A comparison of self-report and health care provider data to assess surveillance definitions of influenza-like illness in outpatients.
Barbara, Angela M. ; Loeb, Mark ; Dolovich, Lisa 等
As part of most influenza surveillance systems, patients who meet
specific symptom criteria will have culture samples taken for laboratory
testing. (1,2) Several surveillance definitions of influenza-like
illness (ILI) have been proposed. (2-6) The Centers for Disease Control
and Prevention (CDC) in the United States defines ILI as the presence of
fever (temperature of 38 degrees Celsius or greater) and one of either
cough or sore throat or both, in the absence of a known cause other than
influenza. (7) Health Canada's Flu Watch uses a variant of the CDC
definition of ILI: fever and cough plus one or more of the
following--sore throat, arthralgia, myalgia, or prostration
(www.phac-aspc.gc.ca/fluwatch). Several studies have found that the
grouping of high fever and cough is the best predictor of influenza.
(8-11) What these ILI definitions have in common is the presence of
fever plus one or more symptoms of respiratory illness.
Data about influenza symptoms can be obtained from multiple
sources. For example, symptoms can be reported by multiple informants,
such as self-reports and health care providers; or by multiple methods,
such as symptom checklists and medical record data. In prior work, we
found that agreements between health record data and self-report varied
for respiratory-related symptoms. (12) Therefore, factors that might
influence the sensitivity, specificity and predictive values of ILI
include the actual surveillance definition, but also the data source
from which the symptom data contained in the ILI definition are taken.
The impact of these factors will be relevant to both public health
researchers and clinicians in determining choice of ILI definitions.
Most studies evaluating the surveillance definitions of influenza
have relied on physician or health record data. (3,8,13,14) Some also
included a patient survey following entry into the study and physician
examination or review of medical records. (15,16) Nicholson and
colleagues (1997) had weekly phone surveillance for symptoms and then
home visits for symptomatic patients. (17) Vaccine effectiveness studies
have also used clinical data, as well as self-report from research
participants. (18-20)
The goal of the current study was to assess the utility of two
sources of data in determining the surveillance definitions for ILI and
their association with laboratory-confirmed influenza. Using data from
the Hutterite Influenza Prevention Study, (18) we compared data
collected retrospectively from medical record extraction, similar data
collected prospectively from research participants, and the combined
data from both sources.
METHODS
Study design and population
Residents of 46 Hutterite colonies in the Canadian provinces of
Alberta (n=22), Saskatchewan (n=22) and Manitoba (n=2) participated in
influenza surveillance for a cluster randomized controlled trial (RCT)
to determine if the vaccination of healthy children and adolescents with
inactivated influenza vaccine would reduce laboratory-confirmed
influenza in other residents of these communities. There were 947
healthy children aged 36 months to 15 years who received either seasonal
influenza vaccine or hepatitis A vaccine and 2,326 other residents of
Hutterite communities who were followed to assess the indirect effect of
vaccinating the children. Full details of the trial are described
elsewhere. (18)
Participant reports of influenza-related symptoms
Study surveillance for influenza took place from December 28, 2008
to June 23, 2009. All participants in the Hutterite Influenza Prevention
Study recorded their influenza-related symptoms (fever, cough, runny
nose, sore throat, headache, sinus problems, muscle ache, fatigue, ear
ache, and chills) using daily diaries. Fever was defined as a
temperature [greater than or equal to] 38 degrees Celsius; each
participating family was given a thermometer to record temperatures.
Trained research nurses visited the Hutterite colonies twice per week to
check diary entries and interviewed individual participants (or parents,
in the case of infants) to confirm their reported symptoms, assess other
symptoms and collect information regarding outpatient visits made to
medical offices and hospital emergency departments for flu-like
symptoms; the latter information included physician name, health care
facility, and location.
Health care provider reports of influenza-related symptoms
For each reported medical visit, a one-page "Patient
Information Request" form was faxed to the medical facility asking
for patient record data regarding presenting symptoms, using the same
list of symptoms as on the participant study diaries. Clinicians were
blinded to the patient's self-reported symptoms. The institutional
review boards at McMaster University, University of Calgary, University
of Saskatchewan, and University of Manitoba approved the study. All
participants gave written consent to allow us to obtain health record
information if they visited a doctor or hospital with flu-related
symptoms during the 2008-2009 influenza season.
Faxes were sent to the physician offices or medical facilities
between March 2009 and September 2009. A response indicating that there
was "no visit" was followed up by (at least one) fax to an
alternative medical facility, based on feedback from the original
responder or geography. Data from the first medical visit reported by
participants and confirmed by the health care provider were included in
the analysis.
