Once a piece of work has been published it may be subject to intense scrutiny and analysis by a wide range of experts, particularly if it generates a lot of attention because of its novelty, its publication in a top international journal or because it contradicts or significantly adds to current wisdom. This can amount to mass peer review by many readers with a wide range of different expertise and experience. If the work is considered important, then other groups may attempt to repeat or extend it. This scrutiny and attempts at repetition may highlight fundamental flaws in the work or show evidence that the data has been fabricated. If work is published in an obscure journal or generates little interest then this scrutiny may be limited and no attempt to repeat the work is likely to be made. One of the fondly held beliefs about science is that it is self-correcting because any false or fabricated data will be identified when other scientists cannot repeat the findings using the methods described. Is this faith in self-correction by repetition justified?
Faith in the efficacy of peer review and repetition
If peer review is seen as the key barrier preventing the initial publication of flawed or fraudulent science then scrutiny by the wider scientific community and attempts at repetition are widely regarded as effective in exposing any offending material that gets through this initial barrier.
Philip Handler, President of the National Academy of Sciences made the following comment about any rare occurrences of scientific fraud before a US congressional committee in 1981:
“It occurs in a system that operates in an effective, democratic and self-correcting mode”
In similar vein:
“Scientists generally trust that fabrication will be uncovered when other scientists cannot replicate (and therefore validate) findings”
Crocker and Cooper (2011) in an editorial in Science.
These and other quotes affirming the faith of their writers in the processes of peer review and replication to police the integrity of the scientific literature were used by Wolfgang Straube and two colleagues in a 2012 article pointedly entitled Scientific Misconduct and the Myth of Self-Correction in Science. This paper suggests that blind faith in self correction may not justified and may encourage complacency.
In their 2012 paper, Stroebe and his colleagues look at the ways in which forty notorious research frauds were discovered starting with William Summerlin in 1974 and ending with Yoshitaka Fujii in 2012 and includes cases from psychology, biomedicine and the physical sciences. Fourteen of their cases are discussed in the case-studies section on this blog. For 21 of these forty cases they list a “whistle-blower” as the mode of discovery and I would classify a further two of the others that I am familiar with (Chandra and Summerlin) in this category. William Summerlin was caught “painting his mice” but it needed the technician who caught him to report this action to trigger Summerlin’s suspension and the formal inquiry into his research conduct i.e. to act as a whistle-blower (see earlier post about whistleblowers). Failure to replicate is listed for only four of the 40 cases and in six cases Stroebe and his colleagues suggest that outside researchers found inconsistencies or problems with the published data that indicated fraud. Only in the case of RK Chandra do they list a journal peer review as a contributor to his discovery and to his public outing as a data fabricator but he was actually accused by a whistleblower, Marilyn Harvey eight years earlier (see later). In the case of Kristin Roovers, which is also discussed later, they suggest that she was unmasked by an editor during the pre-publication process.
None of the cases where they list failure to replicate as the main mode of discovery of a fraudulent researcher are covered in the case studies; several of them are from the physical sciences. Within the case studies on this blog, the suspicions and investigations into the activities of Jatinder Ahluwalia at University College London were brought about by outside researchers reporting their inability to repeat his work. The subsequent failure of others within UCL to repeat the work triggered the investigation into why the results obtained by Ahluwalia differed from those obtained by other members of the UCL team. In the case of Haruko Obokata, the claim that stem cells could be generated by subjecting ordinary skin cells to mild acid shock proved to be unrepeatable by colleagues at the RIKEN Institute but this was only after her papers had been retracted for other acts of misconduct.
Can peer review do more than block the publication of fraudulent data?
