Results
For the analyses, we included 1,363, 165, and 23 warfarin users who had been exposed to simvastatin, atorvastatin, and rosuvastatin, respectively, between 2000 and 2015.
For patients treated with simvastatin, atorvastatin and rosuvastatin, the median ages were 72 years (interquartile range (IQR) 64-79 years), 70 years (IQR 63-76), and 74 years (IQR 64-80), while the proportion of males were 58%, 62%, and 47%, respectively.
INR values increased slightly after initiation of simvastatin treatment with a peak after about 4 weeks (Figure 1). Initiation of simvastatin was associated with an increase in mean INR from 2.40 to 2.71, corresponding to an increase of 0.32 (95%CI: 0.25-0.38, p<0.001) while the median INR change was 0.2 (IQR -0.3-0.8) (Figure 2). During a time window of 1-4 weeks before initiation of statin treatment, 3.4% of patients had at least one INR measurement above 4. This proportion increased to 9.0% during 3-6 weeks after initiation of statin treatment (p<0.01). Similarly, the proportion of the patients with an INR > 5 increased from 1.3% before initiation of statin treatment to 3.2% after (p<0.01).
When stratifying by simvastatin dosage, we found that initiation of both high-dose (≥40mg) simvastatin (0.33, 95%CI 0.25-0.42) and low-dose (<40mg) simvastatin (0.29, 95%CI 0.20-0.38) were associated with a similar modest increase in mean INR. The median change in INR was 0.2 (IQR -0.3-0.9) and 0.3 (IQR -0.2-0.8) for patients receiving high and low simvastatin dose, respectively (Figure 2).
Considering atorvastatin, initiation of treatment was associated with an increase in mean INR from 2.42 to 2.69 (change 0.27, 95%CI 0.12-0.42, p<0.01), while for rosuvastatin was associated with a corresponding increase from 2.31 to 2.61 (change 0.30, 95%CI -0.09-0.69, p=0.121). Analyses of high dose vs. low dose were prohibited by low statistical power for both atorvastatin and rosuvastatin.
Sensitivity analyses excluding patients filling prescriptions for other potentially interacting drugs (n=77), and patients with a mechanical heart valve (n=110), and INR measurements labeled as potentially imprecise (n=81 patients with no alternative measurements) yielded virtually unchanged estimates (data not shown).