Laboratory confirmation of influenza
During the colony visits, research nurses took nasopharyngeal swab
samples from study participants who reported two or more symptoms or
physician-diagnosed otitis media. Specimens were submitted to the public
health laboratories in the respective provinces to be tested for
influenza by Polymerase Chain Reaction (PCR). Influenza was confirmed by
the detection of viral Ribonucleic Acid on the basis of reverse
transcriptase Real Time Polymerase Chain Reaction (RT-PCR) targeting
matrix gene for influenza A and nonstructural gene for influenza B. (18)
PCR has been demonstrated to be more sensitive to viral culture alone;
compared to direct immunofluorescence and cell culture assay, RT-PCR was
95% sensitive and 100% specific for detecting influenza. It is therefore
considered the "gold standard" for detecting influenza.
(15,16,21,22)
Statistical analyses
We calculated the frequency of individual symptoms using three
strategies: a) those reported in the medical records; b) those
self-reported by participants; c) those reported in either the medical
record OR by self-report (combined data sources). For each data
strategy, we used the Pearson chi-square statistic to test for
differences in the number of symptom reports between participants with
and without PCR-confirmed influenza. We then calculated the sensitivity,
specificity, positive predictive value (PPV) and negative predictive
value (NPV) for each symptom by data source, using PCR results as the
gold standard for diagnosis of influenza. We used univariate logistic
regression analysis to evaluate the association of each symptom with
PCR-positive influenza. Odds ratios were calculated to determine the
strength of association between symptom and PCR-confirmed influenza and
95% confidence intervals were calculated to estimate the precision of
each odds ratio. We then included those individual symptoms associated
with laboratory-confirmed influenza (where alpha = 0.05) in the ILI
definitions to be further analyzed.
We analyzed ILI case definitions according to the data source(s)
used to identify the combination of symptoms: a) those reported in the
medical records; b) those self-reported by participants; c) those
reported either in the medical record OR by self-report (combined data
sources). That is, each individual symptom within the ILI definition was
present either in the medical record OR self-reported by the
participant; e.g., a participant had fever and cough if fever was
reported by either self-report OR medical record and cough was reported
by either self-report OR medical record. We excluded combinations of
symptoms with less than 10 PCR-positive cases for each data strategy.
(23)
We used the Pearson chi-square statistic to test for differences in
number of cases for each ILI definition between PCR-positive and
PCR-negative participants, for each data strategy. The sensitivity,
specificity, predictive values, and odds ratios were calculated for ILI
definitions using the three data strategies; laboratory-confirmed
influenza was considered the gold standard. All analyses were conducted
using SPSS 16.0 (SPSS Inc., Chicago, IL).
RESULTS
Of the 3,273 participants in the Hutterite Influenza Study (Table
1), 252 (8%) reported at least one outpatient medical visit during the
influenza season and 176 visits (70%) were confirmed by medical record
information. Twenty-six participants did not meet the criteria for
collecting a swab sample (15 were asymptomatic and 11 reported one
symptom) and 8 participants were excluded from the analysis because swab
samples were collected after eight days of symptom reporting. Therefore,
of the 176 participants with both self-report and physician-recorded
data, 142 (81%) individuals were tested for influenza by PCR; this is
the sample included in the present analyses of ILI symptoms.
The sex and age distributions of all participants in the RCT and of
the sample included in the ILI analyses are shown in Table 1. The age
distributions differ; those included in the ILI analyses were younger.
The mean age was 26.0 years for all RCT participants and 22.1 years for
the 142 individuals included in the present analyses. Sixty-two (44%)
had been vaccinated against influenza. Reported symptoms were not
significantly different between the 62 individuals who were vaccinated
and the 80 individuals not vaccinated.
Of those included in the ILI analyses, 37 individuals (26%) were
PCR positive. Children and adolescents less than 16 years of age
accounted for 52% of the sample and 68% of PCR-confirmed cases of
influenza. PCR-positive cases were younger than PCR-negative cases
(mean, 17.0 versus 23.9 years, p=0.07). We found higher PCR-positivity
rates in children aged 7-15 years (59%) compared to younger children
(23%) and adults aged 23-49 years (18%). The influenza A virus was found
in 19 (51%) of the 37 influenza virus positive participants; influenza B
was found in 18 (49%) participants. Because we used data from
participants' first confirmed medical visits reported during the
influenza season, 117 swab samples (82%) were collected prior to the
introduction of the novel H1N1 pandemic influenza virus in Canada on
April 23, 2009.24 Therefore, only 4 (11%) of the 37 PCR-positive cases
were identified during the H1N1 pandemic.
