Nowcasting (Short-Term Forecasting) of COVID-19 Hospitalizations Using Syndromic Healthcare Data, Sweden, 2020

Abstract

We report on local nowcasting (short-term forecasting) of coronavirus disease (COVID-19) hospitalizations based on syndromic (symptom) data recorded in regular healthcare routines in Östergötland County (population ≈465,000), Sweden, early in the pandemic, when broad laboratory testing was unavailable. Daily nowcasts were supplied to the local healthcare management based on analyses of the time lag between telenursing calls with the chief complaints (cough by adult or fever by adult) and COVID-19 hospitalization. The complaint cough by adult showed satisfactory performance (Pearson correlation coefficient r>0.80; mean absolute percentage error <20%) in nowcasting the incidence of daily COVID-19 hospitalizations 14 days in advance until the incidence decreased to <1.5/100,000 population, whereas the corresponding performance for fever by adult was unsatisfactory. Our results support local nowcasting of hospitalizations on the basis of symptom data recorded in routine healthcare during the initial stage of a pandemic.

Full text

During the initial stage of the coronavirus disease
(COVID-19) pandemic, large variations in vi-
rus dissemination within countries often led to lack
of suf ciently speci c information for local authori-
ties to make accurate decisions about health service
adjustments (1,2). The situation was further worsened
by heterogeneity in virus testing strategies, usually a
result of local differences in laboratory capacities (3),
leading to a need for local-scale COVID-19 forecast-
ing methods based on resources available in the exist-
ing healthcare infrastructure (4). In particular, experts
called for short-term forecasts of incident hospitaliza-
tions to plan staff reallocation and creation of tempo-
rary facilities for intensive or subintensive care with
ventilators (5).
We have previously developed a local in uenza
nowcasting (short-term forecasting) method whereby
syndromic healthcare data are used to nowcast later
diagnostic events (6). The method has shown satis-
factory performance in prospective evaluations (7,8).
We used this experience during the initial stage of the
pandemic in 2020 to nowcast local cases of patients
hospitalized with COVID-19 by modeling associa-
tions with data from Swedish Healthcare Direct’s 24-
hour telenursing service (telephone number 1177) (9).
Telenursing services are available in numerous coun-
tries for health counseling and evaluation of clinical
service needs in the general population (10–12). In
Sweden, the chief complaint for each call is recorded
in an administrative database (13). During the 2009
in uenza pandemic, records of telenursing chief
complaints were used to forecast variations in local
healthcare load, although less accurately than during
regular in uenza seasons (14).
The purpose of our study was to examine the per-
formance of syndromic healthcare data in nowcast-
ing local hospital admissions during the initial stage
of the COVID-19 pandemic, when resources for di-
agnostic laboratory testing were limited. The speci c
aim was to investigate the prospective performance
of symptoms recorded during telenursing calls in
Nowcasting (Short-Term
Forecasting) of COVID-19
Hospitalizations Using Syndromic
Healthcare Data, Sweden, 2020
Armin Spreco, Anna Jöud, Olle Eriksson, Kristian Soltesz, Reidar Källström,
Örjan Dahlström, Henrik Eriksson, Joakim Ekberg, Carl-Oscar Jonson,
Carl-Johan Fraenkel, Torbjörn Lundh, Philip Gerlee, Fredrik Gustafsson, Toomas Timpka
564 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022
RESEARCH
Author a󰀩 liations: Linköping University, Linköping, Sweden
(A. Spreco, O. Eriksson, R. Källström, Ö. Dahlström,
H. Eriksson J. Ekberg, C.-O. Jonson, F. Gustafsson, T. Timpka);
Region Östergötland, Linköping (A. Spreco, R. Källström,
J. Ekberg, C.-O. Jonson, T. Timpka); Lund University, Lund,
Sweden (A. Jöud, K. Soltesz); Skåne University Hospital, Lund
(A. Jöud. C.-J. Fraenkel); Chalmers University of Technology,
Gothenburg, Sweden (T. Lundh, P. Gerlee); Gothenburg
University, Gothenburg (T. Lundh, P. Gerlee)
DOI: https://doi.org/10.3201/eid2803.210267
We report on local nowcasting (short-term forecast-
ing) of coronavirus disease (COVID-19) hospitaliza-
tions based on syndromic (symptom) data recorded
in regular healthcare routines in Östergötland County
(population ≈465,000), Sweden, early in the pandemic,
when broad laboratory testing was unavailable. Daily
nowcasts were supplied to the local healthcare man-
agement based on analyses of the time lag between
telenursing calls with the chief complaints (cough by
adult or fever by adult) and COVID-19 hospitalization.
The complaint cough by adult showed satisfactory per-
formance (Pearson correlation coe󰀩 cient r>0.80; mean
absolute percentage error <20%) in nowcasting the in-
cidence of daily COVID-19 hospitalizations 14 days in
advance until the incidence decreased to <1.5/100,000
population, whereas the corresponding performance
for fever by adult was unsatisfactory. Our results sup-
port local nowcasting of hospitalizations on the basis of
symptom data recorded in routine healthcare during the
initial stage of a pandemic.

