Revisiting Pearl's influenza studies by bootstrapping for forward variable selection with a null factor

Area of Science:

• Epidemiology
• Biostatistics
• Historical Public Health

Background:

• Raymond Pearl's 1919-1921 studies investigated factors influencing the Spanish Flu's severity in US cities.
• Pearl utilized partial correlation coefficients to analyze demographic and disease death rates.
• A need exists to apply modern statistical techniques to historical epidemiological data.

Purpose of the Study:

• To re-evaluate factors predicting Spanish Flu epidemic severity using contemporary statistical methods.
• To compare results with Raymond Pearl's original findings from the early 20th century.
• To demonstrate the utility of advanced variable selection in historical epidemiology.

Main Methods:

• Applied bootstrap simulation with forward variable selection and a null factor for generalized linear regression.
• Utilized AICc (Akaike Information Criterion with correction) for model validation.
• Employed a null factor (a random, independent variable) to assess term significance in model selection.

Main Results:

• Pre-pandemic death rates from organic heart disease and all causes were identified as highly predictive of pandemic severity.
• Results largely corroborate Pearl's original conclusions regarding key predictive factors.
• Substantive nuances were identified, suggesting refinements to historical epidemiological models.

Conclusions:

• State-of-the-art variable selection methodology is valuable for analyzing historical epidemic data.
• Pre-existing health conditions, specifically heart disease mortality, significantly influenced Spanish Flu's impact.
• Modern statistical approaches can enhance understanding of past public health crises.