Table 2 compares the proportions of reported symptoms by data
source between participants who tested positive and those who tested
negative for influenza. Compared to PCR-negative participants,
PCR-positive cases were significantly more likely to have fever
(regardless of the data source) and participant-reported muscle aches.
PCR-positive cases also had significantly more sore throat according to
data from combined sources. The symptoms that were unrelated to
PCR-positivity (runny nose, headache, sinus problems, and chills) and
those negatively related to PCR positivity (earache) were excluded from
subsequent analyses.
Table 3 presents the sensitivity and logistic regression analyses
for the five symptoms found to be related to PCR positivity. Cough had
the highest sensitivity for each data source (76-86%).
Physician-recorded fever had the highest PPV (56%) and odds ratio (8.9,
95% CI 3.81-20.58; p=0.0001). Based on these findings, we further
analyzed five surveillance definitions for ILI: fever and cough; fever
or cough; fever and sore throat; fever and (cough or sore throat), which
meets the CDC criteria; and fever and cough and (sore throat or muscle
aches or fatigue), which meets the Flu Watch criteria. Because of the
low prevalence among PCR-positive cases, we did not analyze the symptom
combinations of fever and fatigue (physician, n=4; participant, n=6;
combined, n=11) or fever and muscle aches (physician, n=3; participant,
n=4; combined, n=10).
Table 4 compares the prevalence of each surveillance definition,
according to each data strategy, between PCR-positive and PCR-negative
participants. PCR-positive individuals had significantly more ILI
according to each surveillance definition, regardless of data source.
Table 5 presents the sensitivity and logistic regression analyses for
the five surveillance definitions. Overall, two symptom
complexes--namely, fever and sore throat; fever and cough, and (sore
throat or muscle aches or fatigue)--based on combined data sources, had
odds ratios over 9.0 and PPVs over 60%. For each ILI definition, the PPV
was higher when based on medical record data. Medical record
documentation of fever or cough had the highest PPV overall (95%). The
case definition of fever or cough had the highest sensitivities (78-98%)
and the lowest specificities (22-67%).
DISCUSSION
In this study, we explored the implications of using two different
data sources independently and jointly as predictor variables to
evaluate surveillance definitions of ILI. As seen in other studies,
(3,9) cough alone had the highest sensitivity, regardless of data source
(76-86%). Unlike self-reported cough and self-reported sore throat,
physician-recorded cough and physician-recorded sore throat were not
more prevalent in PCR-positive subjects compared to PCR-negative
subjects. Cough and sore throat are non-specific symptoms. Participants
were prompted by our research nurses to report symptoms that could
potentially be related to respiratory illness. Physicians, who were
blind to participant responses and interested in participants'
overall health, including symptoms unrelated to influenza, would likely
record cough and sore throat regardless of etiology.
We found that PPVs for ILI based on medical records were the same
or higher than ILI based on self-report data for each surveillance
definition. This was consistent with previous findings by Govaert and
colleagues (16) that predictive values are higher in subpopulations who
consult a general practitioner for influenza symptoms. They found a PPV
of 30% for fever, cough and acute onset based on questionnaire data
compared to 40% for the sample symptom complex according to physician
records. (16) We found a more substantial difference in the medical
records (95%) compared to self-reports (31%) for fever or cough. Family
physicians, having clinical experience with patient consultations for
influenza, may be well placed to infer the significance of symptom
combinations. Indeed, physicians have been found to correctly diagnose
influenza infection in >60-70% of patients on the basis of clinical
symptoms alone. (25)
In contrast to other studies, (8,9,11) we found other symptom
combinations had PPVs that were similar or higher than fever and cough
for medical record data. With the exception of fever or cough, we found
low to modest PPVs (45-62%). It is important to note that PPV is
influenced by influenza prevalence; that is, PPV will improve with
greater circulation of the influenza virus. With lower prevalence rates
of influenza, it would be less likely that a person meeting the ILI
definition will have a PCR-positive test, resulting in lower PPV. (26)
Laboratory confirmation of influenza may have been influenced by
other factors, such as timing of swab sample collection compared to
onset of symptoms. Our study focused on influenza surveillance
definitions, which are based on signs and symptoms rather than other
factors that may give rise to the likelihood of influenza, such as
vaccination uptake, differences in exposure, genetic variation,
comorbidity or other biological factors.
Researchers and public health clinicians should consider the issue
of measurement error and reporting variations when designing studies.
The choice of data source(s) should correspond with the study question
or objective. For example, the data source used will have implications
for studies evaluating the effectiveness of influenza vaccination or
other interventions. Our findings indicate that using medical record
data to determine ILI, due to higher PPVs, will maximize the
effectiveness of an intervention. Using highly specific ILI definitions
will also result in higher estimates of vaccine effectiveness; whereas,
less specific ILI (and highly sensitive) definitions (such as fever or
cough using combined data) will result in lower estimates. (27,28) For
overall disease burden and use of health services, combined information
from both data sources may be more appropriate because of their higher
sensitivities. (5,29) To identify all potential cases of ILI, combining
symptom data from both medical records and self-report will also result
in a higher rate of detection.
