​*Editorial Office Note: An investigation has been launched into the concerns listed in the letter. A Statement of Concern has also been published in Volume 99, Issue 29 of Medicine.*

We applaud the application of a randomized design by Abdelbasset et al. in their trial on the effects of exercise in nonalcoholic fatty liver disease patients (1). However, the results appear to have been presented in duplicate among three separate publications*, and the reporting of the statistical tests is incorrect.

First Major Concern: Apparent Duplicate Publication. The publication by Abdelbasset et al. 2020 in Medicine (2) reports a three-arm trial with a high-intensity interval (HII), moderate-intensity continuous (MIC), and control group. Two other publications report the same outcomes for only two of the three randomized groups. The outcomes presented in (2) are: BMI, intrahepatic triglyceride, visceral adipose fat, total cholesterol, triglycerides, LDLc, HDLc, ALT, HbA1c, and HOMA-IR. A publication in Evidence-Based Complementary and Alternative Medicine (3) reports these same results limited to the MIC and control groups, and a publication in Medicine (4) reports these same results limited to the HII and control groups, with the addition of VO_2peak, VLDL-TG, glucose, and outcomes from a chronic liver disease questionnaire (see our Table A below for a comparison of outcomes among papers). Of note, VO_2peak is the only outcome listed in the retrospective trial registration (1). Using the software iThenticate (5) to detect text overlap, even without considering tables, paper (3) and (4) were 33% and 38% similar with paper (2).

While paper (2) includes a trial registration number (NCT03774511), the other reports do not (3,4). The inclusion of a registration number, as required by International Committee of Medical Journal Editors (ICJME) and CONSORT guidelines (11, 12), can help clarify when reports are related to the same participants.

Additionally, the authors cross-cite their papers as support for one another even though they are the same trial. For example, in the discussion of the most recent paper (3), they write "Similarly, Abdelbasset et al. approved[sic] that cycling exercise with 80% to 85% VO_2max and interval at 50% VO_2max for 40 minutes 3 times weekly for eight weeks showed a fluent decrease of hepatic triglycerides, visceral fats, and insulin resistance in diabetic obese patients with NAFLD [35, 36]", where these references correspond to papers (2) and (4). In another instance, they write in (2): "Previous studies showed that high-intensity interval exercise (HII) improves hepatic function and glucose metabolism [24–26]", where reference 26 corresponds to paper (4).

Duplicate publications raise multiple ethical issues, including a waste of publishing resources and peer reviewer time for the duplication of results. Moreover, because participants were randomized to one of three groups, selectively reporting only two of them in two of the papers is misleading. As described in ICMJE and Committee on Publication Ethics (COPE) guidance, redundant publications may lead to unintentional double-counting of data, for instance in meta-analyses, and therefore they should be retracted (9,10).

Second Major Concern: Invalid Statistical Analysis. Some of the conclusions from (2) are derived from comparing the binary statistical significance of within-group changes over time from one group versus those of another group. This approach, which we call the "Differences in Nominal Significance Error", is invalid (6). As Bland and Altman have written (7,8), doing so for randomized trials may yield drastically higher type I error rates and produce incorrect conclusions. The appropriate test is to compare between the groups directly, such as through a repeated measures ANOVA, ANCOVA, or post-only comparison.

For example, the authors summarize that "The study outcomes approved[sic] that both HII and MIC exercise programs reduce BMI, IHTG, visceral lipids, insulin resistance, and HbA1c" (2) based on within-group rather than between-group tests. We estimated between-group comparisons incorporating both pre and post means and errors (see our Table B below; R code to reproduce our calculations is available at: 10.17605/OSF.IO/D3JHZ). For many of the outcomes, our estimates of between-group tests changed the statistical conclusions compared to the invalid within-group tests. Specifically, the authors reported that hepatic fat, IHTG, visceral adipose fat, BMI, HbA1c, and HOMA-IR differed between at least one of the exercise groups and the control group, but the between-group comparisons were not statistically significant. Because our estimates are based on rounded values as reported in (2), data should be reanalyzed appropriately, and conclusions corrected as needed.

Between-group tests are reported in the other two papers using only the final measure values (3,4), although they do not seem to be correctly calculated. For example, independent t-tests are reported in reference (3) for MIC vs control, and differences between groups at p < 0.05 are reported "between the two groups after intervention" for all outcomes except triglycerides. However, we could not replicate many of the p-values using two-tailed t-tests and the reported means and errors, and half of the statistical conclusions changed (Table C below).

Conclusion. The conclusions reported by Abdelbasset et al. are based on erroneous analyses and are not supported by the data, and the papers seem to be inappropriate duplicate publications. Corrective actions should be taken to preserve the integrity of the scientific record.  

Author Correspondence

Colby J Vorland, Department of Applied Health Science, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA

Greyson Foote, Department of Epidemiology and Biostatistics, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA

Stephanie L Dickinson, Department of Epidemiology and Biostatistics, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA

Evan Mayo-Wilson, Department of Epidemiology and Biostatistics, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA

David B Allison, Department of Epidemiology and Biostatistics, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA

Andrew W Brown, Department of Applied Health Science, School of Public Health - Bloomington, Indiana University, Bloomington, IN, USA