Measuring association among censored antibody titer data

Area of Science:

• Biostatistics
• Medical Data Analysis
• Statistical Modeling

Background:

• Censored data, common in medical studies due to detection limits, poses challenges for traditional analysis.
• Complete case (CC) analysis and substitution methods often yield biased estimates.
• Maximum Likelihood Estimation (MLE) shows promise but requires robust methods for censored data.

Purpose of the Study:

• To develop an MLE approach for estimating associations between two measurements with censored data.
• To extend the MLE method for linear regression models with censored covariates or responses.
• To compare the proposed methods against conventional techniques and other advanced approaches.

Main Methods:

• Utilizing copula functions to link marginal distributions of censored variables.
• Implementing MLE for estimating associations and fitting regression models.
• Conducting simulation studies to evaluate performance against CC analysis, substitution, multiple imputation, and missing indicator models.

Main Results:

• The proposed MLE copula approach significantly outperforms CC and substitution analyses in estimating associations.
• The method accurately estimates associations and is robust to misspecification of copula or marginal distributions.
• The MLE regression approach demonstrates superior performance for censored covariates compared to conventional methods.

Conclusions:

• The MLE approach using copula functions provides a powerful and accurate tool for analyzing censored medical data.
• This method extends to linear regression, offering improved estimates when covariates or responses are censored.
• The proposed techniques offer more reliable statistical inference in medical research involving censored observations.