Abstract
Objectives: Test performance is conventionally expressed by gain in diagnostic certainty. We propose net diagnostic gain and indication area as more appropriate measures of test performance; then, the loss in certainty due to misclassification and the information of "no test" would be performed are taken into account.
Study Design and Setting: A decision analytical model was developed in which two alternative strategies were compared: testing and no testing. Correct diagnostic test results received a positive value; undesired test results received a negative value. Within the "no test" scenario, it was assumed that physicians are more prone to treat as the probability of disease is higher.
Results: Discounting gain and loss in diagnostic certainty results in a concave function of the prior. The indication area is the range of priors with a net diagnostic gain; testing is deleterious beyond this range. The net diagnostic gain reaches a maximum at a specific prior. A freely available Web site-based calculator was developed for easy calculation of the indication area and the maximum diagnostic gain for each combination of sensitivity and specificity.
Conclusion: Medical testing is not indicated when the prior disease probabilities are low (as to screening for a condition) or high (for diagnostic confirmation). Published by Elsevier Inc.
Original language | English |
---|---|
Pages (from-to) | 1120-1128 |
Number of pages | 9 |
Journal | Journal of Clinical Epidemiology |
Volume | 68 |
Issue number | 10 |
DOIs | |
Publication status | Published - Oct 2015 |
Keywords
- Sensitivity
- Specificity
- Prevalence
- Indication area
- Maximum diagnostic gain
- Diagnostic test
- CLINICAL DECISION-MAKING
- HOME PREGNANCY TESTS
- OVER-THE-COUNTER
- TEST ACCURACY
- LYMPH-NODES
- SENSITIVITY
- PREVALENCE
- CANCER
- SPECIFICITY
- PERFORMANCE