See

discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/233948622

A Screening Tool for Dengue Fever in

Children

Article in The Pediatric Infectious Disease Journal · December 2012

DOI: 10.1097/INF.0b013e31827e111e · Source: PubMed

CITATIONS READS

5 218

5 authors, including:

Smile Chien Shih-Bin Su

Chi-Mei Medical Center Chi-Mei Medical Center
37 PUBLICATIONS 199 CITATIONS 95 PUBLICATIONS 547 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

smile@mail.chimei.org.tw View project

All content following this page was uploaded by Shih-Bin Su on 11 October 2014.

The user has requested enhancement of the downloaded file.

 Original Studies

A Screening Tool for Dengue Fever in Children

Wen-Pin Lai, MD,*† Tsair-Wei Chien, MBA,†‡ Hung-Jung Lin, MD, MBA,*§ Shih-Bin Su, MD, PhD,¶ ‖
and Chih-Hung Chang, PhD**††

Background: Dengue fever (DF) is a significant public health issue in Asia.

of dengue hemorrhagic fever every year, >20,000 people, primarily
We aimed to use clinical and laboratory data to derive a rapid and accurate
children, die.5,6
case-finding tool for DF in children.
A challenge encountered by primary care physicians, espe-
Methods: This retrospective study used 24 DF-related characteristics and
cially those who work in emergency departments, is the lack of
clinical features (17 clinical; 7 laboratory) of 177 pediatric patients (69
an accurate diagnostic screening test for DF. The early signs and
diagnosed with DF). Data were psychometrically evaluated using a Rasch
symptoms of DF, such as fever, headache and myalgia, are simi-
measurement model, and their values for predicting DF risk were evaluated.
lar to those of other illnesses.7–9 Some studies using the univari-
Results: The 14-item scale (DF-14) fit the measurement model in assessing
ate analysis report that the presumptive diagnosis of DF is impre-
the likelihood of DF. When a cutoff point of −1.15 (in logit) of the DF-14
cise because its signs and symptoms are not helpful for detecting
scale was used, the sensitivity was 0.76 and the specificity was 0.76. The
DF.7,8 Multivariate regression analyses have also been used to try
area under the curve was 0.93 (95% confidence interval: 0.89–0.97). The
to distinguish patients with dengue from those with other febrile
DF-2 scale, comprised of white blood cell and platelet counts, was simple
illnesses, but none had statistically valid, and none considered sub-
but clinically useful.
stantial changes in clinical features during the course of the illness10
Conclusions: Simple laboratory data, such as those in the DF-2 and DF-14
In the present study, we used a well-developed measurement
scales, are useful for the early detection of DF risk in children. The DF-14
model based on item response theory11,12 to construct a valid and
scale helps discriminate DF from other febrile illnesses and may eliminate
reliable scale using existing clinical data for the early detection of
the need for a costly and time-consuming dengue confirmation test.
DF in children.

Key Words: dengue fever, dengue serologic test, Rasch analysis, receiver
MATERIALS AND METHODS
operating characteristic curve

(Pediatr Infect Dis J 2013;32: 320–324)

Study Population
All children (≤16 years old) at the emergency department
of Chi-Mei Medical Center in southern Taiwan between January 1
and December 31 of 2007 were included. Routine blood samples

