Author + information
- Neil J. Wimmer, MD, MSc†,
- Frederic S. Resnic, MD, MSc‡,
- Laura Mauri, MD, MSc†,
- Michael E. Matheny, MD, MS, MPH§ and
- Robert W. Yeh, MD, MSc‖∗ ()
- †Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts
- ‡Lahey Clinic, Burlington, Massachusetts
- §Vanderbilt University Medical Center, Nashville, Tennessee
- ‖Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts
- ↵∗Division of Cardiology, Massachusetts General Hospital, 55 Fruit Street, GRB 800, Boston, Massachusetts 02114
To the Editor:
There has been growing interest in the use of the transradial approach for percutaneous coronary intervention (PCI), supported by randomized clinical trials involving experienced transradial operators (1,2). However, the assessment of transradial PCI in routine practice has necessarily relied on observational comparisons (3), which may be susceptible to confounding even with modern statistical methods. In any observational study, there is always the possibility that unmeasured confounders bias the intended comparison regardless of the statistical approach used, including the use of advanced multivariable modeling or propensity score methods. Also, observational methods are limited in their inability to demonstrate causality.
Recently, the use of “falsification endpoints” has been highlighted as an underutilized method to assess for residual confounding in observational studies (4). A “falsification hypothesis” is a claim, distinct from the main hypothesis being tested, that researchers believe is highly unlikely to be causally related to the intervention under study, similar to a “negative” control experiment in a laboratory. Most observational studies, including those focused on this clinical question (3), have not reported falsification endpoints. Despite the common parlance of the word “falsification,” the term “falsification endpoints” or “falsification hypothesis” does not imply data falsification or attempts on the part of the investigators to mislead the reader. Rather, these endpoints are intended to test, within a single study, whether significant associations between a treatment and outcome are susceptible to residual confounding.
We sought to compare outcomes for transradial and transfemoral PCI using modern statistical adjustment methods and utilizing a falsification endpoint. Our primary hypothesis was that transradial PCI reduces access site bleeding compared to transfemoral PCI in routine clinical practice. However, because transradial PCI would not be expected to differentially influence bleeding apart from at the arterial access site, nonaccess site bleeding was pre-specified as the falsification endpoint.
Data were obtained from 5 Massachusetts hospitals from 2008 to 2011. De-identified, patient-level data were extracted using National Cardiovascular Data Registry (NCDR) CathPCI forms in patients undergoing PCI without mechanical circulatory support. The dataset contained clinical and procedural elements and follow-up for in-hospital complications. Bleeding endpoints were defined as access versus nonaccess site.
Estimates of the impact of transradial versus transfemoral PCI were determined using 3 methodologies. In method 1, we generated standard logistic regression models adjusting for baseline bleeding risk using a validated PCI bleeding model from NCDR (5). In method 2, we performed 1:1 matching of transradial and transfemoral PCI patients based on propensity score accounting for 19 demographic, clinical, and procedural covariates including anticoagulation strategy. In method 3, we performed adjustment based on inverse probability treatment weights (IPTW) generated from the propensity score model. Analyses were performed using STATA version 11.2 (College Station, Texas).
A total of 17,509 patients underwent PCI, with 17.8% of procedures performed via the transradial approach. A total of 240 (1.4%) patients had either access site bleeding (102 events) or nonaccess site bleeding (144 events). Although there were statistically significant differences in 15 of 17 baseline characteristics examined, patients undergoing transfemoral and transradial PCI were balanced on all characteristics after using propensity score methods (standardized differences <10% for all variables).
The results of the 3 different analyses are presented in Figure 1. Statistically significant reductions in access site bleeding were seen with transradial PCI using all 3 methodologies. However, statistically significant reductions in nonaccess site bleeding were also seen with transradial PCI with all 3 methodologies, suggesting residual confounding.
In these analyses, we identified a relationship between transradial PCI and nonaccess site bleeding that was not identified in randomized clinical trials and that is most likely due to the inability of our methods to fully adjust for unmeasured patient differences. While it would be possible to adjust the results of the primary hypothesis based on the magnitude of confounding of the falsification endpoint, we would not recommend this approach until the validity of such methods has been further explored. These data highlight the dangers of assuming that modern statistical methods alone fully adjust for confounding in nonrandomized data. In observational comparisons such as the comparison of transradial and transfemoral PCI in which differences in unmeasured patient characteristics are likely to be significant based on treatment selection, the use of falsification endpoints may be useful when available.
Please note: Dr. Wimmer is supported by NIH T32-HL00760. Dr. Matheny is supported by Veterans Administration HSR&D Career Development Award CDA-08-020. Dr. Yeh receives research support from the Harvard Clinical Research Institute. Dr. Resnic is a consultant for Medtronic and St. Jude Medical. Dr. Mauri has received research support from Abbott, Boston Scientific, Medtronic, Cordis, Bristol-Myers Squibb, sanofi-aventis, Eli Lilly, and Daichii Sankyo; and serves as a consultant for Medtronic.
- American College of Cardiology Foundation
- Romagnoli E.,
- Biondi-Zoccai G.,
- Sciahbasi A.,
- et al.
- Baklanov D.V.,
- Kaltenbach L.A.,
- Marso S.P.,
- et al.
- Mehta S.K.,
- Frutkin A.D.,
- Lindsey J.B.,
- et al.