Identifying undertreated heart failure patients in a respiratory ward using AI-ECG
E Stenhede, K Berge, O Orstavik, H Schirmer, A RanjbarAbstract
Background/Introduction
Artificial intelligence-enabled electrocardiography (AI-ECG) has demonstrated the ability to detect clinical and subclinical heart failure (HF), and may reveal patients receiving suboptimal treatment in non-cardiology settings.
Purpose
HF is common among patients admitted to non-cardiac wards, yet often remains unrecognised and undertreated.
Methods
We did a retrospective cohort study of adults admitted to the Departments of Cardiology or Pulmonology at a tertiary-care hospital network from January 1, 2023, to June 6, 2025. A previously validated AI-ECG model for HF was applied to the first admission ECG. The primary outcome was all-cause mortality. We compared outcomes and treatment patterns across departments to identify patients with probable HF receiving suboptimal therapy. Guideline-directed medical therapy (GDMT) comprised beta blockers, RAAS inhibitors, MRAs, and SGLT2 inhibitors; loop diuretics were analysed separately. Multivariable Cox proportional hazards models were fitted with continuous AI-ECG risk, GDMT tier (ranging from 0 to 4 medicines), and loop diuretics, adjusted for sex, department, and comorbidities, and stratified by age.
Results
We included 13552 patients, 2864 in pulmonology and 10688 in cardiology. For each patient, the first available ECG recorded during the study period was used for AI-ECG risk assessment. Patients admitted to the pulmonology ward were older, more often female, and had a higher prevalence of chronic obstructive pulmonary disease and lung cancer than those in cardiology. AI-ECG-predicted HF risk was associated with mortality in both departments, see Figure, and absolute mortality was markedly higher in pulmonology. Across increasing AI-ECG risk strata, the use of diagnostic testing and GDMT rose, yet remained lower in pulmonology than in cardiology (p<0.001). Conversely, utilisation of loop diuretics was more frequent in pulmonology (p<0.001). In multivariable Cox models adjusted for age, sex, comorbidities, and department, higher AI-ECG risk was associated with mortality (HR 2.92 [2.50-3.42]), and greater GDMT use was associated with progressively lower mortality (GDMT 4 vs 0, HR 0.46 [0.37-0.56]), see Table.
Conclusions
AI-ECG identifies patients at high risk of HF who are frequently admitted to non-cardiology wards and receive less GDMT. The strong association between AI-ECG risk of HF and mortality, together with the marked survival benefit from comprehensive GDMT, suggests that AI-ECG could help identify unrecognised HF and guide targeted treatment to improve outcomes for HF patients in respiratory and other non-cardiac wards.Estimated mortality by AI-ECG riskFor image description, please refer to the figure legend and surrounding text.Adjusted Cox PH modelFor image description, please refer to the figure legend and surrounding text.