Salami slicing or living off the fat? Justifying multiple publications from a single HIV dataset

Academia has long been mired in a ‘publish or perish’ world 1, 2. The career of an academic who publishes little, or whose publishing rate declines, is threatened, and formal review processes like the Research Assessment Exercise or the current Research Excellence Framework enshrine this reality 3. Since the pressure to publish is so high, academics might be forgiven for gravitating towards the concept of the ‘minimum publishable unit’; the smallest quantum of data or information that can constitute a paper 4. This can lead to (at least) two strategies, both of which are unacceptable: duplicate publication and salami slicing. The first is usually a straightforward issue: substantially the same material is published in more than one journal. The second requires a little explanation and clarification 5. The traditional framework for research is relatively simple. First, the researcher frames the question of interest as a ‘hypothesis’. This is a statement that leads to a testable prediction. An example might be: “Does drug X block sodium channels in nerves?”, which leads to one prediction that if drug X is applied to a nerve then it will induce certain characteristics during voltage clamping, etc. Then, the researcher conducts an experiment to test this prediction. This experiment is designed to focus in on the hypothesis under question, with potentially confounding factors rigorously excluded or controlled for. A suitable number of experimental periods or observations are taken (or subjects recruited in the case of human/patient studies), based on power calculations. In this way, an experimental result consistent with the prediction lends some support to the underlying belief that led to the hypothesis, whereas a result contrary to the prediction destroys the hypothesis 6. Once published, the original data from the experiment itself are of little value in another context for the precise reason that the experiment was so rigorously designed just to test the one hypothesis. Salami slicing involves breaking up data into more than one publication, each of which represents a ‘slice’. For example, a comparison of three laryngoscopes A vs B vs C performed at the same time in the same group of volunteers could be published as A vs B and then separately as B vs C (and then also perhaps A vs C). This might distort the literature if readers or later meta-analysts believe that the data in each ‘slice’ have come from different, independent samples 7. Therefore, such practice is considered to be unacceptable if the studies share the same hypothesis, population and methods 5. However, there may be instances where publishing different analyses of the same database may be fully justified and the researcher is in a position to ‘live long off the fat of the data’. This is especially so where a different and distinctly separate question has been asked or the endpoints reported are very different, and the same dataset can provide the answer to these questions. In such instances, we believe that the manuscripts should each clearly define their separate hypotheses and acknowledge that the data have been presented elsewhere, referencing any other studies from the same database and explaining why such further analysis was justified. The authors should also disclose this information freely to the editors of the journals, as well as enclose any other related papers, whether published yet or not. In terms of ethics, deciding whether the issue is one of salami slicing or not is ever more complicated. Modern research now includes a number of relatively new experimental designs: observational studies; questionnaires; genetic databases; population registries; post-marketing surveillance studies; and meta-analyses 8. Within each of these experimental frameworks, what is regarded as ‘ethical’ for publishing norms has become very complex, since the types of data involved differ substantially. For example, a number of manuscripts have arisen from the Intensive Care National Audit and Research Centre (ICNARC) Case Mix Programme. The expanding pool of patients included in this data registry means that those included towards the beginning of the programme of data collection contribute to the various papers many times, and the papers come from the same group of patients each time, in various combinations 9. At the same time, the number of organisations that may give an opinion on whether a study (and therefore a publication arising from it) is ethical or not has expanded exponentially (for example, Research Ethics Committees; NHS Trusts, including the Caldicott Guardian; Higher Education Institution Research Ethics Committees; and the Medicines and Healthcare products Regulatory Agency). Sometimes a proposal may properly require an application to several of these authorities. With so many organisations involved, it is most unlikely there will be a unitary view on what constitutes ethical behaviour, including the issue of ‘salami slicing’, given the heterogeneity in the type of studies normally handled by each of these bodies. However, ultimately the final decision rests with the journal itself when discussing publication of a manuscript. One type of investigation that may result in multiple publications from the same dataset (by the same or different authors) is the analysis of a ‘patient registry’. A patient registry collects and organises data to help evaluate outcomes for sub-populations defined by a particular disease, condition, or exposure. By definition this serves several scientific or clinical purposes. Types of registries include: product registries (where patients have been exposed to medical devices or drugs); health services registries (patients who have had a common intervention); and disease registries (where patients have the same condition). A single registry can integrate data from various sources. The primary data are collected by the registry for its immediate purpose(s), but secondary data are collected en passant, and could include several types of information related to patient characteristics or habits unrelated to the primary question. Once many registries are created it is possible to link them to provide a wider range of available data. For example, a patient may enter one registry by virtue of having a particular condition (and relevant data exists related to that within this registry). The same patient may also enter another completely different registry, for example, related to his professional status or income, and this registry could contain all manner of additional data not captured by the first. Linking the two datasets thus could address novel hypotheses of interest (e.g. related to whether income status determines disease outcome, etc). Ethical issues are