Syndromic surveillance for local outbreaks of lower-respiratory infections: would it work?

Abstract

Although syndromic surveillance is increasingly used to detect unusual illness, there is a debate whether it is useful for detecting local outbreaks. We evaluated whether syndromic surveillance detects local outbreaks of lower-respiratory infections (LRIs) without swamping true signals by false alarms.
Using retrospective hospitalization data, we simulated prospective surveillance for LRI-elevations. Between 1999-2006, a total of 290762 LRIs were included by date of hospitalization and patients place of residence (>80% coverage, 16 million population). Two large outbreaks of Legionnaires disease in the Netherlands were used as positive controls to test whether these outbreaks could have been detected as local LRI elevations. We used a space-time permutation scan statistic to detect LRI clusters. We evaluated how many LRI-clusters were detected in 1999-2006 and assessed likely causes for the cluster-signals by looking for significantly higher proportions of specific hospital discharge diagnoses (e.g. Legionnaires disease) and overlap with regional influenza elevations. We also evaluated whether the number of space-time signals can be reduced by restricting the scan statistic in space or time. In 1999-2006 the scan-statistic detected 35 local LRI clusters, representing on average 5 clusters per year. The known Legionnaires' disease outbreaks in 1999 and 2006 were detected as LRI-clusters, since cluster-signals were generated with an increased proportion of Legionnaires disease patients (p:<0.0001). 21 other clusters coincided with local influenza and/or respiratory syncytial virus activity, and 1 cluster appeared to be a data artifact. For 11 clusters no likely cause was defined, some possibly representing as yet undetected LRI-outbreaks. With restrictions on time and spatial windows the scan statistic still detected the Legionnaires' disease outbreaks, without loss of timeliness and with less signals generated in time (up to 42% decline).
To our knowledge this is the first study that systematically evaluates the performance of space-time syndromic surveillance with nationwide high coverage data over a longer period. The results show that syndromic surveillance can detect local LRI-outbreaks in a timely manner, independent of laboratory-based outbreak detection. Furthermore, since comparatively few new clusters per year were observed that would prompt investigation, syndromic hospital-surveillance could be a valuable tool for detection of local LRI-outbreaks.

Full text

Page 1

DetectionofExcessInfluenzaSeverity:AssociatingRespiratory
HospitalizationandMortalityDataWithReportsofInfluenza-Like
IllnessbyPrimaryCarePhysicians
Cees C. van den Wijngaard, MSc, Liselotte van Asten, PhD, Adam Meijer, PhD, Wilfrid van Pelt, PhD, Nico J.D. Nagelkerke, PhD, Ge ´ A. Donker, PhD,
Marianne A.B. van der Sande, PhD, and Marion P.G. Koopmans, DVM, PhD
Syndromic surveillance is increasingly used to
monitor symptoms or clinical diagnoses such as
shortness of breathor pneumonia as indicators of
infectiousdisease.The primaryobjectiveofmany
syndromic surveillance systems is the detection
ofunexpecteddiseaseincreasessuchasthosethat
occur as a result of bioterrorism attacks or out-
breaks of emerging diseases such as severe acute
respiratory syndrome (SARS). However, the sig-
nals generated by such syndromic surveillance
also reflect influenza activity.1–4
Worldwide, influenza continues to result in
serious morbidity and mortality.5,6The recur-
rence of influenza epidemics is predominantly
caused by both the antigenic drift of influenza
viruses and changes in the dominant virus types
or subtypes. Antigenic drift occurs during the
replication process of influenza viruses when
mutations in surface proteins lead to declines in
the level of immunity acquired through natural
infection or vaccination.7In addition, the annual
variations in dominant virus types or subtypes,
such as A(H1), A(H3), and B, can lead to dif-
ferences in influenza-related morbidity and
mortality. For example, in recent decades levels
of morbidity and mortality seem to have been
