Are there two logistic regressions for retrospective studies?

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Retrospective studies are crucial for epidemiological research.
• Logistic regression is a common statistical method for analyzing binary outcomes.
• Different modeling approaches can influence risk estimation in retrospective data.

Purpose of the Study:

• To compare two distinct linear logistic regression models for analyzing retrospective study data.
• To evaluate the impact of different dependent variable definitions on relative risk estimation.
• To determine the preferred model for studies with multiple quantitative risk factors.

Main Methods:

• Analysis of retrospective study data using two linear logistic regression approaches.
• Prospective model: dependent variable as case/control status.
• Retrospective model: dependent variable as exposure classification (binary or polychotomous).

Main Results:

• Both models yield increasingly similar relative risk estimates with greater covariate adjustment.
• Identical relative risk estimates are achieved when covariate effects are saturated with parameters.
• The prospective model is generally recommended for studies with multiple quantitative risk factors.

Conclusions:

• The choice of logistic regression model impacts relative risk estimation in retrospective studies.
• Covariate adjustment is key to harmonizing estimates between prospective and retrospective models.
• The prospective logistic regression model offers advantages for complex epidemiological analyses involving quantitative risk factors.