Bootstrap imputation with a disease probability model minimized bias from misclassification due to administrative...

Area of Science:

• Clinical Informatics
• Health Services Research
• Biostatistics

Background:

• Administrative databases frequently use diagnostic codes for patient classification.
• Misclassification of disease status due to inaccurate diagnostic codes introduces bias in research.
• The optimal method for minimizing bias from misclassified disease status remains unclear.

Purpose of the Study:

• To evaluate methods for minimizing bias caused by misclassified disease status in administrative databases.
• To compare the performance of diagnostic codes, categorized probability estimates, and bootstrap imputation for determining severe renal failure status.

Main Methods:

• Utilized serum creatinine measures to define severe renal failure in 50,074 hospitalized patients.
• Compared true severe renal failure prevalence and covariate associations with estimates derived from diagnostic codes, categorized probability estimates, and bootstrap imputation.

Main Results:

• Bootstrap imputation of severe renal failure status using model-derived probability estimates resulted in minimal bias.
• Extensive bias was observed when using diagnostic codes or categorized probability estimates.
• Bootstrap methods demonstrated superior performance in accurately estimating disease prevalence and associations.

Conclusions:

• Bootstrap methods offer a robust approach to minimize bias stemming from misclassified disease status in administrative data.
• Accurate imputation of condition status using multivariable model-derived probability estimates is crucial for reliable health research.
• Findings suggest a shift from traditional diagnostic codes towards advanced imputation techniques for improved data accuracy.