D engue virus (DENV) infection is one of the most common

mosquito-borne viral diseases of humans worldwide.1 It
causes a spectrum of illness from mild dengue fever (DF) to severe
dengue hemorrhagic fever and dengue shock syndrome.2,3 The
prevalence rate of DF has increased significantly in Southeast Asia,
Africa, the Western Pacific and the Americas in the past several
decades.4,5 An estimated 2.5 billion people in >100 countries are at
risk for DENV infection. More than 50 million new infections are
reported annually, and among the 250,000–500,000 recorded cases
From the *Department of Emergency Medicine, Chi-Mei Medical Center;
†Department of Hospital and Health Care Administration, Chia-Nan Uni-
versity of Pharmacy and Science; ‡Department of Administration, Chi-Mei
Medical Center; §Department of Biotechnology; ¶Institute of Biomedical
Engineering, Southern Taiwan University; ‖Department of Family Medicine,
Chi-Mei Medical Center; **Buehler Center on Aging, Health & Society,
Feinberg School of Medicine, Northwestern University, Chicago, IL; and
††Graduate Institute of Biostatistics, China Medical University, Taichung,
Taiwan.
W.-P.L. and T.-W.C. conceived and designed the study, performed the statisti-
cal analyses and were in charge of recruiting study participants. Y.-C.C. and
C.-H.L. helped design the study, collected information and interpreted data.
H.-J.L., S.-B.S. and C.-H.C. helped design and supervise the study, and
helped draft the article. All authors read and approved the final article.
This research was supported by grant Chi-Mei Foundation Hospital Research
9779 from the Chi-Mei Medical Center. The authors have no other funding
or conflicts of interest to disclose.
Supplemental digital content is available for this article. Direct URL citations
appear in the printed text and are provided in the HTML and PDF versions of
this article on the journal’s website (www.pidj.com).
Address for correspondence: Chih-Hung Chang, PhD, Buehler Center on Aging,
Health & Society, Northwestern University Feinberg School of Medicine,
750 N. Lake Shore Dr., Suite 601, Chicago, IL 60611. E-mail: chchang@
northwestern.edu.
Copyright © 2013 by Lippincott Williams & Wilkins
ISSN: 0891-3668/13/3204-0320
DOI: 10.1097/INF.0b013e31827e111e
were collected from patients with febrile illnesses suspected to be
DF based on the guidelines of published literature.7–10 Routine tests
for white blood cell count (WBC), platelet count, aspartate ami-
notransferase (AST), alanine aminotransferase (ALT) and C-reac-
tive protein (CRP) values were done. Serologic confirmation of
DF was then obtained (Dengue Duo IgM & Rapid Strips; Panbio,
Queensland, Australia)13,14 to detect dengue-specific antibodies. Of
the 177 pediatric patients evaluated, 69 were categorized as having
DF based on a positive strip test and 108 as having nondengue fever
(non-DF) based on a negative strip test.
Laboratory and Clinical Characteristics
Guided by the DF literature,7–10 we selected 24 DF-related
clinical features (7 laboratory and 17 clinical) that capture clinical,
historical and laboratory indicators to form the initial set of items
to screen for DENV infection.
From 177 patients clinically suspected in the emergency
department of DENV infection, data were selected to construct
a scale to screen for DENV infection. Some clinical data were
obtained from the patients’ medical records, a personal history of
dengue infection, a family history of dengue infection, a personal
history of mosquito bites within the previous 2 weeks, the patient’s
WBC and platelet counts, AST, ALT, and CRP concentrations
and whether the patient had a high fever (≥39°C), biphasic fever,
rash, petechia, retro-orbital pain, bone pain (arthralgia), headache,
myalgia, abdominal pain, anorexia, occult hematuria, stool occult
blood, cough, sore throat, soft (watery) stool or flushed skin.
These clinical features were further categorized using cut-
offs based on the consensus of a panel of 7 experienced emergency
room pediatricians and on the literature15 for clinical simplicity
and ease of data analysis and interpretation. For example, WBC
and platelet counts were scored on a 0–2 polytomous scale, with

320 | www.pidj.com The Pediatric Infectious Disease Journal  •  Volume 32, Number 4, April 2013
The Pediatric Infectious Disease Journal  •  Volume 32, Number 4, April 2013 Dengue Fever

