methodology used to set up Singapore General Hospital's real-world perioperative registry - the Perioperative and Anesthesia Subject Area Registry (PASAR), which allows seamless interrogation of the patient journey from the outpatient clinic to the operating theatre, ICU or surgical ward, and up to discharge. For the purpose specific to this PhD project, I focused on developing PASAR to fill the specific gaps such as the inclusion of intraoperative data, higher use of clinical data rather than administrative data such as ICD coding and providing a relatively low code interface to provide ease of use for clinical researchers.