A test described as “99% accurate” can sound almost definitive. But accuracy needs a precise definition, and the usefulness of a positive result depends on the setting in which the test is used.
The following is a constructed arithmetic example. It is not data for a real test, a screening recommendation or a way to interpret your own result.
Define the fictional population
Imagine 10,000 people, of whom 100 have the condition being tested for and 9,900 do not. The prevalence in this invented group is therefore 1%.
Assume the test correctly returns positive for 90% of people with the condition. That is a sensitivity of 90%. Assume it correctly returns negative for 99% of people without the condition. That is a specificity of 99%.
Count the results
Among the 100 people with the condition, 90 test positive and ten test negative. Among the 9,900 without it, 9,801 test negative and 99 test positive.
There are therefore 189 positive results in total: 90 true positives plus 99 false positives. In this example, the fraction of positive results that are true positives is 90 divided by 189, or approximately 47.6%.
That is not a contradiction. Sensitivity, specificity and the proportion of positive results that are correct describe different relationships.
Change the setting and the answer changes
Now hold the same assumed sensitivity and specificity, but imagine 1,000 of the 10,000 people have the condition. The test produces 900 true positives and 90 false positives. The fraction of positive results that are true is now 900 divided by 990, approximately 90.9%.
The test assumptions did not change. The population did. This is why a performance claim from one setting should not automatically be applied to another.
Earlier detection is a separate question
A report that a test detects a signal does not, by arithmetic alone, show that using the test improves health outcomes. That broader question requires evidence about what follows testing, including the consequences of correct and incorrect results.
This article does not decide whether any screening program is worthwhile. It explains why a single accuracy number is not enough to answer that question.
What to ask of a diagnostic headline
Ask which metric was reported, who was tested, what reference standard was used and whether the evaluation resembles the intended use. Ask for the actual counts rather than accepting “high accuracy” as a complete result.
Do not enter personal test results into our research tools. They are reading aids, not diagnostic calculators. The useful lesson is that a test percentage has a denominator and a context. Without them, a precise-looking headline can invite an interpretation the data do not support.