A sensitivity analysis to separate bias due to confounding from bias due to predicting misclassification by a...

Area of Science:

• Epidemiology
• Biostatistics
• Oncology

Background:

• Misclassification of exposure, outcome, or confounders requires external adjustments, not just risk-based adjustments.
• A variable predicting misclassification of both exposure and a confounder can also confound the exposure-outcome relationship.

Purpose of the Study:

• To investigate the relationship between less-than-definitive therapy and breast cancer mortality.
• To differentiate between confounding bias and misclassification bias in epidemiological analysis.
• To assess the impact of prognostic evaluation on both therapy misclassification and confounding.

Main Methods:

• Utilized a sensitivity analysis to disentangle misclassification biases from confounding bias.
• Focused on breast cancer mortality within 5 years of diagnosis.
• Examined the role of prognostic evaluation in predicting misclassification of definitive therapy and cancer stage.

Main Results:

• The original multivariable model showed a relative hazard of 1.75 for less-than-definitive therapy.
• Sensitivity analysis yielded a median relative hazard of 1.64, with 90% of estimates between 1.47 and 1.83.
• The original analysis overestimated the relative hazard due to unaddressed misclassification biases.

Conclusions:

• Less-than-definitive therapy is associated with an excess relative hazard of breast cancer mortality.
• Sensitivity analysis provides a more accurate estimate of the therapy-outcome relationship by accounting for misclassification.
• Properly addressing misclassification bias is crucial for accurate epidemiological findings.