Salami publication
A florid example from BJOG
Salami publication is a pernicious type of scientific misconduct where authors publish slightly different aspects of one research study in different papers. For example, publishing baby outcomes in one article and mother outcomes in another. Editors frown on it because it is wasteful, and makes the literature difficult to interpret. Authors do it to pad out their CVs and, by self-citing, increase their citation count. The Committee on Publication Ethics (COPE) (click here) says:
- Related questions, or very closely related, […] should be published as a single paper.
- Salami publication is where papers cover the same population, methods, and question.
- Splitting up papers by outcomes is not legitimate.
Five papers about the effects of alcohol consumption early in pregnancy on fetal development, spread over 56 pages of BJOG this month, are the worst examples I’ve ever seen (Table).
Each paper reported on the same subjects, and same exposure variables. All but one adjusted for the same covariates. The outcomes differed in papers 2, 3 and 4, but paper 1 examined all three, and paper 5 examined the same ones as paper 2 but adjusted for different covariates. Each paper cited between two and four of the co-papers. The textual overlaps amount to about 1/3 of each paper.
| 1 Kesmodel (a) | 2 Falgreen-Erisksen | 3 Skogerbo | 4 Underbjerg | 5 Kesmodel (b) | |
| Subjects | 1628 women sampled from Danish National Birth Cohort | 1628 women sampled from Danish National Birth Cohort | 1628 women sampled from Danish National Birth Cohort | 1628 women sampled from Danish National Birth Cohort | 1628 women sampled from Danish National Birth Cohort |
| Exposure | Average and binge drinking before, and at four time points before 9 weeks in pregnancy | Average and binge drinking before, and at four time points before 9 weeks in pregnancy | Average and binge drinking before, and at four time points before 9 weeks in pregnancy | Average and binge drinking before, and at four time points before 9 weeks in pregnancy | Average and binge drinking before, and at four time points before 9 weeks in pregnancy |
| Co-variates | Parity, smoking, pre-pregnancy BMI, parental education, marital status, child health, drug use, family environment index, maternal depression, parental alcohol at 5 years. | Parity, smoking, pre-pregnancy BMI, parental education, marital status, child health, drug use, family environment index, maternal depression, parental alcohol at 5 years. | Parity, smoking, pre-pregnancy BMI, parental education, marital status, child health, drug use, family environment index, maternal depression, parental alcohol at 5 years. | Parity, smoking, pre-pregnancy BMI, parental education, marital status, child health, drug use, family environment index, maternal depression, parental alcohol at 5 years. | parental education, smoking, child’s gender, age at testing. |
| Outcome variables, all measured at age five years. | IQ by WPPSI-R subdivided into VIQ and PIQ, Attention by TEACH-5, parent measure of child executive function by BRIEF. | IQ by WPPSI-R subdivided into VIQ and PIQ. | Parent and teacher measure of child executive function by BRIEF. | Attention by TEACH-5. | IQ by WPPSI-R subdivided into VIQ and PIQ. |
| Co-papers cited | 2, 3, 4, 5 | 1, 5 | 1, 2, 5 | 1, 2, 5 | 1,2 |
| Conclusions | “The present study suggests that small volumes consumed occasionally may not present serious concern | Maternal consumption of low to moderate quantities of alcohol during pregnancy was not associated with the mean IQ score of preschool children | This study did not observe significant effects of low to moderate alcohol consumption during pregnancy on executive functioning at the age of 5 years | The study detected no effects of lower levels of maternal consumption | We found no systematic association between binge drinking during early pregnancy and child intelligence. However, binge drinking reduced the risk of low, full-scale IQ in gestational weeks 1–2. |
In an earlier report of the study methods (Kesmodel 2010) the authors indicated that they planned to measure another eight baby outcomes at age five.
- Learning/motor skills measured by the Animal house (WPPSI-R),
- Reaction time and speed of information processing measured by the Sternberg task – KVC (Kilburn’s version for children),
- Social skills and behaviour measured by the strengths and difficulties questionnaire (parent and teacher part) – modified version,
- General developmental status measured by Draw a person,
- Behaviour during test session measured by the Behaviour Checklist (clinical rating),
- Vision measured by the Østerberg vision board,
- Hearing measured by Audiometry,
- Motor development (fine and gross) measured by the Movement ABC,
Are we to expect another eight papers reporting these outcomes?
It is particularly disappointing to see this in BJOG, since in the past the journal has taken a robust attitude to this sort of scientific misconduct. The editors not only condoned the redundant publication but wrote an “Editor’s choice” and a commentary about them – salami editorialising!
The message from all this verbiage?
They failed to show any harm from low levels of alcohol consumption but since it is impossible to prove a negative they wisely conclude that to play safe pregnant women should not drink at all. Of course the press took the opposite message. Click here and here for examples. And now the health community is frantically backpedalling.
What a waste of millions of pounds to pad out a few epidemiologists’ CVs and confuse the public!
Jim Thornton
Ripe-tomato.org 
The editor of BJOG, Phil Steer, has written defending the decision to publish on the grounds that the alcohol authors, who originally submitted seven papers, were at least open about the multiple publication. The editors wanted them combined into three papers and eventually compromised on five. He says, “When is salami not salami? When it is joined together to recreate the original sausage.” Good point. They were verbose but not deceptive.
But I still think the editors should have insisted on one paper for one study. The methods were standard – a data base with exposure, co-, and outcome variables. For each analysis they needed to describe how they dealt with missing values, and which statistical test they used. With two more rows the methods of all five papers could be presented in a single table like the one above. With results in another it would have been more readable and saved 30 or 40 pages.
Jim Drife (former BJOG editor) writes, “Your blog is spot-on. Quite a few research teams have been producing finely-sliced salami for many years. Let’s hope your well-illustrated and entirely justified criticism leaves their ears burning. Editors have suffered in (relative) silence for long enough.”