American Journal of Epidemiology
ª The Author 2008. Published by the Johns Hopkins Bloomberg School of Public Health.
All rights reserved. For permissions, please e-mail: [email protected].
Vol. 167, No. 12
DOI: 10.1093/aje/kwn066
Advance Access publication April 3, 2008
Original Contribution
Geographic Prevalence and Multilevel Determination of Community-level Factors
Associated with Herpes Simplex Virus Type 2 Infection in Chennai, India
J. M. Jennings1,2, T. A. Louis3, J. M. Ellen1,2, A. K. Srikrishnan4, S. Sivaram2, K. Mayer5,6,
S. Solomon4, R. Kelly7, and D. D. Celentano2
1
Department of Pediatrics, School of Medicine, Johns Hopkins University, Baltimore, MD.
Department of Epidemiology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD.
3
Department of Biostatistics, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD.
4
Y. R. Gaitonde Center for AIDS Research and Education (Y.R.G. CARE), Chennai, India.
5
Division of Infectious Diseases, School of Medicine, Brown University, Providence, RI.
6
The Miriam Hospital, Providence, RI.
7
Family Health International, Dhaka, Bangladesh.
2
Received for publication April 17, 2007; accepted for publication February 28, 2008.
Herpes simplex virus type 2 (HSV-2) is one of the most prevalent sexually transmitted infections, and it increases
the risk of transmission of human immunodeficiency virus type 1 at least twofold. Individual-level factors are
insufficient to explain geographic and population variation in HSV-2, suggesting the need to identify ecologic
factors. The authors sought to determine the geographic prevalence and community-level factors associated with
HSV-2 after controlling for individual-level factors among slums in Chennai, India. From March to June 2001,
participants aged 18–40 years voluntarily completed a survey and were tested for HSV-2. Community characteristics were assessed through interviews with key informants and other secondary data sources. Multilevel nonlinear analysis was conducted. Eighty-five percent of eligible persons completed the survey; of these, 98%
underwent HSV-2 testing, producing a final sample of 1,275. Participants were of Tamil ethnicity, were predominantly female and married, and were on average 30 years old. Fifteen percent were infected with HSV-2, and there
was significant variation in HSV-2 prevalence among communities. After controlling for individual-level factors, the
authors identified community-level factors, including socioeconomic status and the presence of injection drug
users, that were independently associated with HSV-2 and explained 11% of the variance in prevalence. Future
studies are needed to test mechanisms through which these community-level factors may be operating.
disease transmission; herpesvirus 2, human; residence characteristics; risk factors; sexually transmitted
diseases, viral; simplexvirus
Abbreviations: HIV-1, human immunodeficiency virus type 1; HSV-2, herpes simplex virus type 2; SD, standard deviation; SES,
socioeconomic status; STI, sexually transmitted infection.
Herpes simplex virus type 2 (HSV-2) is a significant public health problem. HSV-2 is one of the most prevalent
sexually transmitted infections (STIs) worldwide (1–3). It
is primarily transmitted through sexual contact, is largely
responsible for genital herpes infection (4), and increases
the risk of human immunodeficiency virus type 1 (HIV-1)
transmission at least twofold (5–8). Symptomatic lesions
provide a portal of entry for HIV-1, and HSV-2 may
accelerate the progression of and increase the infectiousness
of HIV-1 (9). It is estimated that most people worldwide
Correspondence to Dr. Jacky Jennings, Department of Pediatrics, Johns Hopkins School of Medicine, Bayview Medical Center, 5200
Eastern Avenue, Mason F. Lord Building–Center Towers, Room 407, Baltimore, MD 21224 (e-mail: [email protected]).
1495
Am J Epidemiol 2008;167:1495–1503
1496 Jennings et al.
with sexually acquired HIV-1 have active HSV-2 infection (10).
