Universal Clinician Device for improving risk prediction and management of patients with atrial fibrillation: an assumed benefit analysis

Abstract

Aim: Atrial fibrillation (AF) management guidelines advise using risk tools to optimize AF treatment. This study aims to develop a dynamic and clinically applicable digital device to assess stroke and bleeding risk, and to facilitate outcome improvements in AF patients. The device will provide tailored treatment recommendations according to easily attainable individual patient data.

Methods and results: This Universal Clinician Device (UCD) was created using the GARFIELD-AF registry using a split sample approach. The GARFIELD-AF risk tool was adapted with two modifications. First, predictors with ≥1000 missing data points were separated, allowing expected risks estimation. Second, recommendations for modifiable risk factors and associated 2-year outcome estimates were incorporated. Outcomes of interest were all-cause mortality, non-haemorrhagic stroke/systemic embolism (SE), and major bleeding. All patients were randomized to a derivation (n = 34853) and validation cohort (n = 17165). In the derivation cohort, predictors were identified using least absolute shrinkage and selection operator regression. Cox models were fitted with the selected parameters. The UCD demonstrated superior predictive power compared with CHA₂DS₂VASc for all-cause mortality [0.75(0.75-0.76) vs. 0.71(0.70-0.72)] and non-haemorrhagic stroke/SE [0.68(0.66-0.70) vs. 0.65(0.63-0.67)], and with HAS-BLED for major bleeding [0.69(0.67-0.71) vs. 0.64(0.62-0.65)]. Universal Clinician Device recommendations reduced all-cause mortality (8.45-5.42%) and non-haemorrhagic stroke/SE (2.58-1.50%). Patients with concomitant diabetes and chronic kidney disease benefitted further, reducing mortality risk from 13.15% to 8.67%. One-third of patients with a CHA₂DS₂VASc score of >1 had the lowest risk of stroke.

Conclusion: The UCD simultaneously predicts mortality, stroke, and bleeding risk in patients using easily attainable individual clinical data and guideline-based optimized treatment plans.

Clinical trial registration: URL: http://www.clinicaltrials.gov. Unique identifier for GARFIELD-AF: NCT01090362.

Keywords: Atrial fibrillation; Diabetes; Management guidelines; Personalized medicine; Risk prediction; Universal Clinician Device.

Graphical Abstract

The UCD uses patient data, clinical and current treatment characteristics to provide an individual risk assessment of the current risk and the risk with optimal therapy. Applying this device in clinical practice provides significant improvement potential vs. current practice.

Figure 1

Calibration of the Universal Clinician Device withing the derivation set for (A) all-cause mortality, (B) non-haemorrhagic stroke/systemic embolism, and (C) major bleeding, and within the vadlidation set for (A) all-cause mortality, (B) non-haemorrhagic stroke/systemic embolism, and (C) major bleeding at 2 years of follow-up in the GARFIELD-AF population. The dashed line represents the predicted risk over 2 years. The dots represent the observed rate of outcomes over 2 years by decile.

Figure 2

Density function of the Universal Clinician Device risk models for all-cause mortality, non-haemorrhagic stroke/systemic embolism, and major bleeding at 2 years in the full GARFIELD-AF population (not imputed). The X-axis has been truncated at 20%.

Figure 3

Distribution of the CHA₂DS₂VASc score categories according to the deciles of Universal Clinician Device 1 year stroke risk device in the full GARFIELD-AF population.

Figure 4

Estimated average rate for outcomes over 2 years of follow up according to the observed real-world GARFIELD-AF population, if all patients followed CHA₂DS₂VASc guidelines for treatment with oral anticoagulant, and if all patients followed treatment guidelines according to the Universal Clinician Device.

Figure 5

Sample output from the Universal Clinician Device, providing individualized recommendations and the adjusted outcomes risks.