Natural Language Processing to Identify Home Health Care Patients at Risk for Becoming Incapacitated With No Evident Advance Directives or Surrogates

Jiyoun Song; Maxim Topaz; Aviv Y Landau; Robert L Klitzman; Jingjing Shang; Patricia W Stone; Margaret V McDonald; Bevin Cohen

doi:10.1016/j.jamda.2024.105019

Natural Language Processing to Identify Home Health Care Patients at Risk for Becoming Incapacitated With No Evident Advance Directives or Surrogates

J Am Med Dir Assoc. 2024 May 13:105019. doi: 10.1016/j.jamda.2024.105019. Online ahead of print.

Authors

Jiyoun Song¹, Maxim Topaz², Aviv Y Landau³, Robert L Klitzman⁴, Jingjing Shang⁵, Patricia W Stone⁵, Margaret V McDonald⁶, Bevin Cohen⁷

Affiliations

¹ Department of Biobehavioral Health Sciences, University of Pennsylvania School of Nursing, Philadelphia, PA, USA.
² Columbia University School of Nursing, New York, NY, USA; Data Science Institute, Columbia University, New York, NY, USA; Center for Home Care Policy & Research, VNS Health, New York, NY, USA.
³ School of Social Policy and Practice, University of Pennsylvania, Philadelphia, PA, USA.
⁴ Columbia University College of Physicians and Surgeons, New York, NY, USA; Columbia University Joseph Mailman School of Public Health, New York, NY, USA.
⁵ Columbia University School of Nursing, New York, NY, USA.
⁶ Center for Home Care Policy & Research, VNS Health, New York, NY, USA.
⁷ Brookdale Department of Geriatrics and Palliative Medicine, Icahn School of Medicine at Mount Sinai, New York, NY, USA; Center for Nursing Research and Innovation, Mount Sinai Health System, New York, NY, USA. Electronic address: bevin.cohen@mountsinai.org.

PMID: 38754475
DOI: 10.1016/j.jamda.2024.105019

Abstract

Objectives: Home health care patients who are at risk for becoming Incapacitated with No Evident Advance Directives or Surrogates (INEADS) may benefit from timely intervention to assist them with advance care planning. This study aimed to develop natural language processing algorithms for identifying home care patients who do not have advance directives, family members, or close social contacts who can serve as surrogate decision-makers in the event that they lose decisional capacity.

Design: Cross-sectional study of electronic health records.

Setting and participants: Patients receiving post-acute care discharge services from a large home health agency in New York City in 2019 (n = 45,390 enrollment episodes).

Methods: We developed a natural language processing algorithm for identifying information documented in free-text clinical notes (n = 1,429,030 notes) related to 4 categories: evidence of close relationships, evidence of advance directives, evidence suggesting lack of close relationships, and evidence suggesting lack of advance directives. We validated the algorithm against Gold Standard clinician review for 50 patients (n = 314 notes) to calculate precision, recall, and F-score.

Results: Algorithm performance for identifying text related to the 4 categories was excellent (average F-score = 0.91), with the best results for "evidence of close relationships" (F-score = 0.99) and the worst results for "evidence of advance directives" (F-score = 0.86). The algorithm identified 22% of all clinical notes (313,290 of 1,429,030) as having text related to 1 or more categories. More than 98% of enrollment episodes (48,164 of 49,141) included at least 1 clinical note containing text related to 1 or more categories.

Conclusions and implications: This study establishes the feasibility of creating an automated screening algorithm to aid home health care agencies with identifying patients at risk of becoming INEADS. This screening algorithm can be applied as part of a multipronged approach to facilitate clinician support for advance care planning with patients at risk of becoming INEADS.

Keywords: Advance care planning; healthcare proxy; home health care; natural language processing; social isolation; unbefriended.