Related Experiment Videos
Poisson regression analysis of ungrouped data
D Loomis1, D B Richardson, L Elliott
1Department of Epidemiology, University of North Carolina, Chapel Hill, NC 27599-7435, USA. Dana.Loomis@unc.edu
Occupational and Environmental Medicine
|April 20, 2005
Summary
Analyzing epidemiological data with ungrouped person-time in Poisson regression avoids bias from exposure categorization. This method yields results equivalent to proportional hazards regression, offering a more accurate assessment of exposure-response relationships.
Area of Science:
- Epidemiology
- Biostatistics
- Occupational Health
Background:
- Poisson regression is standard for occupational cohort studies, typically using grouped data.
- Grouped analysis involves categorizing exposure and covariate information, tabulating person-time and events.
Purpose of the Study:
- To present an alternative Poisson regression approach using individual units of person-time.
- To evaluate the impact of data grouping on regression analysis outcomes.
Main Methods:
- Poisson regression applied to simulated and empirical cohort data.
- Comparison of ungrouped Poisson regression with proportional hazards regression using simulated data.
- Analysis of a large occupational cohort (138,900 electrical workers) to demonstrate the ungrouped approach.
Main Results:
- Ungrouped Poisson regression yielded unbiased estimates, equivalent to proportional hazards regression for simulated data.
- Grouped and ungrouped analyses produced identical results when models were consistent.
- Bias can occur in grouped analyses when using exposure scores like category means or midpoints.
Conclusions:
- Poisson regression with ungrouped person-time data is a valuable method for epidemiological analysis.
- This approach mitigates bias introduced by exposure data categorization and score assignment.
- Facilitates direct evaluation of how exposure categorization impacts regression findings.