‘Influence of conflicts of interest on public positions in the COVID-19 era, the case of Gilead Sciences’ by Roussel and Raoult (2020)

Abstract

Roussel and Raoult (2020) assessed the correlation between public positions taken by 98 doctors towards hydroxychloroquine and payments received from the Gilead Sciences company. The analysis presented has two major problems, the first of which can probably be fixed retrospectively, but the second of which seems irredeemable.

Roussel and Raoult [1] recently assessed the correlation between public positions taken by 98 medical researchers towards hydroxychloroquine and payments received from Gilead Sciences. The analysis is based on a 5-point ordinal score for the public positions expressed by doctors in the media (academic or not academic) based on systematic search on Google News. The ordinal scale for the opinion scoring ranges from ‘very favourable’ to ‘very unfavourable’. This score placed each doctor in one of the five opinion categories. For each category, the average amount received per person was computed on the basis of payments received as published online (https://transparence.sante.gouv.fr/). The results are presented in their Table 1. For the five average amounts, the authors compute Spearman's rank correlation with the corresponding ordinal opinion score. They find a correlation of 1 with an associated p-value of 0.017.

The analysis presented has two major problems, the first of which can probably be fixed retrospectively but the second of which seems irredeemable.

Firstly, the data analysed are highly aggregated, giving rise to only five aggregate observations. It is unclear why and how this aggregation was done. Clearly the aggregation is the cause of the perfect correlation: The average payment in the most unfavourable category ‘Very unfavourable’ is the highest, the average payment in the second most favourable category ‘Unfavourable’ is second highest, and so forth down to the fifth and most favourable category, ‘Very favourable’, which has the smallest mean payment. Note, however, that only a single rank change would have rendered the correlation insignificant (p =.0833) by an exact test. More importantly, the aggregation is misleading because it exaggerates correlation. At the same time, the aggregation compromises statistical power. The authors have the scores for the individual doctors, so they could have, and should have, computed the rank correlation at the level of individual doctors. The doctor-level correlation is probably much lower than 1. Were the raw data of this study provided with the article, as should be standard procedure, readers could compute this correlation and the associated p-value for themselves.

Secondly, the authors provide a verbal description of the categories for the evaluation of opinion statements. They do not mention any blinding procedure, which would have been critical for this kind of evaluation. It therefore cannot be ruled out that they have seen the payments received by the 98 doctors before evaluating their published opinion statements. Apart from the brief verbal description of the categories, the exact procedure used for evaluating the opinion statements is lacking, so a biased assessment cannot be ruled out. Following common standards, all articles identified in their systematic search should be made publicly available, identifying the passages on which the assessment is based, along with the score assigned to each statement or opinion piece. This would allow readers a transparent evaluation of plausibility of the assessments. Furthermore, it is likely that there were several publications or opinion statements per doctor, and it is not clear from the article how assessment proceeded in such cases. What was done when, for example, in one article the doctor expressed a ‘very favourable’ opinion, and in another article the same doctor expressed a ‘very unfavourable’ opinion towards hydroxychloroquine? How was the overall score for this doctor obtained?

Conflict of interest

None declared.