a higher number indicating a tendency for DF. Some other clinical median age: 5 years old; age range: 0–16 years) who were admitted
features were dichotomized to indicate their presence or absence (0 at a mean of 2.97 days (range: 1–9 days) after fever onset, but had
= “not present” versus 1 = “present”). with no evidence of DENV infection from medical records were
used as a reference group (non-DF group).
Data Analysis A χ2 test at the α level of 0.05 showed that most of the 17
The categorized clinical features, both dichotomous and DF-related symptom characteristics from the 177 pediatric patients
polytomous response items, were analyzed using the Rasch partial (69 diagnosed with DF) were statistically nonsignificant, which
credit model16,17 implemented in the Winsteps computer program.18 indicated that some of these 17 items might be useful for detecting
The accuracy of predicting DF was examined using the receiver DF in children but that they are not sufficient in clinical practice.
operating characteristic (ROC), in which the area under the curve is This means that items with high accuracy in discriminating DF
a graphic plot of the sensitivity versus (1 − specificity) for a binary should be verified using statistically sophisticated methods, such as
classifier system as its discrimination threshold varied.19 Rasch analysis (see Table, Supplemental Digital Content 1, http://
links.lww.com/INF/B424).
Unidimensionality Feature Needed to Measure DF Most patients with abnormal laboratory results had slightly
A Rasch model for partial credit scoring (or partial credit elevated CRP levels (<25 mg/L) (see Table, Supplemental Digital
model) suitable for multiresponse category items, such as WBC, Content 2, http://links.lww.com/INF/B425). As can be seen, a sig-
platelets, AST, ALT and CRP in the DF scale, was used. The Infit nificantly higher proportion of DF group patients had elevated AST
statistic was used to examine whether the data fit the model’s speci- and ALT levels and lowered platelet and WBC counts. In contrast, a
fication. The Infit mean square (MNSQ) statistics are χ2 statistics significantly higher proportion of non-DF group patients had rela-
divided by their degrees of freedom; they have an expected value tively normal WBC and platelet counts.
of 1 when the data fit the Rasch model.20 Items were considered to
measure the same construct (eg, tendency for DF) if their MNSQ Unidimensional Requirement
values were between 0.5 and 1.5. The item “history of dengue” was excluded from further analy-
sis because all patients had the same response: “none.” Nine non-DF–
Statistical Analysis specific items (history of mosquito bites, headache, myalgia, hematu-
The cutoff points of the bimodal distribution for the DF ria, stool blood, cough, sore throat, soft stool and flushed skin) with
and non-DF groups were obtained from ROC curve analysis using vague characteristics (nonspecific symptoms consistent with DF) that
MedCalc for Windows 9.5.0.0 (MedCalc Software, Mariakerke, did not fit the Rasch measurement model (ie, item MNSQ statistics
Belgium). Additionally, we used the ROC curves to examine the fell outside the 0.5–1.5 range) were also removed. Therefore, a total
effectiveness of each unidimensional combination of symptoms of 14 items (5 laboratory and 9 symptom items) were finally retained.
and laboratory characteristics; both bimodal distributions and scat-
tered characteristics by item frequency (ie, item rarely present: the The Unidimensional 14-item (DF-14) Scale
number of responses: the greater the number of responses to an The MNSQ statistics of the 14 retained items (the DF-14
item, the less difficult the item) and the root mean square error (ie, scale) were all within the prespecified (0.5–1.5) range (Fig. 1).
Infit MNSQ20 in Rasch analysis) were examined to verify whether The variance explained by these items was high, which suggested
the constructed scale was valid and reliable for assessing the likeli- a single underlying dengue-like construct. The empirical variance
hood of having DF. The point-biserial correlation coefficient21 was explained by the model was 77.9%, which is greater than the cut-
computed to examine the association between the dichotomous off point of 60%.22 The model’s variance of 79.0% was similar to
variables (eg, DF versus non-DF) and DF clinical features (eg, the empirical variance, which indicated that the data were a good
symptom or laboratory characteristic). fit with the Rasch model’s specifications. The unexplained vari-
ance in the first contrast generated by the Winsteps analysis had
RESULTS an eigenvalue of 1.9 (less than the cutoff point of 3) and accounted
for only 2.9% (less than the cutoff point of 5%) of the unexplained
Clinical Symptoms and Laboratory Characteristics variance,22 which suggested that the DF-14 scale can be used as an
Sixty-nine pediatric patients (40 [58.0%] male; median age: interval unidimensional scale for assessing the tendency for DF.
10 years old; age range: 0–16 years) clinically diagnosed with DF Item difficulties (x-axis) and item point-biserial correlation
and admitted to the hospital after the onset of fever (mean: 4.53 coefficients (y-axis) are plotted against each other (Fig. 2) show 2
days; range: 1–10 days) (DF group) were included in this study distinct clusters of clinical and laboratory characteristics: 5 labo-
(Table 1). One hundred eight pediatric patients (61 [56.5%] male; ratory characteristics at the top and 9 symptom characteristics at
the bottom. The highest 5-point biserial correlation coefficients
were for WBC (0.73), platelets (0.70), AST (0.69), CRP (0.63)
TABLE 1.  Demographic Characteristics of the Study and ALT (0.49). The lowest were for bone pain (arthralgia) (0.19),
Sample retro-orbital pain (0.20), petechia (0.28), biphasic fever (0.28) and
abdominal pain (0.32). Bone pain was the rarest (least chance to
Non-DF DF Total present) symptom with a difficulty of 2.38 logits (a unit with log
Variable N % N % N % P
odds), and CRP (<25 mg/L) was the most common with a difficulty
of −4.33 logits in this sample of patients.
Gender Female 47 43.5 29 42 76 42.9 0.845
Male 61 56.5 40 58 101 57.1 ROC Curves of 3 Possible Scales Proposed for
Age group 0–4 yrs 48 44.4 11 16.2 59 33.5 0.005
5–9 yrs 24 22.2 20 29.4 44 25
Accurately Predicting a Tendency for DF
9–16 yrs 36 33.3 37 54.4 73 41.5 Because simplicity of use in clinical practice is desirable,
P values were determined using the χ2 test.
2 shorter scales were constructed: the DF-2 scale (only WBC and
DF indicates patients with Dengue Duo IgM Rapid Strips test(+); Non-DF, patients platelet counts) and the DF-11 scale (AST, ALT and CRP excluded
with Dengue Duo IgM Rapid Strips test(–) from the DF-14 scale). Three ROC curves, 1 for each scale, were