lower in the influenza A(H1) and B epidemic
seasons than in the A(H3) seasons.8,9
In the Netherlands, as in many countries,
surveillance of influenza is conducted by a net-
work of sentinel general practitioners. Influenza-
like illness (ILI) consultations are reported
weekly, and antigenic properties of isolated
viruses are analyzed to determine their effects
on annual ILI fluctuations.10,1 1Such sentinel
surveillance is considered adequate for moni-
toring the onset and magnitude of annual in-
fluenza epidemics. However, it is not sufficient
for monitoring the incidence of severe influenza
infections leading to hospitalization or death.
Although the relationship between the viru-
lence and transmission capacity of influenza
viruses is still incompletely understood,7varia-
tions in virulence may result in disproportionate
increases in severe illness relative to increases in
the number of patients with ILI consulting their
general practitioners. Such increases might be
captured by monitoring temporal changes in the
association of ILI data obtained from general
practitioners (hereafter GP–ILI data) with hospi-
talization and mortality surveillance data. Such
monitoring is not a part of current global in-
fluenza monitoring activities, although in some
countries ILI data in addition to hospitalization
and mortality data are included in influenza
surveillance.12,13
We explored the potential of this monitoring
strategy to detect excesses in influenza infection
severity by investigating shifts in the annual
association of respiratory hospitalizations and
mortality with GP–ILI incidence data in the
Netherlands between 1999 and 2005. In
addition, we evaluated whether such shifts were
associated with reported circulation of influenza
virus drift variants, mismatches with vaccine
strains, or changes in dominant circulating virus
types or subtypes.
METHODS
We obtained hospital and mortality data
from the Dutch national medical register (99%
coverage of discharge and secondary diagnoses
by date of hospitalization) and the Dutch
causes of death registry (100% coverage of
primary causes of death, as well as complicating
causes and other additional causes of death,
by date of death). We formulated respiratory
hospitalization and mortality syndrome defini-
tions guided by the syndrome definitions of the
Centers for Disease Control and Prevention,
as coded in the International Classification of
Objectives. We explored whether excesses in influenza severity can be
detected by combining respiratory syndromic hospital and mortality data with
data on influenza-like illness (ILI) cases obtained from general practitioners.
Methods. To identify excesses in the severity of influenza infections in the
population of the Netherlands between 1999 and 2005, we looked for increases in
influenza-associated hospitalizations and mortality that were disproportionate to
the number of ILI cases reported by general practitioners. We used generalized
estimating equation regression models to associate syndromic hospital and
mortality data with ILI surveillance data obtained from general practitioners. Virus
isolation and antigenic characterization data were used to interpret the results.
Results. Disproportionate increases in hospitalizations and mortality (relative
to ILI cases reported by general practitioners) were identified in 2003/04 during
the A/Fujian/411/02 (H3N2) drift variant epidemic.
Conclusions. Combined surveillance of respiratory hospitalizations and mor-
tality and ILI data obtained from general practitioners can capture increases in
severe influenza-associated illness that are disproportionate to influenza in-
cidence rates. Therefore, this novel approach should complement traditional
seasonal and pandemic influenza surveillance in efforts to detect increases in
influenza case fatality rates and percentages of patients hospitalized. (Am J
Public Health. Published online ahead of print September 23, 2010: e1–e7. doi:10.
2105/AJPH.2009.168245)
RESEARCH AND PRACTICE
Published online ahead of print September 23, 2010 | American Journal of Public Health
van den Wijngaard et al. | Peer Reviewed | Research and Practice | e1

