Pooled Multinomial Logistic Regression for Parametric G-computation in the Presence of Competing Events

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Parametric g-computation for competing events often requires multiple logistic regression models.
• An alternative approach using a single pooled multinomial logistic model is proposed.

Purpose of the Study:

• To compare two parametric g-computation approaches: multiple pooled logistic regression and pooled multinomial logistic regression.
• To estimate the marginal 2-year risk of highly active antiretroviral therapy (HAART) initiation and AIDS/death before HAART initiation.

Main Methods:

• Utilized data from the Women's Interagency HIV Study (n=1,164).
• Applied both multiple pooled logistic regression and pooled multinomial logistic regression for parametric g-computation.
• Estimated the total effect of historical injection drug use on both event types using the multinomial approach.

Main Results:

• Both g-computation methods produced identical results.
• The 2-year risk difference for HAART initiation was -12.5% (comparing historical injection drug use vs. none).
• The 2-year risk difference for AIDS/death was 13.2% (comparing historical injection drug use vs. none).

Conclusions:

• A pooled multinomial logit nuisance model simplifies parametric g-computation for total effects.
• This method effectively accounts for right-censoring and competing events in survival analysis.
• The findings offer a more streamlined approach to analyzing complex health outcomes in cohort studies.