Turning Three Pet Peeves into Unsolicited Advice

For Journalists, Researchers, and Editors

I’m at the age where pet peeves seem to be easy to come by; no need to go to a breeder or shelter. Three of my newest are discussed below, presented as advice to make this piece more positive than it could have been. One could argue that I lack the standing to write this article. I am not a journalist, a researcher, or a journal editor. That said, I read a lot of medical research articles and the lay press articles that cover them. This advice could make the literature more valuable and patient-centered and improve the coverage of medical research.

For journalists, when reporting on the results of observational studies, please include at least one paragraph describing a plausible interpretation of the data other than that advanced by the authors. We know that when clinical questions are addressed by both experimental and observational studies, the observational study comes to the wrong conclusion at least 20% of the time. I expect these errors are usually due to residual confounding.

We recently discussed an article on Fortnight, one that was widely covered in the lay press, that could have benefited from a discussion of alternative conclusions. The article compared cancer rates in obese people who began a GLP-1 vs. those who were prescribed diet or exercise counseling. Cancer diagnoses were recorded for the 2 years following prescriptions for GLP-1, diet, or exercise. The article showed that those who started the drugs developed fewer cancers (HR 0.59, 0.53-0.67).

Many of the articles in the lay press cautioned that, because this was an observational study, it did not prove causation. None that I read, however, explained what else might be happening. Could it have been that patients destined to develop cancer in the next two years were less likely to be prescribed GLP-1 because the physician was concerned about their health? Could it have been that the patients were already feeling poorly and did want to start a new medication that might leave them nauseated? Given the rapidity of the proposed GLP-1 effect, this alternative interpretation seems more likely than the one advanced.

Providing alternative explanations would not just make articles better; it would make them more interesting.

For those planning to perform an observational study, before you begin work, consider why you are doing the study. If previous observational studies support a hypothesis regarding a treatment, what is probably needed is an RCT. What is probably not needed is another observational trial. Performing an observational trial might benefit your CV, but it is unlikely to benefit patients. Two examples of this come to mind: the first was discussed on the same Fortnight episode.

This study sought to determine whether Amoxicillin-Clavulanate or Amoxicillin is more effective for acute sinusitis. The author wrote the following (my parentheses are added):

A prior (observational) study of nearly 200,000 children with acute sinusitis found no difference in treatment failure between amoxicillin-clavulanate and amoxicillin.

A study in Veterans Health Administration emergency department and urgent care settings (also observational)… reported similar results. To the study investigators’ knowledge, no randomized clinical trial has compared the 2 treatments since a study of 188 children published in 2001, which found no difference in treatment failure.

This paragraph is a perfect lead-in to describing the robust RCT that is about to be presented. Instead, we get another observational trial. Which does nothing to advance knowledge.

Another nearly identical example was published in 2025. This article examined the benefits of early oseltamivir therapy in adults hospitalized with influenza A. Again, this was a well-done observational trial. But it essentially replicated past observational trials, which had themselves already been replicated. To make my point stronger, information is beginning to come out from an RCT that all these observational studies by have missed the mark.

When there is a biologically plausible hypothesis about a therapy that has already been supported by more than one observational trial, and perhaps often further supported by clinical experience, the next study should be an RCT. Only then can we be sure we are doing the right thing.

For those planning a clinical trial, if you are testing a new drug, one for which there is a similar, older, less expensive drug, please compare your fancy new drug not only to a placebo but to that old treatment as well.

A recent RCT in the NEJM studied the effect of finerenone in people with CKD but without Diabetes. Finerenone is a selective nonsteroidal mineralocorticoid receptor antagonist that is purported to have fewer adverse effects than spironolactone. Patients in the study had CKD with an eGFR of 25 to <90 and a urinary albumin-to-creatinine ratio of 200 to ≤3500. The patients were on an ACE or ARB. They were randomized to finenerone or placebo. The finenerone group did a little better than the placebo group with a slower decline in GFR. Wouldn’t it have been interesting to know whether spironolactone was as effective, what the difference in side effects was, and, if there was a difference, whether it was worth the cost?

Again, another nearly identical example was the well-known EMPEROR-Preserved Trial. This trial randomized patients with heart failure with a preserved ejection fraction to empagliflozin or placebo. Here, the question was why there was no 3rd group of patients randomized to an additional (or increased) dose of a diuretic? Given that the trial only revealed an improvement in HF hospitalizations (not even overall hospitalizations), I am suspicious that a third arm would have at least tied with empagliflozin, at lower cost and with fewer UTIs.

These three pieces of unsolicited advice could also be used by journal editors. Ask the authors of observational studies to articulate other explanations of the relationships they find. Reject observational studies that add nothing to medical knowledge, studies performed when the only research warranted is a controlled trial. And lastly, make sure the new drugs are compared to reasonable alternatives at appropriate doses, even if it is ethical to compare only to placebos.

.