http://ajph.aphapublications.org/cgi/doi/10.2105/AJPH.2009.168245The latest version is at
Published Ahead of Print on September 23, 2010, as 10.2105/AJPH.2009.168245

Page 2

Diseases, Ninth Revision, Clinical Modification
(ICD-9-CM; see Appendix A, available as an
online supplement to this article at http://
www.ajph.org).14It has been demonstrated that
these respiratory hospitalization and mortality
syndrome data reflect respiratory pathogen ac-
tivity as measured via laboratory counts.4
We collected ILI data from a sentinel net-
work of general practitioners.10Data on in-
fluenza viruses detected in the Netherlands
between1999 and 2005 were derived from the
Dutch influenza surveillance consortium (com-
prising the National Influenza Centre and the
Netherlands Institute for Health Services Re-
search).1 1,15
We used weekly counts of various respira-
tory pathogens to adjust for the effects on
respiratory hospitalizations and mortality of
pathogen activity other than influenza. We
collected data on respiratory syncytial virus
(RSV), rhinovirus, Mycoplasma pneumoniae,
parainfluenza virus, enterovirus, and adenovi-
rus pathogen counts from a routine laboratory
surveillance system (the Weekly Sentinel Sur-
veillance System of the Dutch Working Group
on Clinical Virology, which covers 38%–73%
of the population of the Netherlands).16We
also used national mandatory notifications to
obtain weekly pertussis counts.
Data Analysis
With the exception of laboratory pathogen
counts (for which data on age were not avail-
able), we aggregated data by week and age
category (0–4 years, 5–19 years, 20–64 years,
65 years or older). In our analyses, we ex-
cluded respiratory mortality among those in
the 0- to 4-year and 5- to 19-year age groups
because of the sporadic counts in these groups.
SAS version 9.1 (SAS Institute Inc, Cary, NC)
was used in conducting the analyses.
For the general practitioner, hospital, and
mortality data, we calculated incidence rates
instead of counts to quantify risk differences
between age categories and to correct for
changes in the age distribution of the popula-
tion and (for the general practitioner sentinel
data) changing registry coverage over time.
After plotting time series of GP–ILI, respiratory
hospitalization, and mortality incidence data,
we looked for increases in hospitalizations and
mortality that seemed disproportionate to in-
creases in ILI cases as a measure of severity of
illness. We also examined time series of re-
spiratory pathogens other than influenza to
assess whether elevations in respiratory hospi-
talizations or mortality might be associated
with other pathogen activity (measured via
routine laboratory surveillance).
We used additive generalized estimating
equation (GEE)17models with a Poisson error
distribution to detect elevations in respiratory
hospitalizations and mortality that were dispro-
portionatetoseasonalrisesinILIincidenceinthe
general practitioner sentinel data. We estimated
hospitalization and mortality time series, strati-
fied by age, according to lagged ILI incidence.
We then used the 95% upper limits of the
models (details on the model variables are pro-
vided in Appendix B, available as an online
supplement to this article at http://www.ajph.
org) to determine distinct episodes in time in
which hospitalizations and mortality increases
were disproportionate to average modeled asso-
ciations with ILI incidence rates.
To adjust for the activity of respiratory
pathogens other than influenza, we considered
RSV, rhinovirus, parainfluenza virus, M pneu-
moniae, adenovirus enterovirus, and pertussis
counts for inclusion in the models as well. We
also assumed a constant basic syndrome level
attributable to factors other than respiratory
pathogen activity. In the summer months,
however, as the basic syndrome level appeared
to be lower with respect to hospitalizations
(possibly as a result of fewer planned hospital-
izations during that period), we used a lower
basic syndrome level (by including a dummy
variable for ‘‘summer’’). The regression model
coefficients for each of the lagged pathogens
and for the GP–ILI incidence data were as-
sumed to be constant in time.
We initially built a generalized linear model
with a Poisson error distribution and an iden-
tity link. To do so we used a forward stepwise
regression approach, selecting the lagged ILI
incidence and lagged pathogen counts that
contributed most to the model fit (5-week lags
were used; e.g., in step1, ILI was included with
a 1-week lag if that exhibited a better model
fit than all other pathogen–lag combinations,
assessed with Akaike’s information criterion18).
We included lagged GP–ILI incidence and the
counts for each pathogen in the model only once
and only if results were significant at the P£.05
level.
We analyzed age-stratified hospitalization,
mortality, and GP–ILI incidence data in the
regression models. Age stratification was not
possible for pathogen counts. We excluded
negative associations for pathogen counts to
avoid spurious model fits due to biologically
implausible associations (e.g., negative associa-
tions between enterovirus, which peaks in
summer, and respiratory syndromes, which
peak in winter). Also, we added seasonal vari-
ables (sine and cosine terms)—guided by
periodograms of the model residuals, which
reflect the importance of specific cyclical pe-
riods (e.g., 26 weeks, 52 weeks) in explaining
the variance in the residuals—to correct for
seasonal variation and used GEEs to correct
the model outcomes for autocorrelation be-
tween observations.
To quantify temporal heterogeneity in asso-
ciations of GP–ILI data with data on hospitali-
zations and mortality respiratory syndromes, we
modified the models by using time-dependent
(by epidemic year, defined as July 1 through
June 30) ILI regression coefficients (instead of
a single regression coefficient for all years).
These annual ILI regression coefficients can be
seen as scaling factors for the number of
hospitalizations or deaths associated with
a one-case increase in ILI incidence per10000
population. We plotted estimates for these
coefficients on a bar chart. The years with the
highest estimated ILI regression coefficients
were considered as those associated with the
most severe illness per ILI case. We conducted
F tests (with a null hypothesis of no differences
in associations over the study period) to de-
termine whether the coefficients differed across
the years of the study (Appendix B, available as
an online supplement to this article at http://
www.ajph.org).
Influenza Virus Isolation and Antigenic
Characterization
To assess whether disproportionate levels of
respiratory hospitalizations and mortality (rel-
ative to GP–ILI incidence rates) might be
related to the circulation of specific influenza
virus variants or subtypes, we explored weekly
reports of influenza virus subtypes A(H1),
A(H3), and B and assessed, on the basis of
antigenic characterization, which influenza
virus drift strains were present in the Nether-
lands during 1999–2005.
RESEARCH AND PRACTICE
e2 | Research and Practice | Peer Reviewed | van den Wijngaard et al.
American Journal of Public Health | Published online ahead of print September 23, 2010

