Two things decide whether a health finding means anything: how clean the comparison was, and how many people actually went through with it. Neither one makes the news. Both matter more than the number that does.

The Synthesis: A randomised trial across 15 hospitals in Spain, published in The Lancet, assigned 57,404 adults aged 50 to 69 either to a single colonoscopy or to a stool test every two years. Ten years on, deaths from colorectal cancer stood at 0.22 percent in the colonoscopy group and 0.24 percent in the stool testing group. The trial had fixed in advance how large a gap it would treat as meaningful, 0.16 percentage points, and the difference came nowhere near it. On the outcome that matters most, the two programs were not meaningfully different.

Part of the reason is that more people went through with the stool test. Among those invited to colonoscopy, 31.8 percent completed any screening; among those invited to stool testing, 39.9 percent did. Colonoscopy is the stronger test when it happens but it happens less often. It also found and removed more precancerous growths, but only among people who chose to go through with it, and people who follow through on a colonoscopy invitation differ from those who don't. That comparison no longer pits the randomised groups against each other, so it carries less weight than the mortality figures.

Whether either test works is not in dispute. The trial asks a different question, and a more useful one. A health system cannot hand people a test, only an invitation. So what matters is not which test is better but which program ends up saving more lives once people decide whether to show up.

The Calibration: Overestimated is the gap between a colonoscopy and a stool test as a population experiences them. Underestimated is what a declined invitation costs. The stool test is widely treated as the compromise option, the one taken by people not up for the real procedure, and on the outcome that matters it holds its own. Most national programs, including those in Canada, the United Kingdom and Australia, already screen average-risk adults using a stool test, and only send people to colonoscopy on an abnormal result or a family history. The United States is among the minority that also screens directly by colonoscopy, according to a 2025 synthesis in the Journal of Translational Medicine. One caveat travels with the result: a single colonoscopy was compared against repeated stool tests, so it measures programs rather than procedures.

The Synthesis: Wales offered its shingles vaccination program to everyone born on or after Sept. 2, 1933, permanently excluding those born just days earlier. Nobody chooses their birthday, so the two groups had no particular reason to differ except in access. That is the hard part of this question. People who turn up for vaccines are healthier to begin with, in ways no statistical adjustment fully removes, and the Welsh rule removed the problem by accident.

A Stanford-led study published in Nature followed 282,541 Welsh adults who were free of dementia when the program began. Over the next seven years, 35,307 were diagnosed, roughly one in eight. Receiving the live-attenuated vaccine lowered that probability by 3.5 percentage points, with uncertainty running from 0.6 to 7.1 points.

The average conceals a split. The reduction was 5.6 points among women and 0.1 among men, whose range runs from a 4.4-point reduction to a 4.1-point increase. The authors stop short of saying the vaccine does nothing in men, since dementia is less common in older men and the estimate is correspondingly imprecise. As it stands, this is a finding in women.

Australia and Ontario had comparable birthdate rules. Measured as the effect of eligibility rather than of vaccination, the reduction was 1.3 percentage points in Wales, 1.8 in Australia over 7.4 years in the Journal of the American Medical Association, and 2.0 in Ontario over 5.5 years in The Lancet Neurology. No trial has tested the question directly. Researchers accept the association and divide on whether it is causal.

Every one of these examined the live-attenuated vaccine, which has since been replaced by a recombinant version everywhere it was in use. The newer vaccine has evidence of its own, including a manufacturer-funded study of more than 500,000 American nursing home residents in Annals of Internal Medicine, where four-year dementia risk was 18.8 percent among the vaccinated and 24.6 percent among the rest. None of it comes from a natural experiment, and those authors report that their own checks for hidden bias suggest some remains.

The Calibration: The finding worth holding is about the evidence, not the vaccine. Three accidents of policy produced the closest thing to a controlled comparison this question has had, and they agree with each other. What does not follow is a reason to act. The vaccine those studies examined is no longer given, the evidence for its replacement rests on the design the natural experiments were built to escape, and shingles vaccination is already recommended for older adults regardless. This is an argument for running a trial, not for changing anything before one exists.

The Synthesis: Prediabetes is usually presented as a waiting room, a place people sit until the real diagnosis arrives. The best long-term data say most of them get up and leave. Pooled individual records from 19 cohort studies covering 76,092 adults, average age 51, followed for a median of 9.8 years, appeared in The Lancet Global Health. Among those meeting the study's fasting glucose criteria for prediabetes, 12.5 percent developed type 2 diabetes within ten years while 36.1 percent returned to normal blood sugar. In the quarter with the highest fasting glucose, 16.1 percent progressed and 13.4 percent reverted. Men and adults over 55 progressed more often.

Observational evidence is the right tool for a question about natural history, and the overall pattern is consistent across the cohorts. The figures are modeled transition probabilities rather than simple counts, so the precision of the decimal places should not be mistaken for certainty.

Where the field differs is in how it draws the category. The American Diabetes Association applies the label from a glycated hemoglobin reading of 5.7 percent, the HbA1c figure that appears on a routine lab report. Diabetes Canada starts at 6.0 percent. So does the United Kingdom, where the National Institute for Health and Care Excellence calls the same range a high risk of type 2 diabetes rather than a diagnosis.

