American Journal of Epidemiology Vol. 136, Mo.

2
CopyrightO 1992 by The Johns Hopkins University School of Hygiene and Pubfc Health Printed in U.S.A.
All rights reserved

The Predisposing and Protective Factors against Dengue

Virus Transmission by Mosquito Vector

Ying-Chin Ko, MekJu Chen, and Shu-Mei Yen

An outbreak of dengue fever occurred in Taiwan between 1987 and 1988. The
highest attack rate among adults was estimated at 5.6% in the city of Kao-hsiung. A
case-control study was carried out to determine the risks of contracting dengue infection
and to identify protective factors against the infection. One hundred dengue patients of

Downloaded from aje.oxfordjournals.org by guest on February 13, 2011

the authors' hospital who were diagnosed by virologic or serologic tests constituted
the case group. Each dengue patient was matched to a control patient of the same age
and sex who had been diagnosed as suffering from a non-vector-bome disease on the
same day as the dengue patient. Of the household protective measures against dengue
infection prior to the occurrence of illness, the adjusted odds ratio, estimated by stratified
analysis, was lower for people who lived in screened houses (odds ratio = 0.58, 95%
confidence interval 0.36-0.92) as compared with inhabitants of unscreened houses.
The odds ratio was as low as 0.18 (95% confidence interval 0.06-0.56) for people
whose homes were fully screened with door screens opening outwardly. Patients who
lived near markets and/or open sewers or ditches were running a risk of dengue
infection 1.8 (95% confidence interval 1.3-2.4) times higher than those who lived
elsewhere. To control dengue outbreaks, the authors recommend that special attention
should be devoted to the reduction of outdoor vector sources. Full screening, especially
outwardly opening screen doors, seems to be an individual's best protection against
dengue fever. Am J Epidemiol 1992;136:214-20.

dengue; mosquitoes

An estimated population of over 1.5 bil-
lion persons is at risk of infection with den-
gue virus (I). Historically, Southeast Asia
has been the area of highest endemicity, but
the Pacific Islands and the Caribbean Basin
have also become endemic areas since the
1970s, as has Africa in the 1980s. A number
of outbreaks in the past decade were docu-
mented in the United States, Mexico, South
America, China, and Australia (2-6). Nu-
merous epidemics were recorded in Taiwan
Received for publication April 25, 1991, and in final
form January 31, 1992.
From the School of Public Health, Kao-hsiung Medical
College, Kao-hsiung, Taiwan, ROC 807. (Reprint requests
to Dr. Y C Ko at this address.)
The authors thank the physicians and nurses of Emer-
gency Service, Kao-hsiung Medical College Hospital, for
their assistance, and they also thank Dr. F F. Czinkotai for
his editorial assistance with this article.
over the last century. Following an island-
wide epidemic in 1942-1943, dengue fever
did not, however, recur on the island for
about 40 years, except for a single outbreak
in 1981 on Liuchiu, an islet off southern
Taiwan. The latest epidemic in Taiwan oc-
curred in the fall of 1987, with sporadic cases
being reported up to the end of 1989 (7).
The highest attack rate among adults of
approximately 5.6 percent observed in the
southern city of Kao-hsiung in 1988 (8) was,
however, significantly lower than the rates
reported in Puerto Rico (18 percent) in 1977
(9) and in Mexico (10.8-45.6 percent) in
1984 (10). This case-control study of age-
and sex-matched pairs was designed to in-
vestigate the circumstances of a dengue
outbreak, especially the predisposing and
protective factors involved.

