Forcing dichotomous disease classification from reference standards leads to bias in diagnostic accuracy estimates: A...

Area of Science:

• Medical Diagnostics
• Biostatistics
• Health Informatics

Background:

• Diagnostic accuracy estimation is crucial for clinical decision-making.
• Uncertainty in disease classification can arise from various diagnostic challenges.
• Current methods may oversimplify disease presence/absence, potentially leading to biased results.

Purpose of the Study:

• To investigate the impact of ignoring diagnostic uncertainty on index test accuracy.
• To evaluate how forcing dichotomous classification affects accuracy estimates.
• To identify factors influencing bias in diagnostic accuracy.

Main Methods:

• Simulated scenarios with varying expert panel compositions and reference test accuracies.
• Varied index test sensitivity, specificity, and target disease prevalence.
• Forced dichotomous classification of target disease presence/absence for each individual.

Main Results:

• Forcing dichotomous disease classification in the presence of uncertainty leads to biased index test accuracy estimates.
• The direction and magnitude of bias are influenced by reference test quality, disease prevalence, and true test performance.
• Bias is evident across simulated scenarios with different data characteristics.

Conclusions:

• Ignoring uncertainty in disease classification can significantly bias diagnostic accuracy.
• Future research should explore methods to incorporate probability or advanced statistics to manage classification uncertainty.
• Empirical studies are needed to validate strategies for reducing bias in diagnostic accuracy estimation.