Nowcasting of COVID-19 Hospitalizations, Sweden
nowcasting daily cases of patients hospitalized with
COVID-19 during March–June 2020 in Östergötland
County, Sweden (population ≈465,000). The Swedish
Ethical Review Authority (dnr. 2020-03183) approved
the study design. Because COVID-19 and inuenza
share characteristic symptoms, we interpreted the
performance of the COVID-19 nowcasting using syn-
dromic symptom data, taking into consideration par-
allel winter inuenza activity in the county.
Methods
We used prospective evaluation design; that is, we
dened the COVID-19 nowcasting procedure and
the evaluation protocol before beginning to col-
lect evaluation data. The management of Region
Östergötland, the public (tax-nanced) healthcare
provider serving Östergötland County, used the dai-
ly nowcasts we created for planning resource alloca-
tion. Nowcasting of COVID-19 hospitalizations was
based on the time lag from telenursing calls with
selected chief complaints (Appendix, https://ww-
wnc.cdc.gov/EID/article/28/3/21-0267-App1.pdf);
we retrieved nowcasting data from the countywide
health information system managed by the health-
care provider (15). Because the COVID-19 pandemic
reached the study county during an ongoing inu-
enza season, we describe the progress of both local
epidemics for comparison.
Data Sources
Syndromic data were recorded from telenursing calls
made by county residents to Swedish Healthcare
Direct. Daily numbers of calls with chief complaints
possibly associated with COVID-19 were retrieved
from Hälsoläge, the national database, using the
xed-eld terminology register service (16). The di-
agnostic data were collected from patients hospital-
ized with the International Classication of Diseases,
10th Revision (ICD-10), code U07.1 (COVID-19, virus
identied). All patients hospitalized with suspected
COVID-19 were given a PCR test for virus identica-
tion and diagnosis.
We retrieved daily numbers of patients diag-
nosed with laboratory-conrmed inuenza (inpatient
and outpatient) for February 20–June 30, 2020. For
comparison, we also retrieved corresponding inuen-
za and telenursing chief complaint data for the same
period for each year during 2015–2019.
Nowcasting Procedure
We began developing the local COVID-19 nowcast-
ing procedure on February 20, 2020. During March
2–6, we examined peer-reviewed scientic reports
on COVID-19 symptoms to select telenursing chief
complaints for the nowcasting, (17–19). The larg-
est study retrieved, involving 1,099 patients from 30
provinces in China, reported fever (89%) and cough
(68%) to be the most common symptoms, followed
by fatigue (38%), shortness of breath (19%), and sore
throat (14%) (17). The study also reported that hos-
pitalized patients were almost exclusively adults.
In the selection of corresponding telenursing chief
complaints for use in nowcasting, we excluded un-
specic symptoms of upper respiratory tract infection
(fatigue and sore throat) and complaints expected to
lead to a recommendation for immediate physical ex-
amination (shortness of breath). We chose the remain-
ing telenursing chief complaints, cough by adult and
fever by adult, as syndromic variables for use in the
nowcasting of COVID-19 hospitalizations. We nal-
ized the procedure on March 20.
Denition of Time Lag
After consultations with local healthcare managers,
we found that we needed short-term forecasts in the
interval of 14–21 days for implementing adjustments
of hospital resources. To select the time lag in the
interval with the highest correlation (i.e. the highest
Pearson correlation coefcient, r) between syndromic
and hospital admission data, we performed analyses
of time series data from the previous 4 weeks for each
of the 2 syndromic variables, leading to 16 possible
outcomes: 8 time lags of 14–21 days for each variable.
To eliminate weekday effects, we smoothed all series
by calculating a 7-day moving average. If correla-
tions for time lags were equal, we chose the longest.
To adjust for the higher daily numbers of telenursing
calls compared with hospitalization cases, we multi-
plied the level for each of the 2 chief telenursing com-
plaints by a ratio calculated by dividing the sum of
hospitalizations during a 14-day period by the sum
of telenursing calls (separately for each syndromic
variable) over a previous 14-day interval at a time
distance, chosen depending on the resulting best time
lag. The length of the interval should be a multiple of
7 days to level out weekday effects and be about the
same as the time lag. Therefore, we chose an interval
of 14 days.
Hospital Admission Nowcasting
We created daily nowcasts and forwarded them to
the healthcare management at Region Östergötland
beginning March 22, 2020. We performed a new cal-
culation of the correlation coefcient each nowcasting
day and chose the time lag with the highest correla-
tion for each of the 2 chief complaints for nowcasts.
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022 565

RESEARCH
We performed daily nowcasts of forthcoming hos-
pitalizations for the period covered by the time lag
between COVID-19 hospitalizations and telenursing
calls for cough by adult and fever by adult through-
out the study period (Appendix).
