Retraction Watch

Tracking retractions as a window into the scientific process

Hypertension retracts paper over data glitch

without comments

The anticipation of having one’s blood pressure measured can cause it to spike.

So, evidently, can errors in data processing — on a national scale.

Hypertension, a journal published by the American Heart Association, has retracted a 2011 paper looking at the implications of blood pressure management guidelines after the authors discovered they had bungled the merging of their data files.

As the notice explains:

For the article by Bertoia et al, “Implications of New Hypertension Guidelines in the United States,”
which published ahead of print on July 18, 2011, and appeared in the September 2011 issue of the
journal (Hypertension. 2011;58:361–366; doi: 10.1161/HYPERTENSIONAHA.111.175463), the
authors discovered an error in the code for analyzing the data. The National Health and Nutrition
Examination Survey (NHANES) medication data file had multiple observations per participant and
was merged incorrectly with the demographic and other data files. Consequently, the sample size was
twice as large as it should have been (24989 instead of 10198). Therefore, the corrected estimates of
the total number of US adults with hypertension, uncontrolled hypertension, and so on, are significantly
different and the percentages are slightly different.

For these reasons, Hypertension requested that the authors resubmit a corrected version of this
manuscript. The new version went through the peer review process and was accepted for
publication.

Because there were several changes due to the correction of the data and addition of data
requested by the reviewers, the editors decided that retracting the original article was appropriate.

The revised article is being published and the original article is being retracted.
The revised version of this article is available at http://hyper.ahajournals.org/lookup/doi/
10.1161/HYPERTENSIONAHA.112.193714.

The authors sincerely regret the incorrect merging of their data and the possible confusion that
this has caused.

The glitch appeared to make a substantial difference in the reported rates of hypertension and related problems. According to the retracted paper,

Using the old definition of hypertension (>140/90 mm Hg), 98 million (21%) Americans have hypertension. Using the updated guidelines, an additional 52 million (11%) American adults now have elevated BP requiring treatment, for a total of 150 million adults (32%).

The corrected article, however, states that:

Using the 2011 American College of Cardiology Foundation/AHA hypertension guidelines (140/90 mm Hg), 72 million Americans (35%) have hypertension. Using the 2007 AHA guidelines, an additional 7 million American adults (5%) have elevated BP requiring treatment, for a total of 79 million adults (40%).

The paper, by a collection of authors from Brown University and the University of Massachusetts in Worcester — the corresponding author, Monica Bertoia, has since moved from Brown to Harvard — has been cited twice, according to Thomson Scientific’s Web of Knowledge. One of the citing papers was an editorial comment in Hypertension that began:

Despite the limitations of this study pointed out by the authors, the article by Bertoia et al is an important contribution to the medical literature.

Written by amarcus41

August 16th, 2012 at 9:30 am

Comments
  • SoRelle, Ruth R August 16, 2012 at 10:03 am

    This is my worst nightmare.

  • CH August 16, 2012 at 11:24 am

    Just reading the bits that were quoted from the abstracts…

    Old paper: 2007 guidelines -> 98 million for 21%, 2011 guidelines -> 150 million for 32% (11% increase)
    New paper: 2007 guidelines -> 79 million for 40%, 2011 guidelines -> 72 million for 35% (5% decrease)

    So, quite the reversal of the observed trend, isn’t it?

    Also

    Old paper: 2007 guidelines define hypertension as >140/90 mm Hg
    New paper: 2011 guidelines define hypertension as >140/90 mm Hg

    So… what should we make of this?

    Finally, the numbers and percentages in the old version of the paper suggested the adult part of the US population is hovering around 450 million people… who disabled the alarm bell there?

  • David Smith August 16, 2012 at 12:30 pm

    This event wasn’t a nightmare. The final estimates did not pass the laugh test. Such a huge change should be treated as implausible without careful examination.

    Examination of the data should not have gotten that far. When the data was merged, yielding about 24,000 observations, this should have been identified as a problem, since this was the larger of two numbers from files being match-merged. This could have been identified and corrected at an early stage.

    The potential nightmare lies much deeper. More subtle analyses are done every day. The potential for error is omnipresent, the potential for discovery is weak. Methods for data quality assurance must be considered in research with a substantial data management and statistical modelling component. Many studies do not incorporate them in a systematic way.

    No data is perfect. We need systematic ways to make it better, but more importantly, to understand how results can err at each stage and how errors influence first estimates, and next, inferences. Common sense at each stage is part of the answer, but only part.

  • vhedwig August 16, 2012 at 1:40 pm

    Regardless of the numbers, if 98 million is 21% (i.e., 1/5th) of the American population, wouldn’t this imply a total US population of 5 times that – i.e., half a billion? The 2010 census puts the number at around 310 million. Are they counting Canada and Mexico, or just the US? Either way, that’s a lot of people with HBP.

    • chirality August 16, 2012 at 7:00 pm

      Whatever the exact numbers, it suggests that hypertension has become the norm in the US. A rather deadly scenario.

  • Carolyn August 17, 2012 at 2:59 pm

    The authors talk about a error in merging ‘their’ data but the problem really is that it wasn’t their data but just public data available on the web, and obviously they didn’t really examine the data carefully before deciding to use it. The authors had nothing to do with collecting or disseminating the data and apparently didn’t know much about it.

  • Post a comment

    Threaded commenting powered by interconnect/it code.