214

MATERIALS AND METHODS
Study population
Kao-hsiung city is located in tropical
southern Taiwan and has had a population
density as high as 8,880 persons/km2 (1.36
million persons/153 km2) in 1987. The cases
in this study have come from the Kao-hsiung
Medical College Hospital, which our govern-
ment regards as a top-grade teaching hospi-
tal, used by patients from all socioeconomic
categories. Patients who were older than 15
years, lived in Kao-hsiung city, and visited
the hospital's emergency department during
October and November of 1988 were de-
fined as the study population. Of the patients
found to suffer from dengue-like illness, 121
were selected for the study group by simple
random sampling. Their diagnoses were
confirmed by virus isolation and/or sero-
logic antibodies. Each dengue patient was
matched to a control patient of the same age
and sex, who had been diagnosed to suffer
from a non-vector-borne disease on the
same day as the dengue patient. A few den-
gue patients could be matched only to con-
trols whose age differed by up to ± 3 years.
Children were excluded, because their dif-
ferential diagnosis of dengue fever was rela-
tively difficult and their responses to ques-
tions could not be relied upon, unless their
parents helped.
Dengue-like illness was clinically defined
as an acute febrile illness with one or more
of the following signs or symptoms (9): head-
ache, retroorbital pain, muscle ache, or rash.
Hemorrhagic signs included one or more of
the following: petechiae, ecchymoses, he-
matemesis, melena, epistaxis, gingival bleed-
ing, menorrhagia, and thrombocytopenia
(platelet count, <100,000/mm3). The con-
ditions the control patients suffered from
were diarrhea, accidents, kidney stones,
heart disease, and cerebrovascular disease.
Case definition
Laboratory confirmation of dengue re-
quired serologic evidence and/or isolation
of a dengue virus. A blood sample was col-
lected from each dengue patient and from
Factors against Dengue 215
each control patient and was tested by virus
isolation and/or their paired sera by hem-
agglutination inhibition. Dengue type 1 vi-
rus was predominant in this outbreak (7)
and was grown by inoculation of the C6/36
clone of Aedes albopictus cells (11). Then
the virus was identified by the indirect im-
munofluorescent assay with monoclonal
antibody (12). A fourfold or greater increase
in hemagglutination inhibition antibody be-
tween the acute and convalescent sera or the
isolation of virus was taken to confirm den-
gue infection. Patients with dengue-like ill-
ness whose diagnoses were not confirmed by
Downloaded from aje.oxfordjournals.org by guest on February 13, 2011
laboratory tests were excluded from the
study together with their matching controls.
Interview and home inspection
All 121 individuals in each group were
interviewed, and their homes were inspec-
ted to collect information relating to their
health and illness. The interviewers sought
historical data such as general demographic
characteristics, past or present existence of
mosquito-breeding sites on the premises (in-
cluding water-storage tanks, flower vases,
tin cans, containers, unused tires, water trays
of the refrigerator), date of eradication of
mosquito-breeding sites, use of screens on
the windows and doors, types of screen fit-
ted, whether or not the patients had traveled
abroad, if pets were present, use of mosquito
nets for sleeping, indoor application of in-
secticides, use of traps or mosquito coils,
household size, type of housing, and height
of house (e.g., either a traditional family
house consisting of no more than four floors
in Taiwan or an apartment in high-rise
blocks with an indication of floor level). A
disease history of the previous 6 months was
taken for each member of the household.
Two survey teams visited the patients in
their homes at different times. Each team
included two trained inspectors. The work
of the first team was to interview patients
according to a structured questionnaire de-
veloped and evaluated by the authors. The
second team, unlike the first team, was un-
aware of patients' diagnostic status, and their
task was mainly to observe and appraise
216 Koetal.
vector-breeding sites, screens, and sanitary
conditions in the patients' living and work-
ing habitats, including possible marketplaces
and/or open sewers or ditches within 50
meters of the residence. With patients' con-
sent, we collected a complete set of data for
100 dengue sufferers and their matching
controls after they had been discharged from
our hospital. Data were not collected for the
remaining 21 pairs either because of an un-
traceable address or because the dengue sta-
tus was not confirmed in the laboratory.
Statistical analysis
Data emanating from different interview-
ers, inspectors, and patients were validated.
The odds ratio and 95 percent confidence
interval for matched case-control pairs
were used to identify the predisposing or
protective factors involved in dengue occur-
rence (13). In case of a zero in discordant
pairs, an unmatched estimation was made.
The Cochran-Mantel-Haenszel statistic was
applied for stratified analyses; the x2 for
trends was performed when the significant
study variable was ranked at a multiple level
of exposure for matched or unmatched data
estimation (14) using the Statistical Analysis
System computer packages (SAS release
6.04). For hypothesis testing, an a-level of
0.05 was chosen.
RESULTS
Of the 100 confirmed dengue cases, 55
were male patients and 45 were females,
with an average age of 32 ± 10 years (range,
17-69 years). The average age of the control
group was 32 ± 11 years (range, 19-67
years). There was no significant difference
between the residential districts or educa-
tional levels of dengue patients and their
matching controls.
The risk of contracting dengue infection
did not appear to be influenced by factors
such as the presence of mosquito-breeding
sites on the premises, if pets were being
raised, household size, and the type of hous-
ing the patients lived in. There was also no
difference between patients with or without
jobs or between students and nonstudents.
Interviewees spending the daytime at work
or at school did not seem to have been
exposed to an increased risk of dengue in-
fection. A few people had traveled abroad.
Only six of the 200 interviewees worked in
a place with screened doors. No significant
differences were found between the protec-
tive measures adopted in the work environ-
ments of dengue patients and their controls.
However, factors such as screens on houses
or the existence of a neighboring market
and/or open sewers or ditches appeared to
be related to dengue infection (table 1).
Downloaded from aje.oxfordjournals.org by guest on February 13, 2011
In all, 173 of the 200 people interviewed
(86.5 percent) lived in houses with screens
on at least some windows and/or doors, 81
dengue patients (81 percent) and 92 controls
(92 percent). Of all preventive measures used
by the patients, living in a screened house
was found to be the only effective measure
against dengue infection. The ratio by pairs
of discordant dengue patients living in
screened houses to controls living in
screened houses was 6 to 17; i.e., the odds
ratio was 0.35 (95 percent confidence inter-
val 0.14-0.85; table 2). Moreover, the odds
ratio was as low as 0.25 (95 percent confi-
dence interval 0.08-0.79) in cases where in-
terviewees lived in houses in which all the
windows and doors were screened (Arena <
0.05; table 2). Of the 173 patients living in
screened houses, in 111 cases all windows
and doors were screened, and in 33 cases the
door screens opened outwardly (for 10 of 52
dengue patients and for 23 of 58 controls).
As for the extent of screening, the odds ratios
were 0.18 for patients living in fully screened
houses with outwardly opening screen doors,
0.48 for those living in fully screened houses
with non-outwardly opening screen doors,
0.37 for those living in partially screened
houses, and 1 (reference category) for those
inhabiting unscreened houses (Arcnd = 0.01;
table 3).
The ratio by pairs of discordant dengue
patients living in the neighborhood of a mar-
ket and/or an open sewer or ditches to con-
trols living under similar conditions was 39
to 14; i.e., the odds ratio was 2.8 (95 percent
confidence interval 1.6-5.0; table 4). More-
 Factors against Dengue 217

