Statistical tests for latent class in censored data due to detection limit.

Area of Science:

• Biostatistics
• Analytical Chemistry
• Biomarker Discovery

Background:

• Assay limit of detection in biological matrices results in censored data.
• Tobit regression (censored normal regression) is standard for data below detection limits.
• Excessive censoring may indicate population heterogeneity with a latent, non-detecting group.

Purpose of the Study:

• To develop and evaluate statistical tests for detecting a latent class in censored biological data.
• To address the gap in testing the necessity of mixture models for such data.

Main Methods:

• Development of Wald, likelihood ratio, and score tests.
• Simulation studies to assess test performance.
• Application to real biological data examples.

Main Results:

• The proposed tests are evaluated for their ability to detect latent classes.
• Performance of the tests is assessed under various simulation conditions.
• The tests are demonstrated on practical biological datasets.

Conclusions:

• The study provides statistical tools to identify population heterogeneity in censored measurements.
• The developed tests can help determine if a mixture model is more appropriate than standard Tobit models.
• This research contributes to more accurate analysis of biological data with non-detectable levels.