Our study sample was made up of a disproportionate number of
younger children under the age of seven years (37%) and adults between
the ages of 23 and 49 years (27%). Compared to older persons, young
children have had higher rates of both seasonal influenza and the 2009
H1N1 pandemic influenza. (30,31) Because our study looked at outpatient
medical visits rather than hospitalizations, the age distribution of our
sample is not surprising. A US population-based surveillance study found
a high burden of influenza infection among outpatients under the age of
five years. (32)
Individuals over 65 years of age are considered to be at high risk
for developing complications of influenza and ILI. However, this age
group was under-represented in our sample. Other studies have found that
older persons reported less influenza symptoms and ILI. (33,34) The
Hutterite population is younger than the overall Canadian population: 5%
of Hutterites are 65 years of age or older in contrast to 13% of the
Canadian population. (35) Older people make up a small proportion of any
Hutterite colony population because of the high fertility rates and
large number of children found in all colonies. (36) A larger sample
size would have allowed us to stratify our analyses by age group and
other demographic characteristics.
A recognized limitation of this study is that it was conducted
within a specific cultural and religious population of outpatients
during a single influenza season. The results may not be generalizable
to all patient populations during other influenza seasons. Hutterites
perceive that good physical health is a gift from God and ill health is
a burden one must bear. (36) This may lead to less awareness of and/or
reluctance to report or complain about bodily symptoms, which may
explain the lower proportions of self-reported ILI and individual
symptoms. Another limitation is that, because we limited the analyses to
participants who had data from both sources, the sample size was modest,
resulting in odds ratios with wide confidence intervals. We are unable
to say how the above limitations have affected our results or the
direction of the bias.
Further research should explore data source use in determining ILI
surveillance definitions and their association with laboratory-confirmed
influenza in other populations with data collected over multiple
influenza seasons.
Acknowledgements: We thank Cassandra Howse, Dominik Mertz, Pardeep
Singh and Lorraine Moss for their feedback on earlier drafts of this
manuscript.
Conflict of Interest: None to declare.
REFERENCES
(1.) Public Health Agency of Canada. Flu watch. Available at:
http://www.phacaspc.gc.ca/fluwatch/ (Accessed June 16, 2010).
(2.) Thursky K, Cordova SP, Smith D, Kelly H. Working towards a
simple case definition for influenza surveillance. J Clin Virol
2003;27(2):170-79.
(3.) Babcock HM, Merz LR, Dubberke ER, Fraser VJ. Case-control
study of clinical features of influenza in hospitalized patients. Infect
Control Hosp Epidemiol 2008;29(10):921-26.
(4.) Aguilera JF, Paget WJ, Mosnier A, Heijnen ML, Uphoff H, van
der Velden J, et al. Heterogeneous case definitions used for the
surveillance of influenza in Europe. Eur J Epidemiol 2003;18(8):751-54.
(5.) Nichol KL, Mendelman P. Influence of clinical case definitions
with differing levels of sensitivity and specificity on estimates of the
relative and absolute health benefits of influenza vaccination among
healthy working adults and implications for economic analyses. Virus Res
2004;103(1-2):3-8.
(6.) Taylor JL, Dwyer DM, Coffman T, Groves C, Patel J, Israel E.
Nursing home outbreak of influenza A (H3N2): Evaluation of vaccine
efficacy and influenza case definitions. Infect Control Hosp Epidemiol
1992;13(2):93-97.
(7.) Centers for Disease Control and Prevention. Prevention and
control of influenza: Recommendation of the Advisory Committee on
Immunization Practices (ACIP). MMWR 2006;55(RR-10):1-42.
(8.) Boivin G, Hardy I, Tellier G, Maziade J. Predicting influenza
infections during epidemics with use of a clinical case definition. Clin
Infect Dis 2000;31(5):1166-69.
(9.) Monto AS, Gravenstein S, Elliott M, Colopy M, Schweinle J.
Clinical signs and symptoms predicting influenza infection. Arch Intern
Med 2000;160(21):3243-47.
(10.) Ohmit SE, Monto AS. Symptomatic predictors of influenza virus
positivity in children during the influenza season. Clin Infect Dis
2006;43(5):564-68.
(11.) Call SA, Vollenweider MA, Hornung CA, Simel DL, McKinney WP.
Does this patient have influenza? JAMA 2005;293(8):987-97.