HSV-2 seroprevalence studies show variation in infection
by geographic location. Some of the highest prevalences of
HSV-2 have been found in Africa and the Americas. Lower
prevalences have been found in Western and Southern
Europe than in Northern Europe and North America, and
although there have been few studies, the lowest prevalence
has been seen in Asia (2, 9, 11–15). There also is a great deal
of variation within regions. In India, HSV-2 prevalences of
over 40 percent have been reported in STI clinics in Pune
(8), but in a study of low-risk blood donors in Vellore,
15 percent of females and 10 percent of males were infected
(16). HSV-2 prevalence has also been found to vary by
individual-level characteristics, including gender, age, sexual
activity level, marital status, socioeconomic status (SES),
education, and race/ethnicity (9, 11, 13). However, these
characteristics are insufficient to explain differences within
and between countries, regions, and population subgroups,
suggesting the need to identify ecologic factors which may
help to explain the differences (11, 17).
To determine whether and to what extent ecologic factors
are causal determinants of differences in HSV-2 risk, a multilevel approach is required wherein individual and ecologic
factors are simultaneously added to a model (18). This approach has gained popularity as the importance of multiple
sources of STI transmission risk has become increasingly
apparent (19). The growth in statistical and computational
tools needed to study multilevel models gives evidence of
their growing popularity and has advanced approaches to
investigation. A stronger influence, however, on expanding
models of STI risk (and other public health issues) to include ecologic factors is the mounting evidence that ecologic factors may help explain disparities in diseases such as
HSV-2, yielding new avenues for disease prevention and
control. A first step in the determination of causal ecologic
factors is to develop hypotheses about community-level factors which may be associated with HSV-2 and ultimately
HIV-1. Developing such hypotheses requires an understanding of the study setting.
The current study was conducted in Chennai, the capital of
Tamil Nadu State in southern India. Chennai is a city of
6 million people containing extensive slums. In Chennai and
similar cities in India, the highest rates of STI and HIV-1 infection have been found among sex workers, men who have
sex with men, and injection drug users (20, 21), attributable in
part to migratory work and related sex-work patterns (22). It is
estimated that approximately 85 percent of HIV-1 transmission occurs through sexual contact, although injection drug
use is an important risk factor in northeastern India (20). STI
and HIV-1 transmission patterns are also related to diverse
cultural, social, religious, and sexual practices in India (22).
Building on this evidence, we developed exploratory hypotheses that community-level factors, such as the presence
of brothels, working men’s quarters, men who have sex with
men, and substance abuse, may serve to increase the likelihood of sexual partnering with an HSV-2-infected person,
resulting in increased risk of individual HSV-2 infection
(figure 1). We explored these hypotheses by collecting individual and community information in selected slums in
FIGURE 1. Hypothesized pathway from community-level factors
(top two boxes) to individual-level risk (bottom three boxes) for herpes
simplex virus type 2 (HSV-2) and ultimately, human immunodeficiency virus type 1 (HIV-1) infection. SES, socioeconomic status.
Chennai. Our objectives in this study were to determine the
geographic prevalence of HSV-2 infection and to determine
community-level factors associated with HSV-2 infection
after controlling for individual-level factors.
MATERIALS AND METHODS
Overview
The study was conducted from March to June of 2001 in
Chennai. The methods of the study have been presented
previously (23). Briefly, 965 urban residential communities
were designated by the Tamil Nadu Slum Clearance Board as
‘‘slum communities.’’ On the basis of power and sample size
calculations conducted for the parent study, a community cluster randomized trial, we determined that 24 communities
would be required from each site (Chennai, China, Peru, Russia, and Zimbabwe). We selected 30 communities in Chennai
to ensure that sufficient candidate communities were available
for the parent study. The communities were selected according
to a number of characteristics to permit comparability among
the communities, to improve our ability to measure intervention efficacy, and to maximize the cost-efficiency of the parent
study. All selected communities had 100–300 residents, were
named, were distinct neighborhoods, were geographically isolated from one another, and had single-story structures.
Trained field-workers enumerated households, and 65
households containing at least one person aged 18–40 years
were selected by systematic random sampling. Among
households with more than one eligible individual, only
one person was selected for participation by simple random
sampling (24). Field-workers invited the selected person to
participate in the study, and the individual and his or her
family were invited to attend a community health camp.
There was no more than a month between household enumeration and attendance at the health camp. The study
methods, procedures, and informed consent forms were approved by institutional review boards at the Johns Hopkins
Bloomberg School of Public Health and the Y. R. Gaitonde
Centre for AIDS Research and Education. Informed consent
was obtained before collection of all data.