Page 3

We also evaluated to what extent these drift
strains were reported to match or not match
the vaccine strains for those years. Individuals
at increased risk for complications of influenza
(elderly people and those with specific comor-
bid conditions) are offered annual influenza
vaccination19(during the study period, vaccine
coverage levels in the Netherlands were above
65% for individuals aged 65 years or older and
approximately80%forthoseagedolderthan75
years). Data on antigenic characteristics and the
match between vaccines and circulating viruses
were derived from annual influenza surveillance
reports.1 1To assess other possible explanations
for disproportionate levels of respiratory hospi-
talizations and mortality relative to GP–ILI in-
cidence rates, we also compared the analysis
results against plotted time series of specific
morbidity patterns associated with respiratory
hospitalizations, as measured via ICD-9-coded
hospital diagnosis incidence rates.
RESULTS
Plots of GP–ILI time series and respiratory
hospitalization and mortality time series
showed approximately concurrent peaks in all
winter seasons. The highest peaks were ob-
served in 1999/00 and 2004/05 (data not
shown). The influenza epidemics in 2000/01,
2001/02, and 2002/03 were relatively mild.
When data on respiratory pathogen counts
(other than influenza) were plotted (data not
shown), RSV showed the clearest winter peaks,
concurrent with elevations in respiratory hos-
pitalizations and mortality.
Therefore, we plotted respiratory hospitali-
zations and mortality against GP–ILI incidence
rates and laboratory RSV counts stratified by
age (Figure 1). Elevations in respiratory hospi-
talizations were highest in the youngest and
oldest age groups (0–4 years and 65 years or
above), and elevations in respiratory mortality
were highest in the oldest age group. Hospi-
talizations in the 0–4-year age group corre-
sponded more with the RSV time series than
with the ILI time series (Figure 1). During the
2003/04 winter season, steep peaks in re-
spiratory hospitalizations were observed
among those aged 5 to 19 years and those 65
years or older, and (although RSV counts
peaked at the same time) these peaks seemed
disproportionately high relative to the ILI time
series during that season (Figure 1). This trend
was also observed for mortality among in-
dividuals 65 years or older (Figure1, indicated
by ellipse).
Note. RSV=respiratory syncytial virus; ILI=influenza-like illness. No mortality time series were plotted for the 0–4-year and 5–19-year age groups because of their low numbers. RSV counts were
plotted for all age groups (because no age data were available), and the counts were scaled to fit the graph.
FIGURE 1—Respiratory hospitalizations and mortality incidences versus ILI incidence rates and RSV laboratory counts, by age group (a) 0–4
years, (b) 5–19 years, (c) 20–64 years, and (d) 65 years or older: the Netherlands, 1999–2005.
RESEARCH AND PRACTICE
Published online ahead of print September 23, 2010 | American Journal of Public Health
van den Wijngaard et al. | Peer Reviewed | Research and Practice | e3

Page 4

Regression Analysis
In the models with constant ILI regression
coefficients over the entire study period,
variations in respiratory hospitalizations and
mortality among individuals 65 years or
older and variations in hospitalizations
among those aged 0 to 4 years were ex-
plained quite well by variations in ILI in-
cidence and respiratory pathogen counts.
The explained variance was lower for the
other age groups.
Periodograms of the model residuals
showed sharp peaks at 1 year along with
smaller harmonics for shorter periods. We
therefore added sine and cosine terms to the
models to adjust for seasonal trends, and we
used GEEs to correct for autocorrelation in
the residuals. These model refinements led to
only minimal changes in the ILI regression
coefficients and the explained variance of
the models; percentages of explained variance
for hospitalizations were 95% among those
aged 0 to 4 years, 47% among those aged 5 to
19 years, 68% among those aged 20 to 64
