Date of Award
Doctor of Philosophy
Interdisciplinary Health Sciences
Dr. Kieran Fogarty
Dr. Amy Curtis
Dr. George White
Dr. Robert Walsh
Epidemiology, public health surveillance, forecasting
The analysis of public health surveillance data to identify departures from historical patterns of disease is required to facilitate the timely identification of potential outbreaks. Using the Box-Jenkins forecasting model, this study examines the potential to predict future disease burden based upon the historical record within local public health jurisdictions. Box-Jenkins forecasting was developed as a direct result of forecast problems in the business, economic, and control-engineering applications, yet it has not been systematically examined for use with public heath surveillance data.
Box-Jenkins forecast models are constructed by stratifying 84,029 disease reports from the State of Utah by year (n = 10), disease type (n = 50), and jurisdiction (n = 13). A disease has to be present in all years and have a rate greater than 0.2/100K to be included in the study. Sixteen diseases have been selected for analysis. Accuracy of the forecasts is determined by conducting 48 forecast trials; within these trials there are 576 monthly forecasts. The results are compared to the actual values for the same period. Accuracy is determined calculating the Mean Absolute Percentage Error (MAPE) for each forecast trial. Forecast predication intervals explore the relationship between actual values and the predication interval associated with each forecast.
Forecasts have an absolute accuracy of 71% (range: 43.4–91.7%). Ten of the 16 forecasts (63%) have an absolute accuracy greater than 75%, four (25%) have an absolute accuracy between 52.6% and 69.6%, and two (12%) have an accuracy of less than 50%. Forecast accuracy is independent of rate of disease (r = –.348, n = 16, p >.05) and jurisdictional size (r =.396, n = 7, p =.380). Eighty-four percent of all forecast values are contained within the first forecast interval, 88% within the second, and 99% within the third.
This study demonstrates that it is possible to predict future disease burden using Box-Jenkins forecasting techniques. The overall accuracy of the forecast and disproportionate number of forecast values contained within the first forecast interval validate this as a method that may be used to monitor disease trends and potentially facilitate the early identification of an outbreak.
Garrett, Larry C., "Using Box-Jenkins Modeling Techniques to Forecast Future Disease Burden and Identify Disease Aberrations in Public Health Surveillance Report" (2012). Dissertations. 106.