Blog

Forecasting Allergy Season

04/24/2017

Seasonal allergies are never fun, and few data sources on the web measure and/or publish actual allergy activity. Sickweather’s illness tracking algorithms have been collecting allergy social media activity for several years, and our results have revealed two important findings:

  1. Sickweather data is comparable to real-world, seasonal allergy encounters. In a blog post by athenahealth, Iyue Sung, Director of athenaResearch, revealed seasonal allergy patterns as a proportion of total primary care provider (PCP) encounters. Compared to an extract of Sickweather data (although not within the same timeframe), both Sickweather and athenahealth show a peak in seasonal allergies in the MAR – MAY timeframe, and a smaller season in the SEPT-OCT range.
    Sickweather Allergy
    athenahealth Allergy
    Source: athenahealth
  2. Aggregate allergy reports not only align well with clinical data, but also with seasonal OTC remedy sales. In a comparison of Sickweather allergy data and Nielsen allergy remedy sales, we observed a Spearman correlation coefficient of 0.9 for adult remedy sales, and 0.85 for child allergy remedy sales.
    Sickweather Nielsen OTC

To further enhance our ability to deliver quality analytics to enterprise clients, Sickweather has developed and launched a 15-week national allergy forecast, available in our Sickweather PRO offering. This forecast uses a variety of machine learning and seasonal models to forecast weekly allergy reports, achieving an 88% week-1 forecast accuracy (or 12% mean absolute percent error). Our forecast is updated daily, to account for new information collected each day, and provides a 15-week historical trend for context.

Sickweather Allergy Forecast

Coupled with our real-time data collection products (including Sickweather Pro and our enterprise APIs), this forecast can provide significant, timely insights into the allergy season.

submitted by: Bogdan Rau, MPH - Lead Epidemiologist & Data Scientist

← Previous entry