(12.) Barbara AM, Loeb M, Dolovich L, Brazil K, Russell ML.
Agreement between self-report and medical records on signs and symptoms
of respiratory illness. Primary Care Resp J (in press).
(13.) Carrat F, Tachet A, Rouzioux C, Housset B, Valleron AJ.
Evaluation of clinical case definitions of influenza: Detailed
investigation of patients during the 1995-1996 epidemic in France. Clin
Infect Dis 1999;28(2):283-90.
(14.) Puzelli S, Valdarchi C, Ciotti M, Dorrucci M, Farchi F,
Babakir-Mina M, et al. Viral causes of influenza-like illness: Insight
from a study during the winters 2004-2007. J Med Virol
2009;81(12):2066-71.
(15.) van Elden LJ, van Essen GA, Boucher CA, van Loon AM, Nijhuis
M, Schipper P, et al. Clinical diagnosis of influenza virus infection:
Evaluation of diagnostic tools in general practice. Br J Gen Pract
2001;51(469):630-34.
(16.) Govaert TM, Dinant GJ, Aretz K, Knottnerus JA. The predictive
value of influenza symptomatology in elderly people. Fam Pract
1998;15(1):16-22.
(17.) Hayden FG, Osterhaus AD, Treanor JJ, Fleming DM, Aoki FY,
Nicholson KG, et al. Efficacy and safety of the neuraminidase inhibitor
zanamivir in the treatment of influenzavirus infections. GG167 Influenza
Study Group. N Engl J Med 1997;337(13):874-80.
(18.) Loeb M, Russell ML, Moss L, Fonseca K, Fox J, Earn DJ, et al.
Effect of influenza vaccination of children on infection rates in
Hutterite communities: A randomized trial. JAMA 2010;303(10):943-50.
(19.) Quach C, Newby D, Daoust G, Rubin E, McDonald J. QuickVue
influenza test for rapid detection of influenza A and B viruses in a
pediatric population. Clin Diagn Lab Immunol 2002;9(4):925-26.
(20.) Rodriguez WJ, Schwartz RH, Thorne MM. Evaluation of
diagnostic tests for influenza in a pediatric practice. Pediatr Infect
Dis J 2002;21(3):193-96.
(21.) Habib-Bein NF, Beckwith WH,3rd, Mayo D, Landry ML. Comparison
of SmartCycler real-time reverse transcription-PCR assay in a public
health laboratory with direct immunofluorescence and cell culture assays
in a medical center for detection of influenza A virus. J Clin Microbiol
2003;41(8):3597-601.
(22.) Schweiger B, Zadow I, Heckler R, Timm H, Pauli G. Application
of a fluorogenic PCR assay for typing and subtyping of influenza viruses
in respiratory samples. J Clin Microbiol 2000;38(4):1552-58.
(23.) Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A
simulation study of the number of events per variable in logistic
regression analysis. J Clin Epidemiol 1996;49(12):1373-79.
(24.) Cutler J, Schleihauf E, Hatchette TF, Billard B, Watson-Creed
G, Davidson R, et al. Investigation of the first cases of human-to-human
infection with the new swine-origin influenza A (H1N1) virus in Canada.
CMAJ 2009;181(3 4):159-63.
(25.) Mandell GL. Essential Atlas of Infectious Diseases, 2nd ed.
Philadelphia, PA: Current Medicine, 2001.
(26.) Altman DG, Bland JM. Diagnostic tests 2: Predictive values.
BMJ 1994;309(6947):102.
(27.) Orenstein EW, De Serres G, Haber MJ, Shay DK, Bridges CB,
Gargiullo P, et al. Methodologic issues regarding the use of three
observational study designs to assess influenza vaccine effectiveness.
Int J Epidemiol 2007;36(3):623-31.
(28.) Beyer WE. Heterogeneity of case definitions used in vaccine
effectiveness studies--and its impact on meta-analysis. Vaccine
2006;24(44-46):6602-4.
(29.) Nichol KL. Heterogeneity of influenza case definitions and
implications for interpreting and comparing study results. Vaccine
2006;24(44-46):6726-28.
(30.) Monto AS, Sullivan KM. Acute respiratory illness in the
community, frequency of illness and the agents involved. Epidemiol
Infect 1993;110(1):145 60.
(31.) National Advisory Committee on Immunization. Statement on
seasonal trivalent inactivated influenza vaccine (TIV) for 2010-2011.
Can Commun Dis Rep 2010;36(ACS-6):1-49.
(32.) Poehling KA, Edwards KM, Weinberg GA, Szilagyi P, Staat MA,
Iwane MK, et al. The underrecognized burden of influenza in young
children. N Engl J Med 2006;355(1):31-40.
