Effect of verification bias on positive and negative predictive values

Area of Science:

• Biostatistics
• Diagnostic Test Accuracy

Background:

• Sensitivity and specificity measure test efficacy.
• Positive and negative predictive values assess diagnostic accuracy in patients.
• True disease status is required for accurate calculation.

Purpose of the Study:

• Evaluate properties of predictive value estimators using only verified disease statuses.
• Assess bias in sensitivity and specificity estimation when not all statuses are verified.
• Examine the impact of conditional independence assumption departures.

Main Methods:

• Analysis of estimators for positive and negative predictive values.
• Derivation of consistent variance estimators.
• Application of Maximum Likelihood (ML) method to assess estimator sensitivity.

Main Results:

• Positive and negative predictive value estimators are unbiased under the specified assumption.
• Consistent estimators for the variances of these predictive values are provided.
• The ML method reveals the sensitivity of naive estimators to assumption departures.

Conclusions:

• The proposed estimators for predictive values are statistically sound.
• Understanding estimator behavior is crucial for accurate diagnostic test evaluation.
• Further research may be needed to address violations of the conditional independence assumption.