years, and 78% among those 65 years or
older. Percentages of explained variance for
mortality were 37% among those aged 20 to
64 years and 76% among those 65 years or
older.
With respect to periods of peak influenza
activity (as measured by peaks in GP–ILI
incidence concurrent with peaks in the
counts of influenza isolates), the time series
of actual hospitalizations among both those
aged 0 to 4 years (data not shown) and those
65 years or older (Figure 2) most clearly
exceeded the 95% upper limit of the models
during winter 2003/04. A subsequent (F-
test) analysis of the model in which year-
specific ILI regression coefficients were used
showed significant annual heterogeneity in
these coefficients for all age categories
(P£.001).
Figure 3a shows the annual GP–ILI re-
gression coefficients for respiratory hospital-
izations. For example, the regression coeffi-
cient value of 3.94 for hospitalizations in the
0- to 4-year age group in 2003/04 indicates
that, for a hypothetical ILI incidence of 100
per 10000 population, the estimated respi-
ratory hospitalization incidence for that age
group is 3.94 (per 10000 population). The
annual GP–ILI regression coefficients for re-
spiratory hospitalizations were highest among
those 65 years or older and those aged 0 to 4
years. In addition, the regression coefficients
for these age groups were significantly higher
in 2003/04 than in any other study year
(P£.001).
Figure 3 (panel b) shows that, as expected,
the mortality regression coefficient was much
higher for those 65 years or older than for
those aged 20 to 64 years. Similar to the data
for hospitalizations, the ILI regression coeffi-
cient for those 65 years or older was clearly
higher in 2003/04 than in any other study
year (P£.03). In 2000/01, some of the esti-
mated ILI regression coefficients were below
zero, reflecting the mild influenza impact in
that season.
Influenza Virus Isolation and Antigenic
Characterization
Figure 2 presents data on influenza virus
subtypes and reported introductions of drift
variants.1 1,20–22All reported influenza drift
strains mismatched to some extent with the
vaccine strains observed over the study period
with the exception of the Caledonia/20/99
(H1N1) strain in 2000/01.
Specific Hospital Diagnoses
In visually exploring respiratory hospital-
ization discharges and diagnoses (data not
shown), we focused on elevations in time that
may have been related to the excess number
of respiratory hospitalizations observed in
2003/04. The elevations in hospitalizations
involving a diagnosis of pneumococcal
pneumonia (ICD-9 code 481) or pneumonia
due to streptococcus (ICD-9 code 4823)
during peak winter influenza activity in
2003/04 and, to a lesser extent, 2002/03
were among the highest observed in the
study period (in 2002/03, as a percentage of
respiratory hospitalizations overall, pneu-
mococcal pneumonia showed the highest
elevation over the study period). The second
highest elevation in hospitalizations involv-
ing an influenza diagnosis (ICD-9 codes
4870 and 4871) was observed in 2003/04
(the highest elevation was in 1999/00; no
significant elevations were observed in
2002/03).
Note. The hospitalization incidence for individuals aged 65 years or older is plotted in a line graph with the predicted value
and the 95% upper limit (Appendix B, available as an online supplement at http://www.ajph.org). Values exceeding the
model’s upper limit are indicated by the ellipse. Below the line graphs, the counts of influenza isolates by subtype—A(H3),
A(H1), and B—are presented as bars on the x-axis, and reports of drift variants are indicated.
aBecause we used generalized estimating equation models, confidence intervals for prediction were not available.
FIGURE 2—Respiratory hospitalization incidence explained by influenza-like illness
incidence versus influenza virus subtype counts and reports of drift variants: the
Netherlands, 1999–2005.
RESEARCH AND PRACTICE
e4 | Research and Practice | Peer Reviewed | van den Wijngaard et al.
American Journal of Public Health | Published online ahead of print September 23, 2010