Descriptive Analyses
Because COVID-19 and inuenza share symptoms
(telenursing chief complaints), we examined the dai-
ly numbers of COVID-19 hospitalizations and cases
of laboratory-conrmed inuenza in Östergötland
County (primary and hospital care) for the period
February 20–June 30, 2020. We also descriptively ana-
lyzed the annual trends for this period in 2015–2019
for cases of laboratory-conrmed inuenza and for
the telenursing chief complaints cough by adult and
fever by adult.
Evaluation Procedure was dened
We evaluated the nowcasting performance during
March 22–June 30, 2020. We dened the evaluation
protocol on March 20 and followed it without altera-
tion throughout the evaluation period. We evaluated
performance by calculating the correlation between
trends in the selected telenursing calls and trends
in later hospitalizations, and by determining the
accuracy of the nowcasted incidence of daily hos-
pitalizations. The outcome measures were the Pear-
son correlation coefcient between the telenursing
and hospitalization data from the nowcasting date
through the period covered by the time lag (denot-
ed as rFND) and the mean absolute percentage error
(MAPE) of the nowcasted hospitalization incidence.
rFND can vary between −1 and 1 (where −1 is perfect
negative correlation and 1 is perfect positive correla-
tion). The lower limit for MAPE is 0; an upper limit
does not exist. Before beginning data collection, we
dened the limits for satisfactory nowcasting perfor-
mance as rFND >0.80 and MAPE <20%.We derived the
limit for rFND from previous nowcasting studies (20)
and determined the MAPE limit, following discus-
sions with health service managers, on the basis of
hospital resources in Sweden, which were overex-
tended before the COVID-19 pandemic (on average,
103 patients occupied 100 administrative hospital
bed units [21]).
Results
COVID-19 Pandemic
Calls by Östergötland county residents to Swedish
Healthcare Direct with the chief complaint of cough
by adult peaked on March 21 (Figure 1, panel A). On
the same day, calls for the complaint fever by adult
reached a plateau that lasted for ≈2 weeks (until April
3) (Figure 1, panel A).
The rst hospitalization in Östergötland County
for COVID-19 occurred on March 8, 2020. At the start
of the evaluation period on March 22, the daily hospi-
talization incidence was 1.8 patients/100,000 popula-
tion; peak incidence (4.9 patients/day/100,000 popu-
lation) was reached on April 2 (Table; Figure 1, panel
B). In mid-May, the daily incidence had declined to
<1.5 hospitalizations/100,000 population; it was 0.6
hospitalizations/100,000 population on June 30, the
end of the study period.
Inuenza Season
The daily incidence of patients with laboratory-con-
rmed inuenza peaked on March 10 (Figure 1, pan-
el C). The recorded incidence decreased thereafter
to a level that was notably below the 5-year histori-
cal trend. Calls to Swedish Healthcare Direct for the
chief complaints cough by adult and fever by adult
did not show a corresponding decrease in March
2020 (Figure 1, panel A). The comparative display of
the historical trends from the previous 5-year period
for these chief complaints showed that the levels
usually increased throughout the month of March
(Figure 1, panels D, E).
Nowcasting Performance
The selected optimal time lag for both the cough
by adult and fever by adult variables was 14 days
throughout the study period, except for cough by
adult during March 26–28, when the time lag was 15
or 16 days (Video, https://wwwnc.cdc.gov/EID/
article/23/3/21-0267-V1.htm). During the ascending
stage of the rst wave of the pandemic (March 22–
April 4), as hospitalizations increased (Figure 2, panel
A), rFND for the Swedish Healthcare Direct chief com-
plaint cough by adult was satisfactory (0.86–0.98),
and MAPE decreased rapidly to a satisfactory level
(from 28% to 3%) (Table; Figure 2, panels B, C; Video).
rFND for the chief complaint fever by adult decreased
during this period to −0.63, and MAPE was mostly
unsatisfactory (14%–47%). At the peak of the wave,
with a daily hospitalization incidence >2.5/100,000
population (April 5–25), rFND (0.74–0.97) and MAPE
(4%−9%) remained satisfactory for cough by adult.
For fever by adult, rFND (−0.63 to 0.95) and MAPE
(14%–52%) stayed at unsatisfactory levels. During
the descending stage, rFND and MAPE for cough by
adult remained satisfactory until hospitalizations de-
clined. When the daily hospitalizations decreased to
<1.5/100,000 population in mid-May, rFND and MAPE
566 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022

Nowcasting of COVID-19 Hospitalizations, Sweden
indicated unsatisfactory performances for both syn-
dromic indicators (Table; Figure 2).
Discussion
This study examined the performance of syndromic
healthcare data (symptoms reported during telenurs-
ing calls) in nowcasting local hospital loads during
the initial stage of the COVID-19 pandemic when re-
sources for diagnostic laboratory testing were limited.
We found that the telenursing chief complaint cough
by adult accurately (rFND 0.74–0.98; MAPE <10%)
nowcasted local hospital loads >14 days in advance
during periods with intense local dissemination of
COVID-19 (corresponding to >2.5 hospitalizations/
day/100,000 population) and continued to provide
reliable nowcasts until the intensity decreased to <1.5
hospitalizations/day/100,000 population.
