Long Life Report

Glossary

Plain-language definitions of the technical terms in our reporting. Every entry is checked against a source, the same as every story.

General

  • Bayesian analysis A statistical framework that treats the quantity being estimated as uncertain: a prior distribution, what was known beforehand, updated by the study data into a posterior distribution.
  • LDL cholesterol The cholesterol carried by low density lipoproteins. As it rises, fatty plaque can build up in the arteries, which is why it is called bad cholesterol.
  • adverse event Any unwanted medical event recorded in a trial participant while on the treatment, whether or not the treatment caused it. Serious is a defined outcome category, not a strength.
  • association A measured relationship between an exposure and an outcome: the two vary together in the data. It says how strongly they travel together, not which one moved the other, or whether either did.
  • case-control study A study that starts from people who already have an outcome, assembles a comparison group without it, and looks backwards at what each group was exposed to.
  • certainty of evidence How much confidence a review has that its estimate is close to the truth, rated high, moderate, low or very low. It rates the evidence, not the treatment.
  • circadian rhythm The roughly 24-hour cycle of physical, mental and behavioural changes a body runs on, kept in step with the outside world mainly by light and dark.
  • cohort study A study that follows groups defined by whether they were exposed to something, and records what happens to them, without the researchers assigning who is exposed.
  • confidence interval The range around a measured result within which the true value can reasonably be expected to lie. A wide interval means the study pinned the answer down loosely.
  • conflict of interest A secondary interest, typically financial, that could influence professional judgement about a primary one such as the validity of research. Journals require it to be declared, not avoided.
  • confounding A distortion of the link between two things by a third factor that is tied to both, so an association appears, disappears or changes size for a reason other than the one being studied.
  • credible interval The Bayesian counterpart of a confidence interval: the range holding 95 percent of the probability for an estimate, and unlike a confidence interval it supports a direct probability statement.
  • hazard ratio The rate at which an event happens in one group divided by the rate in another over the same period, where 1 means no difference, below 1 means fewer events in the first group and above 1 means more.
  • heterogeneity The extent to which the studies pooled in a review disagree with each other by more than chance alone would explain.
  • incidence rate New cases of something divided by the total time the people were watched for it, which is why it is written per 100,000 person-years rather than as a plain percentage.
  • insulin sensitivity How readily a tissue answers insulin's signal to take glucose out of the blood, or in the liver's case to stop putting glucose into it. Its opposite is insulin resistance.
  • intention to treat Analysing every participant in the group they were randomised to, whatever happened next. Analysing only those who finished the trial can flatter a treatment, so the two are often compared.
  • interquartile range The width of the middle half of a set of measurements: the distance between the value a quarter of the way up the ordered list and the value three quarters of the way up.
  • median The middle value once the numbers are put in order, half above and half below. Papers report it rather than an average when extreme values would drag an average away from the typical case.
  • mediation analysis A statistical method that splits an association between two things into the part running through a third thing and the part that does not, and it rests on assumptions that cannot be checked.
  • meta-analysis The statistical combination of results from two or more separate studies, producing one pooled estimate of an effect.
  • metabolic equivalent of task A unit of exercise intensity. One MET is the energy the body uses at rest, so a 4 MET activity uses four times that, and multiplying by minutes gives the MET-minutes used to express a weekly dose.
  • minimal clinically important difference The smallest change in a measurement that a patient would notice as worth having. It is a judgement someone has to set, and different methods of setting it produce very different numbers.
  • multiple comparisons Testing several outcomes in one study, which raises the chance that at least one difference looks real when nothing is there, unless the analysis is planned to allow for it.
  • network meta-analysis A meta-analysis that ranks several treatments at once, combining the trials that compared them head to head with indirect comparisons inferred through a shared comparator.
  • observational study A study that records what people are already exposed to and what happens to them, without researchers deciding who gets what, which is why it can show an association but cannot settle cause.
  • odds ratio The odds of an outcome in one group divided by the odds in another. It approximates relative risk only when the outcome is uncommon, and sits further from 1 when it is not.
  • p-value The probability of getting a result at least as extreme as the one observed, if the treatment truly made no difference. It is not the probability that a finding is true.
  • person-years The unit a study counts exposure in: one person followed for one year is one person-year, so a rate per 1,000 person-years counts events against the time people were actually watched.
  • primary endpoint The outcome a trial names in its protocol, before it begins, as the most important one for judging whether the treatment worked. Most trials have one; some have more than one.
  • publication bias The distortion that follows when whether a study's results are published, and how quickly, depends on what those results showed.
  • randomised controlled trial A study that allocates each participant to a treatment or a comparison group by chance, so that the groups differ only in what they receive.
  • relative risk The risk of an event in one group divided by the risk in another, where 1 means no difference, below 1 means the event was less common in the first group and above 1 means it was more common.
  • reverse causation An association that runs backwards: the outcome, sometimes an illness nobody has diagnosed yet, is what changed the exposure being measured, rather than the exposure changing the outcome.
  • risk of bias An assessment of whether the way a study was designed, conducted or reported could have pushed its result systematically in one direction.
  • sensitivity analysis A re-run of a study's main analysis under a different reasonable choice, to test whether the result holds. Agreement strengthens confidence, and disagreement is itself a finding.
  • standard deviation A measure of how widely the values in a set are spread around their average, expressed in the same units as the values themselves.
  • standardised mean difference An effect size expressing the gap between two groups in units of the spread within them, so that trials using different rating scales can be pooled.
  • statistical model An equation fitted to data that splits a measured outcome into a part explained by other recorded quantities and a part left to chance, estimating how each quantity relates to the outcome.
  • statistical power A study's chance of detecting an effect of a given size if that effect is really there. Where power is low, finding no difference is weak evidence that there is none.
  • surrogate endpoint Something measured in a trial as a stand-in for the outcome that actually matters, used because the real outcome takes longer or more people to observe.
  • systematic review A review that fixes its question and its inclusion rules in advance, then searches for every study that fits them by an explicit method another team could repeat.
  • trial registration A trial's plan, including who will be enrolled and what will be measured, filed in a public register before the first participant joins, under a number that identifies it ever after.

Sleep