TABLE 1. Estimated odds ratio of dengue infection by predisposing factors in case/control groups: Kao-
hslung, Taiwan, 1989
Case Control
Risk factors Odds ratio
(n) in)
Educational level
£12 years 73 64 1.44 (0.78-2.63)*
>12 years 27 34 1.0
Sodoeconomic status
Low 7 10 0.68(0.25-1.86)
Middle + high 93 90 1.0
Household size
£36m 2 /pereon 70 64 1.31(0.73-2.37)
>36 rrr'/person 30 36 1.0
Height of house
<3 floors 85 76 1.79 (0.88-3.66)
>3 floors 15 24 1.0

Downloaded from aje.oxfordjournals.org by guest on February 13, 2011

At work or school
Yes 87 86 1.09(0.48-2.45)
No 13 14 1.0
Breeding sites
Yes 32 36 0.84 (0.47-1.50)
No 68 64 1.0
Mosquito nets
Yes 5 5 1.0 (0.28-3.57)
No 95 95 1.0
Insecticide application
Indoors
Yes 70 67 1.15(0.63-2.09)
No 30 33 1.0
Mosquito traps
Yes 47 43 1.18(0.67-2.05)
No 53 57 1.0
Mosquito coils
Yes 51 45 1.27(0.73-2.22)
No 49 55 1.0
Pets present
Yes 27 28 0.95(0.51-1.77)
No 73 72 1.0
Screen on houses
Yes 81 92 0.37 (0.15-0.89)t
No 19 8 1.0
Markets and/or open
sewers
Yes 70 45 2.85(1.59-5.1 Oft
No 30 55 1.0
* Numbers In parentheses, 95% confidence interval.
t Odds ratio was 0.58 (95% confidence interval 0.36-0.92) when adjusting for the factor "existence of market and/or open
sewer" by stratified analysis.
t Odds ratio was 1.8 (95% confidence Interval 1 3-2.4) when adjusting for the factor "screen on houses" by stratified analysis.