Peer review has been discussed in the previous chapter on protection as one of the barriers to the publication of fraudulent data. It has clearly failed to prevent the publication of hundreds of papers that, on reflection, are deeply flawed and contain what clearly look like fabricated or falsified data when assessed critically with that accusation in mind. One cannot say, however, that peer review offers no protection because one has no idea of how many other papers with fraudulent data were rejected by suspicious referees and editors. In circumstances where referees and editors have substantial grounds for suspecting research fraud they should do more than just reject the manuscript . Authors of rejected papers are given a reasoned case explaining why the paper has been rejected. This may be the tempting option where there is not clear proof that the data has been fabricated or falsified. Accusations of research misconduct may result in costly and time-consuming litigation even when the accused is guilty if they are determined to deny their guilt and mount a legal challenge to the accusations. However, if this easy option of simple rejection is taken then a brazen fraudster may simply submit the rejected paper to another journal perhaps after making adjustments in the light of the criticisms offered in the rejection notice. The first journal has been protected but the problem has simply been passed on to another journal and perhaps the chances of it eventually polluting the scientific record increased by identifying visible flaws. Editors have a responsibility to try to ensure that there is an investigation into the activities of the would be fraudster usually by reporting their suspicions to the employer, the research funding agency or an organisation like the Office of Research Integrity (ORI) in the USA. They must do all they reasonably can to protect the literature as a whole not just their own journal.
Stroebe et al (2012) reported that in only two of their 40 cases of research fraud was the journal reviewing and editing involved in unmasking a notorious fraudster, Ranjit Chandra and Kristen Roovers. A referee suggested in 2000 that the version of Chandra’s paper submitted to the BMJ claiming that micronutrient supplements improved cognitive function in the elderly “had all the hallmarks of being entirely invented”. Richard Smith, who was editor of the BMJ when Chandra’s paper was submitted, discusses this case in a BMJ article in July 2005. The BMJ asked Chandra’s employer Memorial University in Canada to investigate in the light of their concerns. According to Smith, the university held an inquiry but found no serious problem. We now know that Memorial had held an investigation into Chandra’s research in 1994 after complaints made by a whistle-blower, his nurse assistant Marilyn Harvey, and that although this found him guilty no action was taken. The BMJ was unconvinced by Memorial’s response and raised further issues only to be told that Chandra had taken a period of extended leave in early 2002 prior to resigning his position at the university. The university again took no further action because Chandra had left and would not supply any raw data but if the allegation of data fabrication is accepted then there never was any real raw data. The paper was subsequently published in the journal Nutrition who were also contacted by the BMJ but only after the paper had been published and the paper was not finally retracted until 2005. This article by Smith then goes on to discuss the difficult issue of who should investigate the previous studies of a fraudulent author and this is discussed in the final post in this series on disinfection of the scientific record.
Kristin Roovers was discovered not by a referee but by a journal editor during a spot check. She was found to have extensively manipulated a western blot image in one of her manuscripts. The journal notified the US federal ORI and she was subsequently found to have extensively manipulated images in three published papers. This check by the editor occurred after the paper had been accepted for publication when it was undergoing final revisions.
Encouraging and protecting potential whistle-blowers
The findings of Stroebe et al (2012), as well as examples from the case-studies, show that whistle-blowers reporting their suspicions of research misconduct have played a key role in unmasking many notorious research frauds thus curtailing their fraudulent activities. In some cases this has led to the retraction of many of their fraudulent publications (e.g. Dipak Das, Diederick Stapel and Malcolm Pearce). It is vital that those who feel that they have strong evidence that a colleague or collaborator is acting dishonestly should feel confident that they will not be victimised if they report their suspicions; this is particularly important where the suspect is a senior and powerful figure and the accuser a subordinate. Institutions should designate a person or persons to whom those with suspicions can present these suspicions in confidence and with a guarantee that the evidence will be assessed impartially. The accuser should also feel that, if they wish it, their anonymity will be protected whether or not their accusations are accepted provided their accusations were made in good faith and not a vindictive attempt to tarnish the reputation of the accused to settle some personal grievance.
