Three journals, three fixes, one honest map of what peer review missed
JAMA and The Lancet issued four corrections this week, from a misspelled name to a mislabeled drug mechanism.
By The Weekend · · 5 min read
Four corrections landed across two of medicine's most-cited journals this week. One fixed a misspelled surname. Three fixed things that could change how a doctor reads a trial: an axis label on a heart-valve study, a drug class description in a kidney-disease review, and — buried in the driest possible language — an antibiotic trial correction that touches how bacteremia gets treated. None of this is scandal. It is the system working exactly as designed, which is itself the story worth telling.
The correction that matters is the one about a receptor, not a name
David Fiellin's surname being misspelled in a JAMA reply letter about opioid withdrawal is a clerical fix — embarrassing for a proofreader, irrelevant to any patient. Nobody's treatment changes because of it. The other three corrections, all from The Lancet, sit on a different shelf entirely.
The most consequential is buried in a Series paper on chronic kidney disease. A table describing endothelin receptor antagonists — a drug class increasingly used to slow kidney decline — had the mechanism backwards. The corrected text specifies that endothelin receptor A activation, not blockade, drives the harmful effects the drugs are designed to counter. That is not cosmetic. A clinician or researcher skimming that table for how the drug class works could walk away with an inverted model of what the medicine does. Kidney disease affects an estimated 850 million people worldwide, and endothelin antagonists are among the newer additions to a fairly short list of drugs shown to slow progression. Getting the receptor biology backwards in a widely read Lancet Series is the kind of error that gets copied into slide decks, grant applications, and eventually clinical guidelines before anyone catches it.
An axis label sounds trivial until the trial is a coin-flip result
The COMPARE-TAVI 1 correction concerns Figure 2 of a trial comparing two brands of replacement heart valve — SAPIEN 3 and Myval — used in transcatheter aortic valve implantation, a procedure done tens of thousands of times a year for patients too frail for open-heart surgery. The error sits in the positioning of a "–1.2" label on the x-axis and a border between categories. Non-inferiority trials like this one live or die by exactly where a confidence interval crosses zero on a graph. If a reader's eye is drawn to a mispositioned gridline, the visual read of "these valves perform about the same" versus "one edges out the other" can shift without a single number in the actual dataset changing. The correction restores the graph; it doesn't touch the underlying non-inferiority conclusion, but it is a reminder that in trial reporting, the picture often gets read before the text.
The third Lancet correction concerns the SNAP trial, an international study comparing benzylpenicillin against flucloxacillin or cloxacillin for treating Staphylococcus aureus bloodstream infections that remain susceptible to penicillin — a narrower, cheaper, older antibiotic option than the alternatives most hospitals default to. The snippet available doesn't specify which figure or table needed fixing, only that one did. Bacteremia from S. aureus kills roughly 20 percent of patients within 90 days even with treatment, so trials comparing antibiotic choice for it get read closely by infectious-disease pharmacists deciding formulary defaults. A quiet correction to a trial this consequential is worth flagging even before its full content is public.
Where this lands: the reader who never sees the correction notice
Corrections run in the same journal, often months later, in a section almost nobody outside academic medicine opens. A nephrologist who read the kidney-disease Series when it published in a prior issue has no automatic prompt to revisit the corrected table. A cardiologist deciding between valve brands for a specific patient is relying on the original PDF sitting in a reference manager, not on whichever fix ran afterward. The correction exists; the propagation of the correction into the readers who need it does not. That gap — not the original errors — is the actual mechanism worth understanding.
The history: peer review was never built to catch typography
Peer review, as practiced by JAMA and The Lancet, is designed to interrogate methodology — sample size, randomization, statistical approach, conflicts of interest. It is not designed, and was never designed, to proofread axis labels or re-derive receptor pharmacology from first principles in a 60-row table. Reviewers are given weeks, work unpaid, and are asked to judge whether a study's conclusions follow from its data — not to typeset-check figures or fact-check every mechanistic aside in a 6,000-word Series article.
That division of labor has held for decades because most errors it lets through are genuinely inconsequential — a misspelled name, a rounding difference in a supplementary table. The system assumes errors that matter will get caught downstream, by other researchers trying to replicate or build on the finding, and that assumption mostly holds. But "mostly" is doing real work in that sentence. The Lancet issues corrections at a steady clip — several a month across its journal family — and the majority are exactly this kind: not fraud, not retraction-worthy, just human error in a high-volume, fast-moving publishing operation that pushes hundreds of pages of dense technical content through a review cycle measured in weeks.
The catch nobody mentions: corrections aren't indexed the way original papers are
A reader searching PubMed or Google Scholar for the COMPARE-TAVI 1 trial finds the original article first, ranked by citations and relevance. The correction notice is a separate, thinly cited entry that most citation-tracking tools don't automatically surface alongside the paper it fixes. Some journals now append correction flags directly to the original PDF; The Lancet's "Department of Error" notices are freestanding articles, discoverable only if a reader happens to check. There is no industry-standard mechanism forcing a correction to travel with its parent paper into every downstream citation, slide deck, or guideline committee that already used the uncorrected version.
The honest question this leaves open
Nobody tracks, in any systematic way, how many clinical decisions get made using a paper's incorrect version before its correction is issued — or how many keep being made afterward, by readers who cite the original PDF without checking for updates. That number would tell you whether corrections are a housekeeping formality or a live patient-safety gap. It doesn't exist, in either journal's public reporting or anywhere else.
Does a correction notice mean the trial's conclusion changed?
No — in all three Lancet cases here, the correction fixes a table entry, a figure label, or unspecified content without altering the trial's stated primary result.
Who catches these errors, generally?
Overwhelmingly other researchers or readers writing in after publication, not the original peer reviewers or in-house editors.