(33.) Smolderen KG, Vingerhoets AJ, Croon MA, Denollet J.
Personality, psychological stress, and self-reported influenza
symptomatology. BMC Public Health 2007;7:339.
(34.) Thumin F, Wims E. The perception of the common cold, and
other ailments and discomforts, as related to age. Int J Aging Hum Dev
1975;6(1):43-49.
(35.) Statistics Canada. Religion (95A), Age Groups (7A) and Sex
(3) for Population, for Canada, Provinces, Territories, Census
Metropolitan Areas and Census Agglomerations, 1991 and 2001
Censuses--20% Sample Data, 2003 ed. Ottawa: Statistics Canada, 2003.
(36.) Brunt JH, Lindsey E, Hopkinson J. Health promotion in the
Hutterite community and the ethnocentricity of empowerment. Can J Nurs
Res 1997;29(1):17-28.
Received: March 1, 2011
Accepted: August 2, 2011
Angela M. Barbara, PhD, [1] Mark Loeb, MD, [1-3] Lisa Dolovich,
PharmD, [4,5] Kevin Brazil, PhD, [3,6] Margaret L. Russell, MD, PhD [7]
Author Affiliations
[1.] Department of Pathology and Molecular Medicine, McMaster
University, Hamilton, ON
[2.] Department of Medicine, McMaster University, Hamilton, ON
[3.] Department of Clinical Epidemiology and Biostatistics,
McMaster University, Hamilton, ON
[4.] Department of Family Medicine, McMaster University, Hamilton,
ON
[5.] Centre for Evaluation of Medicines, McMaster University,
Hamilton, ON
[6.] St. Joseph's Health System Research Network, Hamilton, ON
[7.] Department of Community Health Sciences, University of
Calgary, Calgary, AB
Correspondence: Angela Barbara, Infectious Diseases Research Unit,
Pathology & Molecular Medicine, McMaster University, 1280 Main
Street West, MDCL 3200, Hamilton, ON L8N 4K1, Tel: 905-525-9140, ext.
21478, Fax: 905-389-5822, E-mail: barbara@mcmaster.ca,
amgbarbara@gmail.com
Table 1. Descriptive Characteristics of All Participants in the
Hutterite Influenza Prevention Study (RCT), Participants With Medical
Record Data Who Provided Swab Specimens (ILI Analyses), PCR-positive
Influenza Cases and PCR-negative Influenza Participants
Characteristic RCT ILI Analyses
PCR Positive
All All Yes
n (%) * n (%) n (%)
Total 3273 142 37
Age group (years)
<7 533 (16.3) 52 (36.6) 12 (32.4)
7-15 835 (25.6) 22 (15.5) 13 (35.1)
16-22 390 (11.9) 11 (7.7) 3 (8.1)
23-49 1053 (32.2) 38 (26.8) 7 (18.9)
50-64 302 (9.2) 11 (7.7) 0
[greater than 160 (4.9) 8 (5.6) 2 (5.4)
or equal to] 65
Sex, female 1858 (56.8) 86 (60.6) 22 (59.5)
Characteristic ILI Analyses
PCR Positive p-value
([dagger])
No
n (%)
Total 105
Age group (years) 0.003
<7 40 (38.1)
7-15 9 (8.6)
16-22 8 (7.6)
23-49 31 (29.5)
50-64 11 (10.5)
[greater than 6 (5.7)
or equal to] 65
Sex, female 64 (61) 0.87
* Percentage per column.
([dagger]) p-value for Pearson chi square test for PCR-positive
participants compared with PCR-negative participants.
