Identification of a real-world clinical model associated with atrial fibrillation in a large french primary care database

Abstract

Abstract Funding Acknowledgements Type of funding sources: Private company. Main funding source(s): Bristol Myers Squibb and Pfizer. Background Undiagnosed atrial fibrillation (AF) is associated with a higher risk of stroke and mortality. To date, the 2020 European Society of Cardiology (ESC) guidelines recommend opportunistic screening for AF in patients ≥65 years (Class I), or with hypertension (HTN) (Class I), or obstructive sleep apnea (OSA) (IIa), and systematic ECG screening in patients ≥75 years or at high risk of stroke defined by the CHA2DS2-VASc score (Class IIa).1 However, in a primary care practice, these criteria cover a large number of patients. Purpose Evaluate the performance of several clinical markers or combination of clinical markers to identify patients at high-risk of AF in a primary care database. Methods This retrospective national study used patients’ data from the French electronic health records THIN® database of 2 000 General Practitioners (GP). Patients included had visited their GP in 2019 and were followed for ≥ 3 years, were aged ≥ 50 with ≥ 1 risk markers potentially associated with AF (HTN, diabetes, obesity, dyslipidaemia, smoking, alcohol dependence, cardiopathy, heart failure (HF), vascular disease, OSA, chronic obstructive pulmonary disease, inflammatory disease, thyroid or renal dysfunction). Non-AF patients with anticoagulant treatment were excluded. The tested variables and models included the 2020 ESC guidelines criteria for AF screening (age ≥65, age ≥ 75, high stroke risk defined as CHA2DS2-VASc ≥ 2 (male) or ≥ 3 (female), HTN, OSA), the CHA2DS2-VASc taken as a score as well as the multivariate analysis of its components (congestive HF, HTN, age ≥ 75, diabetes, stroke, vascular disease, age 65-74, female sex), and an ad hoc model based on the combination of risk markers associated with AF (15 variables based on cardiovascular and non-cardiovascular comorbidities; Figure 1). Multivariate models were performed with logistical regression. Results Among 573 539 screened patients (mean follow-up 13.5 years), 30 476 had AF (5.3%). The ad hoc model using the 15 variables showed the highest performance with AUC 0.80 (Se 69%, Sp 77%) (Figure 1). In this model, all variables were significantly associated with AF except diabetes, dyslipidaemia, and HIV. Age ≥ 75, HF, age 65-74 and male sex were the most significant risk markers associated with AF (Figure 1). In the CHA2DS2-VASc model, AUC was 0.78 for multivariate analysis (Se 69%; Sp 72%), and 0.71 for the score result (Se 77%, Sp 56%). Regarding the ESC guidelines criteria, age ≥ 75 years had the best performance with AUC 0.69 (Se 62%, Sp 75%), then age ≥65 (AUC 0.67), high stroke risk (AUC 0.66), versus AUC < 0.60 for HTN and OSA. Conclusion In this study on a large primary care database, the ad hoc model had the highest performance, sensitivity and specificity. In the future, it might help GPs to identify patients at high-risk of AF in clinical practice and to propose targeted preventive interventions and screening.