over, the odds ratio for dengue infection was
as high as 3.7 (95 percent confidence interval
1.1-12.3) for patients living in the neighbor-
hood of a market as well as of open sewers
or ditches (Arena = 0.001; table 4).
According to stratified analysis used to
adjust for the presence or absence of a mar-
ket and/or open sewers or ditches in pa-
tients' habitat, the adjusted odds ratio was
lower for those living in screened houses
(odds ratio = 0.58, 95 percent confidence
interval 0.36-0.92) as compared with inhab-
itants of unscreened houses. The same
analysis showed that patients living near
markets and/or open sewers or ditches were
more likely to contract dengue infection
218 Koetal.

TABLE 2. Estimated odds ratio of discordant people living In screened houses to unscreened houses, by
pairs, for dengue infection: Kao-hsiung, Taiwan, 1989
No. of cases of dengue fever In
Screen on matched pairs among controls
doors and with screens on doors and Odds ratio* Overall odds ratio
windows wtndows
of cases
Full Partial None

Fun 30 17 0.25 (0.08-0.79)t 0.35(0.14-0.85)

Partial 19 9 0.60(0.15-2.5)
None 12 5 1.0$ 1-0$
• x* for trend: 4.7; p < 0.05.
f Numbers in parentheses, 9 5 % confidence Interval.
$ Referent category.

TABLE 3. Estimated odds ratio of people living in fully screened houses with outwardly opening door

Downloaded from aje.oxfordjournals.org by guest on February 13, 2011

screens for dengue infection: Kao-hsiung, Taiwan, 1989
Screen on doors and
Cases Controls Odds ratio*
windows

Full with outwardly

opening screen
doors 10 23 0.18(0.06-0.56)t
Full with non-outwardly
opening screen door 41 36 0.48(0.19-1.23)
Partial screening 30 33 0.37(0.15-1.01)
None 19 8 1.0|
• x* 'or trend: 6.0, p =• 0.01.
t Numbers In parentheses, 95% confidence interval.
t Referent category.

TABLE 4. Estimated odds ratio of discordant people Irving in the neighborhood of markets and/or open
sewers, by pairs, for dengue infection: Kao-hsiung, Taiwan, 1989

Market No. of cases of dengue fever In

and/or matched pairs among controls with Overall odds
market and/or open sewer Odds ratio'
open sewer ratio
of cases Both Either None

Both 4 11 3.7 (1.1-12.3)t 2.8(1.6-5.0)

Either 21 28 2.5 (1.3-5.0)
None 11 16 1.0$ 1.0*
* x * tor trend: 10.6; p = 0.001.
t Numbers In parentheses, 95% confidence interval.
t Referent category.

(odds ratio = 1.8, 95 percent confidence
interval 1.3-2.4) than were those who did
not live in such a habitat, even when adjust-
ing for screens in their homes.
Family members sharing home with den-
gue patients numbered 398, of whom 52 (13
percent) were reported to have been suffer-
ing from dengue infection that in most cases
was diagnosed by a physician in the preced-
ing months. The households of controls in-
cluded 416 members, of whom only 25 (6
percent) reported dengue infection. The rel-
ative risk ratio was 2.2 (95 percent confi-
dence interval 1.4-3.4). Thus, members of
dengue patients' households were also ex-
posed to an increased risk of dengue infec-
tion, the source of infection being found in
patients' habitat, according to family clusters
versus dengue infection analysis.
DISCUSSION
In this study, dengue was probably con-
tracted at home rather than at work or at