Page 5

Note. The horizontal line in each chart gives the value of an ILI regression coefficient that is constant in time, as an indication of the average value of the ILI regression coefficients over the study
period. The 95% confidence intervals for the regression coefficients are presented in the figure as well.
FIGURE 3—Annual (July 1–June 30) estimates of the association of influenza-like illness (ILI) incidence with (a) respiratory hospitalization
incidence in all age groups and (b) respiratory mortality incidence for individuals aged 20–64 years and individuals 65 years or older: the
Netherlands, 1999–2005.
RESEARCH AND PRACTICE
Published online ahead of print September 23, 2010 | American Journal of Public Health
van den Wijngaard et al. | Peer Reviewed | Research and Practice | e5

Page 6

DISCUSSION
We observed increases in severe illness due
to influenza in the Netherlands between 1999
and 2005 that were disproportionate to ILI
incidence rates. Our observations reveal the
existence of temporal heterogeneities in the
severity of influenza infections, possibly stem-
ming from variations in the virulence of circu-
lating influenza viruses. Several studies have
shown that syndromic data on general respi-
ratory symptoms and clinical diagnoses can be
useful in influenza surveillance.2,3,23–25We
combined respiratory syndrome data on hospi-
talizations and mortality with traditional ILI
surveillance data obtained from general practi-
tioners to determine year-to-year differences in
the number of respiratory hospitalizations and
deaths in proportion to the number of ILI cases.
We linked our observations to virological
changes by visually exploring time series of
influenza subtype counts and reported anti-
genic information about influenza virus strains.
Five drift variants were reported in the
period under study—2 A(H3) variants, 1A(H1)
variant, and 2 B variants (Figure 2)—but only
in 2003/04, in the case of A/Fujian/411/
02(H3N2),20did this reporting of a drift variant
concur with disproportionate levels of hospitali-
zations and mortality.
Although at first glance these results seem to
suggest that it is difficult to predict clinical
effects from virological data, a more thorough
look at our virological findings explains the
absence of excess effects in years other than
2003/04. The relatively low hospitalization
and mortality levels in comparison with ILI
incidence rates in 2000/01 and 2001/02 can
be explained by the relative lack of fitness of
the A/New Caledonia/20/99(H1N1) and B/
Victoria/2/87 variants (respectively), as mor-
bidity and mortality levels tend to be lower in
seasons with predominantly A(H1)21or B22
strains than in A(H3) seasons.8,9
In addition, the A/New Caledonia/20/
99(H1N1) drift variant reported during 2000/
01 had emerged in 1999/00, and the vaccine
for the 2000/01 season contained this strain
and probably provided optimal protection
against this drift variant, thereby reducing
severe illness in elderly people who had been
vaccinated.21In 2004/05, influenza A(H3) and
influenza B drift strains were reported,1 1but their
impact was only moderate. During this season,
the antigenic distance of the dominant A/Cal-
ifornia/7/04-like(H3N2) drift variant virus to-
ward the influenza A(H3N2) virus in 2003/04
(A/Fujian/411/02-like) was relatively small.1 1
This was not the case for A/Fujian/411/02-
like(H3N2) viruses in 2003/04, which were
quite distinct from preceding A(H3N2) vi-
ruses,1 1thereby representing a likely explanation
for the observed excess hospitalizations and
mortality in that flu season. The high influenza
impact among young children and the elderly,
relative to the limited size of the 2003/04
epidemic measured according to GP-ILI data,
seems to be consistent with the high hospitaliza-
tion rates during the 2003 influenza season in
New Zealand in combination with the limited
size of the epidemic also according to GP-ILI
data.26There A/Fujian/411/02(H3N2) was the
dominant subtype as well.
Some other European countries reported
dominant activity or more severe outbreaks
of A/Fujian/411/02(H3N2) in 2002/03, but
there was great variation across Europe in
circulating strains during that winter.27,28In the
Netherlands, A/Fujian/411/02(H3N2) strains
werealsocirculatinginthatperiod,buttheywere
isolated only sporadically; 5 isolates were ob-
served, accounting for 4% of A(H3N2) isolates
overall.29
The introduction of new influenza drift vari-
ants and shifts in influenza subtypes are not the
only possible explanations for the observed
differences in influenza impact. Other viral
factors (e.g., viral replication capacity, virulence,
viral transmissibility) and climatic factors (e.g.,
temperature and relative humidity) may likewise
influence the impact of seasonal influenza on
morbidity and mortality. For instance, studies
have suggested that the antigenic drift of the
A(H3N2) viruses reported in 2003/04 resulted
in declines in the level of population immunity
(leading to A/Fujian/411/02 in 2002) but that
this drift variant became widespread only after
gaining a higher viral replication capacity
through additional reassortment-related changes
in internal genes.30–32
Limitations
A limitation of this study is that it was based
on associating time series of hospitalizations and
mortality with ILI data, and such associations
could be confounded by seasonally circulating
pathogensotherthaninfluenza.Tominimizethis
possibility, we adjusted for the possible impact of
RSV and other respiratory pathogens by in-
cluding them in our regression models. We also
included seasonal terms to correct for possible
confounding by other seasonally varying factors.
Our use of autocorrelation in our models cor-
rected for other, possibly transient causes of
hospitalization and mortality.
Another observation lent additional support
for the association of influenza with excess
elevations. That is, in 2003/04, concurrent
with a moderately high ILI peak, hospitaliza-
tions involving an influenza diagnosis exhibited
the second highest elevation over the study,
and hospital diagnoses of pneumococcal
pneumonia showed a high elevation as well
(which seems to be in line with observations
that influenza infections may predispose pa-
tients for S pneumoniae infections33–35).
Also, to enhance prospective surveillance,
there is a need to further evaluate how in-
creases in hospitalizations and mortality that
are disproportionate to ILI incidence rates can
be detected on a timely basis within a particular
influenza season. Quality control chart ap-
proaches36,37might be developed for the timely
detection of such temporal changes that require
the attention of health authorities.
Conclusions
Our results show that increases in severe
influenza-associated illness that are dispropor-
tionate to the incidence of influenza in the
community can be detected through combined
analyses of GP–ILI data and data on respira-
tory hospitalizations and mortality. This novel
approach should be implemented in global
influenza surveillance programs to provide
better estimates of increases in severe mor-
bidity and mortality due to influenza infections.
Our data also show that there is a possible
relationship between influenza impact and
specific influenza strains. Further research is
needed to better understand the causes of such
relationships. It seems worthwhile to develop
prospective respiratory syndromic surveillance
of hospitalizations and mortality complement-
ing traditional seasonal and pandemic influ-
enza surveillance to allowdetection of increases
in influenza case fatalityrates and percentages of
patientshospitalized. Duringongoing(pandemic)
influenza epidemics, such surveillance
RESEARCH AND PRACTICE
e6 | Research and Practice | Peer Reviewed | van den Wijngaard et al.
American Journal of Public Health | Published online ahead of print September 23, 2010

