Predicting heart failure with preserved ejection fraction in a tertiary hospital in the Netherlands: the IDENTIFY-HFpEF study
A Uijl, M J Boonstra, S Van Der Zwaard, M C Van De Veerdonk, J W J Beulens, F W AsselbergsAbstract
Background
Heart failure with preserved ejection fraction (HFpEF) represents a substantial proportion of heart failure cases but is frequently under-recognised in clinical practice. Diagnostic challenges, together with incomplete or non-standardised documentation of key clinical and imaging data in electronic health records (EHRs), contribute to delayed identification and suboptimal management, even in patients regularly seen in hospital settings.
Purpose
This study aimed to develop a prediction model for HFpEF identification using routinely available clinical variables in a tertiary hospital population, to support early recognition and potential electronic alert systems for at-risk patients.
Methods
The study population included all adult patients with an echocardiogram with at least 2 echo parameters recorded in a tertiary hospital in the Netherlands between 1/1/2018 - 13/11/2025. We excluded all patients with a clinician verified reduced ejection fraction (EF) or mildly reduced EF diagnosis, any patients with an echocardiogram with EF measured <50% or ARNI (Angiotensin Receptor Neprilysin Inhibitor) users. We identified patients with HFpEF based on 1) a clinician-verified HFpEF diagnosis recorded for the Dutch heart failure registry; 2) a data-approximated diagnosis including any patient with a heart failure diagnosis without prior recorded EF <50%. Multivariable logistic regression was used to assess the association between 26 predictors and HFpEF. The final model included predictors selected using bidirectional stepwise selection minimising Akaike’s Information Criterion. Missing data in predictors was imputed using multiple imputation in 10 imputations. Model results were presented as odds ratios (OR) with 95% confidence intervals (95%CI) in a forest plot. Discriminative performance was assessed using the receiver operating characteristic curve and corresponding c-statistic.
Results
Out of 16.091 patients with an echocardiogram, 1396 patients were identified with HFpEF. Strongest predictors were use of treatments including loop diuretics (OR 3.5 [95%CI 3.0-4.1]) and MRA (OR 2.1 [95%CI 1.7-2.5]), presence of comorbidities including CKD (OR 2.2 [95%CI 1.9-2.6]), COPD (OR 1.8 [95%CI 1.6-2.1]) and AF (OR 1.7 [95%CI 1.5-2.0]) (Figure 1). The model showed moderate discrimination (c-statistic 0.757 [95%CI 0.755 – 0.760]). The threshold maximizing sensitivity and specificity (Figure 2) yielded a sensitivity of 82%, specificity of 74%, and a number-needed-to-alert of 4.3, out of 4 potential alerts based on this prediction model, 1 patient will have HFpEF.
Conclusions
Routinely available clinical variables provide moderate predictive accuracy for HFpEF suggesting that simple models, without echocardiographic information, can support case identification in hospital populations. While performance is promising, further refinement with internal and external validation are needed to improve reliability and generalisability.Forestplot of predictors for HFpEFFor image description, please refer to the figure legend and surrounding text.Threshold of HFpEF predictionFor image description, please refer to the figure legend and surrounding text.