Biomarker validation with an imperfect reference: Issues and bounds.

Area of Science:

• Biomarker discovery and validation
• Diagnostic test evaluation
• Medical statistics

Background:

• Assessing new biomarkers for acute kidney injury (AKI) is crucial but difficult without a definitive gold standard.
• Existing methods often rely on imperfect reference tests, introducing potential bias in performance evaluation.
• Latent class analysis and similar techniques use strong, often unverifiable assumptions.

Purpose of the Study:

• To investigate the impact of reference test errors on new biomarker performance assessment.
• To analyze the conditional independence assumption in biomarker evaluation.
• To establish bounds for the true sensitivity and specificity of a new biomarker.

Main Methods:

• Exploration of the conditional independence assumption in statistical models for biomarker assessment.
• Analysis of information content derived from comparing a new biomarker with a reference test.
• Derivation of bounds for sensitivity and specificity using known reference test operating characteristics.

Main Results:

• Conditional independence is only feasible within a limited range of disease prevalence values for observed data.
• Bounds for the new biomarker's sensitivity and specificity can be derived.
• The utility of these bounds varies, being tight and informative in some scenarios and wide in others.

Conclusions:

• Reference test inaccuracies significantly bias new biomarker evaluations.
• While bounds for sensitivity and specificity can be informative, their width depends on the data and reference test performance.
• Careful consideration of assumptions and potential biases is necessary when evaluating biomarkers without a gold standard.