Regression with interval-censored covariates: Application to cross-sectional incidence estimation

Area of Science:

• Statistics
• Biostatistics
• Regression Analysis

Background:

• Generalized linear regression is a powerful tool for analyzing various data types.
• Interval-censored covariates, where an exact value is unknown but falls within a range, present unique analytical challenges.
• Existing methods for handling interval-censored covariates have limitations.

Purpose of the Study:

• To develop and evaluate a novel method for generalized linear regression incorporating interval-censored covariates.
• To address scenarios where the interval-censored covariate is a function of other variables.
• To compare the proposed method's performance against traditional approaches like midpoint imputation and uniform imputation.

Main Methods:

• A new approach models the distributions of variables determining the interval-censored covariate, inferring its distribution indirectly.
• The Expectation-Maximization (EM) algorithm is employed for parameter estimation.
• A simulation framework was developed to rigorously assess accuracy across diverse scenarios.

Main Results:

• The proposed indirect modeling approach demonstrates significantly less bias compared to using censoring interval midpoints.
• Uniform imputation also showed higher bias than the novel method.
• The proposed method incurred only minor increases in standard error relative to the alternatives.

Conclusions:

• The indirect modeling approach offers a more accurate and less biased method for generalized linear regression with interval-censored covariates.
• This technique provides a valuable extension to existing regression methodologies, particularly when covariates are functions of other variables.
• The findings suggest this method is a robust alternative for handling complex covariate censoring in statistical modeling.