important in registries and perhaps there has been a change in attitudes to allow greater use and interrogation of registries. A fundamental principle is that patients should provide information to the registry voluntarily, and know the purpose for which they are providing that information. Thereafter, it is the flow of information identifiable to a specific patient that is a matter of ethical concern, and not the flow of any information not identifiable. Thus if Mr Smith is identified as having heart failure in one registry, then the registry should transmit only the information that patient A has heart failure (taking care also to remove other identifiable information such as address, date of birth, etc) if the registry is to be interrogated for a use not known to Mr Smith. This issue of Anaesthesia contains a publication from Redman et al. that is relevant to this debate 10. In this article, the authors acknowledge that they have conducted further analysis of the same database used in an earlier publication 11. The authors explain this was justified to examine a particular subset of patients with HIV about whom few data are published. Furthermore, the same database was employed for an earlier publication 12, but this study looked at different endpoints and the data analysis was conducted differently. Finally, the data were also included in a meta-analysis 13, along with that from six other studies. The question is, does this number of publications from a single database represent salami slicing or is it justified? The authors’ explanation was that within each publication the hypothesis was different, as were the analyses and outcomes, or a specific reason for further analysis existed (in the case of the most recent publication about HIV patients). On balance, reviewers and editors decided that the value gained justified this further publication. With respect to the specifics of this manuscript and its relevance to HIV, it has been assumed for many years that HIV-positive patients with access to antiretroviral therapy will have a normal or near normal life expectancy if the virus is kept under control 14. However, as the HIV population starts to age, there are concerns that cardiovascular disease may manifest at an earlier age than in control populations, as HIV-positive patients appear to be at an increased risk of vascular events. Uncontrolled HIV disease is associated with raised vascular inflammatory markers, and increased risk of cardiovascular events is seen in patients who interrupt their treatment and then become viraemic again 15. The degree to which associated immune dysfunction plays a role in this phenomenon is not known. However, it is debatable whether patients who are well controlled on antiretroviral therapy with a supressed viral load and immune reconstitution also have an increased risk of cardiovascular disease related to their chronic HIV infection. There are concerns that in patients controlled on antiretroviral drugs, toxicities may influence the incidence of cardiovascular disease. Dyslipidaemia is seen in untreated and treated patients and persistently low high-density lipoprotein levels are often recorded, as are antiretroviral drug-induced increases in triglycerides. Interestingly, large observational cohort studies have shown that the use of specific drugs, in particular abacavir and lopinavir, is associated with an increased risk of myocardial infarction not explained by lipid abnormalities 16. Conversely, this effect of abacavir was not seen in a meta-analysis of recent randomised clinical trials 17. Much of the data coming from cohort studies have the problem of potential unmeasured confounders when calculating rates of cardiovascular disease and comparing them with a normal population. Although an increased incidence of cardiovascular disease could be the result of HIV infection, any increase could simply reflect the background rate expected with the demographic characteristics and risk factor profile of an HIV-positive population. Patients who are HIV-positive are more likely to smoke and therefore at risk of smoking-relating diseases, vascular co-morbidities associated with other co-infections such as syphilis and hepatitis C, and recreational drug use. Social class and poverty would have an important impact on many of these factors. In fact, a study of mortality rates within an HIV-positive population has shown showed significantly different mortality depending on sex, educational status, and race/ethnicity 18. However, a recent study tried to control for these factors and showed a 50% increase in myocardial infarction rates in HIV-positive patients compared with controls 19. Redman et al.'s study has shown that HIV-positive patients undergoing vascular surgery are younger and have fewer cardiovascular risk factors than HIV-negative patients, but their mortality after surgery is very similar 10. It is not surprising that the HIV-positive patients had fewer cardiovascular risk factors when looking at their age distribution. So why was their outcome similar to older patients who did not have HIV? The number of patients studied was small but was HIV the major determining factor? It is possible that the findings of this paper may be independent of the HIV disease itself and it is interesting to note that those on or off HIV therapy had similar outcomes. It is also conceivable that unmeasured factors may have played a role in the outcome, including nutritional status and co-morbid conditions. The indications for the procedures themselves may have an impact on these data, especially as the HIV-positive patients had twice the rate of amputations. If these were because of road traffic injuries or gangrene secondary to intravenous drug or cocaine use, for example, then it would have had a major impact on outcomes and explain the differences. In order to unravel the effect of HIV from other causal factors, well-designed prospective trials or large cohort studies examining outcomes in patients with HIV are required. At least based on the evidence reported here, being HIV-positive should not be a reason for exclusion from vascular surgery, but as the authors suggest, optimisation of cardiovascular medical therapy and postoperative troponin surveillance should be considered. In conclusion, Redman et al. were justified in undertaking further analysis of their data registry because of the particular nature of the research and the difficult questions they asked. However, all authors are urged to be scrupulous in avoiding salami-slicing when conducting and publishing repeated analysis of data. Authors are urged especially to acknowledge previous analyses clearly, and to ensure that they have ethical permission and consent for using and publishing the data in such a manner. No external funding or competing interests declared.