Evaluation of Nowcasting for Detecting and Predicting Local Influenza Epidemics, Sweden, 2009–2014

Abstract

The growing availability of big data in healthcare and public health opens possibilities for infectious disease control in local settings. We prospectively evaluated a method for integrated local detection and prediction (nowcasting) of influenza epidemics over 5 years, using the total population in Östergötland County, Sweden. We used routine health information system data on influenza-diagnosis cases and syndromic telenursing data for July 2009–June 2014 to evaluate epidemic detection, peak-timing prediction, and peak-intensity prediction. Detection performance was satisfactory throughout the period, except for the 2011–12 influenza A(H3N2) season, which followed a season with influenza B and pandemic influenza A(H1N1)pdm09 virus activity. Peak-timing prediction performance was satisfactory for the 4 influenza seasons but not the pandemic. Peak-intensity levels were correctly categorized for the pandemic and 2 of 4 influenza seasons. We recommend using versions of this method modified with regard to local use context for further evaluations using standard methods.

Nowcasting (Short-Term Forecasting) of Influenza Epidemics in Local Settings, Sweden, 2008–2019

“…‡Positive value means that the algorithm issued an alarm before the local epidemic had started; negative value means that the alarm was raised after the start of the epidemic. §Actual start is the date when the retrospectively calculated intensity level reached the predefined threshold for start of an epidemic (6.3 influenza-diagnosis cases/100,000 population recorded during a floating 7-day period) ( 7 , 11 ). ¶No update of threshold before this seasonal influenza because the previous outbreak was a pandemic.…”

“…Longitudinal prospective evaluations might be needed to draw valid conclusions concerning the performance of local epidemic nowcasting, and inclusion of data from urban counties might be required for generalizability ( 7 ). We found in our study that the performance of seasonal influenza nowcasting was satisfactory during a 10-year period in 3 urban counties regarding local detection and peak-timing prediction performance.…”

“…We used an open cohort design based on the total population in 3 urban counties: Stockholm County (population 2,231,000), West Gothia County (population 1,649,000), and Scania County (population 1,304,000) ( Figure 1 ). We used retrospective data from January 1, 2008, through June 30, 2009, and prospective data from July 1 through February 28, 2019, from 2 sources in the countywide health information system: daily numbers of clinically diagnosed influenza cases ( Figure 2 ) and daily syndromic chief complaint data from a telenursing service ( Figure 3 ) ( 6 , 7 ). The clinical influenza case data were used for detection of the local start of the epidemic and prediction of its peak intensity, and the syndromic data were used to detect the peak timing.…”

“…Short-term forecasting is denoted as nowcasting ( 5 ). Recently, a prospective 5-year appraisal of a local nowcasting method ( 6 ) in a county in Sweden (county population ≈460,000) indicated promising results with regard to detection of the local start of the epidemic, prediction of peak timing, and prediction of peak intensity ( 7 ). The appraisal concluded that a longer prospective evaluation was needed to ascertain the validity of the results and that data from larger urban counties were required to draw reliable conclusions about generalizability.…”

Developing a Heart Transplantation Self-Management Support Mobile Health App in Taiwan: Qualitative Study

“…The growing availability of big data in health care and public health opens up possibilities for infectious disease control in local settings. The detection and prediction (nowcasting) of influenza epidemics are now becoming possible [ 31 ].…”