The Calibration: Inevitability is the miscalibration. The familiar claim that most people with prediabetes go on to develop diabetes does not survive contact with long-term cohort data. What gets too little weight is where in the range a reading falls, since risk climbs steeply across the band rather than switching on at its lower edge. The number is more informative than the label attached to it.

A striking result lands. Two years later a larger study finds a smaller effect, then a review finds an even smaller one. Readers experience this as science contradicting itself. Usually what has changed is the size of the finding rather than its direction, though sometimes the direction changes too, and both are regular enough to expect.

The mechanism is selection. Small studies produce unstable estimates, and the ones that land high are the ones that look worth publishing and worth covering. Economists call this the winner's curse, and it has been measured inside health research: an analysis in Health Policy found that higher-impact journals systematically reported larger estimates than lower-impact ones for the same underlying quantities.

The consequences show up downstream. Of 45 highly cited clinical studies reporting that an intervention worked, examined in the Journal of the American Medical Association, seven were later contradicted outright and another seven had reported effects stronger than subsequent research found; 20 were replicated. Nonrandomised studies fared worst: five of six were later contradicted or scaled back, against nine of 39 randomised trials.

The reframe is to read a first striking result as a ceiling rather than an estimate, and to expect the ceiling to come down. Roughly a third of even the most prominent findings did not survive in their original form. That is not a scandal about scientists. It is what the publication process does to a number on its way to the reader.

The Claim: Canola, soybean and sunflower oils drive chronic inflammation and should be replaced with butter or beef tallow.

The Distortion: Seed oils are high in linoleic acid, which the body can convert into compounds that feed inflammation. That chemistry is real, and it is where the argument stops. Measured in people, the pattern runs the other way: a pooled analysis of 30 observational cohort studies in Circulation found that those with more linoleic acid in their blood and tissue had lower rates of cardiovascular disease, cardiovascular death and ischemic stroke. Trials since have not found that higher linoleic acid intake raises inflammatory markers either.

The Calibration: A chemical pathway shows where to look for harm. It does not establish that harm is there. The studies that looked did not find it.

The Claim: Taping the mouth shut overnight forces nasal breathing and improves sleep and snoring.

The Distortion: A systematic review in PLOS ONE found only ten studies in the entire literature, covering 213 patients, and rated all of them poor quality. What benefit appeared was confined to mild obstructive sleep apnea, four of the studies had excluded anyone with nasal obstruction, and the reviewers flagged a risk of asphyxiation for precisely that excluded group.

The Calibration: The evidence is too thin to support the general claim, and the people most inclined to try this are the ones the studies deliberately left out.

Stay calibrated.

If The Calibration brought you a clearer perspective today, forward it to someone who values the same.

Received this from a friend or colleague? Subscribe here to get it directly.

Effect of Invitation to Colonoscopy Versus Faecal Immunochemical Test Screening on Colorectal Cancer Mortality (COLONPREV): A Pragmatic, Randomised, Controlled, Non-Inferiority Trial. The Lancet, 2025.

Global Colorectal Cancer Screening Programs and Coverage Rate Estimation: An Evidence Synthesis. Journal of Translational Medicine, 2025.

A Natural Experiment on the Effect of Herpes Zoster Vaccination on Dementia. Nature, 2025.

Herpes Zoster Vaccination and Dementia Occurrence. The Journal of the American Medical Association, 2025.

Herpes Zoster Vaccination and Incident Dementia in Canada: An Analysis of Natural Experiments. The Lancet Neurology, 2026.

Dementia Risk After Recombinant Herpes Zoster Vaccination in Older Adults With a Recent Skilled-Nursing Facility Stay: A Target Trial Emulation. Annals of Internal Medicine, 2026.

Prediabetes Transitions to Normoglycaemia or Type 2 Diabetes and Associated Risk Factors in the Obesity, Diabetes and Cardiovascular Disease Collaboration: An Individual-Level Pooled Analysis of 19 Prospective Cohort Studies. The Lancet Global Health, 2025.

Diagnosis and Classification of Diabetes: Standards of Care in Diabetes—2026. Diabetes Care, 2026.

Definition, Classification and Diagnosis of Diabetes, Prediabetes and Metabolic Syndrome. Canadian Journal of Diabetes, 2018.

Type 2 Diabetes: Prevention in People at High Risk. National Institute for Health and Care Excellence, 2017.

Publication Selection in Health Policy Research: The Winner's Curse Hypothesis. Health Policy, 2013.

Contradicted and Initially Stronger Effects in Highly Cited Clinical Research. The Journal of the American Medical Association, 2005.

Biomarkers of Dietary Omega-6 Fatty Acids and Incident Cardiovascular Disease and Mortality: An Individual-Level Pooled Analysis of 30 Cohort Studies. Circulation, 2019.

Breaking Social Media Fads and Uncovering the Safety and Efficacy of Mouth Taping in Patients With Mouth Breathing, Sleep Disordered Breathing, or Obstructive Sleep Apnea: A Systematic Review. PLOS ONE, 2025.

The Calibration is an editorially curated health briefing. Content is drawn from published research and credible sources for informational purposes only. It does not constitute medical advice, clinical guidance, diagnosis, or treatment, and is not a substitute for consultation with a qualified healthcare professional. Nothing in this briefing should be used to make health or treatment decisions or to delay seeking professional medical advice.

Sponsored content, when present, will always be clearly identified.