Table 2. Symptoms Experienced by PCR-positive Influenza Cases and
PCR-negative Influenza Participants, According to Each Data Strategy
Symptoms, All PCR Positive
Data Source n (% *) n (%)
All 142 37
Fever
Medical record 45 (31.7) 25 (67.6)
Participant 22 (15.5) 10 (27.0)
Combined ([dagger]) 51 (35.9) 26 (70.3)
Cough
Medical record 92 (64.8) 28 (75.7)
Participant 89 (62.7) 29 (78.4)
Combined 111 (78.2) 32 (86.5)
Sore throat
Medical record 65 (45.8) 19 (51.4)
Participant 55 (38.7) 19 (51.4)
Combined 89 (62.7) 30 (81.1)
Runny nose
Medical record 44 (31.0) 9 (24.3)
Participant 51 (35.9) 15 (40.5)
Combined 75 (52.8) 19 (51.4)
Headache
Medical record 16 (11.3) 4 (10.8)
Participant 23 (16.2) 6 (16.2)
Combined 32 (22.5) 8 (21.6)
Sinus problems
Medical record 26 (18.3) 5 (13.5)
Participant 25 (17.6) 5 (13.5)
Combined 38 (26.8) 7 (18.9)
Muscle aches
Medical record 9 (6.3) 3 (8.1)
Participant 15 (10.6) 8 (21.6)
Combined 20 (14.1) 9 (24.3)
Fatigue
Medical record 12 (8.5) 5 (13.5)
Participant 20 (14.1) 9 (24.3)
Combined 29 (20.4) 12 (32.0)
Earache
Medical record 26 (18.3) 2 (5.4)
Participant 17 (12.0) 2 (5.4)
Combined 30 (21.0) 3 (8.1)
Chills
Medical record 9 (6.3) 2 (5.4)
Participant 15 (10.6) 4 (10.8)
Combined 22 (15.5) 6 (16.2)
Symptoms, PCR Negative p-value
Data Source n (%) ([dagger])
All 105
Fever
Medical record 20 (19.5) <0.0001
Participant 12 (11.4) 0.02
Combined ([dagger]) 25 (23.8) <0.0001
Cough
Medical record 64 (60.9) 0.10
Participant 60 (57.1) 0.02
Combined 79 (75.2) 0.16
Sore throat
Medical record 46 (43.8) 0.43
Participant 36 (34.3) 0.07
Combined 59 (56.2) 0.01
Runny nose
Medical record 35 (33.3) 0.30
Participant 36 (34.3) 0.50
Combined 56 (53.3) 0.84
Headache
Medical record 12 (11.5) 0.92
Participant 17 (16.2) 1.00
Combined 24 (22.8) 0.88
Sinus problems
Medical record 21 (20.0) 0.35
Participant 20 (19.1) 0.45
Combined 31 (29.5) 0.18
Muscle aches
Medical record 6 (5.7) 0.61
Participant 7 (6.7) 0.05
Combined 11 (10.5) 0.08
Fatigue
Medical record 7 (6.7) 0.28
Participant 11 (10.5) 0.08
Combined 17 (16.0) 0.06
Earache
Medical record 24 (22.9) 0.002
Participant 15 (14.3) 0.09
Combined 27 (25.7) 0.006
Chills
Medical record 7 (6.7) 0.79
Participant 11 (10.5) 0.96
Combined 16 (15.2) 0.89
* Percentage of total in column.
([dagger]) p-value for Pearson chi square test for PCR-positive
participants compared with PCR-negative participants.
([double dagger]) Combined = symptom identified by medical record OR
self-report.
Table 3. Symptoms, as Reported by Data Source, Predicting Influenza
Symptoms,
Data Source n Sensitivity Specificity PPV NPV
Fever
Medical record 45 0.68 0.81 0.56 0.88
Participant 22 0.27 0.89 0.45 0.78
Combined * 51 0.70 0.76 0.51 0.88
Cough
Medical record 92 0.76 0.39 0.30 0.82
Participant 89 0.78 0.43 0.33 0.85
Combined 111 0.86 0.29 0.25 0.84
Sore throat
Medical record 65 0.51 0.56 0.29 0.77
Participant 55 0.51 0.66 0.35 0.79
Combined 89 0.81 0.44 0.34 0.87
Muscle aches
Medical record 9 0.08 0.94 0.33 0.74
Participant 15 0.22 0.93 0.53 0.77
Combined 20 0.24 0.90 0.45 0.77
Fatigue
Medical record 12 0.14 0.93 0.42 0.75
Participant 20 0.24 0.90 0.45 0.77
Combined 29 032 0.84 0.42 0.78
Logistic Regression
Symptoms,
Data Source Odds 95% p-value
Ratio Confidence
Intervals
Fever
Medical record 8.90 3.81-20.58 <0.0001
Participant 2.87 1.12-7.37 0.03
Combined * 7.56 3.28-17.45 <0.0001
Cough
Medical record 1.99 0.85-4.65 0.11
Participant 2.72 1.13-6.51 0.03
Combined 2.11 0.74-5.97 0.16
Sore throat
Medical record 1.35 0.64-2.87 0.43
Participant 2.02 0.95-4.33 0.07
Combined 3.34 1.35-8.29 0.01
Muscle aches
Medical record 1.46 0.35-6.14 0.61
Participant 3.86 1.29-11.55 0.02
Combined 2.75 1.03-7.30 0.04
Fatigue
Medical record 2.19 0.65-7.37 0.21
Participant 2.75 1.03-7.30 0.04
Combined 2.49 1.05-5.89 0.04
PPV = positive predictive value; NPV = negative predictive value.
* Combined = symptom identified by medical record OR self-report.