Am J Epidemiol 2008;167:1495–1503
Community-level Determinants of HSV-2 Infection
Procedures
At the community health camps, participants answered
a survey using computer-assisted personal interviewing
and were given a physical examination that included testing
for HSV-2. The surveys were conducted in private areas in
the Tamil language by same-sex, trained interviewers. The
survey included items on sociodemographic factors, sexual
behavior, history of STIs, drug use, and other topics. HSV-2
testing was conducted in the Chennai study laboratory by
means of the HerpeSelect 2 enzyme immunoassay (Focus
Diagnostics, Inc., Cypress, California). A registered nurse
conducted the blood draw for HSV-2 antibody testing in
a setting conducive to infection control procedures. Twenty
percent of all biologic specimens were sent to Johns
Hopkins University School of Medicine (Baltimore, Maryland)
for quality control. All services were free, and no monetary
incentive was offered for participation.
A second survey was conducted among key informants to
assess community characteristics such as SES and evidence
of community-level STI risk factors (25). Staff of the Y. R.
Gaitonde Centre for AIDS Research and Education, who
were familiar with the study areas, were trained in ethnographic field methods. Briefly, staff identified approximately
10 key informants—influential representatives and other
persons knowledgeable about the community. Verbal informed consent was elicited from the key informants; no
one refused participation. Key informants were asked questions on a standardized list of community-level measures.
Data on two additional community-level measures
were also obtained. Staff counted the number of wine shops
in each community, and they obtained from the slum clearance board the number of years each slum had been in
existence.
1497
thatch. Numbers of people per household were averaged
across the key informant responses.
We conducted a principal-components analysis of the five
items plus years of existence of the community to identify
one community-level measure of SES. One factor with an
eigenvalue of 2.78 emerged from four out of six of the
community-level SES items, including electricity and water
structures and housing and roofing materials. The factor
showed good reliability, with a Cronbach’s alpha of 0.85.
From those four items, using regression methods, we created one scale with a mean of 0 and a variance equal to the
squared multiple correlation between the estimated factor
scores and the true factor values. The remaining two SES
measures—average number of people per household and
years of existence of the slum—were analyzed as separate
measures.
Community-level STI risk factors. We assessed six
community-level STI risk factors. Five were based on key
informant reports and the sixth was based on field staff
observations. Key informants in each community were
asked to report on the existence of working men’s quarters,
brothels, men who had sex with men, injection drug users,
and public intoxication. Key informants’ responses were
coded as present or absent. For each slum, a summary measure for each risk factor was created on the basis of the
percentage of key informants from each slum reporting
the presence of each risk factor (the implicit assumption
being that the higher the percentage of key informants endorsing the presence of a risk factor, the more likely the
factor was to actually be present in the slum). The sixth item
was the number of wine shops in the community; wine
shops were counted by staff, and the raw count was used.
Statistical analysis
Measures
Level 1: individual-level measures. Trained interviewers
asked participants about the following selected sociodemographic factors: ethnicity, age, sex, marital status, participation in migratory work during most years, circumcision
status (males only), and educational level. Participants were
also asked about the following selected STI risk behaviors:
STI history, number of casual sex partners in the past
3 months, and drug abuse (‘‘In the past 3 months, have
you used drugs, e.g., marijuana, cocaine, heroin, amphetamines, barbiturates, inhaled glue or solvents, or any other
drug to get high?’’).
Level 2: community-level measures. Information on
community-level measures, including measures of SES
and STI risk factors, was collected in each community.
Socioeconomic status. Key informants were asked five
SES-related questions regarding electricity structures, water
structures, housing and roofing materials, and number of
persons per household. For the creation of summary measures by community, the first four items were summarized
separately as the proportion of key informants agreeing on
the presence of electricity and water structures and on
whether building and roofing materials were thatch or nonAm J Epidemiol 2008;167:1495–1503
The data in the study formed a nested structure; that is,
the participants (level 1) were nested within the communities (level 2). Because of this data structure, multilevel models represented the most appropriate method of analysis.
Multilevel analysis accounts for nonindependence of observations within groups, uses empirical Bayes adjustments for
the group means, and allows for significance testing of the
between- and within-group (levels) variances on the outcome, HSV-2.