school. Researchers from Puerto Rico ar-
rived at a similar conclusion (9). We found
that poor outdoor environmental sanitation
was the main predisposing factor, while
screens on doors and windows were the main
protective factor against dengue infection by
outdoor vector. Other possible predisposing
and protective factors did not seem to differ
between patients' and controls' households,
despite several earlier reports to the contrary
(9, 10, 15). In Taiwan, environmental sani-
tation is poor in areas adjacent to traditional
markets, and mosquito-breeding sites (dis-
carded bottles, tin cans, boxes) here tend to
be abundant. Visits to the market by persons
suffering from viremia would certainly ex-
pose local residents to an increased risk of
dengue infection via vector. In general, peo-
ple tend to be unconcerned about outdoor
sanitation, which they feel is a public rather
than a private affair, especially in common
basements beneath residential buildings.
In general, Aedes aegypti does not utilize
sewers as a larval habitat. Seasonal dengue
epidemics tend to be associated with rainy
periods in most tropical areas. Gubler (1)
found this to be somewhat surprising, be-
cause the principal larval habitats for A.
aegypli were clay water-storage jars usually
kept indoors in Southeast Asia. He suggested
that the rainy season was more conducive to
the adult mosquito, thus enhancing the
probability that infected mosquitoes survive
the extrinsic incubation period and transmit
the virus to other persons (1).
However, reports from Singapore (16) in-
dicated that water reservoirs such as sewer
inlets, catch basins, cesspools, storm drains,
and street and roof gutters can also be A.
aegypti breeding sites. In this study, a rela-
tion between dengue cases and open sewers
or ditches was observed, indicating that out-
door sites in the rainy season cannot be
ignored. Summer and autumn typhoons in
Taiwan fill up a shallow open sewer or the
basement of a building with clean rain water,
providing an ideal larval habitat for A. ae-
gypti. We observed from multivariate analy-
sis in an earlier ecologic study (7) that,
among the usual risk factors affecting the
proliferation, a prior month of precipitation
Factors against Dengue 219
can best predict an outbreak of dengue fever.
A. aegypti larvae were found in water col-
lecting in discarded automobile tires, of
which there were more than 100,000 gath-
ered from the streets of Kao-hsiung city
during the peak period of the 1988 epidemic
(7). Moreover, another report indicated that
the outdoor density of A. aegypti larvae was
77 percent of the total density (17). There-
fore, we suggested that the 1988 outbreak
was sustained from outdoor sources suitable
for the growth of A. aegypti and that the
presence of screens on doors and windows
confers a certain measure of protection
Downloaded from aje.oxfordjournals.org by guest on February 13, 2011
against transmission of the illness. Others in
Puerto Rico (9) and Mexico (10) have also
found that the absence of screens could be
collated with the occurrence of dengue
symptoms. In the present case-control study,
we were able to show that screens in general
had a protective effect, which was highest in
those cases where all windows were fully
screened and where door screens opened
outwardly. (Outwardly opening screens ap-
pear to impede mosquitoes perching on the
screen outside from entering the house with
an entering person.)
Thus, screens, which are common in Kao-
hsiung, seem to play a significant role in the
prevention of dengue fever. Of the 200 pa-
tients who participated in this study, 173
(86.5 percent) lived in houses that were at
least partially fitted with screens. Although
these statistics may not necessarily apply to
all residents, the use of screens in Kao-
hsiung was probably significantly more fre-
quent than in Puerto Rico (no data given)
or Mexico (from 13 percent to 34 percent)
and, hence, the relatively low attack rate
observed in Kao-hsiung. In all, 110 inter-
viewees (55 percent) lived in a house with
all doors and windows fully screened but,
surprisingly, the houses of only 33 persons
were fitted with outwardly opening door
screens. The data on people living in un-
screened houses and their dengue infection
rates (table 3) yielded an attributable risk
percentage of 63 percent, suggesting that
nearly 63 percent of dengue infection could
be eliminated if all people lived in ade-
quately screened houses.
220 Ko et ai.
