Article Text

Download PDFPDF

P3-6 Additional remarks: randomized trial of a novel artificial intelligence/machine learning model to predict the need for specialty palliative care in hospitalized patients
  1. Jacob Strand1,2,
  2. Alisha Morgan1,2,
  3. Jordan Karow2,
  4. Emily Olson2,
  5. Claudia Anderson2,
  6. Curtis Storlie3,4 and
  7. Patrick M Wilson3
  1. 1Division of Community Internal Medicine, Geriatrics and Palliative Care, Mayo Clinic, USA
  2. 2Section of Palliative Care, Department of Medicine, Mayo Clinic, USA
  3. 3Department of Health Sciences Research, Mayo Clinic, USA
  4. 4Mayo Clinic Kern Center for the Science of Health Care Delivery, Mayo Clinic, USA


I. Original Research Background Despite the potential for artificial intelligence/machine learning (AI/ML) models to target hospitalized patients most likely to benefit from specialty palliative care (PC), the successful implementation of such models into practice remain complex challenges independent of the potential clinical impacts which have been sparsely reported.

II. Research Objectives First, to develop a predictive analytic model into a comprehensive clinical framework predictive of specialty PC need with the aims to (i) identify in a timely fashion, hospitalized patients likely to benefit from a PC consult, and (ii) intervene on such patients by contacting their care team. Subsequently, to test the ability of this machine learning model to identify hospitalized patients, in a timely fashion, likely to benefit from a PC consult. This objective involved treating PC consultation itself as a time-to-event outcome and was analyzed in a two-armed, stepped-wedge cluster randomized trial.

III. Methods Initially, electronic health record data for 68,349 inpatient specialty PC consultations in 2017 at a large hospital were used to train the AI/ML model. After validating its training performance and integrating it into clinical workflow, patients were identified as high-risk for specialty PC need by the AI/ML model from 16 hospital units (including 8 cardiology units) in a large medical center from August 2019–November 2020. Patients flagged by the algorithm were reviewed by the PC consult team and if deemed appropriate by reviewer, the patient’s care team was contacted to discuss PC involvement. Primary outcome intervention effect on timely PC consultation. Secondary outcomes included 30/60/90-day readmissions, ICU transfers, inpatient length of stay (LOS).

IV. Results Over 1679 encounters (733 intervention, 946 control), patients within the intervention arm were 40% more likely to see timely PC (OR 1.42, CI 1.06–1.90) with a reduced likelihood of readmission at 60/90-days (OR 0.74, CI 0.56–0.96 and OR 0.711, CI 0.56–0.91). There were no significant differences between groups in 30-day readmissions (OR 0.80, CI 0.60–1.08), ICU transfers (OR 1.03, CI 0.70–1.52), LOS (OR 1.08, CI 0.96–1.20).

V. Conclusion A novel AI/ML model predicting need for PC consultation met its primary endpoint of providing increased timely PC for hospitalized patients and demonstrated a reduction in 60/90-day hospital readmissions.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.