Page 7

informationcouldbeusedtodeterminetheneed
for control measures such as additional vaccina-
tion or prophylactic treatment. j
About the Authors
Cees C. van den Wijngaard, Liselotte van Asten, Adam
Meijer, Wilfrid van Pelt, Marianne A.B. van der Sande, and
Marion P.G. Koopmans are with the National Institute for
Public Health and the Environment, Centre for Infectious
Disease Control, Bilthoven, the Netherlands. Marion P.G.
Koopmans is also with the Erasmus Medical Centre, Rotter-
dam, the Netherlands. Nico J.D. Nagelkerke is with the
Department of Community Medicine, United Arab Emirates
University, Al-Ain. Ge ´ A. Donker is with the Netherlands
Institute for Health Services Research, Utrecht.
Correspondence should be sent to Cees C. van den
Wijngaard, MSc, Centre for Infectious Disease Control,
Netherlands National Institute for Public Health and the
Environment, PO Box 1, 3720 BA Bilthoven, the Nether-
lands (e-mail: kees.van.den.wijngaard@rivm.nl). Reprints
can be ordered at http://www.ajph.org by clicking on the
‘‘Reprints/Eprints’’ link.
This article was accepted November 4, 2009.
Contributors
C.C. van den Wijngaard contributed to the study design,
analyzed the data, and wrote the article. L. van Asten
contributed to the study design, data analysis, and drafting
of the article. A. Meijer interpreted the virological data
and helped draft the article. W. van Pelt and M.P.G.
Koopmans contributed to the study design and the in-
terpretation of data and reviewed drafts of the article.
N.J.D. Nagelkerke contributed to the statistical design and
the interpretation of data and reviewed drafts of the article.
G. Donker collected surveillance data and reviewed
drafts of the article. M.A.B. van der Sande helped interpret
the results and reviewed drafts of the article.
Acknowledgments
We thank Statistics Netherlands, the Dutch National
Medical Register, the Netherlands Institute for Health
Services Research, and the National Influenza Centre of
the Netherlands for providing data; the members of the
Dutch Working Group on Clinical Virology for allowing
us access to their registry and for collecting and providing
weekly diagnostic results; Hans van Vliet and Mirjam
Kretzschmar for reading and commenting on the article;
and Hendriek Boshuizen and Jan van de Kassteele for
providing feedback on the statistical methods used.
Human Participant Protection
Because the data examined in this study were obtained
from surveillance or medical research registries, no pro-
tocol approval was needed.
References
1.Buehler JW, Berkelman RL, Hartley DM, et al.
Syndromic surveillance and bioterrorism-related epi-
demics. Emerg Infect Dis. 2003;9(10):1197–1204.
2.
automatedmedicalrecordsforrapididentificationofillness
syndromes (syndromic surveillance): the example of lower
respiratory infection. BMC Public Health. 2001;1:9.
Lazarus R, Kleinman KP, Dashevsky I, et al. Using
3.
surveillance in public health practice, New York City.
Emerg Infect Dis. 2004;10(5):858–864.
Heffernan R, Mostashari F, Das D, et al. Syndromic
4.
Validation of syndromic surveillance for respiratory path-
ogen activity. Emerg Infect Dis. 2008;14(6):917–925.
Van den Wijngaard C, van Asten L, van Pelt W, et al.
5.
Lancet. 2003;362(9397):1733–1745.
Nicholson KG, Wood JM, Zambon M. Influenza.
6.
2004;10(suppl 12):S82–S87.
Palese P. Influenza: old and new threats. Nat Med.
7.
of antigenic drift. Vaccine. 2007;25(39–40):6852–6862.
Carrat F, Flahault A. Influenza vaccine: the challenge
8.
impact of influenza epidemics on mortality: introducing
a severity index. Am J Public Health. 1997;87(12):1944–
1950.
Simonsen L, Clarke MJ, Williamson GD, et al. The
9.
in the United States, France, and Australia: transmission
and prospects for control. Epidemiol Infect. 2008;136(6):
852–864.
Chowell G, Miller MA, Viboud C. Seasonal influenza
10. Donker GA. Continuous morbidity registration
sentinels: Netherlands 2006 [in Dutch]. Available at
http://www.nivel.nl/pdf/CMR-Peilstations-2006.pdf.
Accessed August 5, 2010.
11. De Jong JC, Rimmelzwaan GF, Bartelds AI, et al. The
influenza season 2004/’05 in the Netherlands with the
largest epidemic of the last 5 years caused by the virus
variant A/California and the composition of the vaccine
for the season 2005/’06 [in Dutch]. Ned Tijdschr Gen-
eeskd. 2005;149(42):2355–2361.
12. Thompson WW, Comanor L, Shay DK. Epidemiol-
ogy of seasonal influenza: use of surveillance data and
statistical models to estimate the burden of disease.
J Infect Dis. 2006;194(suppl 2):S82–S91.
13. Huang QS, Lopez L, Adlam B. Influenza surveillance
in New Zealand in 2005. N Z Med J. 2007;120(1256):
U2581.
