A comparison of methods to correct for misclassification bias from administrative database diagnostic codes

Area of Science:

• Epidemiology
• Health Informatics
• Biostatistics

Background:

• Administrative databases are crucial for health research.
• Diagnostic codes may not accurately reflect true disease status, leading to misclassification bias.
• Methods to correct for this bias are not well-established.

Purpose of the Study:

• To evaluate methods for correcting misclassification bias in administrative database research.
• To compare quantitative bias analysis (QBA) and bootstrap imputation for bias correction.

Main Methods:

• Two cohorts studied: severe renal failure and Colles' fracture.
• Disease status determined by gold standard methods and compared to diagnostic codes.
• Misclassification bias adjusted using QBA with varying bias parameters and bootstrap imputation with probability models.

Main Results:

• Diagnostic codes caused significant bias in prevalence and disease-covariable associations.
• QBA's effectiveness depended on bias parameter accuracy and specificity.
• Bootstrap imputation corrected bias but required well-calibrated disease models.

Conclusions:

• Inaccurate diagnostic codes in administrative data can lead to substantial bias.
• QBA with accurate parameters or bootstrap imputation with calibrated models can address misclassification bias.