MAINTAINING THE INTEGRITY OF ELECTRONIC PUBLICATIONS: POTENTIAL PROBLEMS AND POSSIBLE SOLUTIONS I
Richard Smith
British Medical Journal

Fraud and misconduct in science are as old as science itself, but until comparatively recently they were not seen as presenting a serious problem.  They occurred only rarely, and, people argued, science is anyway "self correcting." This is still the state of affairs in many countries: fraud is hardly considered in the national scientific debate, and there are no ways to deal with the problem. In other countries, particularly the United States, a series of high profile cases has forced the scientific community to take the problem of fraud seriously.

The international scientific community has no option but to take fraud and misconduct seriously. The issue has risen high on many scientific and editorial agendas just at the time when scientific publishing is becoming increasingly electronic. A question naturally arises whether electronic publishing will make the problem better or worse. It might allow new and more efficient forms of fraud that are harder to detect, or it might make the process of publishing science more complete and transparent and so help reduce the problem. My view is that it will mostly bring benefit.

WHAT IS RESEARCH MISCONDUCT?
The term "scientific fraud" has tended to give way to "research misconduct" partly because "fraud" is such an emotive word and partly because "misconduct" recognises that misbehaviour ranges from the completely dishonest through to the much more minor but still important. We have no complete taxonomy of  research misconduct, but the following list can help us consider the problem in more detail. The list ranges from the serious to the minor and is far from complete.

SERIOUS RESEARCH MISCONDUCT:
Fabrication: invention of data or cases
Falsification: wilful distortion of data
Plagiarism: copying large chunks of data and words without attribution, doing studies of a new drug and an old drug against placebo and then presenting the results as if you have done a trial of the two drugs against each other,
ignoring outliers without declaring it, not admitting that some data are missing
Plagiarism: copying of words and ideas without attribution, not including data on side effects in a clinical trial, undeclared post hoc analyses
Gift authorship
Not attributing other authors
Redundant publication
Not publishing research
Not disclosing a conflict of interest
Failing to present complete data in a study
Minor research misconduct

HOW COMMON IS RESEARCH MISCONDUCT?
We cannot know with confidence the prevalence of these forms of misconduct, but we have some data to help us. Unsurprisingly we have the least data on the most serious forms of misconduct, but a congressional inquiry in the United States heard of hundreds of cases of misconduct, many of them fabrication, falsification, or plagiarism. A survey of professors of medicine showed that most had personal experience of fraud. The survey also showed that in most cases the fraud had not been made public.

Several studies have shown that about a fifth of authors of studies in major medical journals do not meet the criteria for authorship laid down by the International Committee of Medical Journal Editors. They were honorary or gift authors. About a fifth of scientific papers in medical journals are published more than once in a closely related form. There is good evidence that "positive" trials that support a new drug are likely to be published more than once, whereas "negative " trials are often not published at all. A recent study showed that the financial conflict of interest of 70 authors was disclosed in only two cases.

Finally, many studies have shown that many papers in medical journals fail to give basic data like the age, sex, and diagnosis of participants, the comparison between responders and non-responders, information on adverse effects of drugs, the method of randomisation, etc. This failure to give basic information is rarely intentional misconduct, but is a good example of where misconduct merges with poor practice and sloppiness.

WILL ELECTRONIC PUBLISHING MAKE THINGS WORSE OR BETTER?
The data I have given above together with my own experience as an editor lead me to think that the integrity of current scientific publishing is currently low. There is much room for improvement. We can imagine ways in which computers and electronic publishing might make things worse. It may be easier to copy and fabricate data, to alter the data of others, to manipulate images, or to disguise attempts at fabrication after they are suspected. Some of these crimes depend on electronic publishing but many result simply from the wide availability of computers and could be carried on even if publishing were to remain largely in paper form (which it won't).

The sad truth is that if a researcher wants to deceive editors and readers then he or she can do so without much difficulty. Science operates on trust. Peer review does not protect against fraud. No system could eradicate fraud within science. If the managers of science were to adopt the philosophy of managers of casinos (that fraud will occur inevitably and with increasing ingenuity and that everybody must be assumed to be a potential fraudster) then it might be reduced. But it would not be eradicated--and science would be changed for ever. Nevertheless, I can see many ways in which electronic publishing could increase the integrity of the scientific record.

1. Scientific papers could be published together with the full raw data plus the software used to analyse the data.
2. Much fuller methods can be published than is usually possible in paper journals.
3. Authors could be obliged to complete standard forms--for instance, the CONSORT criteria for the publication of randomised controlled trials--to increase the chance that all essential information will be included.
4. Study protocols could be peer reviewed and published. Journals that have accepted the protocols would then be obliged either to publish the final study results or give a reason why not.
5. There need be no problem getting "boring negative" results published because space would not be at a premium.  
6. The peer review process could be conducted openly on the web, increasing, for instance, the chances that somebody will recognise something being published twice.
7. All discussions that took place in the peer review could be posted on the web, allowing those interested to reassure themselves on the integrity of the process.
8. Corrections could be posted almost immediately. Indeed, the study could be modified--with some record that a change had taken place.
9. Much more space would be available to describe who contributed what to studies, avoiding the problem of gift authorship. Standard proforma might be used.  
10. As the full text of studies becomes available on large databases it will become more difficult to publish the same data twice and redundant or duplicate publication will become easier to detect.  
11. The linking of papers to full references will make it easier to detect the many cases where the "supporting references" do not actually support what is published.

CONCLUSION:
We are still in the early days of electronic publishing. We are only just beginning to see all its possibilities, and I'm sure that there will be many other ways that I cannot currently envisage to increase the integrity of the published scientific record. There may also be new ways to corrupt it.