Assessing agreement between malaria slide density readings.

Area of Science:

• Biostatistics
• Infectious Disease Epidemiology
• Medical Diagnostics

Background:

• Current criteria for assessing agreement in malaria parasite density slide readings (e.g., percent difference, absolute difference) lack clear justification for parameter selection.
• Existing methods may overestimate or underestimate discrepancies at low or high parasite densities, respectively.

Purpose of the Study:

• To develop a procedure for assessing agreement between replicate slide readings that avoids the disadvantages of current methods.
• To provide clearer justification for the parameters used in defining acceptable agreement in malaria parasite density measurements.

Main Methods:

• Utilized the Poisson distribution to model expected variation in parasite density within a slide.
• Proposed a criterion based on a constant difference in square root counts, which accounts for density-dependent variability.
• Applied the method to example datasets from Tanzania and a microscopist training trial.

Main Results:

• A criterion based on square root counts, when back-transformed, provides a wider range of acceptable agreement at higher average parasite densities.
• Observed agreement limits in datasets were wider and less symmetrical than those predicted by the Poisson distribution, indicating issues beyond random variation.
• The proposed method highlights that disagreements at low parasite densities may stem from within-slide variation rather than reading errors.

Conclusions:

• The proposed method establishes limits of agreement that naturally accommodate increasing variability with parasite density.
• It aids in identifying between-reader variability consistent with random variation, reducing unnecessary additional readings.
• Further investigation in diverse settings is recommended to validate applicability and refine numerical limits of agreement.