Analysis of surveillance data with Poisson regression: a case study.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Legal abortion restrictions impact abortion-related mortality.
• Assessing statistical methods for detecting changes in public health surveillance data is crucial.

Purpose of the Study:

• To evaluate the sensitivity of Poisson regression in detecting changes in abortion-associated mortality.
• To determine the data requirements for reliably detecting such changes at national and state levels.

Main Methods:

• Analysis of abortion-associated mortality data from 1962 to 1984 for the United States and individual states.
• Application of Poisson regression to assess the detection of mortality changes following the elimination of legal abortion restrictions.
• Comparison of statistical tests, including those based on model deviance and observed vs. expected event counts.

Main Results:

• Poisson regression detected the expected decrease in abortion mortality nationally.
• Detection of this trend in state-level data was inconsistent, requiring larger datasets (approx. 370 events over 23 years).
• Smaller state datasets (e.g., 1 event/year) failed to detect the legal change, suggesting a minimum threshold of 2-3 events/year for outlier detection.

Conclusions:

• Poisson regression's ability to detect changes in abortion mortality is dependent on data volume, especially at the state level.
• A minimum of 2-3 events per year is recommended for Poisson regression to reliably identify outliers in surveillance data.
• Tests based on model deviance are more effective than those comparing observed and expected counts for this type of analysis.