There are also several examples in the case studies of whistle-blowers who feel that their, ultimately proven, accusations of misconduct were not initially taken seriously enough or who felt that their own careers were adversely affected by their acting as a whistle-blower e.g. in the Stephen Breuning, Ranjit Chandra and Michael Briggs case studies.
In the Stephen Breuning case-study, Professor Robert Sprague of the University of Illinois played a central whistle-blower’s role in unmasking the fraudulent activities of the psychologist who had been a research collaborator of Sprague’s. Sprague formally reported his suspicions and evidence to the National Institute of Mental Health in the USA in December 1983 but it was not until April 1987 that the NIMH finally produced a report that unequivocally supported his accusations. In late 1986, after 17 years of funding of his work by the NIMH, this funding was abruptly stopped and Sprague clearly believes that his role as a whistle-blower in the Breuning affair was a major factor in his loss of funding; a US Congressional committee seemed to share this suspicion. After making his report to the NIMH Spague suggests that one of their first actions was to investigate him rather than Breuning. He was also threatened with legal action by a Vice-President of the University of Pittsburgh for making slanderous and libelous comments (critical of the University’s handling of the Breuning affair) to a congressional committee even though comments made before Congress are protected from such challenges; he later received an apology from the President of the university.
Dr James Rossiter, then chair of the ethics committee at Deakin University in Australia was the first to report his suspicions about the fraudulent research activities of biochemist Michael Briggs on oral contraceptive safety. He summed up his experiences of being a whistle-blower in a 1992 article in Nature entitled Reflections of a whistle-blower. Rossiter’s first action in this matter was in October 1983 when he reported his suspicions to the chancellor of the university along with a file of supporting evidence. A final report on the Briggs affair was eventually produced by the university in 1988 – Briggs resigned from the university in 1985 and died in Spain in 1986. In this Nature article, Rossiter describes a prolonged campaign of harassment and intimidation against him during the period between his first reporting his suspicions and the 1988 report including:
- 200 threatening, obscene or silent phone calls; many at night
- Letters containing obscene allegations about his private life
- Expressions of doubt about his sanity from a psychologist on the Deakin staff which he believes adversely affected his private medical practice
- Attempts to remove him as chair of the ethics committee at Deakin
- Attempts to blame him for leaking details of the affair to a journalist (he says that in fact the confidential files had been removed from his locked office).
Marilyn Harvey was a nurse working for Ranjit Chandra in the early 1990s. Her job was to recruit new parents who had a family history of allergy to take part in a trial to see if a particular type of infant formula reduced the risk of their babies developing allergies. She reported that Chandra had in fact published these studies when she had still only recruited a small fraction of the subjects required. It transpired in a series of documentaries broadcast by the Canadian Broadcasting Corporation in 2006 entitled The secret life of Dr Chandra 2006 that an internal inquiry at Memorial University had concluded in 1994 that “scientific misconduct has been committed by Dr Chandra” but no action was taken against him. In July 2000, Marilyn Harvey was served with notification that she was being sued by Chandra for allegedly stealing data from one of his studies. The lawsuit was dropped a few months later but this must have caused her considerable distress at the time. In 2007, after details of Chandra’s activities were broadcast in Canada, Harvey filed her own lawsuit against the university which claimed that they had failed to properly investigate the allegations that she had made more than a decade earlier. The lawsuit contended that the university had led the medical community to believe that her accusations were unjustified and she thus acquired a reputation as a troublemaker. I have not been able to find out how this legal case between Ms Harvey and Memorial University was resolved but during this search, I did find that in 2014 Memorial University had introduced a Marilyn Harvey Award to recognise the importance of research ethics. In the accompanying description of the award it states that the university:
“has named this award in honour of Marilyn Harvey, BN, a research nurse who brought forward her concerns regarding research ethics to senior administrators at the university”
This seems like a quite a turnaround in attitude and one can only hope that the university regrets its past actions and has tried to learn from any mistakes that were made in its handling of the Chandra case. This seems like a belated recognition of the correctness of Harvey’s actions and something akin to an apology by the university.
