Study Results Explained: Risk, Outcomes and Certainty

Are we counting events or people?

This is a hypothetical example, not a study result. In group A, two of ten people each fall three times. In group B, three of ten people each fall once.

Group A has fewer people falling but more fall events. An actual event rate also accounts for observation time; raw counts alone may not be comparable. Injurious falls, serious injuries and balance-test scores are other distinct outcomes.

A 20% relative reduction is not 20 percentage points

Another hypothetical example: over the same year, ten in every 100 people in a control group fall, compared with eight in an exercise group. Risk changes from 10% to 8%: a 20% relative reduction, but an absolute reduction of two percentage points, or two fewer people per 100.

With a different baseline risk, the same relative reduction corresponds to a different absolute number. Read a percentage alongside the comparator, participants, outcome and observation period. It is not a fixed prediction for you.

This example is not the absolute effect of a tai chi trial. In the 2019 review, the tai chi risk ratio of 0.80 concerns people with at least one fall. The rate ratio of 0.81 concerns fall events. They cannot be swapped.

What does a confidence interval describe?

The 2019 tai chi fall-rate estimate was 0.81, with a 95% confidence interval of 0.67–0.99. The estimate describes an average effect; the interval expresses uncertainty around that estimate. It does not mean 95% of participants improved by amounts within the interval.

For a ratio, no difference corresponds to 1. For a difference in scores, it corresponds to 0. An interval containing that value is generally compatible with no difference, and may also be compatible with some benefit or harm. It does not automatically prove no effect or equivalence.

Questionnaire results need units and a sense of practical importance. A standardised mean difference is not a score on the original questionnaire or a percentage improvement. A statistically detectable difference need not mean a noticeable change in daily life.

What does evidence certainty add?

The 2019 review rated the tai chi fall-rate evidence as low certainty, and evidence for people with at least one fall as high certainty. Ratings apply to a particular outcome and comparison. They do not label every tai chi course as highly supported.

Assessment considers risk of bias, consistency, relevance to the question and population, precision, and possible missing unpublished findings. A large total sample or a well-known journal does not replace those checks.

High certainty means greater confidence in the effect estimate. It is not a guarantee of individual benefit or a safety certificate. Low certainty means further evidence may change the assessment; it does not establish that a practice is ineffective.

Check these four lines in a real example

1. In the community falls review, identify participants and controls before reading the largest reduction.
2. Record the fall-rate analysis separately: seven studies, 2,655 participants. Then record the people-falling analysis: eight studies, 2,677 participants. Do not add the samples.
3. Keep each ratio, confidence interval and certainty rating together.
4. Open the abstract and check its scope and search date. The 23,407 participants belong to the entire exercise review, not the tai chi subgroup.

You can record your own experience, but one improvement does not establish that a practice caused it or translate directly into a trial’s average difference.