Purpose: To develop a clinical decision support tool to assist physicians with atrial fibrillation therapy design for primary prevention of stroke.
Method: The long-term antithrombotic therapy with warfarin reduces the relative risk of stroke from atrial fibrillation by approximately 65%, while increasing the bleeding risk. Aspirin constitutes a less aggressive therapy option with lesser impact on the stroke and bleeding risks. Given the potential benefits and risks of these options, and recognizing each patient's case-mix and clinical variables, the therapy choice decisions are critical. We formulate the problem as a Markov Decision Process (MDP) so as to maximize the patient's expected remaining QALY (i.e., quality-adjusted life years). The decision alternatives at each time epoch are warfarin, aspirin and “no medication”. The risk score and the bleeding score of the patient are the state variables of the MDP. The former is estimated using the CHADS scale, and the latter is accessed via the Beyth framework. The MDP transition probabilities are estimated using charts of the 950+ atrial fibrillation patients from the anticoagulant clinic of Montreal Jewish General Hospital.
Result: Using the MDP model, we identified the optimal therapy choice based on the patient's age, stroke score (s) and bleeding score (b). The Table below depicts our results. For example, our framework suggests that an 85 year old patient with s=1 and b=1 should be started on Warfarin (W), whereas if the patient was a year older with the same risk factor Aspirin (A) would be the better choice. We estimated the expected remaining QALY of the 950+ patients by the use of our methodology. Comparing the results with the patient files showed that significant improvements in may be achieved. Especially, for the younger group of patients with moderate stroke risk (with stroke score 1 or 2) and high bleeding risk (with bleeding score 2 and 3), these potential improvements varied between 16% and 37%.
Conclusion: The current clinical guidelines for atrial fibrillation therapy are based on only the stroke risk of the patient. Our research shows that it is possible to improve the health outcomes by also incorporating the patient's bleeding risk in the decision process.
