Fitting general relative risk models for survival time and matched case-control analysis.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Cox proportional hazards and conditional logistic regression are standard epidemiological tools.
• Current statistical software limits analysis to log-linear models, implying exponential exposure-response and multiplicative interactions.
• This limitation restricts the accurate modeling of complex relationships in epidemiological data.

Purpose of the Study:

• To describe methods for fitting non-log-linear Cox and conditional logistic regression models.
• To extend the flexibility of survival and matched case-control data analysis.
• To enable the fitting of general relative risk models beyond standard log-linear forms.

Main Methods:

• Developed and illustrated methods for fitting non-log-linear Cox and conditional logistic regression models.
• Utilized data from a lung cancer mortality study of Colorado Plateau uranium miners (1950-1982).
• Applied methods to matched case-control data, countermatched data with weights, d:m matching, and full cohort Cox regression using SAS.

Main Results:

• Successfully demonstrated the fitting of non-log-linear models for Cox and conditional logistic regression.
• Showcased the application of these flexible models to real-world epidemiological data.
• Validated the utility of the described methods in analyzing complex exposure-response relationships.

Conclusions:

• The described methods provide a more flexible approach to analyzing survival and matched case-control data.
• These techniques overcome the limitations of standard log-linear models in epidemiological research.
• The study facilitates more accurate risk assessment in epidemiological studies by allowing for general relative risk models.