Exploratory analysis was conducted and summary statistics were generated for the individual-level (level 1) variables and the community-level (level 2) variables using Stata
Intercooled, version 8.0 (Stata Corporation, College Station,
Texas), and HLM, version 6.02 (Scientific Software International, Inc., Lincolnwood, Illinois). Maps of HSV-2 prevalence by community were created using MapInfo, version
7.0 (MapInfo Corporation, Troy, New York). A series of
models was generated to determine whether and to what
extent community-level factors significantly explained
HSV-2 prevalence after controlling for individual-level factors in multilevel nonlinear modeling. First, random-effects
analysis of variance was conducted to assess the extent of
variation in HSV-2 prevalence between the communities
and to test the significance of the variation. Next, an
1498 Jennings et al.
TABLE 1. Individual-level characteristics of study participants (n ¼ 1,275) from selected slum communities in Chennai, India,
according to HSV-2* infection status, 2001y
Total
HSV-2-positive
HSV-2-negative
Individual-level characteristic
Tamil ethnicity
No.
%
1,275
100
Age (years)
Mean (SD*)
No.
%
196
15.37
30.00 (6.14)
Mean (SD)
No.
%
1,079
84.63
32.77 (6.06)
Mean (SD)
29.51 (6.03)
Sex
Female
729
57
129
66
600
56
Male
546
43
67
34
479
44
Marital status
Married
1,096
86
161
82
935
87
Widowed/separated/divorced
99
8
29
15
70
6
Single/never married
80
6
6
3
74
7
College degree or more
360
28
38
19
322
30
High school diploma
301
24
38
19
263
24
Primary school or less
614
48
120
61
494
46
117
26
18
15
99
85
51
10
7
14
44
86
Ever having a sexually transmitted disease
72
6
15
8
57
6
Using drugs to get high in the past 3 months
90
7
24
12
66
6
Education
Engaging in migratory work during
most years (n ¼ 598)
Circumcised (males only) (n ¼ 535)
No. of casual sex partners in the past year
0.20 (4.69)
0.57 (4.69)
0.14 (1.34)
* HSV-2, herpes simplex virus type 2; SD, standard deviation.
y For some variables, the percentage was based on a smaller number of participants because of missing data.
exploratory ecologic analysis was conducted to determine
the association of the community-level variables to HSV-2
prevalence. Then analysis of covariance was conducted to
determine which of the level 1 variables were significantly
associated (two requirements—the 95 percent confidence
interval did not include 1 and the p value was less than
0.05) with individual-level HSV-2. Finally, a means-asoutcomes regression model was developed which began
with the significantly associated level 1 factors and then
added in all level 2 factors. Level 2 variables which were
the least significant were subtracted one by one from the
model until a final model which contained only variables
that were significant at levels 1 and 2 remained.
By generating Akaike and Bayesian information criteria
values, we assessed the goodness of fit and parsimony of the
model. Akaike’s information criterion considers the number
of model parameters, and the Bayesian information criterion
considers the effect of model parameters and sample size
(26, 27). To date, HLM software does not produce values for
Akaike’s information criterion or the Bayesian information
criterion; we used Stata for these calculations. Actual fit was
assessed by residual analysis. At the community level, we
created a Q-Q plot of the Mahalanobis distance—that is, the
standardized squared distance of a unit from the center of
a v-dimensional distribution, where v is the number of random effects per unit used to diagnose fit, versus the expected
values of the order statistics for a sample of size J from
a population that is distributed v(v)2.
RESULTS
Study population
Of an estimated 14,000 adults living in the 30 communities, 1,947 (13 percent; 969 men and 978 women) were
selected to participate in the study. Among selected individuals, 1,656 (85 percent) completed the survey, and of these,
1,631 (98 percent) provided samples for STI testing. Persons
less likely to complete the survey were more often male
(34 percent male vs. 54 percent female; v2 p < 0.01), more
often married (77 percent married vs. 71 percent unmarried;
v2 p < 0.01), and slightly older (mean age ¼ 29.4 years vs.
28.5 years). Those who completed the survey but did not
provide a biologic sample (vs. those that did) were more
likely to be female, married, and slightly older, although
they did not differ from the study population in terms of
education, health status, or lifetime number of partners (data
not included). Additionally, 309 participants (19 percent)
reported no lifetime sexual activity and were excluded because they were not at risk for HSV-2.
