Multiple linear regression modeling with values below a lower limit of quantification - a statistical method...

Area of Science:

• Biostatistics
• Medical Data Analysis
• Regression Modeling

Background:

• Missing values are common in medical data, particularly laboratory measurements below the lower limit of quantification (LLOQ).
• Handling these left-censored values is crucial for accurate multivariable linear regression analysis.
• Limited comparative studies exist for methods addressing LLOQ data in regression.

Purpose of the Study:

• To compare the performance of statistical methods for handling values below the LLOQ in multiple linear regression.
• To evaluate methods when LLOQ data is an independent or dependent variable.
• To provide guidance on selecting appropriate methods for left-censored data in regression.

Main Methods:

• A simulation study was conducted to compare statistical methods.
• Methods were evaluated for scenarios where LLOQ data served as independent or dependent variables.
• Variations in data distributions, sample sizes, missing proportions, correlations, and linearity were explored.

Main Results:

• The two compartment model demonstrated superior performance (bias, coverage) when LLOQ data was an independent variable without significant collinearity.
• Tobit regression exhibited the lowest bias and highest coverage for LLOQ data as the dependent variable, up to 0.8 censoring proportions.

Conclusions:

• Choosing an appropriate method for handling left-censored data below the LLOQ is essential in multiple linear regression.
• This study offers guidance on the performance of established methods for such data challenges.