Two species of vector for dengue trans-
mission are found in Taiwan: A. albopictus
and A. aegypti. The latter was demonstrated
to be the only vector to transmit dengue
virus in this outbreak (18). In general, A.
aegypti is a domestic mosquito, and its dis-
tribution is confined to urban areas of the
city. Vector control, including killing the
adult mosquito and eliminating larval habi-
tats, is essential for dengue prevention. At
present, insecticide spraying is used as a
routine measure during outbreaks. It is,
however, a futile exercise to keep on killing
mosquitoes in the presence of an almost
unlimited number of breeding sites, for the
larvae laid at these sites soon grow into adult
mosquitoes. Reduction of breeding sources
is the most effective method of vector con-
trol of the Aedes genus. Singapore research-
ers (16) showed the indoor density of A.
aegypti larvae to be 78.9 percent of the total,
indicating that special attention (including
health education) has to be paid to indoor
breeding sites. In Taiwan or the Americas,
however, where dengue activity was sus-
tained from outdoor sources, eliminating
outdoor breeding sites seems to be more
important for dengue control than eliminat-
ing indoor sites, although the latter practice
should not be abandoned. Since effective
vector control takes a long time to develop,
the best prevention for an individual during
an outbreak is the utilization of screens,
particularly outwardly opening door screens.
REFERENCES
1. Gubler DJ. Dengue. San Juan, Puerto Rico: San
Juan Laboratories, Dengue Branch, Division of
Vector-borne Viral Diseases, Center for Infectious
Diseases, 1986:233^0.
2. Hafkin B, Kaplan JE, Reed C, et al. Reintroduction
of dengue fever into the continental United States.
I. Dengue surveillance in Texas, 1980. Am J Trop
MedHyg 1982;31:1222-8.
3. Kaplan JE, Eliason DA, Moore M, et al. Epidemi-
ologic investigations of dengue infection in Mexico,
1980. Am J Epidemiol 1983;117:335-44.
4. Centers for Disease Control. Dengue in the Amer-
icas, 1985. (Editorial). JAMA 1987;257:166.
5. Guard RW, Stallman ND, Wiemers MA. Dengue
in the northern region of Queensland, 1981-1982.
MedJAust 1984; 140:765-9.
6. Li FS, Yang FR, Song JC, et al. Etiologic and
serologic investigations of the 1980 epidemic of
dengue fever on Hainan Island, China. Am J Trop
MedHyg 1986;35:1051—4.
7. Ko YC. Epidemiology of dengue fever in Taiwan.
(In Chinese). Kao Hsiung I Hsueh Ko Hsueh Tsa
Chih 1989;5:1-11.
8. Ko YC, Chen JW, Chang IC. Attack rate of dengue-
like illness among teachers in Kao-hsiung city,
1988. Kao Hsiung I Hsueh Ko Hsueh Tsa Chih
1989;5:129-31.
9. Morens DM, Rigan-Perez JG, Lopez-Correa RH,
Downloaded from aje.oxfordjournals.org by guest on February 13, 2011
et al. Dengue in Puerto Rico, 1977: public health
response to characterize and control an epidemic
of multiple serotypes. Am J Trop Med Hyg 1986;
35:197-211.
10. Dantes HG, Koopman JS, Addy CL, et al. Dengue
epidemics on the Pacific Coast of Mexico. Int J
Epidemiol 1988; 17:178-86.
11. Igarashi A. Isolation of a Singh's Aedes albopictus
cells clone sensitive to dengue and chikungunya
viruses. J Gen Virol 1978;40:531^M.
12. Henchal E, McCown JM, Seguim MC, et al. Rapid
identification of dengue virus isolates by using
monoclonal antibodies in an indirect immunoflu-
orescence assay. Am J Trop Med Hyg 1983;32:
164-9.
13. Rothman KJ. Matching. In: Rothman KJ, ed.
Modern epidemiology. Boston: Little, Brown and
Company, 1986:237-83.
14. Breslow NE, Day NE, eds. Classical methods of
analysis of matched data. In: Statistical methods in
cancer research. Vol 1. The analysis of case-control
studies. Lyon: International Agency for Research
on Cancer, 1980:162-88. (IARC scientific publi-
cation no. 32).
15. Foy HM, Limpakarnjanarat K, Korprasertisri S, et
al. Dengue fever in Thailand in the 1980s. Pre-
sented at the 61st annual meeting of the American
Epidemiological Society, San Diego, California,
March 14-15, 1988.
16. Chan KL. Singapore's dengue hemorrhagic fever
control program: a case study on the successful
control of Aedes aegypti and Aedes aJbopictus using
mainly environmental measures as a part of inte-
grated vector control. Tokyo: Southeast Asian
Medical Information Center, 1985:9-60. (SEAMIC
publication).
17. Lien JC. Entomological aspects of dengue fever.
Presented at the Kao-hsiung Medical College 1988
symposium on dengue fever, Kao-hsiung, Taiwan,
April 1988.
18. The epidemics of dengue in Taiwan, 1988. Epide-
miol Bull 1988,4:87-94.