© 2013 Lippincott Williams & Wilkins www.pidj.com | 321

Lai et al The Pediatric Infectious Disease Journal  •  Volume 32, Number 4, April 2013
FIGURE 1.  Item-person map incorporated with Infit MNSQ for items. SD, standard deviation.
plotted for comparisons (Fig. 3). The DF-14 curve surpassed the
DF-2 curve but did not significantly differ from the DF-11 curve.
The sensitivity and specificity at the cutoff point
of −1.15 logit (ie, the probability of a tendency for DF = 0.76 =
exp(log odds) exp( −1.15)
1− =1− = 1 − 0.24 ) for the
(1 + exp(log odds)) (1 + exp( −1.15))
DF-14 scale was 0.76 and 0.86, respectively. The area under the
ROC curve was 0.93 (95% confidence interval: 0.89–0.97) for
the DF-14 scale, 0.92 (95% confidence interval: 0.87–0.96) for
the DF-11 scale and 0.86 (95% confidence interval: 0.79–0.91)
for the DF-2 scale.
A bimodal diagnosis distinguishing DF and non-DF using
the ROC curve analysis was plotted (see Fig., Supplemental Digi-
tal Content 3, http://links.lww.com/INF/B426); the cutoff point
was −1.15 logits.
DISCUSSION
Unique Feature of This Study
We found that using a Rasch model to verify a simple but clini-
cally useful screening tool for detecting DF in children is promising.
Unlike a traditional univariate approach, which does not indicate that
any sign or symptom is definitively predictive of DF, the Rasch model
used in this study did indicate signs and symptoms of a tendency for
DF. For instance, patients with anorexia, abdominal discomfort or
diarrhea and signs of pharyngitis were deemed to have gastrointestinal
problems and respiratory tract infections when using the traditional
univariate approach.7–9,23–25 In contrast, patients with anorexia, abdom-
inal pain, high fever and biphasic fever would be deemed as having a
tendency for DF when using the Rasch model proposed in this study.
Main Findings
WBC and platelet counts were the 2 items in the DF-14 scale
that had the most power to predict DF in children. The DF-14, DF-11
322 | www.pidj.com
and DF-2 scales, despite their having a different number of items,
were found to be measuring the same “tendency for DF” construct.
The DF-2 scale is considered acceptable for its simplicity in practice,
even though it is less accurate than the DF-11 and DF-14 scales.
Early Detection of DF
Bone pain with 2.43 logits was the least common symptom,
and CRP (<25  mg/L, a laboratory characteristic) with −4.77
logits was the most common on the DF-14 scale. The symptom
characteristics, with item difficulties ranging from −1.00 to +3.00
logits, for DF detection are less common than changes in laboratory
characteristics, with item difficulties ranging from −4.77 to +1.00
logits. This suggests that using signs or symptoms, but not both, to
distinguish DF from non-DF is insufficient before delivering initial
therapy.7,8 Considering and evaluating the symptom characteristics
and the laboratory characteristics together (DF-11 and DF-14)
are more effective for the early detection of DF. This is consistent
with other study findings23,26 that used different variables together
to distinguish DF from non-DF. Similar to other studies,6,23,25,26 we
found that thrombocytopenia and leukopenia were highly associated
with DF. WBC and platelet counts are easily obtainable in primary
care settings; thus, a combination of laboratory variables (low
platelet or WBC counts) for early prediction of DENV infection
is feasible.
To examine which items provide the highest accuracy in
detecting DF, Rasch analysis provides a way (the nearer to the mean
of the patients, the better the predictive power of an item) to explore
them.27–29 First, we found that the estimated patient measures had a
mean of −1.66 logits and a standard deviation of 1.54 logits. Sec-
ond, we examined the item difficulties (rare occurrence) in Figure
2. WBC (−1.12 logits) and rash (−1.17 logits) are close to the cutoff
point of −1.15 logits (slightly lower than the mean) for discriminat-
ing DF from other febrile diseases (Fig. 4). This means that many
