Data integrity, reliability and fraud in medical research.

Abstract

BACKGROUND: Data reliability in original research requires collective trust from the academic community. Standards exist to ensure data integrity, but these safeguards are applied non-uniformly so errors or even fraud may still exist in the literature. OBJECTIVE: To examine the prevalence and consequences of data errors, data reliability safeguards and fraudulent data among medical academics. METHODOLOGY: Corresponding authors of every fourth primary research paper published in the Journal of the American Medical Association (2001-2003), Canadian Medical Association Journal (2001-2003), British Medical Journal (1998-2000), and Lancet (1998-2000) were surveyed electronically. Questions focused on each author's personal experience with data reliability, data errors and data interpretation. RESULTS: Sixty-five percent (127/195) of corresponding authors responded. Ninety-four percent of respondents accepted full responsibility for the integrity of the last manuscript on which they were listed as co-author; however, 21% had discovered incorrect data after publication in previous manuscripts they had co-authored. Fraudulent data was discovered by 4% of respondents in their previous work. Four percent also noted 'smudged' data. Eighty-seven percent of respondents used data reliability safeguards in their last published manuscript, typically data review by multiple authors or double data entry. Twenty-one percent were involved in a paper that was submitted despite disagreement about the interpretation of the results, although the disagreeing author commonly withdrew from authorship. CONCLUSIONS: Data reliability remains a difficult issue in medical literature. A significant proportion of respondents did not use data reliability safeguards. Research fraud does exist in academia; however, it was not reported to be highly prevalent.