Clinical claims have a shelf life. Almost none of them carry the date they were last checked, and a claim that has been withdrawn reads exactly like one that still holds. This is a method for telling the difference before you act on it.
Medical knowledge is usually described as growing. It also decays, and the decay is fast enough to matter inside a single illness. Researchers tracked 100 quantitative systematic reviews for the appearance of new evidence important enough to warrant updating them, and found that such a signal appeared within two years for 23 percent of the reviews, with a median survival free of any such signal of 5.5 years (Shojania et al., Annals of Internal Medicine, 2007). Those reviews were not wrong when published. The ground moved underneath them.
Over longer horizons the picture is similar. Investigators assessed 474 conclusions drawn from the cirrhosis and hepatitis literature published between 1945 and 1999 and judged that, as of 2000, 60 percent were still true, 19 percent had become obsolete, and 21 percent were false. They put the half-life of a clinical conclusion at roughly 45 years (Poynard et al., Annals of Internal Medicine, 2002). Two fifths of what a field once accepted did not survive.
What makes this a practical problem rather than a philosophical one is that expiry leaves no mark on the page. A printed handout, a cached search result, a patient-education leaflet, a hospital intranet page, and a model trained on all of the above will state a withdrawn recommendation in exactly the tone they use for a current one. The reader has no surface cue. The date the claim was made is often missing, and the date it was last reviewed is almost always missing, which is the more useful of the two.
So the question worth asking about any clinical statement is not only whether it is well sourced. It is when the source was last checked, by whom, and against what.
The unit that expires is not the article. It is the individual claim. One paper can contain a finding that still holds, a dosage that has been revised, and a recommendation that has been withdrawn, and the paper itself carries a single publication date that tells you about none of them.
Any given claim sits in one of four states. It is current, meaning someone has looked at it recently against newer evidence and it survived. It is aged, meaning no one has looked, so its status is unknown rather than confirmed. It is superseded, meaning a later authoritative document has replaced it, whether or not the older document has been taken down. Or it is retracted, meaning the record itself has been withdrawn.
The trap is the second state. Silence reads like endorsement. A claim that has sat unexamined for fifteen years looks identical to one that was reaffirmed last month, and it is often the older claim that has propagated furthest, because it has had longer to be quoted, translated, reprinted, and absorbed into training data. Distinguishing "still true" from "still posted" is most of the work.
On 22 August 2007, the National Institute for Health and Care Excellence published clinical guideline CG53 on chronic fatigue syndrome. It stated that cognitive behavioral therapy and graded exercise therapy "should be offered to people with mild or moderate CFS/ME... because currently these are the interventions for which there is the clearest research evidence of benefit" (CG53, recommendation 1.6.2.4). On 29 October 2021, NICE published guideline NG206, which updates and replaces CG53. Fourteen years separate them, and the second reverses the central management advice of the first.
The replacement is specific, and the specifics matter more than the summaries of it that circulate. NG206 tells clinicians not to offer people with ME/CFS "any therapy based on physical activity or exercise as a cure for ME/CFS," not to offer "generalised physical activity or exercise programmes," and not to offer any programme "that uses fixed incremental increases in physical activity or exercise, for example, graded exercise therapy," or programmes "based on deconditioning and exercise avoidance theories as perpetuating ME/CFS" (NG206, recommendation 1.11.14). On the other therapy, the guideline states that cognitive behavioral therapy "has sometimes been assumed to be a cure for ME/CFS" and that "it should only be offered to support people who live with ME/CFS to manage their symptoms, improve their functioning and reduce the distress associated with having a chronic illness" (NG206, box 5). The guideline also makes post-exertional malaise central to diagnosis.