At level 2, data were collected from 29 communities, and
the average number of key informants was 10 (standard
deviation (SD), 0.57; range, 9–11). Upon inspection of the
data, 47 participants (4 percent) were excluded from further
analysis because one slum had only one respondent and
another slum did not have any data because it was dissolved
before data collection. Thus, the final study sample included
Am J Epidemiol 2008;167:1495–1503
Community-level Determinants of HSV-2 Infection
TABLE 2. Percentages of key informants reporting the
presence of various community-level characteristics, averaged
across selected slum communities (n ¼ 28), Chennai, India,
2001
Community-level characteristic
Working men’s quarters
Mean % (SD*)
Range of %’s
3 (12)
0–60
Brothels
14 (36)
0–100
Men who have sex with men
28 (30)
0–100
Injection drug users
47 (41)
0–100
Public intoxication (very often
and frequently)
43 (39)
0–100
* SD, standard deviation.
1,275 participants from 28 communities with an average of
46 participants per community.
Characteristics of the study population
Individual-level demographic and behavioral characteristics are shown in table 1, summarized and stratified by
HSV-2 status. In terms of individual-level characteristics,
100 percent of participants were of the Tamil ethnic group,
and women constituted 57 percent of the population. Fifteen
percent of the population was infected with HSV-2. The
average age of the study participants was 30 years; 86 percent of participants were married, and 28 percent had a college degree. Twenty-six percent of participants participated
in migratory work during most years, and 10 percent of
males were circumcised. Four percent reported having had
at least one casual sex partner in the past year, with an overall
reported average of 0.20 casual partners in the past year.
Seven percent of participants had used a drug to get high
in the past 3 months, and 6 percent had a history of an STI.
Community-level characteristics were summarized
across communities. In terms of SES, electricity and water structures existed in 86 percent and 43 percent of
communities, respectively. Thatch material was used in
home and roof construction in 61 percent and 57 percent
of communities, respectively. The average number of
persons per household was 5.25 (SD, 0.67), and the mean
of the variances within each community was 0.30 (SD,
0.51). The average number of years a slum had been in
existence was 28.71 (SD, 14.42; range, 1–60). Table 2
provides the percentage of key informants reporting on
the existence of five community-level STI risk factors
averaged across the slum communities. On average, 3
percent of key informants reported their slums as having
working men’s quarters, 14 percent reported brothels,
28 percent reported men who had sex with men,
47 percent reported injection drug users, and 43 percent
reported very often or frequent public intoxication. On
average, 3.25 (SD, 2.94; range, 1–12) wine shops were
counted per community.
Analysis-of-variance model
Figure 2 shows a map of Chennai with the HSV-2 prevalence for each community. The mean HSV-2 prevalence
was 15.37 (SD, 36; range, 4.17–49.06). The multilevel
analysis-of-variance model showed a significant amount of
variation (30 percent; v2 p < 0.00) in HSV-2 prevalence
between the communities. Given these findings, we proceeded to the ecologic model.
Ecologic model
To determine which community-level factors were associated with HSV-2 prevalence, we generated an ecologic
model (table 3). Since this model was intended to be
FIGURE 2. Prevalence of herpes simplex virus type 2 in 28 slum communities in Chennai, India, 2001.
