CardioExchange Archive

Disclosure: Larry Husten works with Dr. Harlan M. Krumholz, who is mentioned in this post, on CardioExchange. Dr. Krumholz has recused himself from all editorial discussions and decisions about CardioExchange’s coverage of this story.

Last year, in what may have been an unprecedented action, a paper on the potential cardiovascular benefits of transcendental meditation in African Americans was withdrawn by the editors of the Archives of Internal Medicine only 12 minutes before the paper’s scheduled publication in that journal. No definitive explanation was ever provided, although the journal editors and the paper’s authors said that the action was prompted by last-minute questions from reviewers at the NIH, which had helped to fund the study.

Now, a new version of the paper has been published in Circulation: Cardiovascular Quality and Outcomes. The first author is Robert H. Schneider, of the Institute for Natural Medicine and Prevention at the Maharishi University of Management, in Fairfield, Iowa. Schneider’s coauthors are from the same institution and from the Department of Medicine at the Medical College of Wisconsin in Milwaukee. It is clearly the same study of 201 African American patients randomized to transcendental meditation or health education and followed for 5.4 years, although some of the numbers have changed in important ways between the earlier and later versions. (I had obtained the earlier version as part of Archives‘ routine early release of articles to the media.)

One change involves the primary endpoint. The new paper in Circulation:CV Quality & Outcomes reports that of 52 primary-endpoint events (a composite of death, MI, or stroke), 20 occurred in the meditation group and 32 in the health-education control group. By contrast, the previous Archives version identified 51 primary-endpoint events: 20 in the meditation group and 31 in the control group.

In both papers, the difference in the primary endpoint was reported not to achieve statistical significance until after the investigators adjusted for baseline differences. The unadjusted hazard ratio (HR) for the new study was 0.64 (95% CI, 0.37–1.12; P=0.12). After adjustment for age, gender, and use of lipid-lowering medications, the HR dropped to 0.52 (95% CI, 0.29–0.92; P=0.025), apparently a statistically significant advantage of meditation.

Here are some other questions that I put directly to Circulation’s editors:

A. Were staff at AHA or Circulation: CV Quality & Outcomes ever aware that an earlier version of this paper was to appear in the Archives of Internal Medicine and withdrawn only 12 minutes before it was scheduled to go live? Should this information have been disclosed by the study authors when submitting the paper?

B. The original publication in Archives appeared to have been cancelled because of questions raised by NIH reviewers. Have the Circulation: CV Quality & Outcomes editors received any assurance that these questions have been addressed in the new paper?

C. This trial was started back in 1998 but was not registered on ClinicalTrials.gov until February 2011 (shortly before the intended Archives publication). Why wasn’t it registered earlier?

D. ClinicalTrials.gov identifies June 2007 as the final data-collection date for the primary-outcome measure. Why do the primary-endpoint numbers differ between the Archives version and the newer version of this paper?

I asked Dr. Harlan M. Krumholz, editor of Circulation: Cadiovascular Quality and Outcomes and of CardioExchange, to comment on this story. Here is the statement he offered on November 13:

We had no prior knowledge of what transpired with the Archives of Internal Medicine. The Schneider paper went through rigorous peer review, statistical review, and editorial discussions — and the authors of the article were responsive to the review process. As a result, the paper was accepted for publication, and we are going ahead as planned. If you have further questions, we suggest you contact the researchers directly.

Also on November 13, I separately asked Sanjay Kaul, of UCLA’s School of Medicine, for perspective on the paper. Not having seen Dr. Krumholz’s above response, Dr. Kaul enumerated the following points:

1. All-cause mortality constituted nearly 80% (41 of 52 events) of the primary composite endpoint. This is highly unusual, as most trials yield a higher proportion of nonfatal events than fatal events. This leads me to question the fidelity of the adjudication process to pick up nonfatal events.

2. It is difficult to attribute a nearly 50% reduction in hard CV outcomes to a <5-mm-Hg difference in systolic BP. Total anger was significantly reduced, but attributable risk of anger or emotional distress for MI or death is very small.

3. It is interesting to note that the study was originally powered for a 36% reduction in a broader endpoint but for a larger 50% reduction in the revised narrower endpoint. This is difficult to justify.

4. It is not clear whether the adjusted analyses were adjusted for multiple comparisons. Given the wide confidence interval and less than robust P-value, it is likely the results would be no longer significant.

5. Even if one accepts the adjusted P-value, the strength of evidence is not robust.

Update, November 18: An earlier version of this article contained several mistakes. First, the difference in the number of primary-endpoint events between the two papers is much smaller than I had stated. The original paper reported 51 primary-endpoint events (not 40, as I had misstated). The new paper reports 52 primary-endpoint events. I also incorrectly claimed that some  data reported in the original Archives paper were not included in the published Circulation: CV Quality & Outcomes paper. In fact, these data were reported in Figure 1. I apologize for this mistake and regret the errors. –Larry Husten