Author + information
- Timothy M. Fernandes, MD, MPH∗ ( and )
- Richard H. White, MD
- ↵∗Division of Pulmonary, Critical Care and Sleep Medicine, University of California, San Diego, 9500 Gilman Drive, Mail Code #0643, La Jolla, California 92093-0643
We commend the RIETE (Computerized Registry of Patients with Venous Thromboembolism) group for attempting to analyze the benefits of use of an inferior vena cava (IVC) filter after acute pulmonary embolism (PE) using their large prospectively collected registry (1). Their propensity-matched analysis showed a lower risk-adjusted PE-related mortality rate in patients who received an IVC filter compared those who did not. However, we wish to point out a significant source of bias in this observational study that cannot be adjusted for with the use of propensity matching.
The bias we refer to is termed “immortal time bias” (2). This bias is frequently encountered when analyzing the effectiveness of an intervention, such as use of an IVC filter, when using observational data. The placement of a filter also depends on meeting some clinical criteria, having a physician available to place the filter, and the patient’s clinical condition. In more critically ill patients, insertion of a filter may not be possible and death may occur before placement of the filter. In analyzing the data, if simply comparing the outcomes of patients with an IVC filter versus those without, the results are biased. All patients who received a filter were alive at the time of the procedure (hence, they were “immortal” up to the time the filter was placed), whereas the patients who did not receive a filter included those who may have died before a filter could be placed. This distinction between the 2 groups is important; upwards of 30% of PE-related mortality occurs within the first 24 h of hospitalization (3). In the study by Muriel et al. (1), in the patients who received an IVC filter, the authors started the clock on the primary outcome of 30-day mortality on the day the filter was placed; however, in the patients who did not receive a filter, it was considered to be when anticoagulation was started. A better way of analyzing these data would be to use the admission date or the date of diagnosis of PE as the anchor time and model death due to PE by using a Cox proportional hazard model, entering use of an IVC filter as a time-dependent covariate. The hazard associated with use of a filter is compared with the hazard of not using a filter in patients who are alive on the same day. Using this methodology, all patients who were not treated with a filter but who died early after diagnosis of PE are excluded from calculation of the relative hazard associated with use of a filter. Alternatively, patients could be matched on both propensity score and being alive on the day of placement of the filter, thereby excluding those who died before the intervention could be completed. We suggest that the authors use one of these suggested methodologies to determine if the hazard ratio for PE-related mortality is similar or significantly different between those with and without a filter.
The authors raise several interesting points, but we ultimately agree with the conclusion of Dr. Morris in the accompanying editorial (4). The only way to truly settle the controversial issue of whether filters are beneficial in patients who cannot be anticoagulated is by conducting a well-designed clinical trial. However, performing such a trial will be difficult. Informed consent will have to be obtained quickly, followed by rapid randomization and expeditious insertion of the IVC filter. Until that time, clinicians must try to make inferences from observational studies that have unmeasured confounders as well as immortal time bias. We believe the results of such analyses should not have any major impact on venous thromboembolism guidelines or clinical practice.
- American College of Cardiology Foundation
- Muriel A.,
- Jiménez D.,
- Aujesky D.,
- et al.
- Linda E.L.,
- James A.H.,
- Abbas K.,
- Samy S.
- Morris T.A.