Predicting Unscheduled Emergency Department Revisits Leading to Acute Hospital Admissions Among Older Adults

Abstract Background Unscheduled emergency department (ED) revisits leading to acute hospital admission (RVA) are tantamount to a failed discharge, associated with physician error, mis-prognosis, and inadequate care planning. Previous research has shown RVA to be associated with adverse outcomes such as ICU admissions, long hospitalizations and mortality. Given the limited impact of pre-existing screening tools for older adults, we developed and validated a machine learning model to predict individual patient risk of RVA within 72 hours and 9 days of index ED visits. Method: A machine learning model was applied to retrospective electronic health record (EHR) data of patients presenting to 2 geographically and demographically divergent urban EDs in 2019. 478 clinically meaningful EHR data variables were included: socio-demographics, ED and comorbidity diagnoses, therapeutics, laboratory test orders and test results, diagnostic imaging test orders, vital signs, and utilization and operational data. Multiple machine learning algorithms were constructed; models were compared against a pre-existing adult ED-RVA risk score as a baseline. Results A total of 62,154 patients were included in the analysis, with 508 (0.82%) and 889 (1.4%) having 72-hour and 9-day RVA. The best-performing model, combining deep significance clustering (DICE) and regularized logistic regression, achieved AUC of 0.86 and 0.79 for 72-hour and 9-day ED-RVA for older adult patients, respectively, outperforming the pre-existing RVA risk score (0.704 and 0.694). Discussion Machine learning models to screen for and predict older adults at high-risk for ED-RVA may be useful in directing interventions to reduce adverse events in older adults discharged from the ED.

Rahul Sharma, 3 and Peter Steel, 3 , 1. New York, New York,  United States, 2. Weill Cornell Medicine, New York, New  York, United States, 3. NewYork-Presbyterian Hospital  Weill Cornell Medical Center, New York, New York, United  States, 4. Weill Cornell Medical College, New York, New  York, United States, 5. NewYork-Presbyterian Hospital,  New York, New York, United States  Background: Unscheduled emergency department (ED) revisits leading to acute hospital admission (RVA) are tantamount to a failed discharge, associated with physician error, mis-prognosis, and inadequate care planning.Previous research has shown RVA to be associated with adverse outcomes such as ICU admissions, long hospitalizations and mortality.Given the limited impact of pre-existing screening tools for older adults, we developed and validated a machine learning model to predict individual patient risk of RVA within 72 hours and 9 days of index ED visits.
Method: A machine learning model was applied to retrospective electronic health record (EHR) data of patients presenting to 2 geographically and demographically divergent urban EDs in 2019.478 clinically meaningful EHR data variables were included: socio-demographics, ED and comorbidity diagnoses, therapeutics, laboratory test orders and test results, diagnostic imaging test orders, vital signs, and utilization and operational data.Multiple machine learning algorithms were constructed; models were compared against a pre-existing adult ED-RVA risk score as a baseline.
Results: A total of 62,154 patients were included in the analysis, with 508 (0.82%) and 889 (1.4%) having 72-hour and 9-day RVA.The best-performing model, combining deep significance clustering (DICE) and regularized logistic regression, achieved AUC of 0.86 and 0.79 for 72-hour and 9-day ED-RVA for older adult patients, respectively, outperforming the pre-existing RVA risk score (0.704 and 0.694).
Discussion: Machine learning models to screen for and predict older adults at high-risk for ED-RVA may be useful in directing interventions to reduce adverse events in older adults discharged from the ED.

REJECTION OF CARE IN HOSPITALIZED PERSONS LIVING WITH DEMENTIA: THE IMPACT OF NURSE COMMUNICATION
Clarissa Shaw, 1 Caitlin Ward, 1 Jean Gordon, 1 Kristine Williams, 2 and Keela Herr, 1 , 1. University of Iowa, Iowa City, Iowa, United States, 2. University of Kansas School of Nursing, Kansas City, Kansas, United States Rejection of care (RoC) by persons living with dementia (PLWD) has yet to be measured in the hospital setting.Elderspeak communication (i.e., baby talk or infantilization) is an established antecedent to RoC in nursing home dementia care.The purpose of this study was to determine the impact of elderspeak communication by nursing staff on RoC by hospitalized PLWD.Eighty-eight care encounters between 16 PLWD and 53 nursing staff were observed for RoC using the Resistiveness to Care scale in one Midwestern hospital.Audio-recordings of the care encounters were transcribed verbatim and coded for semantic, pragmatic, and prosodic features of elderspeak.Over one-quarter (28.7%) of the duration of nursing staff speech towards PLWD constituted elderspeak and nearly all (96.6%) of the 88 care encounters included some elderspeak.Almost half of the observations (48.9%) included RoC behaviors by PLWD.Rejection of care was modeled as 584 Innovation in Aging, 2021, Vol. 5, No. S1 GSA 2021 Annual Scientific Meeting