Variability in diagnostic accuracy can be estimated using simple population weighting

Abstract 

Objective

Diagnostic accuracy of a quantitative diagnostic test at a given numeric cutoff is dependent on the type of population (e.g., chronic, referrals, or screening) under investigation. Simple weighted averages calculated from a single study sample may be used to assess variability in accuracy in different types of populations.

Study Design And Setting

We evaluated the accuracy of the 4-variable Modification of Diet in Renal Disease (MDRD) Study equation as a diagnostic test to separate stage 1 and 2 chronic kidney disease (≥60mL/min per 1.73m²) from stage 3–5 (<60mL/min per 1.73m² requiring treatment to prevent progression) in a sample of 850 patients referred for determination of glomerular filtration rate (GFR). Using population weighting, we also estimated the accuracy of the MDRD equation when the GFR distribution typically found in screening situations was mimicked.

Results

Estimated diagnostic accuracy of the MDRD equation varied substantially for different population types (sensitivity range 82%–97%, specificity 67%–93%; figures include the original MDRD study).

Conclusions

Reports of diagnostic accuracy should include estimates of the variability of diagnostic accuracy, using different real or tentative population distributions. Population weighting is a useful tool for this purpose.

Keywords: Diagnostic tests, Routine sensitivity and specificity