Machine Learning-Driven Decision-Support System for Nursing Risk Assessment in Post-Discharge Care: A Design Science Approach

Hospital readmissions represent a persistent challenge for healthcare systems, often stemming from inadequate post-discharge monitoring. This study presents a Machine Learning (ML)-driven Clinical Decision Support System (CDSS) designed to enhance nursing risk assessment in post-discharge care. Developed using a Design Science Research Methodology, the artefact integrates a digital questionnaire, an ML-based risk stratification model, and a real-time dashboard to optimise follow-up processes. The system was developed at a medical-surgical inpatient unit of a private hospital in Portugal. A retrospective dataset of 10,134 structured telephone follow-up records, classified by nurses into three risk levels--stable (A), requiring reassessment (B), and clinically concerning (C)--was used to train and evaluate the ML model. Among the evaluated classifiers, Logistic Regression was selected for deployment based on its high specificity (0.9993), precision (0.9879), and absence of critical false negatives, despite slightly lower recall compared to XGBoost. The CDSS enables real-time risk classification based on patient-reported outcomes, supporting timely identification and prioritisation of patients requiring clinical attention. Simulation results indicate a potential reduction of up to 79% in nurse follow-up workload, while preserving care quality by focusing resources on moderate- and high-risk patients. Although direct evidence of reduced readmissions is not yet available, the system aligns with established best practices in transitional care. This study demonstrates the feasibility and utility of ML-based dynamic risk stratification for post-discharge monitoring, offering a scalable and interpretable solution that enhances clinical decision-making and resource allocation in nursing practice.