Although fever is a characteristic COVID-19
symptom, the performance of the Swedish Health-
care Direct chief complaint fever by adult in nowcast-
ing was less satisfactory. This observation could be
cause by the co-circulation of inuenza virus strains
and severe acute respiratory syndrome coronavirus
2 (SARS-CoV-2); fever by adult was recorded as a
chief complaint from telenursing calls resulting from
both inuenza infection and COVID-19 (22). Even
though cough was also a representative symptom for
inuenza, it appeared to be more uniquely record-
ed as the chief complaint from telenursing calls for
COVID-19. We also observed that the incidence of pa-
tients with a laboratory-conrmed diagnosis of inu-
enza peaked on March 10, just before the COVID-19
pandemic reached Östergötland County, and there-
after decreased to a level notably below the 5-year
historical trend. It is unclear whether this decrease in
the recorded incidence of inuenza represents a true
decline in infections or due to changes in healthcare-
seeking behaviors (23). These observations suggest
that COVID-19 nowcasting based on symptom data
should be performed with caution during periods in
which SARS-CoV-2 is co-circulating with inuenza
and other respiratory viruses.
Poor forecasting reliability during the rst wave
of the COVID-19 pandemic led to demands on invest-
ments in developing task-specic models and quality
data collection (24,25). One explanation for the satisfac-
tory local nowcasting performance we observed is the
rapid and stable access to syndromic and diagnostic
data throughout the emerging rst wave of the pan-
demic. Most methods for COVID-19 nowcasting have
used diagnostic data to model the near-future progress
(typically 2–6 days) of the corresponding events (26);
A. Altmejd, et al., unpub. data, https://arxiv.org/
pdf/2006.06840.pdf). In contrast to such autoregres-
sive models, we used a separate syndromic data source
to nowcast COVID-19 hospitalizations 14–21 days in
advance. This time lag to hospitalizations was needed
to rearrange the local healthcare organization to care
for patients with COVID-19 while minimizing collat-
eral effects on other patient groups. We collected the
syndromic and diagnostic data used for the nowcast-
ing from a regular health information system (15) and
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022 567
Figure 1. Daily incidence of telenursing calls for 2 chief
complaints, COVID-19 hospitalizations, and laboratory-conrmed
inuenza plus reference data from before the COVID-19
pandemic, Östergötland County, Sweden. A) Telenursing calls per
100,000 population for chief complaints of cough by adult (blue
line) and fever by adult (red line), February 20–June 30, 2020.
B) COVID-19 hospitalizations per 100,000 population, February
20–June 30, 2020. C) Cases of laboratory-conrmed inuenza
per 100,000 population February 20–June 30, 2020 (black line).
Light gray line indicates the average for cases of laboratory-
conrmed inuenza in 2015–2019; dark gray shaded area is the
corresponding range. D) Telenursing calls per 100,000 population
for the chief complaint cough by adult in 2015–2019 (light grey
line) with corresponding range (dark grey shaded area). E)
Telenursing calls per 100,000 population for the chief complaint
fever by adult in 2015–2019 (light grey line) with corresponding
range (dark grey shaded area).

RESEARCH
analyzed the data using experiences from nowcasting
the 2009 inuenza pandemic and subsequent winter
inuenza seasons (6,14,27). The syndromic data were
recorded by telenurses specially trained in assessment
of adults and children who experienced infectious-
disease symptoms (13). At the time of the outbreak of
COVID-19 in Sweden (February 2020), telenursing had
evolved from a triage practice within primary care (28–
31) into a key resource in healthcare provision staffed
by experienced nursing professionals (9). The diagnos-
tic data we used for the nowcasting in this study were
recorded using standardized coding routines (32) by
physicians with clinical responsibility for patients hos-
pitalized with COVID-19.
