LAUSR

March 19, 2019 1561 McVin Hua Heng Cheen, BSc (Pharm) (Hons), PhD To the Editor: I read with great interest the article by Siontis et al1 published in Circulation. As mentioned in the commentary, current randomized trials of apixaban to date have excluded patients with severe and end-stage kidney disease.2 This study provides new insights into the potential superior safety and effectiveness of apixaban in comparison with warfarin for stroke prevention in patients with end-stage kidney disease who have atrial fibrillation. In addition to the limitations highlighted by Hylek,2 an important bias, known as immortal time bias, could have exaggerated the benefits of apixaban. This was highlighted recently by Suissa,3 who illustrated how immortal time bias could have exaggerated the real-world benefits of sodium-glucose cotransporter 2 inhibitor in the CVD-REAL study (Comparative Effectiveness of Cardiovascular Outcomes in New Users of SGLT-2 Inhibitors).4 Siontis et al1 could have introduced the same bias to their study by the approach used to form the cohorts. To maximize the capture of new apixaban prescriptions, the authors assigned patients who were originally prescribed warfarin and then switched to apixaban to the apixaban group with time 0 as the date of the apixaban prescription. This introduced immortal time bias by omitting the time between the first warfarin prescription and the first apixaban prescription from the design and analysis. This time is termed immortal because patients had to be alive to receive their first apixaban prescription. Instead of omitting this immortal time, it should be classified as warfarin exposed until the first apixaban prescription. Omitting this immortal time could have led to the increased rate of stroke/systemic embolism, bleeding, and death in the warfarin group, because the denominator to compute the person-time exposed did not include the immortal time. Although Siontis et al have performed sensitivity analysis by excluding patients who switched from warfarin to apixaban, the problem of immortal time bias could not be eliminated because the immortal time on warfarin was still omitted from the analysis. Moreover, exclusion of 24.7% of patients in the apixaban group who were switched from warfarin could have reduced the statistical power of the analysis. To overcome potential immortal time bias, I propose 2 approaches to strengthen the analysis: 1. Modify the Cox regression analyses to incorporate time-dependent exposure to allow classifying the immortal time as warfarin-exposed until the start of apixaban. 2. Adopt a prevalent new-user design with time-conditional propensity scores.5 As observed by Siontis et al,1 there was a steep increase in the number of new apixaban prescriptions during the study period, with >1 in 4 new anticoagulant prescriptions in 2015 for apixaban. Although observational studies can provide important real-world evidence for the benefits of apixaban, careful handling of immortal time bias is warranted to avoid overexaggerating its benefits. © 2019 American Heart Association, Inc. LETTER TO THE EDITOR