Another example of a would-be whistle-blower who has suffered for her activities is that of Dr Helene Hill who at 86 years of age (March 2015) is still Professor of Radiology at Rutgers University Medical School. This is despite being threatened with termination of her employment for her relentless pursuit of what she believes to be a case involving multiple acts of data fabrication. In 1999, she accused a postdoctoral fellow at the university (Dr Anupam Bishayee) of reporting cell counts from culture dishes that she claimed had been empty. She reported her observations to the university ethics committee but they decided that her case was not proven. She then reported her concerns to the ORI who conducted a statistical analysis of the postdoctoral fellow’s data but despite finding some suspicious features they also concluded that there was insufficient evidence to prove misconduct. Dr Hunt has also pursued her case through the courts on two occasions at a cost of over $200,000 but on both occasions she was unsuccessful in proving her case to the satisfaction of the judge. She obtained access to Bishayee’s laboratory notebooks in which some of the his raw data was recorded. She and a statistician have analysed the last digit frequency of machine counts recorded by hand in these notebooks. If numbers are generated by a machine then all 10 digits should occur with roughly equal frequency but when people make up data then tend to favor certain digits. According to her the actual frequencies recorded in these notebooks had a chance of being generated randomly of less than 1 in 100 billion which if true would indicate that they were fabricated. A similar analysis was commissioned by the BMJ on raw data submitted by Dr Ram B Singh. Dr Hill said in a personal correspondence to me that she wants to retire but is determined to get her statistical paper published before retirement even though it has been rejected by at least nine journals so far; usually by the editorial office before being sent for peer review. She expected to publish a book about the affair about a month after our correspondence in March 2015.
Once a paper has been published it will be exposed to a wider audience who collectively may have skills and expertise way beyond those of the initial peer review team. If the paper generates sufficient interest and is published in a traditional journal then many of the journal subscribers and library or internet browsers may read the paper and consider its content critically. They may see errors and flaws that went unnoticed during the peer and co-author review processes. If the paper is in an obscure journal or an open access journal with a very low impact factor then this process of reader scrutiny may be very limited or may not take place at all. Very few people may actually read the paper, it may simply be absorbed into a growing mass of obscure and largely invisible papers that serve little purpose beyond adding another line or two to the author’s CV; major faults may thus go unnoticed. Even if the obscure paper is critically read by someone who has found it in the course of a specific literature search and some of the faults are spotted, such readers may be less inclined to take up the issue with the journal and make their criticisms known especially if a long time has elapsed since publication. They may simply disregard its findings and become even more mistrustful of material published in some obscure journals. If the person doing the electronic search in order to review a topic or conduct a meta-analysis does not see the flaws in the paper then it may distort their conclusions. Those with experience of reviewing the literature tend to disregard isolated or anomalous reports from certain low quality journals in the absence of any more authoritative substantiation.
If critical readers are sufficiently motivated, they may voice their concerns in a letter to the editor who may then publish the letter in a subsequent issue of the journal. The original author of the criticised paper will probably be shown the criticisms prior to their being published and offered a chance to respond to them; their responses are sometimes more damning than the original allegation. If the criticisms of the paper are serious enough and apparently substantiated, they may be the spark that ignites a process of critical debate, deeper investigation and retraction of a flawed paper. It may even lead to a critical re-examination of other work published by the criticised authors. This process may not only lead to retraction of the flawed paper but maybe a general discrediting of its author and ultimately the retraction of other fabricated papers.
One of the most obvious things a reader might spot is that they are a co-author on a paper they were unaware of. Scientists may well check their own publications and their citation rates intermittently or when updating their CV to apply for a new job, promotion or research funding. They may come across publications for which they are listed amongst the authors without their knowledge or overt consent. Anyone in this situation who feels that their contribution was not sufficient to justify co-authorship and when they feel unable to accept responsibility for the data and its interpretation has a duty to inform the editor of the journal of this. Even if the unwitting co-author feels that they did make a real contribution to the work and are confident enough about the work to accept responsibility for it then this publication without consent would still be a breach of research ethics and etiquette and should be handled as such.
