The emergency department is a fast-paced environment responsible for large volumes of patients with varied acuity. Operationalpressures on emergency departments are increasing, which creates the imperative to efficiently identify patients at risk ofacute deterioration and minimise patients’ risk of morbidity and mortality. We apply state-of-the-art machine learning methodsto predict patient deterioration early, based on their first recorded vital signs, observations, laboratory results, and otherpredictors documented in Electronic Patient Records. We build on prior work by incorporating interpretable machine learningand fairness-aware modelling, and achieve improved classification performance over the current standard, the National EarlyWarning Score 2, measured by average precision and daily alert rate. We use a cross-sectional Electronic Patient Recorddataset comprising 121, 058 unplanned admissions at Salford Royal Hospital, UK, to systematically compare model variationsfor predicting mortality and critical care utilisation within 24 hours of admission. Our models yield up to 0.375 increase inaverage precision, up to 18.5% reduction in daily alert rate, and a median 0.577 reduction in differential bias amplification acrossthe protected demographics of age and sex. We use Shapely Additive exPlanations to justify the models’ outputs, verify thatthe captured data associations align with domain knowledge, and pair predictions with the causal context of each patient’smost influential characteristics. We encourage future research to follow a systematised approach to data-driven risk modellingand help obtain clinically applicable support tools.