Simultaneous alleviation of verification and reference standard biases in a community-based tuberculosis screening...

Area of Science:

• Epidemiology
• Biostatistics
• Infectious Disease Research

Background:

• Tuberculosis (TB) prevalence surveys face challenges from reference standard and verification biases.
• Reference standard bias stems from imperfect diagnostic tests, while verification bias arises from selective testing of symptomatic individuals.
• Bayesian latent class analysis (LCA) can address reference standard bias but may still be affected by verification bias.

Purpose of the Study:

• To identify optimal methods for simultaneously reducing reference standard and verification biases in TB prevalence surveys.
• To improve the accuracy of pulmonary TB prevalence and diagnostic test performance estimates.
• To evaluate advanced Bayesian approaches for handling missing data in TB prevalence surveys.

Main Methods:

• Secondary analysis of 9869 participants from a South African community-based screening study.
• Bayesian LCA was performed using five approaches to handle unverified individuals, including complete-case analysis, assuming negative results, and various imputation methods (MICE, MAR, MNAR).
• Simulations were conducted with a 2.0% true prevalence to assess bias alleviation.

Main Results:

• Bayesian LCA with simultaneous imputation under MAR and MNAR assumptions effectively alleviated both reference standard and verification biases.
• Composite reference standard (CRS) analysis and Bayesian LCA assuming negative results only alleviated biases when true prevalence was <3.0%.
• In the Vukuzazi study, Bayesian LCA with simultaneous imputation yielded PTB prevalence estimates of 0.9% (MAR) and 0.7% (MNAR), with realistic diagnostic accuracy.

Conclusions:

• Advanced Bayesian techniques, particularly simultaneous imputation of missing data, are effective in mitigating biases in TB prevalence surveys.
• These methods enhance the reliability of community-based screening programs for TB.
• Imputing missing bacteriological test results as negative is a plausible strategy under realistic assumptions.