Colleagues may also spot other statements in a paper that they know are untrue e.g. they may sit on the ethics committee and know that a study did not get the ethical approval that it claims to have or they may know that it would have been impossible for the study to have been conducted in the manner and using the resources described in the paper. e.g. use of unavailable apparatus or materials. This is when they need enough confidence in their institutional procedures to act as whistle-blower.
Features and characteristics of fraudulent papers and scientists
Discussed below are a number of individual problems with past papers, submitted manuscripts or the output of particular authors that have led to allegations and eventually to acceptance that research misconduct has occurred. This list is not claimed to be definitive but where reviewers, critical readers, employers or colleagues notice such features or characteristics in something they read then this might encourage them to consider the possibility that some sort of fraudulent activity may have taken place. Once fraud is suspected then detailed analysis of published work and hopefully of the author’s notebooks or raw data should yield more conclusive evidence and maybe even proof of fraud. I have somewhat arbitrarily separated those factors that lead to initial suspicion from those that might be used to audit the veracity of an accused person’s research output which I have discussed in the next chapter on disinfection; in practice, of course some of these methods may be used for both purposes.
Very high publication rate and huge throughput of (rare) subjects
If someone is fabricating research data then this will take much less time and effort than actually generating it honestly. The fraudster can also generate clean data that is likely to be readily accepted by unsuspecting reviewers and editors. Some fraudulent authors have been greedy in their publication of fraudulent papers and have aroused suspicion because of their improbably prolific generation of apparently high quality research data such as in the examples below.
- In his article describing his experiences of being a whistle-blower, Professor Robert Sprague says that after two years of collaborating with Stephen Breuning he became suspicious of Breuning’s prolific output. He knew that he could not be this productive himself and he began to question how Breuning managed it.
- The suspicions of Professors Paul Mckeigue and George Davey Smith about the activities of the Indian physician Ram B Singh were aroused by his apparently remarkable productivity. They pointed out in a letter to the BM J that he had been first author on 28 full articles published in a four year period (1989-93) and that he had published at least five major intervention trials in an eighteen month period. A subsequent literature search confirmed that between 1990 and 1994 he was first author on no less than 25 clinical research trials or case-control studies.
- The classic case of the over-producer was Yoshitaka Fujii who published clinical trials at an astonishing rate. In his peak year (1998) he published 30 clinical trials.
A related issue is that some fraudsters have apparently recruited very high numbers of subjects in a short period of time, in some cases subjects with very specific and quite uncommon inclusion criteria as in the examples below.
- In his peak year for publishing clinical trials, Fujii apparently recruited 3000 post-operative subjects. In an attempt to disguise this impossibly large throughput of patient subjects he claimed that many of his studies were multi-site studies and he added the names of co-authors from other institutions, usually without their permission, to add credibility to these claims.
- Sir Cyril Burt not only claimed to have studied 53 pairs of identical twins reared apart but also claimed that they were separated within six months of birth and reared in totally different social environments. This seems impossible and is made even more unlikely because policy or practice was to place separated twins with other family members or in social environment close to that in their birth families.
- Gynaecologist Malcolm Pearce published a double blind trial involving 200 women with polycystic ovary syndrome who were trying to have a baby after 3 previous miscarriages in their first trimesters. It would have taken a major referral centre a decade to recruit this number of women meeting these uncommon inclusion criteria.
Claimed use of resources that were not available
Some fraudsters have carelessly claimed to have used a resource that was not available to them when the work was said to have been conducted as in the examples below.