Syndromic symptom data have been used for sev-
eral purposes in the early response to the COVID-19
pandemic. Using web-based data collection from the
general public, the EPICOVID19 study in Italy found a
strong association between olfactory and taste symp-
toms and laboratory-conrmed COVID-19 (33). Loss
of smell and taste have also been reported as a charac-
teristic COVID-19 symptom from similar research in
568 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022
Table. Weekly nowcasting performance for 2 syndromic variables in the first wave of the coronavirus pandemic, Östergötland County,
Sweden, 2020*
Nowcasting dates
Hospitalizations/day/
100,000 population
Cough by adult
Fever by adult
rFND
MAPE
rFND
MAPE
Week 1 (Mar 22–28)
1.8–3.4
0.86–0.97
9 –28
0.01–0.99
14–20
Week 2 (Mar 29–Apr 4)
3.4–4.9
0.93–0.98
3–5
−0.63 to −0.32
17–47
Week 3 (Apr 5–11)†
3.2–4.5
0.89–0.95
4–6
−0.20 to 0.79
39–52
Week 4 (Apr 12–18)
2.6–3.2
0.92–0.97
4–6
0.87–0.95
16–45
Week 5 (Apr 19–25)
2.1–2.6
0.74–0.94
6–9
0.70–0.93
15–21
Week 6 (Apr 26–May 2)
1.4–2.1
0.46–0.73
10–13
0.58–0.73
9–13
Week 7 (May 3–9)
1.4–1.6
0.64–0.91
7–13
0.65–0.82
8–11
Week 8 (May 10–16)
1.1–1.5
0.53–0.74
8–17
0.45–0.65
9–11
Week 9 (May 17–23)
0.9–1.1
−0.28 to 0.57
19–41
−0.08 to 0.44
9–14
Week 10 (May 24–30)
0.9–1.1
−0.87 to −0.46
38–47
−0.57 to −0.16
14–18
Week 11 (May 31–Jun 6)
0.8–1.1
−0.86 to −0.26
19–32
−0.90 to 0.63
17–28
Week 12 (Jun 7–13)
0.8–1.0
−0.03 to 0.48
29–55
0.74–0.78
21–34
Week 13 (Jun 14–20)
0.6–1.0
−0.41 to 0.36
17–48
−0.53 to 0.60
12–32
Week 14 (Jun 21–27)
0.5–0.7
−0.20 to 0.58
15–28
0.13–0.78
10–23
Week 15 (Jun 28–30)‡
0.6–0.7
0.42 to 0.50
24 to 25
0.66 to 0.70
20–22
*MAPE, mean absolute percentage error; rFND, Pearson correlation coefficient between the telenursing and hospitalization data from the nowcasting date
through the period covered by the time lag.
†Includes local peak of the first pandemic wave.
‡Only 3 days because it is the end of the study period.
Figure 2. Local nowcasting
performance in Östergötland
County, Sweden, during the rst
wave of coronavirus disease
(COVID-19), March 22–June
30, 2020. A) Weekly average of
daily incidence of COVID-19,
hospitalizations/week/100,000
population. The horizontal line
indicates lowest incidence
for reliable predictions (1.5
daily hospitalizations/100,000
population). B) Weekly average
of daily correlation between
telenursing data and COVID-19
hospitalizations from the
nowcasting date through the
period covered by the time lag
for cough by adult (blue line)
and fever by adult (red line). C)
Weekly average of daily MAPE
per week for cough by adult
(blue line) and fever by adult
(red line). MAPE, mean absolute
percentage error.

Nowcasting of COVID-19 Hospitalizations, Sweden
the United Kingdom and the United States (34), Italy
(35), and France (36). These symptom-tracking studies
have provided important insights into the spectrum
of COVID-19 symptoms, the rate of these symptoms
in nonhospitalized persons, and the natural history
of the infection. Nonetheless, for local nowcasting
of hospital admissions during the early stages of a
pandemic, rapid initiation of data collection and rep-
resentative population coverage are required. Stud-
ies conducted in April and May 2020 showed that
willingness to use a mobile application to support
COVID-19 surveillance was 55%–70% in countries
such as the United States, Switzerland, and Italy (37).
However, by November 2020, the use of such mobile
applications was still limited in nations where govern-
ments had promoted their development and dissemi-
nation; for example, 26% in Australia, 13% in Italy,
and 2% in France (38). These proportions indicate that
achievement of representative population coverage
and continuity in data collection are challenging for
COVID-19 forecasting using mobile applications. One
reason for the low use of mobile applications is that le-
gal and condentiality issues have not been resolved
for data collection from personal Internet devices in
public health practice (39). Our nowcasting approach
used trends in routinely recorded healthcare data for
short-term forecasts of hospitalization cases. The ap-
proach did not require data normally unavailable for
local healthcare providers and did thereby allow early
initiation of nowcasting to support the local health-
care managers in their decision making.
The aim of this study was to assess hospital ad-
mission nowcasting during the early pandemic stage
when broad laboratory testing still was unavailable.
The syndromic variables (telenursing chief com-
plaint codes) were thus determined in mid-March
2020 based on the information available. A limita-
tion of the study is that it is possible that later selec-
tion of codes would have inuenced the nowcasting
outcomes. Also, use of individual-level telenursing
data and sociodemographic data may have enabled
detailed detection of municipality-level clusters dur-
ing the initial stage of a pandemic. However, reports
of variations in telenursing outreach and use across
geographic areas and population groups, for example,
among immigrants and the elderly (12,40), imply that
further studies are needed to establish whether a more
detailed version of our nowcasting procedure would
be suitable for more specic early detection. More-
over, the outcome measures used in the study may not
cover all aspects of healthcare load during pandemics.
The coefcient rFND shows correspondences between
the nowcasted and observed series of hospitalization
incidences over time, and MAPE displays how much
the nowcasted incidences deviated as a percentage
from the observed incidences. In future studies of CO-
VID-19 hospitalizations, nowcasting the prevalence
of hospitalized patients can be considered, which will
require considering the length of hospital stay for dif-
ferent categories of COVID-19 patients. Moreover, the
study did not use accuracy metrics such as uncertainty
bounds around the point predictions because the pub-
lic health practitioners did not request such bounds.
