Diagnostic Accuracy: Choose the Right Denominator

Label the reference classification first

Write down the target condition, index test, positivity threshold, and reference standard before copying numbers. A diagnostic table is easy to transpose when the paper and calculator arrange rows differently. Spell out true positive, false negative, false positive, and true negative in your extraction notes.

FDA's diagnostic-test reporting guidance distinguishes evaluation against a reference standard from agreement with a non-reference comparator. When the comparator does not establish condition status, calling agreement “sensitivity” can misrepresent what was measured.

Follow each denominator

Sensitivity uses everyone with the condition according to the reference standard: true positives divided by true positives plus false negatives. Specificity uses everyone without it: true negatives divided by true negatives plus false positives.

Positive predictive value instead starts with positive index-test results. Negative predictive value starts with negative results. These answer different questions, and predictive values depend on the population's condition prevalence and selection process. An enriched diagnostic sample is not automatically representative of routine practice.

Recalculate a fictional table

Imagine 80 true positives, 20 false negatives, 30 false positives, and 170 true negatives. The table includes 300 participants, of whom 100 have the condition and 200 do not.

Sensitivity is 80/100 = 80%; specificity is 170/200 = 85%. Positive predictive value is 80/110 = 72.73%, and negative predictive value is 170/190 = 89.47%. Use the diagnostic-accuracy calculator to check these point estimates. The numbers are fictional and do not validate a real test. Point estimates alone do not describe precision.

Account for people outside the table

Compare the table total with the recruitment flow. Record uninterpretable index tests, missing reference tests, exclusions, and timing between tests. Ask whether every participant received the same reference procedure and whether knowledge of one result could influence the other assessment.

The diagnostic extraction worksheet provides a place to preserve these decisions. Do not remove an indeterminate category merely to force a binary calculation; document the handling rule and its implications first.

Before reporting, verify that the table describes one threshold and one relevant observation per participant. Repeated specimens, several thresholds, or clustered observations require analysis that respects dependence. A clean four-cell table is the beginning of an accuracy assessment, not the whole assessment.

Source checked on 30 October 2026; numerical example independently recalculated.