Comparative diagnostic accuracy studies with an imperfect reference standard - a comparison of correction methods.

Area of Science:

• Diagnostic accuracy studies
• Biostatistics
• Medical decision-making

Background:

• Two correction methods exist to estimate index test accuracy with imperfect reference standards: Staquet et al. and Brenner.
• These methods assume known sensitivity and specificity of the reference standard.
• No prior studies have statistically compared these widely used correction methods.

Purpose of the Study:

• To statistically compare the performance of the Staquet et al. and Brenner correction methods.
• To evaluate their accuracy in estimating sensitivity and specificity of index tests.

Main Methods:

• Comparative analysis using simulation techniques.
• Assessed methods under conditional independence and dependence between index test and reference standard.
• Validated findings on three clinical datasets.

Main Results:

• Under conditional independence, Staquet et al. method generally outperformed Brenner's.
• Staquet et al. method yielded illogical results (outside [0,1]) at very high or low disease prevalence.
• Both methods failed under conditional dependence, especially with non-zero covariance.

Conclusions:

• The Staquet et al. method is preferable to Brenner's when tests are conditionally independent and reference standard accuracy is known.
• Alternative statistical approaches, like latent class analysis, are recommended for high/low prevalence or conditionally dependent tests.