Pharmaceutical Companies and Misleading Claims in the Literature

I remember reading somewhere—I forget where—that these companies will often publish their material in-house (i.e. rather than going to an independent, well-established journal, they publish under their own company). Surely, there’s decent material even there, but if one desires to read good evidence, why bother; just read independent journals.

Nevertheless, even some of these papers may be colored by the industry. A discussion of six recommendations by Montori et al.¹ follows.

Simply read the methods and results sections; disregard the discussion; interpret the stats on your own. Even if the methods are sound (which is a dangerous assumption), the discussion and the figures that the authors choose to present (e.g. RRR over ARR) may delude readers into judging the drug/therapy far more generously than what truth condones.

Read preappraised sources like the ACP Journal Club. The abstracts presented within these journals provide a more objective review of such papers, often including critical information such as blinding and completeness of follow-up. The abstracts are written by people without fiscal engagement or personal interest in the matter.

Faulty comparators: why are the authors using a placebo rather than the currently established drug/therapy? This would apply to active-comparator studies that look at whether the new drug is better than an existing treatment.

The ambiguity of composite endpoints allows researchers to make bolder statistical claims by muddling the most important endpoint. As an example, the paper presents the following endpoints combined: death, end-stage renal disease, and doubling of serum creatinine concentration in the study of diabetic nephropathy. Clinically significant endpoints and surrogates are mixed together, even though these endpoints should, in reality, be sorted out and studied individually—in this paper at least, since it conspires to use composites as a statistical farce.

Most composites are used because measuring one endpoint alone—such as death—can be statistically inefficient. That is, if death occurs infrequently, you, as a researcher, need a lot of people to die before making any strong claims. By combining the aforementioned example endpoints and counting each endpoint as one event, researchers effectively increase statistical efficiency.

Small treatment effects may also be indicative of questionable interpretation. Choosing to display RRR rather than ARR, for instance, often de-emphasizes or completely ignores the latter.