It would have been possible to change the evaluation
metrics afterwards, but doing so would have neutral-
ized the prospective evaluation design. In the future,
the nowcasting method can be further developed by
including uncertainty bounds or probability estimates
(41). The current approach has at least 2 uncertainties
that can be quantied; uncertainty about how many
persons with symptoms call the telenursing service,
and uncertainty about the proportion of calls for a spe-
cic chief complaint that is constituted by COVID-19
cases. Finally, the nowcasting method was intended
for use during the initial stage of a pandemic when
broad laboratory testing is unavailable. The results
are mainly generalizable to other early pandemic set-
tings in which comparable infrastructural resources
are available. Generalization of our results and ap-
plication of the nowcasting method to later pandemic
phases, when population-level laboratory testing is
available, warrants more research.
We conclude that symptom data regularly re-
corded in healthcare can be used for local nowcasting
of hospital loads during the initial stage of a pandem-
ic when broad laboratory testing still is unavailable.
The telenursing chief complaint cough by adult dis-
played satisfactory nowcasting performance during
initial pandemic periods with high community dis-
semination of COVID-19 (>1.5 hospitalization cases/
day/100,000 population). The study also indicates
that symptom data should be used with caution for
pandemic nowcasting when the novel virus is co-cir-
culating with competing viruses. Our results support
local nowcasting of hospitalizations on the basis of
regularly recorded syndromic data during the initial
stage of a pandemic.
This study was supported by grants from the Swedish
Civil Contingencies Agency (grant no. 2010-2788), the
Swedish Research Council (grant no. 2021-05608), ALF
grants from Region Östergötland (dnr. 936190), and by
grants from the Research Council of Southeast Sweden
(grant no. FORSS-940915). The funders had no role in the
study design, data collection and analysis, decision to
publish, or preparation of the manuscript.
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022 569

RESEARCH
Authors’ contributions: A.S., A.J., O.E., Ö.D., and T.T
conceived and designed the study. A.S. and O.E. analyzed
the data. O.E., K.S., and F.G. veried the results. A.S. and
T.T. wrote the article. A.J., O.E., R.K., K.S., Ö.D., H.E., J.E.,
C.O.J., C.J.F., T.L., P.G., and F.G. revised the article and
provided intellectual content. A.J., O.E., R.K., Ö.D., H.E.,
J.E., C.O.J., C.J.F., K.S., T.L., P.G., and F.G. gave nal
approval of the version to be published. T.T. is the
guarantor of the content.
About the Author
Dr. Spreco is a researcher in the eld of syndromic
infectious disease surveillance at Linköping University,
and Region Östergötland, Sweden. His main research
focus is on evaluation and development of algorithms
for local detection and prediction of infectious diseases.
During the 2020 pandemic, he provided local forecasts of
COVID-19 hospitalizations and healthcare capacity needs
to Swedish healthcare regions.
References
1 Paul R, Arif AA, Adeyemi O, Ghosh S, Han D. Progression
of COVID-19 from urban to rural areas in the United States:
a spatiotemporal analysis of prevalence rates. J Rural Health.
2020;36:591-601. https://doi.org/10.1111/jrh.12486
2. García-Basteiro AL, Chaccour C, Guinovart C, Llupià A,
Brew J, Trilla A, et al. Monitoring the COVID-19 epidemic
in the context of widespread local transmission. Lancet
Respir Med. 2020;8:440–2. https://doi.org/10.1016/
S2213-2600(20)30162-4
3. Gill M, Sridhar D, Godlee F. Lessons from Leicester: a
COVID-19 testing system that’s not t for purpose. BMJ.
2020;370:m2690. https://doi.org/10.1136/bmj.m2690
4. Chiolero A. Predicting COVID-19 resurgence: do it locally.
BMJ. 2020;370:m2731. https://doi.org/10.1136/bmj.m2731
5. Grasselli G, Pesenti A, Cecconi M. Critical care utilization for
the COVID-19 outbreak in Lombardy, Italy: early experience
and forecast during an emergency response. JAMA.
2020;323:1545–6. https://doi.org/10.1001/jama.2020.4031
6. Spreco A, Eriksson O, Dahlström Ö, Cowling BJ, Timpka T.
Integrated detection and prediction of inuenza activity for
real-time surveillance: algorithm design. J Med Internet Res.
2017;19:e211. https://doi.org/10.2196/jmir.7101
7. Spreco A, Eriksson O, Dahlström Ö, Cowling BJ, Timpka T.
Evaluation of nowcasting for detecting and predicting local
inuenza epidemics, Sweden, 2009–2014. Emerg Infect Dis.
2018;24:1868–73. https://doi.org/10.3201/eid2410.171940
8. Spreco A, Eriksson O, Dahlström Ö, Cowling BJ, Biggerstaff
M, Ljunggren G, et al. Nowcasting (short-term forecasting)
of inuenza epidemics in local settings, Sweden, 2008–2019.
