1818. Validating the use of Google enquiry metadata to enhance pertussis surveillance in Australia
Session: Poster Abstract Session: Respiratory Infections: Potpourri
Saturday, October 10, 2015
Room: Poster Hall
Background:  Australia has experienced several recent pertussis epidemics. Timely surveillance is vital to guide public health responses but reporting of sentinel surveillance pertussis incidence data in Australia is not real-time.  We sought to develop a Google-based model for the real-time detection and prediction of pertussis epidemics in Australia. Methods:  Weekly national pertussis incidence was determined using molecular confirmed sentinel surveillance case data from the Communicable Diseases Network Australia.  Weekly trends in candidate predictor Google search terms related to pertussis were obtained using the Google Trends and Google Correlate websites.  A linear regression model was fit using a training period of high pertussis variability February 2008 - April 2010. Candidate Google predictors were added sequentially to the model and retained if they improved the adjusted R2 of the model and also improved the model performance as assessed by a ten-fold internal cross-validation. Skewed variables were log-transformed where appropriate to improve model fit and highly collinear predictors were excluded. The model was then externally validated on a one year hold-out period of data corresponding to the largest recent epidemic wave of pertussis in Australia (May 2010 - April 2011). Results: A four-term linear model was fit with an adjusted R2 of 0.82 (r = 0.90) and internal ten-fold cross-validation r = 0.90 (p<0.001). External validation on the hold-out data showed correlation between predicted and actual pertussis incidence of r = 0.87, p<0.001 (R2 = 0.75). The Google model predicted the time of peak pertussis incidence by one week. Conclusion: A Google-based real-time surveillance model was moderately to highly accurate in the real-time detection of pertussis in Australia and predicted peak pertussis activity by one week. This model may offer a complementary signal to enhance pertussis surveillance in Australia during epidemic periods.
Simon Pollett, MBBS, BMedSci, DTMH, FRACP1,2, Nicholas Wood, MBBS DCH MPH FRACP PhD3, W. John Boscardin, PhD2, Henrik Bengtsson, PhD2 and George Rutherford, MD MPH2, (1)Marie Bashir Institute for Infectious Diseases and Biosecurity, University of Sydney, Sydney, NSW, Australia, (2)Department of Epidemiology and Biostatistics, University of California at San Francisco, San Francisco, CA, (3)National Centre for Immunisation Research and Surveillance, Westmead, Australia


S. Pollett, None

N. Wood, None

W. J. Boscardin, None

H. Bengtsson, None

G. Rutherford, None

Findings in the abstracts are embargoed until 12:01 a.m. PDT, Wednesday Oct. 7th with the exception of research findings presented at the IDWeek press conferences.