Am J Epidemiol 2008;167:1495–1503
1499
1500 Jennings et al.
TABLE 3. Individual- and community-level characteristics associated with individual-level herpes simplex virus type 2 infection in
selected slum communities, Chennai, India, 2001y
Analysis of
covariance
OR
95% CI
MAORz
adjusted
OR
95% CI
Final model
adjusted
OR
95% CI
Intercept
0.15**
0.06, 0.37
0.01**
0.00, 0.02
0.01**
0.00, 0.02
Age (years)
1.05**
1.03, 1.07
1.10**
1.07, 1.14
1.10**
1.07, 1.13
0.66, 0.95
0.48**
0.32, 0.71
0.48**
0.32, 0.72
1.04*
1.00, 1.08
1.04*
1.00, 1.08
2.33*
1.16, 4.67
2.35*
1.17, 4.72
0.75*
0.59, 0.96
1.95*
1.15, 3.30
Ecologic
ORz
95% CIz
Individual-level characteristics
(n ¼ 1,275)
Sex
Female
Referent
Male
0.79*
Marital status
Married
Referent
Widowed/separated/divorced
1.55**
1.15, 2.08
Single/never married
0.73
0.50, 1.08
Education
College degree or more
Referent
High school diploma
1.05
0.85, 1.30
Primary school or less
1.32**
1.10, 1.58
Engaging in migratory work during
most years (n ¼ 598)
1.11
0.69, 1.80
Circumcised (males only) (n ¼ 535)
1.12
0.47, 2.68
No. of casual sex partners in
the past year
1.03*
1.01, 1.05
Ever having a sexually transmitted
infection
1.31
0.88, 1.96
Using drugs to get high in the
past 3 months
1.44*
1.02, 2.03
Community-level characteristics,
as reported by key
informants (n ¼ 28)
Socioeconomic status
0.70*
0.55, 0.90
0.81*
0.70, 0.95
Average no. of persons per
household
1.03
0.73, 1.46
1.00
0.70, 1.43
Average no. of years of slum’s
existence
1.01*
1.00, 1.02
1.01*
1.00, 1.02
Working men’s quarters
1.35
0.82, 2.23
0.72
0.26, 2.0
Brothels
1.12
0.76, 1.66
1.68
0.86, 3.27
Men who had sex with men
0.57
0.31, 1.01
0.48*
0.24, 0.97
Injection drug users
1.76*
1.03, 3.03
1.70
0.99, 2.93
% of key informants reporting:
Public intoxication (very often
and frequently)
1.63
0.94, 2.83
0.99
0.97, 1.02
Average no. of wine shops
0.96
0.92, 1.00
1.64
0.94, 2.89
Tau
0.18
0.27
0.17
0.16
* p < 0.05; **p < 0.01.
y All models represent multilevel nonlinear models.
z OR, odds ratio; CI, confidence interval; MAOR, means as outcomes regression.
exploratory, we entered all community-level factors into
the model, including SES, number of persons per household,
years of slum existence, working men’s quarters, presence
of brothels, presence of men who had sex with men,
presence of injection drug users, frequency of public intoxication, and number of wine shops. SES and the presence
of injection drug users had 95 percent confidence
intervals that did not include 1 and p values that were less
than 0.05.
Analysis-of-covariance model
To determine which individual-level factors were associated with individual-level HSV-2 infection after accounting
Am J Epidemiol 2008;167:1495–1503
Community-level Determinants of HSV-2 Infection
1501
each 30th-percentile decrease in SES. Within low- and
medium-SES slums, HSV-2 prevalence increased for each
30th-percentile increase in percentage of injection drug use.
In high-SES slums, HSV-2 prevalence remained the same or
decreased slightly (not statistically significant) for each
30th-percentile increase in percentage of injection drug use.
The Akaike and Bayesian information criterion values
for the model with only level 1 factors were 1,013.59 and
45.72, respectively. The Akaike and Bayesian information
criterion values for the final model with significant level 1
and level 2 factors were 987.05 and 61.97, respectively.
The lower values for the final model with significant level 1
and level 2 factors suggest a better fit and greater parsimony
than was seen for the model with level 1 factors alone. The
Q-Q plot also suggests a good-fit model with no serious
departures from normality.
FIGURE 3. Prevalence of herpes simplex virus type 2 (HSV-2) by
30th percentile of socioeconomic status (SES) and percentage of key
informants reporting the presence of injection drug users in selected
slum communities (n ¼ 28), Chennai, India, 2001. N/A, not applicable.
for the clustering of participants within communities, we constructed a series of analysis-of-covariance models (table 3).
Individual-level factors, including age, sex, marital status,
education, migratory work, circumcision status (males
only), number of casual sex partners in the past year, STI
history, and drug use (use of drugs to get high in the past 3
months) were entered into the model. Increased age, male
sex, higher numbers of casual sex partners, and likelihood of
drug use were all significantly related to individual-level
infection with HSV-2 and were retained at level 1.
