RPD323 - Novel NLP Methods for a Goals of Care Quality Measure among Patients with Advanced Cancer Receiving Intensive Care

Key Message : We applied novel NLP methods to capture goals of care conversations (GOCC) not documented in templated notes among a national cohort of Veterans with advanced cancer admitted to intensive care. This work enables rapid quality measurement and improvement efforts that were previously not feasible due to lengthy manual chart abstraction.

Abstract: Context: Ensuring timely GOCC for patients with advanced cancer receiving intensive care is a validated quality measure supported by VA’s life-sustaining-treatment (LST) initiative since 2018 (1).

Objectives: Apply novel NLP methods to measure GOCC among patients with advanced cancer receiving intensive care to allow for rapid quality improvement because manual chart abstraction takes over 2 hours per chart.

Methods: Among a national cohort of Veterans with advanced cancer, we examined documentation of GOCC for patients admitted to ICU for >2 days using structured data from templated LST notes. Next, in a random subset of 1000 Veterans we used ClinicalRegex, a natural language processing (NLP) software, to identify GOCC in unstructured data from the electronic health record. We optimized and validated an NLP algorithm using iterative comparison to 100 manually abstracted charts using validated methods (2) and applied the algorithm to 1000 randomly selected charts. GOCC before and after national LST implementation were examined.

Results: Among 48,240 nationally representative Veterans (average 68 years old, 97% male, 71% white, and 29% rural) with advanced cancer from 1/1/2016-12/31/2019, 6,741 had an ICU episode of > 2 days. From 2016-2019, 22% of veterans had an LST note (1455/6741). Of the patients diagnosed in 2019 (after LST policy implementation), 37% (312/845) had an LST note.

Among the NLP cohort of 1000 Veterans with advanced cancer and ICU episode of >2 days, ClinicalRegex identified GOCC in 49% of veterans, LST notes were present in 22%, and 47% had neither. Of the 126 patients diagnosed in 2019 (after LST policy implementation), 59% (74/126) had a GOCC, 41% (51/126) had a LST note, and 37% (46/126) had neither.

Conclusions: Novel NLP methods efficiently capture GOCC not documented in templated notes and allow for rapid quality measurement to facilitate improvement in rates of GOCC among patients with advanced cancer.

References: 1. Walling AM, Tisnado D, Asch SM, et al. The quality of supportive cancer care in the veterans affairs health system and targets for improvement. JAMA Intern Med. 2013;173(22):2071-9.
2. Lindvall C, Deng C, Moseley E et al. Natural Language Processing to Identify Advance Care Planning Documentation in a Multisite Pragmatic Clinical Trial. J Pain Symptom. 2022;63(1):e29-e36.