What moved underneath the guidance was not a single new trial. It was a re-examination of evidence that already existed. The PACE trial's recovery paper reported that 22 percent of participants met the trial's criteria for recovery after cognitive behavioral therapy and 22 percent after graded exercise therapy, against 8 percent and 7 percent in the comparison arms (White et al., Psychological Medicine, 2013). Those criteria had been revised from the ones set out in the trial's published protocol. A later reanalysis applied the protocol's original definition instead, and found recovery rates of 7 percent for cognitive behavioral therapy, 4 percent for graded exercise therapy, and 3 percent for the control group, a difference that was not statistically significant (Wilshire et al., BMC Psychology, 2018). Separately, a federal evidence review ran a sensitivity analysis removing the three trials that used the broad Oxford case definition, and reported that on that basis there "would be insufficient evidence of the effectiveness of GET on any outcome," with a single small trial left standing (AHRQ, addendum to Evidence Report No. 219, July 2016). That is an absence of evidence rather than a demonstration of ineffectiveness, and it is exactly the finding a recommendation cannot survive: once the population was narrowed to people who met stricter criteria, almost nothing was left to support it.
The correction then arrived at different institutions at different times. The AHRQ addendum came in July 2016. In July 2017 the Centers for Disease Control and Prevention removed both therapies from its public ME/CFS pages. The agency did not announce the change, but its own archived pages record it: the version captured in June 2017 lists cognitive behavioral therapy and an activity and exercise program, and the version captured a month later contains neither. The current CDC page recommends neither, states that there is no cure or approved treatment, and warns that standard exercise recommendations for healthy people can be harmful in ME/CFS (CDC, last reviewed May 2024; read July 2026). NICE replaced CG53 in October 2021.
And one major source did not move with the others. The Cochrane review of exercise therapy for chronic fatigue syndrome was amended in October 2019 after a formal complaint, with added emphasis on the limited applicability of the evidence to ME/CFS as currently defined and on the limitations of the data about harms. Cochrane announced an intention to update it further. On 16 December 2024, Cochrane stated that, because of insufficient new research in the field and a lack of resources to oversee the work, the update would not be proceeding. The review was republished days later with an editorial note attached, so the standing version now carries a 2024 date over an evidence base amended in 2019 and assembled well before that.
That is the shape of the problem in one illness. A patient searching today can reach a guideline that withdrew the recommendation, an agency page that quietly dropped it, a systematic review that still carries it in amended form and is no longer being updated, and an unlimited supply of older material that never changed at all. All of it is real. None of it is labelled with its position in the sequence.
None of these requires clinical training. They require reading the document's metadata as carefully as its prose, which most readers never do because nothing invites them to.
| What to check | What it tells you |
|---|---|
| Date of the claim, not the page | A page updated last month can restate a study from 2004. Find the date attached to the underlying evidence, which usually means finding the citation. A page with no dates anywhere is not neutral; it is a page that has declined to tell you. |
| Superseding language | Guidelines announce their own replacements. NG206 states that it updates and replaces CG53. Search the issuing body's site for the guideline number rather than the topic, because the withdrawn document often still ranks well and reads as authoritative on its own. |
| Retraction and correction status | Check the paper itself in PubMed or PubMed Central for a retraction or correction notice, and check the Retraction Watch database. Check the source, not the repetition: a claim can be quoted widely for years after the paper behind it was withdrawn, and the quotations carry no notice. |
| The population behind the evidence | Ask who was enrolled, under which case definition. The ME/CFS reversal turned substantially on this: restrict the analysis to trials using stricter criteria and almost no evidence remains. A conclusion can expire because the definition of the illness moved, not because any new data arrived. |
| Whether the outcome was pre-specified | Compare what the paper measures against what its registered protocol said it would measure. A recovery rate of 22 percent and one of 7 percent came from the same trial and differed in which definition was applied. Outcome changes made after a trial begins are a standard reason later reviewers discount a result. |
| Who is still watching | Find the review clock. Bodies that maintain guidance publish review or surveillance dates; reviews that are no longer maintained sometimes say so, as Cochrane did in 2024. A source that has stopped being updated has not been confirmed. It has been left where it was. |
Run in order, these six take a few minutes per claim and settle most cases. They also fail usefully: when the checks cannot establish currency, the correct conclusion is that the claim's status is unknown, which is a usable answer and a very different thing from treating it as confirmed.
