Artificial Intelligence Methods for Improved Detection of Undiagnosed Heart Failure with Preserved Ejection Fraction
Key result
NLP reveals ~75% of HF patients with LVEF ≥50% meet HFpEF criteria without a clinical diagnosis.
- n=8,606
Why the study?
Heart failure with preserved ejection fraction remains underdiagnosed despite accounting for nearly half of all heart failure cases, prompting exploration of natural language processing to improve detection per ESC criteria.
Can a natural language processing (NLP) pipeline applied to electronic health records improve the detection of undiagnosed HFpEF compared to clinician-assigned diagnosis?
Population
8606 patients with HF, including 3727 consecutive patients with HF and LVEF >=50%
Comparison
Clinician-assigned HFpEF diagnosis vs meeting ESC criteria without formal diagnosis
Design
Retrospective cohort study validated in a second independent centre
Authors
Discussion
Loading...
Member takes
Implication
Natural language processing applied to electronic health records can identify a large proportion of patients with undiagnosed HFpEF who meet ESC criteria and are at high risk of mortality.
Study Design
Type
Cohort (n=8,606)
Multicenter
Yes
Structured PICO
Can a natural language processing (NLP) pipeline applied to electronic health records improve the detection of undiagnosed HFpEF compared to clinician-assigned diagnosis?
P
Population
8,606 patients with a clinical diagnosis of heart failure between 2010 and 2022, including 3,727 with LVEF ≥50% evaluated for HFpEF.
E
Exposure
Natural language processing (NLP) pipeline applied to electronic health records to identify and categorize HFpEF according to European Society of Cardiology (ESC) diagnostic criteria.
C
Comparator
Clinician-assigned diagnosis of HFpEF.
O
Outcome
Detection rate of HFpEF, hospitalization frequency, and 5-year mortality.
Main Result
Absolute Event Rate: 75.4% vs 8.3%
Natural language processing applied to electronic health records can identify a large proportion of patients with undiagnosed HFpEF who meet ESC criteria and are at high risk of mortality.
Cite This Study
Wu et al. (2023) conducted a cohort in Heart failure with preserved ejection fraction (HFpEF) (n=8,606). Natural language processing (NLP) pipeline vs. Clinician-assigned diagnosis was evaluated on Identification of HFpEF among patients with HF and LVEF ≥50%. An NLP pipeline applied to EHR data revealed that 75.4% of patients with HF and LVEF ≥50% met ESC criteria for HFpEF without a formal diagnosis, whereas only 8.3% had a clinician-assigned diagnosis.