Applying artificial neural network in predicting sepsis mortality in the emergency department based on clinical features and complete blood count parameters

Abstract

Abstract A complete blood count (CBC) is routinely ordered for emergency department (ED) patients with infections. Certain parameters, such as the neutrophil-to-lymphocyte ratio (NLR), might have prognostic value. We aimed to evaluate the prognostic value of the presenting CBC parameters combined with clinical variables in predicting 30-day mortality in adult ED patients with infections using an artificial neural network (ANN). We conducted a retrospective study of ED patients with infections between 17 December 2021 and 16 February 2022. Clinical variables and CBC parameters were collected from patient records, with NLR, monocyte-to-lymphocyte ratio (MLR), and platelet-to-lymphocyte ratio (PLR) calculated. We determined the discriminatory performance using the area under the receiver operating characteristic curve (AUROC) and performed a 70/30 random data split and supervised ANN machine learning. We analyzed 558 patients, of whom 144 (25.8%) had sepsis and 60 (10.8%) died at 30 days. The AUROCs of NLR, MLR, PLR, and their sum were 0.644 (95% CI 0.573–0.716), 0.555 (95% CI 0.482–0.628), 0.606 (95% CI 0.529–0.682), and 0.610 (95% CI 0.534–0.686), respectively. The ANN model based on twelve variables including clinical variables, hemoglobin, red cell distribution width, NLR, and PLR achieved an AUROC of 0.811 in the testing dataset.