Unusual clustering of coefficients of variation in published articles from a medical biochemistry department in India

Abstract

A simple statistical method is described to test whether data are consistent with minimum statistical variability expected in a biological experiment. The method is applied to data presented in data tables in a subset of 84 articles among more than 200 published by 3 investigators in a small medical biochemistry department at a major university in India and to 29 "control" articles selected by key word PubMed searches. Major conclusions include: 1) unusual clustering of coefficients of variation (CVs) was observed for data from the majority of articles analyzed that were published by the 3 investigators from 2000-2007; unusual clustering was not observed for data from any of their articles examined that were published between 1992 and 1999; and 2) among a group of 29 control articles retrieved by PubMed key word, title, or title/abstract searches, unusually clustered CVs were observed in 3 articles. Two of these articles were coauthored by 1 of the 3 investigators, and 1 was from the same university but a different department. We are unable to offer a statistical or biological explanation for the unusual clustering observed.

Analysis of data integrity

“…Constructed 50% limits are illustrated by horizontal lines. Reproduced with permission from Hudes et al [2].…”

“…We analysed a subset of 84 articles among more than 200 published collectively by these investigators over a period of only about five years. With the assistance of the Editor-in-Chief at FASEB Journal, we were able to publish, back-to-back in the same issue, our analysis [2], a critique co-authored by the three investigators [3], and our reply to them [4]. That reply included an attempt to explain our analysis in simple language, similar to Professor Pandit's goal in his editorial [5] discussing Carlisle's analysis [1].…”

“…We would like to congratulate Dr Carlisle on the Herculean feat of analysing 168 randomised controlled trials (RCTs) [1], having ourselves recently completed a statistical analysis of a case involving all three principal investigators in a small medical biochemistry department at a major university in India [2][3][4]. After spending almost three years and many hundreds of hours, we understand how much time and effort it takes to undertake this type of analysis.…”

“…Twenty-nine 'control' articles were also selected, all from different laboratories, using unbiased PubMed key word searches [2]. Only one of these 29 papers demonstrated 'unusual clustering', which to our surprise turned out to be a paper from the same institution as the three investigators, but from a different department.…”

Calculating the probability of random sampling for continuous variables in submitted or published randomised controlled trials

“…Mathematical analysis of clinical trial data has suggested that some data may be unreliable or even fabricated [1,2]. The 2012 analysis of randomised controlled trials by Dr Yoshitaka Fujii et al showed that baseline group data, such as age, height or weight, were very unlikely to have been generated through random allocation [3][4][5].…”