Correlation of school absenteeism and laboratory results for Flu A in Alberta, Canada

Elizabeth Birk-Urovitz, Ye Li, Steven Drews, Christopher Sikora, Deena Hinshaw, Rita K. Biel, Faiza Habib, Laura Riviera, Hussain Usman, David Strong, Ian Johnson

Abstract


ObjectiveTo assess the correlations between weekly rates of elementaryschool absenteeism due to illness (SAi) and percent positivity forinfluenza A from laboratory testing (PPFluA) when conducted at acity level from September to December over multiple years.IntroductionRates of student absenteeism in schools have been mainly used todetect outbreaks in schools and prompt public health action to stoplocal transmission1,2. A report by Mogto et al.3stated that aggregatedcounts of school absenteeism (SAi) were correlated with PPFluA, butthe sample may have been biased. The purpose of this study was toassess the correlation between aggregated rates of SAi and PPFluAfor two cities, Calgary and Edmonton, in Alberta. In such situations,SAi could potentially be used as a proxy for PPFluA when there arenot enough samples for stable laboratory estimates.MethodsThe Alberta Real-Time Syndromic Surveillance Net (ARTSSN)4collects elementary SA data from the two major school boards intwo cities in Alberta with populations >800,000. Since reasons forSA are stated, rates of SAi can be calculated. Data were obtained forthree years, 2012 to 2014, for each city. Laboratory data on tests ofrespiratory agents using a standardized protocol were obtained fromAlberta’s Provincial Laboratory for Public Health for the same timeperiod and locations. The dates of the specimens being received bythe laboratory were used in this analysis. For each data source, therelative proportions (SAi and PPFluA) were calculated. Data forthe first week of school in September and for the last two weeks ofDecember were removed for each year due to the SAi rates beingunstable. Linear regression models were constructed, with rates ofSAi predicted by PPFluA. Separate models were run for each cityand for each year, resulting in a total of 6 models. Percent positivityfor entero-rhinoviruses (PPERV) was added to see if it improved themodel. The regression models were created using Excel and checkedin the statistical programs, SAS and R. An analysis to assess theinfluence of a lag period was assessed using R.ResultsFor each city, the provincial lab tested between 4,000 and 6,000specimens each fall and SAi rates were based on denominators ofbetween 20,000 and 36,000 children. The R2, betas, and p-valuesfor all 6 regression models are shown in Table 1. The minimumcorrelation value was 0.693 and the maximum was 0.935. Dueto the strong negative correlations between PPERV and PPFluA,PPERV was not retained in the models. Looking at the lag periods,the maximum correlations occurred at a zero week lag in two years(2012 and 2014) and at a -1 week lag in 2013. The two years with azero lag were both dominated by a H3N2 strain while the year withmainly a H1N1 strain showed a lag of -1. Only one year of H1N1 datawas available for analysis.ConclusionsWe observed strong correlations between the weekly rates ofelementary SAi and PPFluA at the city level over three years, fromSeptember to December. The reasons for the difference in lag timesbetween the H1N1 and H3N2 seasons are being investigated.

Full Text:

PDF


DOI: https://doi.org/10.5210/ojphi.v9i1.7679



Online Journal of Public Health Informatics * ISSN 1947-2579 * http://ojphi.org