Most sources handle expiry by deletion. The old page comes down, or the old sentence is edited away, and what remains gives no indication that anything was ever different. That is tidy and it destroys information. A reader who was told something in 2015 and reads the revised page in 2026 has no way to see that the advice changed, when it changed, or on what basis, which is precisely what they need in order to make sense of care they already received.
VictorOS is built the other way. Each claim in a condition map carries the identifier of the study it came from and the record of a human confirming it, so currency is checkable rather than assumed. When a claim is displaced, it is marked as superseded and kept alongside what replaced it, so the change itself remains part of the record. Where the evidence does not settle a question, the map says so instead of smoothing the gap.
This is also why a claim's age is treated as information rather than embarrassment. A statement grounded in a 2009 study is not automatically wrong, and one grounded in a 2025 study is not automatically right. What matters is that the reader can see the date, the source, and whether anyone has looked since.
A patient and a specialist can read the same sentence and take away different things, not because the specialist knows more biology but because the specialist knows the sentence is contested, or superseded, or resting on a trial the field stopped trusting. That context lives in guideline committees, journal correspondence, and professional networks. It rarely reaches the page the patient is reading, and it never reaches the printout in the folder.
Closing that gap does not require a patient to adjudicate the science. It requires the date, the source, and the review status to travel with the claim, so the question "is this still true" becomes answerable by looking rather than by asking someone with access. The six checks are how a reader recovers that information when a source withholds it. Carrying it on every claim is how a source stops withholding it.
Fourteen years passed between the two ME/CFS guidelines. People were treated, assessed, believed, and disbelieved on the strength of the first one throughout. The evidence that eventually overturned it was largely available the whole time. What was missing was not the data. It was any visible sign, on the documents people were handed, that the question was still open.
Verification note: each figure was checked against its primary source, and several were narrowed after checking. Shojania et al. (2007) measured the appearance of a "signal for updating" across 100 quantitative systematic reviews, not obsolescence as such; the paper reports a signal within two years for 23 percent of reviews and a median survival free of a signal of 5.5 years, and the text is worded to match. Poynard et al. (2002) assessed 474 conclusions from cirrhosis and hepatitis literature published 1945 to 1999 and reported 60 percent true, 19 percent obsolete, and 21 percent false as of 2000, with a 45-year half-life; those judgments were the investigators' own, which is why the text attributes them. The CG53 quotation is recommendation 1.6.2.4 of the 2007 guideline, read from an archived capture, since the current NICE page for CG53 carries only the replacement notice. NG206 wording is quoted from recommendation 1.11.14 and from box 5 on cognitive behavioral therapy; the CBT statement appears in a box rather than a numbered recommendation. The 22 percent recovery figures are from White et al. (2013) and reflect that trial's revised recovery criteria, which the text says plainly; the 7, 4, and 3 percent figures and the non-significant result are the intention-to-treat analysis in Wilshire et al. (2018), whose available-case analysis gave 8, 5, and 3 percent. The AHRQ finding is quoted in its conditional form, "would be insufficient evidence," because it was a sensitivity analysis, and the addendum attributes the downgrade partly to reduced statistical power; the text says this is absence of evidence rather than evidence of ineffectiveness. The CDC change was not announced by the agency and is documented here from CDC's own archived pages of June and July 2017, with the current position taken from the CDC management page (last reviewed May 2024, read July 2026). Cochrane's notice that the update will not proceed is dated 16 December 2024; the review was republished on 19 December 2024 as CD003200.pub9 with an editorial note, so the version number is current while the underlying evidence base is not. Characterizations of NICE's reasoning come from the guideline text rather than from secondary summaries, several of which paraphrase it more strongly than the document does.
VictorOS organizes evidence; it does not practice medicine. This note describes a method for judging whether a clinical claim is current. It does not give medical advice, diagnose, or recommend or discourage any treatment for any person. Guidance differs between countries and changes over time; the documents cited here are named with their dates so you can check their status yourself. Disease facts in VictorOS guides are based on articles retrieved from PubMed and cited with stable identifiers. These materials support your medical team; your clinicians remain the ones who diagnose and treat.