Table 4. Symptom Combinations, According to Data Source, of All
Participants, PCR-positive Influenza Cases and PCR-negative
Participants
Symptoms, All Influenza
Data Source n (%) Positive
n (%)
All 142 37
Fever and cough
Medical record 32 (22.5) 18 (48.6)
Participant 20 (14.1) 10 (27.0)
Combined data 45 (31.7) 23 (62.2)
([double dagger])
Fever or cough
Medical record 105 (73.9) 35 (94.6)
Participant 91 (64.1) 29 (78.4)
Combined data 117 (82.4) 35 (94.6)
([double dagger])
Fever and sore throat
Medical record 32 (22.5) 18 (48.6)
Participant 20 (14.1) 10 (27.0)
Combined data 45 (31.7) 23 (62.2)
Fever and (cough or
sore throat)
Medical record 38 (26.8) 22 (59.5)
Participant 22 (15.5) 10 (27.0)
Combined data 51 (35.9) 26 (70.3)
Fever and cough and
(sore throat or muscle
aches or fatigue)
Medical record 20 (14.1) 12 (32.4)
Participant 15 (10.6) 9 (24.3)
Combined data 34 (24) 21 (56.8)
Symptoms, Influenza p-value
Data Source Negative ([dagger])
n (%)
All 105
Fever and cough
Medical record 14 (13.3) <0.0001
Participant 10 (9.5) 0.01
Combined data 22 (21.0) <0.0001
([double dagger])
Fever or cough
Medical record 70 (0.67) 0.001
Participant 62 (59.0) 0.04
Combined data 23 (21.9) 0.02
([double dagger])
Fever and sore throat
Medical record 14 (13.3) <0.0001
Participant 10 (9.5) 0.01
Combined data 22 (21.0) <0.0001
Fever and (cough or
sore throat)
Medical record 16 (15.2) <0.0001
Participant 12 (11.4) 0.02
Combined data 25 (23.8) <0.0001
Fever and cough and
(sore throat or muscle
aches or fatigue)
Medical record 8 (7.6) <0.0001
Participant 6 (5.7) 0.002
Combined data 13 (12.4) <0.0001
* Percentage of total per row.
([dagger]) p-value for Pearson chi square test for PCR-positive
participants compared with PCR-negative participants.
([double dagger]) Combined data = individual symptoms within the ILI
case definition identified by medical record OR self-report.
Table 5. Surveillance Definitions, as Reported by Data Source,
Predicting Influenza
Symptoms,
Data Source n Sensitivity Specificity PPV
Fever and cough
Medical record 32 0.49 0.87 0.56
Participant 20 0.27 0.90 0.50
Combined data * 45 0.62 0.79 0.51
Fever or cough
Medical record 105 0.95 0.67 0.95
Participant 91 0.78 0.59 0.31
Combined data 117 0.95 0.22 0.30
Fever and sore throat
Medical record 21 0.35 0.94 0.62
Participant 14 0.22 0.95 0.57
Combined data 36 0.59 0.87 0.61
Fever and (cough or
sore throat)
Medical record 38 0.59 0.85 0.58
Participant 22 0.27 0.89 0.45
Combined data 51 0.70 0.76 0.51
Fever and cough and
(sore throat or muscle
aches or fatigue)
Medical record 20 0.32 0.92 0.60
Participant 15 0.24 0.94 0.60
Combined data 34 0.57 0.88 0.62
Logistic Regression
Symptoms,
Data Source NPV Odds 95% p-value
Ratio Confidence
Intervals
Fever and cough
Medical record 0.83 6.16 2.61-14.49 <0.0001
Participant 0.78 3.52 1.33-9.33 0.01
Combined data * 0.86 6.20 2.75-13.99 <0.0001
Fever or cough
Medical record 0.67 8.75 1.99-38.50 0.004
Participant 0.84 2.51 1.05-6.03 0.039
Combined data 0.92 4.91 1.10-21.96 0.037
Fever and sore throat
Medical record 0.80 6.57 2.45-17.63 <0.0001
Participant 0.77 4.55 1.46-14.18 0.01
Combined data 0.86 9.53 4.01-22.63 <0.0001
Fever and (cough or
sore throat)
Medical record 0.86 8.16 3.51-18.99 <0.0001
Participant 0.78 2.87 1.12-7.37 0.03
Combined data 0.88 7.56 3.28-17.45 <0.0001
Fever and cough and
(sore throat or muscle
aches or fatigue)
Medical record 0.80 5.82 2.15-15.77 0.001
Participant 0.78 5.30 1.74-16.17 0.003
Combined data 0.85 9.29 3.88-22.21 <0.0001
* Combined data = individual symptoms within the ILI case definition
identified by medical record OR self-report.