Emerg Infect Dis. 2020;26:2669–77. https://doi.org/10.3201/
eid2611.200448
9. Online counselling support [in Swedish]. 2020 [cited 2020
Mar 10]. https://www.1177.se/om-1177-vardguiden/
om-1177-vardguiden/radgivningsstodet-webb%2D%2
Dett-stod-i-din-verksamhet
10. Kvedar J, Coye MJ, Everett W. Connected health: a review of
technologies and strategies to improve patient care with
telemedicine and telehealth. Health Aff (Millwood).
2014;33:194–9. https://doi.org/10.1377/hlthaff.2013.0992
11. Cunningham PN, Grant-Pearce C, Green L, Miles ID,
Rigby J, Uyarra E. In sickness, in health, and in innovation:
NHS DIRECT—a health sector innovation study. Presented
at: Breaking New Ground: Innovation in the Public Sector
International Conference; September 22–23, 2005;
University College, Cork, Ireland.
12. Blakoe M, Gamst-Jensen H, von Euler-Chelpin M,
Collatz Christensen H, Møller T. Sociodemographic and
health-related determinants for making repeated calls
to a medical helpline: a prospective cohort study.
BMJ Open. 2019;9:e030173. https://doi.org/10.1136/
bmjopen-2019-030173
13. Kaminsky E, Aurin IE, Hedin K, Andersson L, André M.
Registered nurses’ views on telephone nursing for patients
with respiratory tract infections in primary healthcare—a
qualitative interview study. BMC Nurs. 2020;19:65.
https://doi.org/10.1186/s12912-020-00459-1
14. Timpka T, Spreco A, Eriksson O, Dahlström Ö, Gursky EA,
Strömgren M, et al. Predictive performance of telenursing
complaints in inuenza surveillance: a prospective
cohort study in Sweden. Euro Surveill. 2014;19:20966.
https://doi.org/10.2807/1560-7917.ES2014.19.46.20966
15. Timpka T, Eriksson H, Gursky EA, Strömgren M, Holm E,
Ekberg J, et al. Requirements and design of the PROSPER
protocol for implementation of information infrastructures
supporting pandemic response: a Nominal Group study.
PLoS One. 2011;6:e17941. https://doi.org/10.1371/
journal.pone.0017941
16. Folkhälsomyndigheten. Syndrome monitoring [in Swedish]
[cited 2020 Mar 15]. https://www.folkhalsomyndigheten.
se/smittskydd-beredskap/overvakning-och-rapportering/
syndromovervakning
17. Guan WJ, Ni ZY, Hu Y, Liang WH, Ou CQ, He JX, et al.;
China Medical Treatment Expert Group for COVID-19.
Clinical characteristics of coronavirus disease 2019 in China.
N Engl J Med. 2020;382:1708–20. https://doi.org/10.1056/
NEJMoa2002032
18. Wu Z, McGoogan JM. Characteristics of and important
lessons from the coronavirus disease 2019 (COVID-19)
outbreak in China: summary of a report of 72,314 cases
from the Chinese Center for Disease Control and
Prevention. JAMA. 2020;323:1239–42. https://doi.org/
10.1001/jama.2020.2648
19. Chen N, Zhou M, Dong X, Qu J, Gong F, Han Y, et al.
Epidemiological and clinical characteristics of 99 cases of
2019 novel coronavirus pneumonia in Wuhan, China: a
descriptive study. Lancet. 2020;395:507–13. https://doi.org/
10.1016/S0140-6736(20)30211-7
20. Spreco A, Eriksson O, Dahlström Ö, Timpka T. Inuenza
detection and prediction algorithms: comparative accuracy
trial in Östergötland county, Sweden, 2008–2012.
Epidemiol Infect. 2017;145:2166–75. https://doi.
org/10.1017/S0950268817001005
21. Sveriges Kommuner och landsting (SKL). No one unnecessarily
at hospital [in Swedish]. 2016 [cited 2021 Oct 10].
https://webbutik.skr.se/bilder/artiklar/pdf/7585-421-2.pdf
22. US Centers for Disease Control and Prevention. Flu
symptoms and complications. 2020 [cited 2020 Oct 19].
https://www.cdc.gov/u/symptoms/symptoms.htm
23. Melidou A, Pereyaslov D, Hungnes O, Prosenc K, Alm E,
Adlhoch C, et al.; World Health Organization European
Region Inuenza Surveillance Network. Virological
surveillance of inuenza viruses in the WHO European
Region in 2019/20—impact of the COVID-19 pandemic.
Euro Surveill. 2020;25:2001822. https://doi.org/
10.2807/1560-7917.ES.2020.25.46.2001822
570 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022

Nowcasting of COVID-19 Hospitalizations, Sweden
24. Chin V, Samia NI, Marchant R, Rosen O, Ioannidis JPA,
Tanner MA, et al. A case study in model failure?