- Jon Sudbo, the Norwegian dentist, published a case-control study in the Lancet comparing smoking behavior and the use of anti-inflammatory drugs in people with and without oral cancer. He claimed to have used a cancer register in this study which was not open at the time his study was conducted.
- Michael Briggs, the expert in oral contraceptive safety, claimed to have used a compound called desogestrol in one of his studies in Australia in 1980 but this compound was not available in Australia at this time and Briggs did not have a licence to import it.
- Ranjit Chandra in his now retracted paper on the effects of a dietary supplement on cognitive function claimed to have used without any attribution, a test of long term memory based upon recall of events in the person’s life. According to experts in psychology, no such validated test existed at this time.
- My ex-colleague Jatinder Ahluwalia published a paper within a month of starting at the University of East London (UEL) that involved the use of a sophisticated and expensive patch clamp apparatus. The paper gives his affiliation as UEL and makes no mention of the use of any outside resources anywhere in the paper. UEL did not have a functioning patch clamp at that time and so the work could not have been conducted solely at UEL. This does not, of course, prove that the data was fabricated but failure to acknowledge where the work was actually carried out would be a breach of research ethics. It was said that at Cambridge University he reported use of animals and radioactive substances that were not record in the official log of their use.
Irregularities and inconsistencies in the published data
In this section, I am going to confine my discussion to flaws and irregularities that might be spotted by a vigilant reviewer or reader of the published paper who does not have access to the raw data for the study. It should be a condition of publication that raw data must be available for scrutiny if requested or perhaps deposited in a secure electronic location prior to publication. There are several tools that can be employed to test the likely veracity of raw data and these are discussed in the next chapter. Rather than try and make an exhaustive list of all the possible things that could lead a reader to suspect fraud, I am going to give a range of examples from the case studies. In many cases the flaws in papers now known or believed to be fraudulent are very obvious once they have been pointed out.
- A decade before he was confirmed as a serial fraudster and broke the record for the number of retracted papers, it was noted that the incidence of headaches as a side-effect in the clinical trials of Japanese anesthesiologist was so consistent as to be statistically impossible. A later detailed statistical analysis of the background data in his numerous published clinical trials proved beyond any reasonable doubt that these values could not possibly have been generated by random allocation of subjects to control and treatment groups.
- Another anesthesiologist Joachim Boldt from Germany held the record for retracted papers until that was shattered by Fujii. In a letter to the editor of a journal, one reader suggested that the variability recorded in one of the Boldt’s papers was too small to be realistically feasible and upon further scrutiny the editor agreed with this assessment and began the process that led to his unmasking as a serial fabricator of research data.
- Sir Cyril Burt published 3 data sets comparing the IQ of identical twins reared apart and together and for each of these reports he quoted identical correlation coefficients (r values) despite large changes in the numbers of subjects (r=0771 for those reared apart and 0.944 for those reared together). This is essentially statistically impossible and in a wider assessment of Burt’s publications it was noted that multiple use of exactly the same correlation coefficient was not uncommon. As another example from Burt, two tables in one of his papers contain data sets giving the IQs and social class of fathers and their sons. When frequency distributions were plotted, the fathers and sons curves were perfect and identical normal distributions which were also identical to a theoretical distribution with a mean of 100 and a standard deviation of 15; this is despite there being general agreement (including by Burt himself) that IQ distributions tend to be skewed rather than perfectly normal. The data in these tables was generated dishonestly from a theoretical normal curve.
- The retracted paper of Ranjit Chandra which suggested that dietary supplements improved cognitive function in the elderly has many glaring statistical and other anomalies. He tabulates before and after supplementation values for the subjects’ performance on a range of cognitive function tests. Some of the values in this table suggest that many subjects were seriously cognitively impaired (demented) at the outset despite claims in the text that all subjects were psychologically normal at the outset. The change in some of the values as a result of the supplementation are so great that they could be said to indicate that the supplement cures many cases of dementia. The cognitive function results in the table are given as mean values ± standard error and this would mean that many individuals had scores of less than zero on some tests and other subjects more than the maximum score. He subsequently claimed that the use of standard error was a typographical error and should have read standard deviation but as discussed in the case-study this change simply created even more statistical impossibilities. In comments about another of his papers it was noted that he got 100% agreement from potential subjects to participate in the study which is unheard of.
