Historical Epidemiology with Newspapers

The high resolution of newspapers shows seasonal patterns

These patterns co-exist with strong historical signals

Cholera has strong seasonality, and strong signals around world epidemics (in 1886 and 1892)

Seasonal patterns give strong elements to analyze

This chart, for instance, show Christmas coming earlier and stronger over time as advertising increases.

We can use this to track diseases

Measles shows a spike in the spring months

There are Historical Patterns

Pneumonia's peaks shift by decade

Statistical techniques like Fourier Transforms and Wavelet Analysis can reveal substantial seasonal trends...

Here, wavelet analysis shows a seasonal pattern for Pneumonia peaking every year

...even in cases, like Pertussis, where seasonality is an open question in the medical literature

Even without data normalization, there is strong evidence of a seasonal pattern to Pertussis, and possibly some cycles of > 1 year

This contradicts the limited ground-truth data we have on Pertussis

Hand entered data from US mortality reports (blue) shows much more incidence in the summer months than newspapers (red)

Some variation displays a step-wise pattern at the month breaks, including Whooping Cough...

(Trend lines added for each month)

And a disease with similar symptoms, croup. But why?

Trend lines added for each month

Answer: Advertising!

Advertisers often purchase ads by the month or quarter; cough remedies run in the winter or fall.

This highlights the importance of fully understanding source materials

By understanding these effects, it may be possible to overcome them

And some diseases of greater public health importance don't show advertising patterns, suggesting other promising routes for studying public health in the era before a public health infrastructure