Robustness of prevalence estimates derived from misclassified data from administrative databases

Area of Science:

• Health Services Research
• Biostatistics
• Epidemiology

Background:

• Medical administrative databases are increasingly used for research due to cost-effectiveness.
• These databases often contain misclassification errors, particularly in diagnostic codes, affecting data accuracy.
• Inaccurate diagnostic codes can bias prevalence estimates and associations with other health variables.

Purpose of the Study:

• To estimate the prevalence of osteoarthritis (OA) in elderly Quebeckers using a government administrative database.
• To compare a naive prevalence estimate with those derived from Bayesian latent class models that adjust for diagnostic code errors.

Main Methods:

• Utilized a government administrative database for elderly Quebeckers.
• Employed Bayesian latent class models to adjust for misclassified physician diagnostic codes.
• Incorporated other available diagnostic clues to refine estimates.

Main Results:

• Prevalence estimates for osteoarthritis varied significantly based on the model and assumptions used.
• Naive estimates relying solely on listed diagnoses differed substantially from adjusted estimates.
• The study highlights the impact of diagnostic code accuracy on prevalence calculations.

Conclusions:

• Inferences drawn from medical administrative databases require cautious interpretation.
• Further research is needed to assess the reliability of data items within these databases.
• Adjusting for misclassification errors is essential for accurate epidemiological research using administrative data.