Related Experiment Videos
Bias in discrepant analysis: when two wrongs don't make a right
1Robert Wood Johnson Clinical Scholars Program, University of North Carolina at Chapel Hill 27599, USA.
Journal of Clinical Epidemiology
|March 12, 1998
Summary
Discrepant analysis, used to evaluate new diagnostic tests with imperfect standards, overestimates test accuracy. This method, even with a perfect gold standard, introduces bias and should not be routinely used for diagnostic test evaluation.
Area of Science:
- Medical Diagnostics
- Biostatistics
- Health Services Research
Background:
- Imperfect reference standards, such as alloyed standards, pose challenges in evaluating new diagnostic tests.
- Discrepant analysis is a modified evaluation method involving additional testing to resolve discrepancies between a new test and a reference standard.
Purpose of the Study:
- To algebraically demonstrate the bias introduced by discrepant analysis in diagnostic test evaluation.
- To assess the impact of using alloyed standards in both stages of discrepant analysis.
Main Methods:
- Algebraic demonstration of bias in sensitivity and specificity calculations.
- Analysis of factors influencing bias magnitude, including true test performance, disease prevalence, and error concordance.
Main Results:
- Discrepant analysis systematically overestimates sensitivity and specificity, even when a perfect gold standard is used for resolution.
- Substantial bias is evident when alloyed standard tests are employed in both stages of the discrepant analysis.
Conclusions:
- Discrepant analysis is not a reliable method for evaluating diagnostic tests due to inherent overestimation of accuracy.
- The use of alloyed standards within discrepant analysis exacerbates bias and should be avoided for routine diagnostic test evaluation.