COVID-19 daily deaths and ICU bed utilization predictions
in New York state. Eur J Epidemiol. 2020;35:733–42.
https://doi.org/10.1007/s10654-020-00669-6
25. Press WH, Levin RC. Modeling, post COVID-19. Science.
2020;370:1015. https://doi.org/10.1126/science.abf7914
26. Guenther F, Bender A, Katz K, Kuechenhoff H, Höhle M.
Nowcasting the COVID-19 pandemic in Bavaria. Biom J.
2021;63:490–502 https://doi.org/10.1002/bimj.202000112
27. Timpka T, Spreco A, Dahlström Ö, Eriksson O, Gursky E,
Ekberg J, et al. Performance of eHealth data sources in local
inuenza surveillance: a 5-year open cohort study. J Med
Internet Res. 2014;16:e116. https://doi.org/10.2196/
jmir.3099
28. Timpka T, Arborelius E. The primary-care nurse’s
dilemmas: a study of knowledge use and need during
telephone consultations. J Adv Nurs. 1990;15:1457–65.
https://doi.org/10.1111/j.1365-2648.1990.tb01789.x
29. Marklund B, Bengtsson C, Blomkvist S, Furunes B,
Gäcke-Herbst R, Silfverhielm B, et al. Evaluation of the
telephone advisory activity at Swedish primary health care
centres. Fam Pract. 1990;7:184–9. https://doi.org/10.1093/
fampra/7.3.184
30. Marklund B, Koritz P, Bjorkander E, Bengtsson C. How well
do nurse-run telephone consultations and consultations in
the surgery agree? Experience in Swedish primary health
care. Br J Gen Pract. 1991;41:462–5.
31. Timpka T. The patient and the primary care team: a
small-scale critical theory. J Adv Nurs. 2000;31:558–64.
https://doi.org/10.1046/j.1365-2648.2000.01310.x
32. Socialstyrelsen. Coding of COVID-19 [in Swedish]. 2020
[cited 2021 Jan 1]. https://www.socialstyrelsen.se/
globalassets/sharepoint-dokument/dokument-webb/
klassikationer-och-koder/kodning-av-covid-19.pdf
33. Adorni F, Prinelli F, Bianchi F, Giacomelli A, Pagani G,
Bernacchia D, et al. Self-reported symptoms of SARS-CoV-2
infection in a nonhospitalized population in Italy:
cross-sectional study of the EPICOVID19 web-based
survey. JMIR Public Health Surveill. 2020;6:e21866.
https://doi.org/10.2196/21866
34. Menni C, Valdes AM, Freidin MB, Sudre CH, Nguyen LH,
Drew DA, et al. Real-time tracking of self-reported
symptoms to predict potential COVID-19. Nat Med.
2020;26:1037–40. https://doi.org/10.1038/s41591-020-0916-2
35. Popovic M, Moccia C, Isaevska E, Moirano G, Pizzi C,
Zugna D. COVID-19-like symptoms and their relation to
SARS-CoV-2 epidemic in children and adults of the Italian
birth cohort. Research Square. 2020. https://doi.org/
10.21203/rs.3.rs-34027/v1
36. Denis F, Galmiche S, Dinh A, Fontanet A, Scherpereel A,
Benezit F, et al. Epidemiological observations on the
association between anosmia and COVID-19 infection:
analysis of data from a self-assessment web application.
J Med Internet Res. 2020;22:e19855. https://doi.org/10.2196/
19855
37. Hargittai E, Redmiles E. Will Americans be willing to install
COVID-19 tracking apps? Sci Am. 2020 Apr 28 [cited 2022
Jan 4]. https://blogs.scienticamerican.com/observations/
will-americans-be-willing-to-install-covid-19-tracking-apps
38. Blasimme A, Vayena E. What’s next for COVID-19 apps?
Governance and oversight. Science. 2020;370:760–2.
https://doi.org/10.1126/science.abd9006
39. Bernard R, Bowsher G, Sullivan R. COVID-19 and the
rise of participatory SIGINT: an examination of the rise in
government surveillance through mobile applications. Am
J Public Health. 2020;110:1780–5. https://doi.org/10.2105/
AJPH.2020.305912
40. Cook EJ, Sharp C, Randhawa G, Guppy A, Gangotra R,
Cox J. Who uses NHS health checks? Investigating the
impact of ethnicity and gender and method of invitation on
uptake of NHS health checks. Int J Equity Health. 2016;15:13.
https://doi.org/10.1186/s12939-016-0303-2
41. Gneiting T, Balabdaoui F, Raftery AE. Probabilistic
forecasts, calibration and sharpness. J R Stat Soc Series
B Stat Methodol. 2007;69:243–68. https://doi.org/10.1111/
j.1467-9868.2007.00587.x
Address for correspondence: Armin Spreco, Department of
Health, Medicine and Caring Sciences, Linköping University,
s-581 83 Linköping, Sweden; email: armin.spreco@liu.se
Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 28, No. 3, March 2022 571