Applying Health Informatics Approaches to Support Public Health Risk Communication - Temporal and Spatial Analysis of H1N1 Human Infection Distribution in the U.S.
Abstract
Background: we propose a prototype of an informatics-facilitated public health risk communication system for H1N1 surveillance. This system is enabled by incorporating spatial and temporal analysis to evaluate the nature of emergency relevance of reported human cases of novel H1N1 Influenza human infection and its severity. Human cases reported in the U.S. were analyzed as of June 12, 2009, on the day when the WHO declared Level 6 pandemic flu of the H1N1 virus. It seeks to determine emergency nature of the case by excessive human cases by spatial and temporal distribution in the U.S. We evaluated the distribution trend of historical and current excess cases, and their associated geographic location and time period of occurrence by excess level. We also measured temporal variation pattern of case distribution to understand potential temporal trend of emergency relevance. Methods: Data were collected from the CDC H1N1 program Website. Daily reported confirmed/probable human H1N1 infection cases were tallied from April 22, 2009 to June 12th, 2009. Totally 18,080 cases reported by 51 states (including the District of Columbia) were analyzed for the more than 7 weeks period. We employed the Scan Statistic using both temporal only option Poisson model and Space-Time Permutation Scan Statistic to conduct both retrospective and prospective analyses and to detect high excess case clusters. Time series plots and periodicity tests were used to investigate temporal pattern of case distribution. Results: Temporal only scan reveals that, between 4/22 and 6/12, cases reported on 6/12 increased by thirteen times (obs/exp=13.32, RR=17.57) as many as the period before 6/12. Spatial temporal analysis identifies twelve statistically significant clusters involving 39 states or regions. Retrospective analysis detected that California, Nevada, Arizona as the most likely excess case cluster on 05/14, followed by Texas cluster on 05/22 – 05/27, 16 states surrounding Connecticut on 6/5 – 6/12, 5 states surrounding Illinois on 05/05 – 05/08, and Wisconsin on 5/27 – 5/29. Three hot-spot clusters were identified with one centered around California, Nevada, Arizona (5/14, obs/exp=4.9), one in the state of Texas (5/22-5/27, obs/exp=3.2), and one surrounded Illinois, Indiana, Missouri, Kentucky, Tennessee (5/5-5/8, obs/exp=2.5). There was no hot-spot cluster identified by prospective analysis. In addition, there was substantial increase in cases and periodicity occurred on Thursdays and Fridays. Discussion and Conclusion: Spatiotemporal clusters and temporal periodicity exist in H1N1 Influenza human infections in the U.S. Spatiotemporal analysis identified several potential emergency-relevant case clusters by retrospective analysis that warrants further attention. The results render lessons that based on the limitation of case reporting, backlog of case confirmation, and findings from the present study that some of the most likely clusters happened in the past, it is advisable that decision of raising pandemic alert level should not solely base on visual inspection of increased cases but should also fully consider emergency relevance information for risk communication.