Corrections
What we changed after publishing it, and why.
13 changes
Almost no site does this, which is exactly why it is here. An article that quietly rewrites itself is a blog. An article that shows its own history is a document you can trust twice.
We separate two things that get run together everywhere. A correction is us having been wrong. An update is the evidence having moved. Calling everything an update hides the corrections among them.
Kind
Subject
Year
Showing 13 of 13
Correction
The database file had 349 rows while the article says 341 studies. Eleven rows were a second dose of a study already listed, and three PubMed records were missing. The file now has 341 rows, one per study. The study-type column was guessed from the title and was often wrong, a case report of a brain haemorrhage in a patient was filed as an animal study, and it is replaced by PubMed MeSH terms. The doses had been pulled out of the abstracts by a regex that ate thousands separators and decimal points, so 4.7 FU was stored as 700 FU; they are now printed exactly as the abstract has them. The file is also available in English.
Update
And back to reasonable, this time with a rule rather than a judgement. The scale was built for claims about an effect, and whether a measurement predicts something cannot be randomised: nobody can be assigned an age or a blood value. There a second population counts as the second test. PhenoAge was built on NHANES III and holds up in NHANES IV, so reasonable. Strong stays out of reach, because what earns that rung cannot exist here in principle. The exception is written on the methodology page and on the evidence list.
Correction
Confidence moved from weak to limited. The trials do not all agree, which is not the same as barely having a conclusion.
Correction
Removed a claim that almost no seller publishes a transmission figure. Most do publish a percentage. What they do not publish is a spectrum, and that is what the sentence should have said.
Update
The article listed a piece on blue light and the circadian rhythm as still to come. It was published, so this now links to it.
Correction
The title said "what 341 studies show". PubMed returned 341 records, 107 of which have a person in them, and 13 are cited. It now says "from 341 studies".
Update
Given a confidence rating of weak, which is what "nobody has ever tested the strap being sold" deserves.
Correction
A heading read "Where it did go wrong" without saying what "it" was. The section is about a quartz heat lamp somebody looked into, not about the therapy trials, and it now says so.
Correction
Confidence moved from moderate to limited. PhenoAge rests on cohort data and no trial, and our own rule reserves the higher rungs for randomised evidence.
Correction
Confidence moved from strong to limited. We had it at strong because the prediction is about as solid as prediction gets, but our own rule reserves strong for randomised trials. This research is observational from beginning to end.
Correction
Confidence moved from moderate to limited: the evidence is strong for some uses and thin for others, which is not a body that agrees with itself.
Correction
The title said "what 2,225 studies show". The search returned 2,225 records and roughly 430 were read in full. It now says a 2,225-study search.
Update
Reclassified from a verdict to a measurement. A laboratory analysis of what is inside a tub is a reading, not an estimate of an effect, so it carries no confidence rating.