Case Study

Using machine learning to predict anticoagulation control in atrial fibrillation


Warfarin remains a widely used anticoagulant for the prevention of stroke and systemic embolism in patients with atrial fibrillation in the UK, despite the availability of newer direct oral anticoagulants (DOAC).

Patients treated with warfarin require regular and frequent monitoring, which is resource intensive and a burden for patients.

Our client, who market a DOAC, wanted to develop a method to accurately predict which patients are likely to have suboptimal control with warfarin and may be better suited to DOAC therapy.


Retrospective cohort study of adult patients treated with warfarin using linked primary and secondary care data.

Various machine learning techniques were explored to predict suboptimal anticoagulation control.

Baseline and time-varying data were employed.

Patient records representing unique lines of warfarin therapy were separated into training (70%) and holdout sets (30%) for model training and testing.


Machine learning algorithms displayed clinically useful ability to predict patients who are at greater risk of suboptimal control.

Addition of time-varying data to the algorithm improved predictive performance.

The algorithms provide improved predictive tools for identifying patients who may benefit from more frequent monitoring or switching to alternative therapies not requiring dose adjustments.

The study culminated in publication in peer-reviewed journals:

  1. Hill et al. PLoS ONE 2019;14(11): e0224582
  2. Hill et al. J Med Econ 2020;23:4, 386-393
  3. Hill et al. Contemp Clin Trials 2020;99:106191
  4. Gordon et al. Inform Med Unlocked 2021;25:100688
  5. Hill et al. Value Health 2022;25(1):S111