Quantifying temporal trends of age-standardized rates with odds

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Traditional methods for quantifying disease rate trends often use linear regression on log-transformed data.
• Log-transformation may not always be appropriate for analyzing disease rates.
• Accurate quantification of temporal trends is crucial for public health surveillance.

Purpose of the Study:

• To introduce and validate the rank-ordered logit (ROL) model as an alternative to log-transformed linear regression for analyzing temporal trends in disease rates.
• To compare the ROL model's performance against traditional methods using real-world cancer incidence data.
• To provide a transformation-invariant method for epidemiological trend analysis.

Main Methods:

• Proposed the rank-ordered logit (ROL) model, which quantifies trends using odds and is implemented via proportional hazards regression commands.
• Applied the ROL method to age-standardized cancer incidence rates from the Cancer Incidence in Five Continents plus (CI5plus) database (1953-2007).
• Compared ROL model estimates with those from linear regression with and without log-transformation.

Main Results:

• Demonstrated strong concordance in the direction and significance of temporal trends across all three methods (ROL, log-transformed linear regression, untransformed linear regression).
• Showcased that the ROL model's estimate is comparable to a scaled slope parameter from linear regression.
• Validated the ROL model's applicability to large-scale epidemiological datasets.

Conclusions:

• The ROL model provides a reliable, transformation-invariant alternative for quantifying temporal trends in disease incidence or mortality rates.
• This method aligns with epidemiological practices by using odds for trend quantification.
• The ROL model offers a valuable tool for researchers and public health professionals analyzing population health trends.