Development of a prediction model for heart failure with preserved ejection fraction (HFpEF ) in patients with type 2 diabetes in primary care
M J C Van Den Berg, A Uijl, R Meer, A G Hoek, K A Kortekaas, E Wierda, M Hollander, P J M Elders, J W J Beulens, M T BlomAbstract
Background
Patients with type 2 diabetes (T2D) have a twofold increased risk of heart failure (HF), particularly HF with preserved ejection fraction (HFpEF). In primary care, HFpEF is often only diagnosed after overt clinical signs develop, even in high-risk individuals. Early recognition of HFpEF in T2D patients is crucial in order to start treatment earlier and reduce adverse events and health care costs. The aim of this study is to create a prediction model for early detection of HFpEF and its precursor left ventricular diastolic dysfunction (LVDD) in patients with T2D in primary care.
Methods
A total of 844 T2D primary care patients without known HF from the Dutch Diabetes Care System cohort underwent transthoracic echocardiography between 2019 and 2022. Fifteen routinely collected clinical variables related to lifestyle, medical history and biomarkers were evaluated as potential predictors using a five-year look-back window. Logistic regression was employed to predict HFpEF as defined by the European Society of Cardiology (ESC)’21 criteria (model 1) and the Dutch General Practitioner guidelines for heart failure (NHG)’24 (model 2), whereas multinomial regression was utilized for LVDD according to the American Society of Echocardiography and European Association of Cardiovascular Imaging ASE/EACVI’25 guideline (model 3). Predictor selection was performed via backward selection (p.crit=0.10), and internal validation was conducted using bootstrapping.
Results
HFpEF prevalence was 37% (ESC) and 12% (NHG), while LVDD was present in 71%. The predictors included in Model 1 are current and former smokers, female sex, higher body mass index (BMI), older age, and a history of myocardial infarction. Model 2 additionally incorporates elevated systolic blood pressure and preserved kidney function as protective factors. The predictors for Model 3 comprise female sex, history of myocardial infarction, age, and systolic blood pressure, as well as the use of oral blood glucose-lowering medications and longer duration of diabetes. Models 1 and 2 demonstrated moderate discrimination (AUC 0.70 (95%-CI: 0.66-0.74) and 0.73 (95% CI: 0.68-0.78)) and excellent calibration (brier scores 0.205 and 0.096 respectively), both maintained after internal validation. Model 3 showed lower discrimination (AUC 0.68 (95%-CI: 0.64-0.72) for normal, 0.61 (95%-CI: 0.57-0.65) for mild and 0.61 (95%-CI: 0.57-0.66) for severe LVDD) and good calibration for normal and severe, but not mild, categories with an overall brier score of 0.208.
Conclusion
In conclusion, routinely available clinical variables enable moderate prediction of HFpEF, but not LVDD, in primary care T2D patients, while also providing insight into HFpEF risk factors in this population.For image description, please refer to the figure legend and surrounding text.For image description, please refer to the figure legend and surrounding text.