Using Machine Learning to Establish Predictors of Mortality in Patients Undergoing Laparotomy for Emergency General Surgical Conditions.

Abstract

Patients undergoing laparotomy for emergency general surgery (EGS) conditions, constitute a high-risk group with poor outcomes. These patients have a high prevalence of comorbidities. This study aims to identify patient factors, physiological and time-related factors, which place patients into a group at increased risk of mortality. In a retrospective analysis of all patients undergoing an emergency laparotomy at Greys Hospital from December 2012 to 2018, we used decision tree discrimination to identify high-risk groups. Our cohort included 1461 patients undergoing a laparotomy for an EGS condition. The mortality rate was 12.4% (181). Nine hundred and ten patients (62.3%) had at least one known comorbidity on admission. There was a higher rate of comorbidities among those that died (154; 85.1%). Patient factors found to be associated with mortality were the age of 46years or greater (p <0.001), current tuberculosis (p <0.001), hypertension (p = 0.014), at least one comorbidity (0.006), and malignancy (0.033). Significant physiological risk factors for mortality were base excess less than -6.8mmol/L (p <0.001), serum urea greater than 7.0mmol/L (p <0.001) and waiting time from admission to operation (p = 0.014). In patients with an enteric breach, those younger than 46years and a Shock Index of more than 1.0 were high-risk. Patients without an enteric breach were high-risk if operative duration exceeded 90min (p = 0.004) and serum urea exceeding 7mmol/dl (p = 0.016). In EGS patients, patient factors as well as physiological factors place patients into a high-risk group. Identifying a high-risk group should prompt consideration for an adjusted approach that ameliorates outcomes.