Regression models for censored serological data

Area of Science:

• Epidemiology
• Biostatistics
• Immunology

Background:

• Standardizing serosurvey results across different laboratories and assays is crucial for accurate epidemiological analysis.
• Censored serological measurements, common in serosurveys, can introduce bias into regression analyses.
• The European Sero-Epidemiology Network 2 project highlighted the need for robust methods to handle such data.

Purpose of the Study:

• To assess the impact of censored serological measurements on regression equations used for standardizing serosurvey results.
• To compare the performance of various statistical methods for adjusting censored data.
• To identify the most accurate method for standardizing serological measurements from different laboratories.

Main Methods:

• Comparison of statistical methods including deletion, simple substitution, multiple imputation, and censored regression.
• Simulations generated from scenarios based on serological panel comparisons from multiple national laboratories and assays.
• Evaluation of methods under varying proportions of censored data and different regression assumptions.

Main Results:

• Simple substitution and deletion methods performed adequately with low censoring (<20%).
• Censored regression generally yielded estimates closer to the true values across various scenarios.
• Interval-censored regression provided the least biased estimates specifically for assay data from dilution series.

Conclusions:

• Censored regression methods offer superior accuracy for standardizing serosurvey data compared to simpler methods.
• Interval-censored regression is recommended for assay data derived from dilution series due to its minimal bias.
• Accurate standardization of serological data is essential for reliable epidemiological surveillance.