- In the Ram B Singh case-study, a variety of critics and reviewers have suggested a number of highly improbable findings in his published papers and submitted manuscripts. These include: extraordinary levels of compliance with prescribed diets; baseline intakes of dietary variables that were extremely low and showed exceptionally low variability; much greater improvements in risk factors than those recorded by other authors; an incredible lack of deaths from other causes in a study relating diet to heart disease etc.
Findings that are at variance with others
It was noted earlier that undue faith has been placed in the idea that repetition will rapidly show up false or fabricated data and thus that science rapidly corrects itself. Almost a mirror image of repetition is comparison of newly published data with that already in the literature. Are the new findings compatible with existing findings or are they totally different to those reported previously? Where a treatment is being tested, are the claimed treatment effects within the bounds of what are realistically credible? Thus one of Ram Singh’s submissions to the BMJ indicated that Spirulina, a widely available algal food supplement, was more effective than statins at lowering blood cholesterol levels. Similarly Ranjit Chandra’s retracted paper suggested that a simple vitamin and mineral preparation was 6 times more effective than the most potent then known drug in increasing cognitive function scores in elderly people. The spark that triggered the auditing of Werner Bezwoda’s studies suggesting that high dose chemotherapy greatly extended life expectancy in women with advanced breast cancer was the difference between his findings and those of four other groups from Europe and the USA who all presented their results at a conference in Atlanta in May 1999.
In an earlier post distortion of meta-analyses is discussed as a major harmful effect of fraudulent research because this is now such a popular research tool and because of its position at the pinnacle of the pyramid of evidence. Meta-analysis can also paradoxically be used as an indicator that the work of one author or group is producing results that are inconsistent with those of all other authors. In this previous post, it was shown how data from the Chandra group distorted meta-analyses of the effects of micronutrient supplements on immune function in the elderly. It is tempting to hope that this meta-analysis might have triggered suspicion about the veracity of Chandra’s data if he had not already been accused of research fraud. For example, why did these supplements apparently reduce dramatically the number of infection days during the year but did not affect the chances of having an infection? In 2001 Kranke et al reported in their meta-analysis, the work of prolific fraudster Yoshitaka Fujii was altering conclusions about the effectiveness of a drug used in treating post-operative nausea and vomiting. Although they did not directly accuse Fujii of fabricating data the implication is clear from their analysis and discussion; an implicit accusation that was made more than a decade before his final exposure as the most prolific fraudster to date.
Irregularities with images used in publications
Most serious photographers now make use of sophisticated computer software like Adobe Photoshop to improve or modify the photographs that they have taken. They can be used for a number of useful or entertaining purposes e.g. to remove “red eye” from portraits, to remove an unpleasant item caught at the edge of an otherwise good picture, to insert or remove images of individuals and even to restore treasured old creased or faded family photographs. They can also be used for more sinister purposes e.g. to insert innocent individuals into photographs of illegal activities or to embarrass someone by transposing their head onto a pornographic image. The old adage “the camera never lies” no longer holds.
This technology can also be used to commit research fraud. Images are often submitted as part of the supporting evidence in a manuscript just like summaries of numerical data. These images often involve a series of bands or peaks to represent the presence of specific chemical compounds or traces from electronic recorders to indicate the magnitude of the response of biological specimens to a particular stimulus. Western blotting is a very widely used technique to detect the presence of specific proteins within biological samples. The final western blot image is seen as a series of dark bands on a gel and each band represents a specific protein. These banded gels are then photographed and often submitted as part of a paper to confirm the presence or absence of particular proteins in different samples. The same software used to alter ordinary photographs can also be used to alter these western blot images so that the new image tells a very different story to that of the original e.g. the presence of a specific protein in the sample can be hidden or added.
