Keywords: predictive rule, EHR, utilization, AI (PubMed Search)
Author- Steve Schenkel, MD MPP Professor of Emergency Mediciner at UMEM:
A recent Annals of Emergency Medicine Publication (here https://www.annemergmed.com/article/S0196-0644(22)01276-8/fulltext) tested a predictive rule for Likelihood to Occupy an Inpatient Bed associated with a common Electronic Health Record.
At the individual patient level, the score performed ok. Depending on the chosen threshold, it traded off sensitivity and specificity and generally became more accurate the longer the patient was in the ED.
The authors and the associated editorial (here https://www.annemergmed.com/article/S0196-0644(22)01401-9/fulltext) suggest a different, potentially more beneficial use: to allow aggregate prediction of admissions across an entire department and therefore prompt earlier planning to prevent crowding on account of boarding.
The takeaway: Administrative prediction rules oriented toward individual patients may be more meaningfully used to predict resource needs, including in-patient beds, across the ED population.