A GEE approach to estimating accuracy and its confidence intervals for correlated data

Area of Science:

• Biostatistics
• Medical Diagnostics
• Statistical Modeling

Background:

• Assessing diagnostic test performance is crucial in clinical practice.
• Existing methods for confidence intervals of diagnostic measures like sensitivity and specificity are available for correlated data.
• Confidence intervals for diagnostic accuracy in correlated observations were previously lacking.

Purpose of the Study:

• To develop a statistical method for constructing confidence intervals for diagnostic accuracy.
• To address the challenge of correlated observations in diagnostic test evaluations.
• To provide a reliable measure of accuracy for studies involving multiple tests on the same patient sample.

Main Methods:

• Utilized the generalized estimating equations (GEE) method, previously applied to other diagnostic measures.
• Employed the delta-method in conjunction with GEE to derive confidence intervals for accuracy.
• Applied the method to scenarios with correlated observations from diagnostic tests.

Main Results:

• Successfully constructed confidence intervals for diagnostic accuracy in the context of correlated observations.
• Simulation studies confirmed that the developed confidence intervals achieve appropriate coverage rates.
• The proposed method provides a statistically sound approach for accuracy estimation.

Conclusions:

• The GEE and delta-method provide a valid approach for confidence intervals of diagnostic accuracy with correlated data.
• This method enhances the evaluation of diagnostic tests when multiple correlated measurements are involved.
• The findings support improved statistical rigor in diagnostic accuracy assessment.