key items surrounding the cutoff value make the standard error of
the measurement smallest (ie, the variance will be the largest owing
© 2013 Lippincott Williams & Wilkins
The Pediatric Infectious Disease Journal  •  Volume 32, Number 4, April 2013 Dengue Fever
FIGURE 2.  Characteristics scattered by item difficulties and
point-biserial correlation coefficients for symptoms and labo-
ratory characteristics. Petechia and biphasic fever have the
same values for item difficulties and point-biserial correlation
coefficients.
to SE = 1/variance in Rasch analysis) and the magnitude of the
information the greatest; thus, the predictive criteria used in this
study are precise enough.
Cutoff Thresholds Determined for Distinct Scales
Before conducting a costly and complicated DF confirma-
tion test, it is to contain costs by using some easily obtainable labo-
ratory data, such as the WBC and the platelet counts (DF-2 scale)
to quickly and reasonably predict the tendency for DF in children.
This is particularly important for healthcare professionals working
for a medical system with constrained resources. Raw scores of 2,
FIGURE 3. Receiver operating curves of the DF-2, DF-11,
and DF-14 scales for detecting dengue fever. AUC indicates
area under the curve.
© 2013 Lippincott Williams & Wilkins
FIGURE 4. A decision tree showing both conservative and
aggressive approaches to diagnosing dengue fever before
doing an expensive and time-consuming confirmation test.
AUC indicates area under the curve.
3 and 5 for DF-2, DF-11 and DF-14, respectively, were considered
reasonable thresholds for predicting DF.
Decision Tree Used for Practitioners in DF
Discrimination
A diagnosis based on a single criterion (only one of the DF
tests is met) is recommended for endemic areas, but a diagnosis
based on at least 2 criteria (2 of the DF tests are met) is recom-
mended for nonendemic areas (Fig. 4). We conclude that using the
measure-based information will be more efficient and cost-effec-
tive than the traditional univariate approach to DF screening and
diagnosis.
This study has some limitations. First, the study lacks sero-
type identification for all patients. Second, it is a hospital-based
design. The physical signs, symptoms and laboratory data of the
patients were observed and recorded only during a patient’s first hos-
pital visit before therapy. Studies on the natural history of dengue
and nondengue disease are needed. Third, the sample size was rela-
tively small; even we were able to construct a highly accurate dis-
criminatory screening feature with an accuracy of >0.80 area under
the curve for all 3 simplified scales (DF-2, DF-11 and DF-14). How-
ever, our findings need confirmation by prospective studies in other
areas affected by DF. The identification of potentially useful distin-
guishing features in patients with a DENV infection in this study
has implications for healthcare professionals who practice at clinics
in rural areas with limited resources. They often demand a simple
and efficient method to help detect DF at an early stage because the
www.pidj.com | 323
Lai et al The Pediatric Infectious Disease Journal  •  Volume 32, Number 4, April 2013
dengue IgM/IgG antibody, polymerase chain reaction and virus iso-
lation tests are not readily available in such remote areas.
Future Research and Prospects
Additional investigation is needed of the effects of the cat-
egorization of the scales for WBC and platelet counts because it
is important to understand whether fewer or more categories (eg,
0–1, 0–2, 0–3 or even 0–5 points of a Likert-type scale) will make
the screening tool more accurate or not. The decision-support rule
for adult DF detection, similar to that shown in Figure 4, would
also be a logical next step to establish and evaluate in larger-scale
studies.
REFERENCES
1. World Health Organization. Dengue and Dengue Haemorrhagic Fever. Fact
sheet No. 117. Geneva: World Health Organization; 2002.
