Methods
ORCHARDS : The ORegon CHild Absenteeism due to Respiratory
Disease Study (ORCHARDS) is a prospective, observational study of
kindergarten through 12th grade (K-12) student
absenteeism and influenza in the Oregon School District (OSD), Dane
County, located in southcentral Wisconsin. The primary goal of ORCHARDS
is to develop a system to monitor cause-specific, K-12 absenteeism on a
daily basis and to assess its usability for early detection of influenza
and ILI transmission in schools and in the community. The overall
methodology of ORCHARDS is detailed elsewhere20.
Population: The OSD comprises six public schools with a growing
enrollment, estimated at 4,091 students (18% of total population)
during the 2018-2019 school year21. The district’s
overall population is estimated at 23,000 and is less racially and
ethnically diverse, wealthier, and more educated than the average
community in the United States22.
Data Collection: Parents/guardians are required to report
absences using an automated telephone system and are prompted to report
respiratory symptoms. The OSD records absenteeism in Infinite Campus®
(https://www.infinitecampus.com), a commercially available
electronic student information system. For the purpose of this study, in
2014 the OSD Information Technology staff added an option within the
system that allowed entry of student absenteeism characterized as a-ILI.
We defined absence as missing any part of the school day. We defined ILI
as the presence of fever and at least one of the following symptoms:
cough, sore throat, nasal congestion, or runny nose20as reported by a parent/guardian on the telephone system. The daily
count of a-ILI was the primary outcome measure for this study.
Data Extraction: The OSD developed an automated process to
extract daily counts of student absences by school, grade, and type of
absence. Data were sent on a daily basis to ORCHARDS researchers using a
secure file transfer (ftp) site. No personal identifiable information
was included, and the data were fully compliant with the Family
Educational Rights and Privacy Act (FERPA 20 U.S.C. § 1232g; 34 CFR Part
99).
Community Risk: The Wisconsin component of the Influenza
Incidence Surveillance Project (W-IISP) is a long-standing, independent
influenza surveillance system that assesses MAI in and around
OSD23. The system has been in continuous operation
since October 2009 and is organized by the ORCHARDS research team.
W-IISP includes five primary care clinics, one of which is located in
the OSD and four that are located in communities surrounding OSD. The
clinics conduct active laboratory-supported surveillance for influenza
and other respiratory viruses in patients presenting with acute
respiratory illnesses. Weekly counts of laboratory-confirmed MAI served
as a proxy for underlying community influenza risk in this analysis.
Timing of regularly scheduled major school breaks : The winter
holiday (including Christmas and New Year’s Day) at the OSD is
relatively fixed in time, occurring in late December and early January,
extending between 10 and 16 days, including weekend days (Table 1). The
timing of spring break is more variable depending upon year but is fixed
in length at 9 days (including weekend days).
Pseudo-breaks as a control: We introduced into the analysis
“pseudo-breaks” of 9 days’ duration between winter and spring break
each year and starting five weeks before the spring break, when school
was actually in session, to support any findings of the true impact of
the planned breaks. The timing of regularly scheduled winter and spring
breaks, along with pseudo-breaks, is presented in Figure 2 against the
backdrop of statewide laboratory-confirmed influenza detections.
Statistical analysis: Analyses were performed on absentee data
from five consecutive academic school years (September 2, 2014, to June
12, 2019). The primary outcome measure was the number of a-ILI days in
the two weeks before and after the regularly scheduled school break.
Absenteeism due to ILI has been validated as an acceptable marker for
influenza through the home visit component in
ORCHARDS20.
The Cochran-Mantel-Haenszel (CMH) Test was used to measure the crude
association between a-ILI and the two-week period before versus after
each break period, stratified by school year. Exposure was defined as
whether the a-ILI count occurred during the two-week period before a
break (not exposed) or after a break (exposed). Cases were defined as
the sum of a-ILI counts for each pairing of school year and the two-week
time periods before versus after a break. The number of controls (not
absent due to ILI) in each of these two-week periods was defined as the
number of students enrolled in OSD for that school year, multiplied by
the number of school days in attendance during those two-week periods,
minus the number of a-ILI cases in that same period.
Generalized Linear Regression Models (GLM) were used to assess the
relationship between a-ILI counts and before- versus after-break
periods, while accounting for the community’s underlying influenza risk.
A Poisson distribution was assumed with the outcome of daily a-ILI
counts and the canonical natural log link function used. The natural log
of the OSD enrollment number for that school year was used as an offset
to account for varying enrollment numbers. Time from break was accounted
for within these two-week periods, measured in days. For the period
before a break, days were counted going backwards in time from the first
date of the break. For the period after a break, days were counted going
forward in time from the last date of the break (i.e., break periods are
“day 0”, school days before a break are negative days, and school days
after a break are positive days). Covariates used in this model included
community-level influenza risk, linear effect of time, quadratic effect
of time, indicator of before or after break, an interaction between the
break indicator and linear time effect, and an interaction between the
break indicator and the quadratic time effect.
The community-level influenza risk was represented as a weekly measure
of influenza risk in the community. This was calculated by summing the
number of MAI instances in the community data set for the first 7-day
period (week 1) before and after break, and the second 7-day period
(week 2) before and after break, for a total of 20 such calculations at
the community level (4 weeks calculated for each of the 5 school years
analyzed), for each break type (winter and spring). To assess if there
was an association between a-ILI and the period indicator of before and
after break, a 3 degree-of-freedom Likelihood Ratio Test (LRT) was
performed. The results from the GLM were compared to a null model
without the period indicator and its interactions to assess the
significance of comparing before versus after breaks.
The analysis approach above for winter and spring break was also applied
to the pseudo-break period that was introduced five weeks prior to each
spring break when school remained in session. This pseudo-break serves
as a control to assess for time as a potential confounder. The estimates
predicted by the model for the pseudo-break were compared to the results
from the winter and spring break analyses to help assess the true impact
of school breaks on a-ILI.