An Improved Pooled Logistic Regression Implementation

Area of Science:

• Epidemiology
• Biostatistics
• Computational Statistics

Background:

• Pooled logistic regression is widely used for survival analysis in epidemiology.
• Computational challenges often lead to approximations like widening time intervals or using parametric time forms.
• These approximations can limit flexibility in modeling time-dependent effects.

Purpose of the Study:

• To present a novel computational approach for pooled logistic regression.
• To reduce the computational burden without constraining the functional form of time.
• To offer an alternative to existing methods for handling computational challenges.

Main Methods:

• The proposed algorithm processes only unique event times from the dataset.
• This method is compatible with flexible time modeling using disjoint indicators.
• Implementations were compared in SAS, R, and Python using a public dataset.

Main Results:

• The proposed implementation yielded identical point estimates to the standard method.
• Computational speed improvements ranged from 6 to 68 times faster, depending on the software.
• This demonstrates significant efficiency gains for the new algorithm.

Conclusions:

• The novel implementation simplifies the estimation of pooled logistic regression models.
• This simplification is particularly beneficial when using bootstrap methods for statistical inference.
• The approach offers a computationally efficient alternative for epidemiological survival analyses.