Zero-inflated models for adjusting varying exposures: a cautionary note on the pitfalls of using offset

Area of Science:

• Public Health and Epidemiology
• Statistical Modeling

Background:

• Zero-inflated count data are common in public health and epidemiology.
• Traditional two-part models use an offset for exposure, assuming a fixed effect.
• This assumption is restrictive, and offsets are often only in the count component.

Purpose of the Study:

• To propose a novel zero-inflated model incorporating varying exposure as a covariate.
• To address limitations of the traditional offset approach in zero-inflated models.
• To investigate the impact of exposure on both excessive zeros and event counts.

Main Methods:

• Developed a modified zero-inflated model.
• Incorporated exposure as a covariate in both the zero-probability and count components.
• Utilized a real-world public health example and simulation studies for validation.

Main Results:

• The proposed method offers a more flexible approach to modeling exposure effects.
• Simulation studies demonstrate the method's performance across various scenarios.
• The real-world example illustrates practical application and benefits.

Conclusions:

• Treating exposure as a covariate in zero-inflated models enhances analytical flexibility.
• This approach better captures the influence of exposure on both zero-inflation and event rates.
• The proposed method provides a valuable tool for epidemiological and public health research.