Early warning systems in inpatient anorexia nervosa: A validation of the MARSIPAN-based modified early warning system

Area of Science:

• Clinical Medicine
• Medical Technology
• Psychiatry

Background:

• Early detection of patient deterioration is crucial in inpatient settings.
• Existing early warning systems may not be optimized for the unique physiological presentations in eating disorders.
• The MARSIPAN-guidance-adapted Early Warning System (MARSI MEWS) was developed to address this gap.

Purpose of the Study:

• To evaluate the validity of the MARSI MEWS in identifying critical deterioration in anorexia nervosa inpatients.
• To compare the performance of MARSI MEWS against the National Early Warning Score (NEWS) and the Physical Risk in Eating Disorders Index (PREDIX).
• To explore the feasibility of a machine-learning-based early-warning-system (ML-EWS).

Main Methods:

• A retrospective analysis of 3,937 patient observations from 36 inpatients at Addenbrookes Hospital (2017-2018).
• Establishment of a "gold standard" for deteriorating cases by three independent raters.
• Performance evaluation using Receiver Operating Characteristic Area Under the curve (ROC AUC) for MARSI MEWS, NEWS, and PREDIX.
• Testing of an ML-EWS (random forest) using cross-validation and out-of-sample prediction.

Main Results:

• MARSI MEWS demonstrated superior performance with a ROC AUC of 0.916, compared to NEWS (0.828) and PREDIX (0.865).
• The ML-EWS achieved high performance metrics, with ROC AUC of 0.980 in independent samples and 0.922 in multilevel analysis.
• These findings suggest MARSI MEWS is highly suitable for inpatient anorexia nervosa settings.

Conclusions:

• MARSI MEWS is recommended for identifying critically deteriorating cases in the inpatient anorexia nervosa population.
• The study provides a proof of concept for developing artificial intelligence-based early warning systems in anorexia nervosa.
• Further research may be needed to assess community practice, where PREDIX might remain optimal.