Intended for healthcare professionals

CCBYNC Open access
Research

Risk of stroke and transient ischaemic attack in patients with a diagnosis of resolved atrial fibrillation: retrospective cohort studies

BMJ 2018; 361 doi: https://doi.org/10.1136/bmj.k1717 (Published 09 May 2018) Cite this as: BMJ 2018;361:k1717

Opinion: The importance of asking the right research question

Re: Risk of stroke and transient ischaemic attack in patients with a diagnosis of resolved atrial fibrillation: retrospective cohort studies

One of the goals of this study was to determine whether the rate of stroke and TIA differed between two groups of patients with "resolved" atrial fibrillation: those with and without recurrent atrial fibrillation. However, the authors did not specify the analytic approach used to try to achieve this goal. This is a concern because of the potential for "immortal time bias" (1) when dealing with definition of exposure categories based on changes in exposure status that can occur after the beginning of follow-up for outcome events.

One means of avoiding this source of bias would have been to tabulate the incidence of stroke and TIA in patients with a recurrence beginning at the time of the recurrence, with the experience of such patients prior to that time combined with the experience of patients in whom no recurrence had occurred. Because follow-up for stroke and TIA did not commence until 180 days had elapsed from the time the atrial fibrillation had been judged to have been "resolved", another successful approach would have been to define the patients with and without recurrence based only on those recurrences which took place during those 180 days, However, if for patients with a recurrence of atrial fibrillation that took place beyond 180 days the calculated rate of stroke and TIA included their time at risk after 180 days but PRIOR to the recurrence, that rate would be spuriously low: no strokes or TIAs could have occurred prior to the recurrence. (Had such an event occurred, it would have been assigned to the group of patients who had not had a recurrence.) By the same token, the failure to assign the person-time accrued after 180 days but prior to the development of recurrence would have led to a spuriously high rate in patients with no recurrence of atrial fibrillation.

I would encourage Adderley et al. to clarify how the rates of stroke and TIA were calculated in patients with "resolved" atrial fibrillation who did and did not have recurrent fibrillation.

1. Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol 2008;167:492-9.

Competing interests: No competing interests

24 May 2018
Noel S Weiss
Professor
University of Washington
Box 357236, Seattle, WA USA