Publication fraud, dishonesty, and deceit

Let me initiate this editorial by stating that intentional publication fraud (knowingly publishing falsified data) is much less common than unintentional publication error (unknowingly making a mistake in the publication);^1,2 fraud typically involves only 1% or less of all studies.³ Within the majority of biomedical literature the data reported are valid and the results of the manuscript accurately reflect factual findings from an investigation. Also worth noting (to the best of my knowledge), to date I know of no publications that have been retracted related to physical therapy or manual therapy literature that have any association to fraud, dishonesty, or deceit. However, this is not to say that it has not happened and has yet to be identified or that it is not going to happen. In fact, retractions associated with fraud in the general biomedical literature have increased sharply over the last decade⁴ and most are from studies that originated within the United States of America.⁵

Publication fraud, dishonesty, and deceit can occur in a number of ways and can have minor or major consequences. Stewart and Feder⁶ identified two major sub-categories of errors in publication. The first category (Type A) is associated with errors or inconsistencies, failure to obtain relevant data, and honorary authorship, all elements that they claimed reflect carelessness more so than malfeasance. A more serious form (Type B) includes intentional errors, involves premeditated misleading, incorrect citations, failure to acknowledge how data were collected, and other truly fraud-related actions. This includes fabricating or falsifying data,^7,8 questionable data quality,⁹ or smudged (fixed) data.² Both forms of errors can lead to changes in clinical practice that are not beneficial, may lead to higher healthcare costs, and may lead to an unnecessary focus on an ineffectual treatment approach. Type A errors may be associated with study biases, mode of administration biases,¹⁰ and other unknowingly influential elements; often occurring to an unwilling or unknowing researcher. It is worth mentioning this because Type B errors are dubious, potentially disastrous, and extreme, and do occur within the knowledge and oversight of the researcher.

It is a lot easier to discuss fraud than it is to detect it.¹¹ The Committee on Publication Ethics (COPE) guidelines¹² outline a number of potential fraud-related errors that are designed to assist authors during the reporting of their paper. Various elements COPE identifies include: (i) data fabrication, (ii) data manipulation/falsification, (iii) misleading reporting, or (iv) selective reporting. Certainly data fabrication, manipulation, or falsification can lead to bogus results and is never tolerated. This is likely the most extreme form of fraud in publication and may involve changing numbers, creative false results, or changing values to influence an outcome. Selective reporting occurs when data are calculated, analyzed, and reported in a way to influence the desired finding. One example involves the use of a particular outcome value that inflates the study result, or the capture of three or more outcomes measures, and using only a single measure that gives the greatest effect.

COPE guidelines¹³ provide flow charts for editors as well when suspicious data within a study are submitted. COPE recommends the use of a second opinion from a reviewer, contact with the author to discuss the information, and potentially requesting the raw data from the study. When a paper has already been published it is recommended to contact to regulatory research body of the institution and request an investigation. Unsatisfactory results from an inquiry may demand a publication of concern regarding the suspected paper.

As a reader of medical literature, you may ask: doesn’t the peer review process reduce the likelihood of this? The answer is yes and no. It has been stated that peer review creates a self-serving cliques of reviewers who function to approve others’ works that are similar to their own.¹⁴ One can see this on some journal editorial boards where like-type thinkers or research constituents congregate and control information flow within selected journals. The journal editor in chief is equally, potentially at fault. The editor picks the reviewers and can deeply influence the review process based on the obstacles stacked before the paper. In addition, peer review is fraught with personal opinions, and different reviewer levels of interest, ability, and ire. Peer review truly is a ‘first step’ of an imperfect process.¹⁵

With these imposing challenges, how does the lay clinician recognize potentially fraudulent results on data already published? Let’s face it, there is likely no chance that a clinician can request raw data, discuss the results of a study with the original author or statistician, or perform many of the COPE-recommended processes available to an editor, without offending a number of good and honest researchers along the way. However, if one has suspicions, it is best to use the following factors that hint toward potentially fraudulent data.

First, effect sizes are a standard magnitude measure and are a universal value across interventional studies. If a single study reports an effect size that is several standard deviations above the mean of the majority of like-type studies, and the design of the study is not compellingly different, one should question the results of that paper. Second, if the study results cannot be replicated by independent investigators, who use a similar population of patients, a similar level of study control, and similar analyses, then one should have concern regarding the findings of the original study. Third, consider the ‘time share’ rule. Just as there are strings attached to a ‘free television or vacation’ just for attending a short 30-minute conference, there are strings attached to a study that demonstrates benefit on nearly all patients, from a unique and dedicated treatment, regardless of patient background or severity. If it looks too good to be true and it is not something you routinely experience in your clinical practice, it likely is too good to be true. Fourth, recognition of poor study design and the influence toward equally poorly transferable results is critical when dissecting literature. Lastly, evidence does suggest that most fraudulent papers are written by a first author that has been fraudulent previously, are published in journal with high impact factors, and involve numerous authors for a single publication.⁵

Just recently, a group of cancer researchers from Duke University have been criticized within the science realms for a number of publication-based misconducts.¹⁶ External investigators have found discrepancies in the researchers’ datasets, alleged errors within their statistical analyses, and inability to reproduce any of the findings that the original researchers have reported in a number of trials. Allegedly, the problems are related to data smudging, corruption, and deceitful summaries. These problems, which by their nature are compelling on a grand, international stage, are likely more common than we think on smaller stages, such as those associated with rehabilitation, and in less significant scale.

Despite the tone of this editorial, it is important to note once more that the majority of publications in biomedical research are legit, valid, and were performed and reported with great care. As stated, I know of no single physical or manual therapy related study that has been identified as fraudulent. One should not automatically suspect bogus results from a novel finding, yet understanding the potential for this and the mechanisms associated with fraud can improve the safe and appropriate consumption of quality research.