A new statistical test for latent class in censored data due to detection limit

Area of Science:

• Biostatistics
• Biomarker Analysis
• Public Health Research

Background:

• Biomarker measurements in biological samples frequently fall below assay detection limits, resulting in censored data.
• Censored data are prevalent in public health and medical research, posing analytical challenges.
• Standard Tobit regression models assume normality and may not adequately handle excessive censoring due to population heterogeneity.

Purpose of the Study:

• To develop a novel statistical test for detecting latent subpopulations within biomarker data analyzed by Tobit regression.
• To address situations where observed censored data exceed expectations under a standard Tobit model.
• To improve the accuracy of biomarker analysis in heterogeneous populations.

Main Methods:

• Development of a new test statistic based on comparing observed censored data with Tobit model expectations.
• Derivation of the test statistic's closed form and asymptotic properties using estimating equations.
• Evaluation of the new test's performance through simulation studies and comparison with existing methods (Wald, likelihood ratio, score tests).

Main Results:

• The proposed test effectively identifies latent classes in censored biomarker data.
• Simulation studies demonstrate the new test's performance and its advantages over existing methods.
• The test provides a robust approach for analyzing biomarker data with potential population heterogeneity.

Conclusions:

• The developed test offers a valuable tool for accurately analyzing censored biomarker data, particularly in the presence of unobserved subpopulations.
• This method enhances the reliability of findings in public health and medical research where biomarker detection limits are common.
• The study contributes a statistically sound approach to handling complex censored data scenarios.