New artificial intelligence prediction model using serial prothrombin time international normalized ratio measurements in atrial fibrillation patients on vitamin K antagonists: GARFIELD-AF
Key result
A new AI model using serial PT-INR measurements within 30 days of starting VKAs predicted major bleed (c-statistic 0.75), stroke/SE (0.70), and death (0.61) better than time in therapeutic range.
Why the study?
Most clinical risk stratification models rely on single time-point measurements rather than serial measurements, whereas AI can predict outcomes from multi-dimensional datasets.
Does an artificial intelligence prediction model using serial PT-INR measurements improve the prediction of clinical outcomes in atrial fibrillation patients on vitamin K antagonists compared to time in therapeutic range?
Population
Patients with newly diagnosed AF treated with VKAs having at least three PT-INR measurements within 30 days
Comparison
AI prediction model using serial PT-INR measurements vs time in therapeutic range
Design
Registry-based prediction model derivation and validation study
Follow-up
Up to 1 year
Authors
Discussion
Loading...
Member takes
Implication
May aid early risk stratification in new VKA users; hypothesis-generating and requires prospective validation before practice change.
Study Design
Type
Cohort (n=4,708)
Multicenter
Yes
Structured PICO
Does an artificial intelligence prediction model using serial PT-INR measurements improve the prediction of clinical outcomes in atrial fibrillation patients on vitamin K antagonists compared to time in therapeutic range?
P
Population
4,708 patients with newly diagnosed atrial fibrillation treated with vitamin K antagonists who had at least three PT-INR measurements in the first 30 days, followed for up to 1 year.
E
Exposure
Artificial intelligence prediction model using serial PT-INR measurements within days 0-30 after starting treatment
C
Comparator
Time in therapeutic range (TTR)
O
Outcome
Prediction of major bleed, stroke/systemic embolism (SE), and all-cause death over days 31-365hard clinical
Main Result
Effect estimate: c-statistic 0.75 for major bleed, 0.70 for stroke/SE, 0.61 for death
An AI model using serial PT-INR values within the first month of VKA therapy predicts major bleeding, stroke/SE, and death better than time in therapeutic range in patients with atrial fibrillation.
Cite This Study
Goto et al. (2019) conducted a cohort in Atrial fibrillation (n=4,708). AI prediction model using serial PT-INR measurements vs. Time in therapeutic range was evaluated on Major bleed, stroke/systemic embolism (SE), and all-cause death (c-statistic 0.75 for major bleed, 0.70 for stroke/SE, 0.61 for death). A new AI model using serial PT-INR measurements within 30 days of starting VKAs predicted major bleed (c-statistic 0.75), stroke/SE (0.70), and death (0.61) better than time in therapeutic range.