2. Henchal EA, Putnak JR. The dengue viruses. Clin Microbiol Rev.
1990;3:376–396.
3. Gubler DJ. Dengue and dengue hemorrhagic fever. Clin Microbiol Rev.
1998;11:480–496.
4. Calisher CH. Persistent emergence of dengue. Emerging Infect Dis.
2005;11:738–739.
5. Guzmán MG, Kourí G. Dengue: an update. Lancet Infect Dis. 2002;2:33–42.
6. Wilder-Smith A, Schwartz E. Dengue in travelers. N Engl J Med.
2005;353:924–932.
7. Phuong HL, de Vries PJ, Nga TT, et al. Dengue as a cause of acute undif-
ferentiated fever in Vietnam. BMC Infect Dis. 2006;6:123.
8. Nunes-Araújo FR, Ferreira MS, Nishioka SD. Dengue fever in Brazilian
adults and children: assessment of clinical findings and their validity for
diagnosis. Ann Trop Med Parasitol. 2003;97:415–419.
9. Hammond SN, Balmaseda A, Pérez L, et al. Differences in dengue severity
in infants, children, and adults in a 3-year hospital-based study in Nicara-
gua. Am J Trop Med Hyg. 2005;73:1063–1070.
10. Potts JA, Rothman AL. Clinical and laboratory features that distinguish
dengue from other febrile illnesses in endemic populations. Trop Med Int
Health. 2008;13:1328–1340.
28. Chien TW, Wang WC, Huang SY, et al. A web-based computerized adaptive
11. Lord FM. Applications of Item Response Theory to Practical Testing Prob-
lems. Hillsdale, NJ: Lawrence Erlbaum; 1980.
12. Stout WF. A new item response theory modeling approach with applications
to unidimensionality assessment and ability estimation. Psychometrika.
1990;55:293–325.
324 | www.pidj.com
View publication stats
13. Kittigul L, Suankeow K. Use of a rapid immunochromatographic test for
early diagnosis of dengue virus infection. Eur J Clin Microbiol Infect Dis.
2002;21:224–226.
14. Vaughn DW, Nisalak A, Kalayanarooj S, et al. Evaluation of a rapid immu-
nochromatographic test for diagnosis of dengue virus infection. J Clin
Microbiol. 1998;36:234–238.
15. Lai WP, Chien TW, Lin HJ, et al. Clinical prediction of pediatric dengue virus
infection in Taiwan—a Rasch scaling approach. HealthMED. 2011;5(suppl
6):1977–1985.
16. Masters GN. A Rasch model for partial credit scoring. Psychometrika.
1982;47:149–174.
17. Rasch G. Probabilistic Models for Some Intelligent and Attainment Tests.
Copenhagen, Denmark: Institute of Educational Research; 1960.
18. Linacre JM. WINSTEPS [computer program]. Available at: http://www.
WINSTEPS.com. Accessed November 9, 2012.
19. Fawcett T. An introduction to ROC analysis. Pattern Recognition Lett.
2006;27:861–874.
20. Wright BD, Linacre JM. Reasonable mean-square fit values. Rasch Measure
Trans. 1994;8:370–371.
21. Glass GV, Hopkins KD. Statistical Methods in Education and Psychology.
3rd ed. Boston: Allyn & Bacon; 1995.
22. Linacre JM. A User’s Guide to WINSTEPS. Chicago, IL: MESA Press;
2009.
23. Wilder-Smith A, Earnest A, Paton NI. Use of simple laboratory features to
distinguish the early stage of severe acute respiratory syndrome from den-
gue fever. Clin Infect Dis. 2004;39:1818–1823.
24. Lee MS, Hwang KP, Chen TC, et al. Clinical characteristics of dengue and
dengue hemorrhagic fever in a medical center of southern Taiwan during the
2002 epidemic. J Microbiol Immunol Infect. 2006;39:121–129.
25. Liu HW, Ho TL, Hwang CS, et al. Clinical observations of virologically
confirmed dengue fever in the 1987 outbreak in southern Taiwan. Gaoxiong
Yi Xue Ke Xue Za Zhi. 1989;5:42–49.
26. Watt G, Jongsakul K, Chouriyagune C, et al. Differentiating dengue virus
infection from scrub typhus in Thai adults with fever. Am J Trop Med Hyg.
2003;68:536–538.
27. Chien TW, Hsu SY, Tai C, et al. Using Rasch analysis to validate the revised
PSQI to assess sleep disorders in Taiwan’s hi-tech workers. Community
Ment Health J. 2008;44:417–425.
testing (CAT) to assess patient perception in hospitalization. J Med Internet
Res. 2011;13:e61.
29. Chien TW, Lai WP, Lu CW, et al. Web-based computer adaptive assessment
of individual perceptions of job satisfaction for hospital workplace employ-
ees. BMC Med Res Methodol. 2011;11:47.
© 2013 Lippincott Williams & Wilkins