Artificial neural network‐based prediction of prolonged length of stay and need for post‐acute care in acute coronary syndrome patients undergoing percutaneous coronary intervention

• DOI: 10.1111/eci.13406
• Publication Date: 2020-10-12T05:18:27Z
• Authors: Hemant Kulkarni; Manoj Thangam; Amit P. Amin
• Keywords: Acute care; Coronary care unit

Abstract Background Prolonged length of stay (LOS) and post‐acute care after percutaneous coronary intervention (PCI) is common costly. Risk models for predicting prolonged LOS have limited accuracy. Our goal was to develop validate using artificial neural networks (ANN) predict > 7days need PCI. Methods We defined as ≥7 days patients discharged to: extended care, transitional unit, rehabilitation, other acute hospital, nursing home or hospice care. Data from 22 675 who presented with ACS underwent PCI shuffled split into a derivation set (75% dataset) validation dataset (25% dataset). Calibration plots were used examine the overall predictive performance MLP by plotting observed expected risk deciles fitting lowess smoother data. Classification accuracy assessed receiver‐operating characteristic (ROC) area under ROC curve (AUC). Results MLP‐based model predicted an 90.87% 88.36% in training test sets, respectively. The had 90.22% 86.31% This achieved quick convergence. Predicted probabilities showed good (prolonged LOS) excellent calibration (post‐acute care). Conclusions ANN‐based accurately Larger studies replicability longitudinal evidence impact are needed establish these current practice.