Abstract: TH-PO240
Comparison of Approaches to the Identification of Symptom Burden in Hemodialysis Patients Utilizing Electronic Health Records
Session Information
- Hemodialysis and Frequent Dialysis - II
November 07, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 701 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Chan, Lili, Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Beers, Kelly H., Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Chauhan, Kinsuk, Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Debnath, Neha, Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Pattharanitima, Pattharawin, Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Coca, Steven G., Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Van Vleck, Tielman T., Icahn School of Medicine at Mount Sinai, New York, New York, United States
- Nadkarni, Girish N., Icahn School of Medicine at Mount Sinai, New York, New York, United States
Background
Symptoms are common in patients on maintenance hemodialysis (HD), however identification within the electronic medical record (EMR) is challenging. Natural language processing (NLP) can be utilized to identify symptoms from narrative clinical documentation by physicians and other providers.
Methods
We utilized NLP to identify 7 patient symptoms from clinical notes of HD patients from the BioMe Biobank and validated our findings using the MIMIC-III database. We compared NLP performance with ICD codes and validated the performance of NLP and ICD codes vs. manual chart review.
Results
We identified 1034 and 519 HD patients from BioMe and MIMIC-III, respectively. In BioMe, the most frequent symptoms identified were pain (NLP 93% vs. ICD 46%, P<0.001), fatigue (NLP 84% vs. ICD 41%, P<0.001), and nausea and/or vomiting (NLP 74% vs. ICD 19%, P<0.001). Sensitivity for NLP ranged from 0.85 (95% CI 0.65-96) for depression to 0.99 (95% CI 0.93-1) for fatigue while sensitivity for ICD ranged from 0.09 (95% CI 0.01-0.29) for cramps to 0.59 (95% CI 0.43-0.73) for fatigue. Results were similar in MIMIC-III. ICD codes were significantly more specific for nausea and/or vomiting in BioMe and for fatigue, depression, and pain in MIMIC-III. A majority of patients in both cohorts had ≥4 symptoms. Patients with more symptoms identified by NLP, ICD, and chart review had more clinical encounters. Results were similar in a subgroup of 608 patients who had ≥2 years of follow up with only 1 year of notes reviewed.
Conclusion
NLP had higher sensitivity compared to ICD codes for identification, with comparable specificity for most symptoms and may be useful for the high-throughput identification of patient-centered outcomes.
Funding
- NIDDK Support