A sensitivity analysis approach for informative dropout using shared parameter models.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Shared parameter models (SPMs) are utilized to mitigate bias caused by informative dropout in longitudinal studies.
• A key assumption in SPMs is the conditional independence between the longitudinal outcome and dropout time, which remains unverifiable.
• Current sensitivity analysis methods for SPMs are insufficient for this unverifiable assumption.

Purpose of the Study:

• To propose a novel approach for transparent sensitivity analyses in the presence of informative dropout.
• To clearly separate the observed data likelihood from the extrapolation distribution within SPMs.
• To enhance the reliability of statistical inferences from longitudinal studies with missing data.

Main Methods:

• Developed a new sensitivity analysis framework for SPMs that distinguishes observed data likelihood from the extrapolation distribution.
• Employed a typical SPM as a model for the complete data generating mechanism.
• Utilized a skew-normal distribution for the default extrapolation distribution and calibrated sensitivity parameters using observed dropout data.

Main Results:

• The proposed method provides a transparent approach to sensitivity analyses for informative dropout.
• Successfully separated the observed data likelihood from the extrapolation distribution.
• Applied the methodology to address informative dropout in the HIV Epidemiology Research Study.

Conclusions:

• The new approach enhances transparency in sensitivity analyses for SPMs with informative dropout.
• This method offers a more robust way to handle unverifiable assumptions in longitudinal data analysis.
• The findings have implications for improving the accuracy of statistical models in various research fields dealing with missing data.