Global Odds Model with Proportional Odds and Trend Odds Applied to Gross and Microscopic Brain Infarcts

Area of Science:

• Epidemiology
• Biostatistics
• Medical Research

Background:

• Ordinal measures of symptoms or pathology are common in medical and epidemiological studies.
• Analyzing two correlated ordinal measures often involves modeling one as a predictor for the other, leading to issues like biased probabilities and reduced statistical power due to multicollinearity.
• These limitations necessitate examining both variables as simultaneous outcomes.

Purpose of the Study:

• To address limitations in analyzing correlated ordinal measures in medical research.
• To propose an extended statistical model for simultaneous analysis of two correlated ordinal outcomes.
• To offer a parsimonious modeling option when standard assumptions are not met.

Main Methods:

• Utilized proportional odds models for marginal probabilities and constant global odds models for associations.
• Extended existing models to accommodate situations where standard assumptions of proportional marginal odds and constant global odds do not hold.
• Employed simulation studies and analyzed real-world data on brain infarcts in older adults to compare approaches.

Main Results:

• Identified issues with traditional methods, including biased probability estimates and decreased power.
• Demonstrated the utility of simultaneous outcome modeling for correlated ordinal measures.
• In the brain infarcts data, age at death was a marginal predictor for both gross and microscopic infarcts but did not influence their association.

Conclusions:

• Simultaneous modeling of correlated ordinal outcomes offers advantages over traditional predictor-outcome approaches.
• The proposed extended model provides a flexible and parsimonious solution when standard assumptions are violated.
• The findings have implications for analyzing complex health data, such as in studies of aging and neurological conditions.