14. International Classification of Diseases, Ninth Re-
vision, Clinical Modification. Hyattsville, MD: National
Center for Health Statistics; 1980. DHHS publication
PHS 80-1260.
15. DijkstraF,vanderPlasSM,MeijerA,etal. Respiratory
surveillancein2004/2005[inDutch].Availableat:http://
www.rivm.nl/infectieziektenbulletin/bul1606/art_surveil-
lance.html. Accessed August 5, 2010.
16. Van den Brandhof WE, Kroes ACM, Bosman A,
et al. Reporting virus diagnostics in the Netherlands:
representativeness of the virological weekly reports
[in Dutch]. Available at: http://www.rivm.nl/
infectieziektenbulletin/bul1304/vir_diagnostiek.html.
Accessed August 5, 2010.
17. Liang KY, Zeger SL. Longitudinal data analysis using
generalized linear models. Biometrika. 1986;73(1):13–22.
18. Akaike H. A new look at statistical model identifica-
tion. IEEE Trans Automat Contr. 1974;19(6):716–723.
19. Hak E, van Loon S, Buskens E, et al. Design of the
Dutch Prevention of Influenza, Surveillance and Manage-
ment (PRISMA) study. Vaccine. 2003;21(15):1719–1724.
20. Rimmelzwaan GF, de Jong JC, Bartelds AI, et al. The
2003/2004 influenza season in the Netherlands with
a limited epidemic of the virus variant A/Fujian, and the
vaccinecompositionforthe2004/2005season[inDutch].
Ned Tijdschr Geneeskd. 2004;148(40):1984–1988.
21. De Jong JC, Rimmelzwaan GF, Bartelds AI, et al.
2000/01 influenza season and the vaccine composition
for the season 2001/’02 [in Dutch]. Ned Tijdschr Gen-
eeskd. 2001;145(40):1945–1950.
22. Rimmelzwaan GF, de Jong JC, Bartelds AI, et al. The
2001/2002 influenza season and the vaccine composi-
tion for the 2002/2003 season [in Dutch]. Ned Tijdschr
Geneeskd. 2002;146(39):1846–1850.
23. Quenel P, Dab W, Hannoun C, et al. Sensitivity,
specificity and predictive values of health service based
indicators for the surveillance of influenza A epidemics.
Int J Epidemiol. 1994;23(4):849–855.
24. Cooper DL, Smith GE, Hollyoak VA, et al. Use of
NHS Direct calls for surveillance of influenza—a second
year’s experience. Commun Dis Public Health. 2002;5(2):
127–131.
25. Olson DR, Heffernan RT, Paladini M, et al. Moni-
toring the impact of influenza by age: emergency de-
partment fever and respiratory complaint surveillance in
New York City. PLoS Med. 2007;4(8):e247.
26. Lopez L, Huang S, Baker M. Influenza in New
Zealand—2003. Available at: http://www.surv.esr.cri.nz/
PDF_surveillance/Virology/FluAnnRpt/InfluenzaAnn2003.
pdf. Accessed August 5, 2010.
27. Paget WJ, Meerhoff TJ, Rebelo de Andrade H.
Heterogeneous influenza activity across Europe during
the winter of 2002–2003. Euro Surveill. 2003;8(12):
230–239.
28. Bragstad K, Jorgensen PH, Handberg KJ, et al. New
avian influenza A virus subtype combination H5N7
identified in Danish mallard ducks. Virus Res. 2005;
109(2):181–190.
29. De Jong JC, Rimmelzwaan GF, Bartelds AI, et al. The
2002/2003 influenza season in the Netherlands and the
vaccinecompositionforthe2003/2004season.[inDutch]
Ned Tijdschr Geneeskd. 2003;147(40):1971–1975.
30. HolmesEC, Ghedin E, Miller N, et al. Whole-genome
analysis of human influenza A virus reveals multiple
persistent lineages and reassortment among recent H3N2
viruses. PLoS Biol. 2005;3(9):e300.
31. Barr IG, Komadina N, Hurt AC, et al. An influenza
A(H3) reassortant was epidemic in Australia and New
Zealand in 2003. J Med Virol. 2005;76(3):391–397.
32. Jin H, Zhou H, Liu H, et al. Two residues in the
hemagglutinin of A/Fujian/411/02-like influenza vi-
ruses are responsible for antigenic drift from A/Panama/
2007/99. Virology. 2005;336(1):113–119.
33. Kim PE, Musher DM, Glezen WP, et al. Association of
invasive pneumococcal disease with season, atmospheric
conditions, air pollution, and the isolation of respiratory
viruses. Clin Infect Dis. 1996;22(1):100–106.
34. Avadhanula V, Rodriguez CA, Devincenzo JP, et al.
Respiratory viruses augment the adhesion of bacterial
pathogenstorespiratoryepitheliuminaviralspecies-andcell
type-dependent manner. J Virol. 2006;80(4):1629–1636.
35. Jansen AG, Sanders EA, van der Ende A, et al.
Invasive pneumococcal and meningococcal disease: asso-
ciation with influenza virus and respiratory syncytial virus
activity? Epidemiol Infect. 2008;136(11):1448–1454.
36. Jarrett JE, Pan X. The quality control chart for
monitoring multivariate autocorrelated processes. Com-
put Stat Data Anal. 2007;51(8):3862–3870.
37. Testik MC. Model inadequacy and residuals control
charts for autocorrelated processes. Qual Reliab Eng Int.
2005;21(2):115–130.
RESEARCH AND PRACTICE
Published online ahead of print September 23, 2010 | American Journal of Public Health
van den Wijngaard et al. | Peer Reviewed | Research and Practice | e7