Several examples of this type of data fabrication or falsification have been discussed in the case studies such as those listed below.
- The papers by Haruko Obokata and her colleagues in Nature in early 2014 claiming a simple method of generating stem cells quickly attracted allegations that images had been manipulated and mislabeled. This led to retraction of the articles and subsequent work at the RIKEN institute where she worked failed to replicate the findings. It also later transpired that the papers were originally submitted to the journal Science who spotted the manipulated images, rejected the papers and warned Obokata.
- In their investigation into the activities of Dipak Das, the University of Connecticut review board concluded that at least 145 western blot images spread over 25 publications and 3 research grant applications had been manipulated in a manner that constituted research misconduct. He was found to have spliced together separate images to create a single image, pasted extra bands into otherwise normal blots and erased or partly erased bands from other blots.
- Dr Jatinder Ahluwalia was dismissed from its graduate studies programme by the University of Cambridge when it was noticed that he had falsified chart recorder traces produced to support his thesis for his Certificate of Postgraduate Studies.
The case of Dr Kristin Roovers was mentioned earlier because her manipulation of an image was detected prior to publication by a journal editor. She was found to have also manipulated images in three already published papers. She was found to have cut and pasted either within a figure or between figures so that the submitted image was very different from the original. The manipulated image sent with the manuscript and the original are shown in figure 1.
Figure 1 The original and the manipulated image submitted by Kristin Roovers to the Journal of Clinical Investigation
In a 2008 article in The Chronicle of Higher Education, Jeffrey Young talks about this case as an example of the major and growing problem of image manipulation in scientific papers. He noted that in 2006, The Journal of Cell Biology had started to check all images that it publishes and suggested that this added about 30 minutes to the production process. He also discusses a software tool developed by Hany Farid,, a computer-science professor at Dartmouth College, that detects manipulated images. Young wrote that 44% of the cases handled by the ORI in 2005-6 involved accusations of image fraud compared to 6% a decade earlier. In an article by Natasha Gilbert in Nature in October 2009 she produced a graph showing the exponential growth in the number of cases reported to the ORI for alleged image manipulation (see figure 2)
Figure 2 The growth in image fraud cases reported to the ORI
Other publishers, including the publishers of Nature, have since introduced a similar image checking process to the Journal of Cell Biology but this has not prevented papers with manipulated images being published and retracted from Nature e.g. Kim et al (2009) which was first corrected and then retracted because of multiple inappropriate manipulations of images. Most recently the original issues raised about the retracted papers of Haruko Obokata related to image manipulation.
The Rockerfeller University Press whose list of prestigious journal includes the Journal of Cell Biology list 4 basic guidelines for the handling of image data:
- No specific feature within an image may be enhanced, obscured, moved, removed, or introduced
- Adjustments of brightness, contrast, or colour balance are acceptable if they are applied to the whole image and as long as they do not obscure, eliminate, or misrepresent any information present in the original
- The grouping of images from different parts of the same gel, or from different gels, fields, or exposures must be made explicit by the arrangement of the figure (e.g. dividing lines) and in the text of the figure legend
- If the original data cannot be produced by an author when asked to provide it, acceptance of the manuscript may be revoked.
An account of some of the methods available for detecting image manipulation is given in an editorial in the November 2013 issue of the Elsevier journal Biochemical Pharmacology but detailed discussion of these methods is beyond the scope of this article.
Once there are substantial grounds for suspecting that an author may have committed a serious act of research fraud then this should trigger an investigation, firstly into the initial allegations and then if these are found to be proven then there should be an analysis of as much of that author’s previous work as is practical. A whole range of extra tools and indicators can then be used by those conducting that investigation and this should lead to a cleansing of the literature of any other suspect data by this author; this disinfection phase is discussed in the final post in this series.