AtlasSession 4L12: Introduction to measurement errors and diagnostic test accuracy studiesp.24–30, 57–60
Diagnostic test accuracy (2×2)
- 1 earlier topics
- What is systematic error
- Accuracy vs precision
- Random vs systematic error
- Assessing measurement error
- Diagnostic test accuracy (2×2)
- Next: Thresholds and ROC curves
Fig. 1
OCT screening of 200 tissue-engineered cartilage constructs for delamination, histology as the reference standard (constructs)
Fig. 1. Four boxes of constructs, sorted by OCT (rows) and Delaminated on histology (columns). Of 56 truly delaminated, 46 were caught and 10 missed. Of 144 truly sound, 126 were cleared and 18 were false alarms. Caught is called sensitivity; cleared is specificity.
- Caught (truly delaminated)
- 46 of 56, about 82%
- Cleared (truly sound)
- 126 of 144, about 88%
2×2 Workbench · First look
Level 1 · First look · Lecture 12 · p.24–30, 57–60
Sort test results into four boxes and say, in words, how many real cases the test caught and how many sound ones it cleared.
Build the diagnostic 2×2 (index test rows, reference standard columns) from a narrative and compute sensitivity, specificity, PPV and NPV with the right denominators.
Smoke alarm (the step you are on)
It can ring for burnt toast (false alarm) or stay silent in a fire (miss).
A smoke alarm can ring for burnt toast (a false alarm) or stay silent in a fire (a miss). OCT does both: it raised 18 false alarms on sound constructs and missed 10 delaminated ones.