Final model
Next, to determine the extent to which community-level
variables were associated with HSV-2 infection, we constructed a series of means-as-outcomes-regression models
(table 3). In the first model, we included all communitylevel factors. Insignificant factors were subtracted, resulting
in a final model with factors (SES and the presence of injection drug users) that had 95 percent confidence intervals
which did not include 1 and p values less than 0.05. In the
final model controlling for individual-level factors, a oneunit increase in community-level SES was associated with
an odds ratio of 0.75 (95 percent confidence interval: 0.59,
0.96), which indicates a reduction of 25 percent in the odds
of HSV-2 infection in the community. A one-unit increase in
the proportion of injection drug users in the community, on
the other hand, was associated with an odds ratio of 1.95
(95 percent confidence interval: 1.15, 3.30), which indicates
an increase of 95 percent in the odds of HSV-2 infection in
the community. In the final model, the significant level 2
factors helped to explain an additional 11 percent of the
variance associated with HSV-2 infection. Figure 3 further
elucidates the results of the final model, showing HSV-2
prevalences by 30th percentile of SES and percentage of
key informants reporting on the presence of injection drug
users. Overall, the mean HSV-2 prevalence decreased with
Am J Epidemiol 2008;167:1495–1503
DISCUSSION
Among these communities, we identified considerable
differences in the prevalences of HSV-2 infection, which
ranged from 4.17 percent to 49.06 percent. These findings
are similar to those of a large number of studies that have
found geographic heterogeneity in HSV-2 and other STIs
(2, 9, 11–15). After controlling for significant individual
factors for HSV-2, including age, sex, number of casual
sex partners in the past year, and drug use, we identified
community-level factors (including SES and the presence
of injection drug users) which were independently associated with HSV-2 prevalence and which explained 11 percent
of the variance.
The current study was not designed to test the mechanisms through which potential STI risk factors at the community level may increase individual-level risks of HSV-2
infection, yet it represents the first step in the determination
of community-level factors and their independent associations with HSV-2. The potential importance of communitylevel risk factors for HSV-2 and other STIs begins with the
recognition that sexual partnerships are often formed in
tightly confined social settings that have spatial and cultural
boundaries (28, 29). These settings become, in a sense, sexual marketplaces, since they provide the types of individuals
and opportunities available for partnering, as well as the
social rules that guide the selection of sex partners (28).
Research in India and other Asian countries suggests that
STI and HIV-1 transmission patterns are partly related to
sexual marketplaces and their players, such as commercial
sex workers and men who have sex with men (20, 30–32).
For example, sex work in Chennai is not structurally organized as is the case in other cities in India, such as Mumbai
or Calcutta. Rather than a ‘‘red light district’’ with large
brothels, sex work in Chennai is based on small groups of
women either working the streets under the protection of
a pimp or working ‘‘freelance’’ (as in the case of housewives). While men may seek sex partners at some distance
from their homes to avoid unwarranted recognition, the majority of the men included in the survey reported selecting
sex partners near their place of residence.
A number of studies have also found structural-level community factors such as SES to be associated with STI
1502 Jennings et al.
prevalence, and other studies have found the same with drug
users. One hypothesis is that impoverished slums may have
more drug users, and concentration of drug users within a
community may alter sexual marketplaces by altering sex
partner selection patterns and sex network structures. Drug
markets, like sex marketplaces, have been described as having dense networks with specific mixing patterns and local
geographic characteristics (33), and they overlap with the
networks of commercial sex workers and men who have sex
with men (34–37), providing the potential for transmission
and acquisition of STIs within and between these networks.
There is good evidence suggesting connections between
high-risk groups and lower-risk groups within communities.
For example, in Chennai, approximately one quarter of
HIV-1-infected patients are housewives who are otherwise at
low risk (20). Our finding of an association between individuallevel HSV-2 infection and the presence of injection drug users
at the community level suggests that drug networks may increase the likelihood of selecting a sex partner who is infected
with an STI and may play an important role in transmission of
STIs and HIV-1 to lower-risk community members.
The current study had a number of limitations. The selection of communities was restricted to allow for comparability
in the parent trial, possibly resulting in a biased sample of
communities. We were unable to assess this possible bias, as
there are no statistics maintained by the slum clearance board
on the slums. Thus, results may only be generalizable to slum
communities with a similar profile. Women were slightly
more likely to participate in this study than males, since they
are more likely to be in residence in the slums. In addition, the
current study was not designed to directly test causality or
the mechanisms through which potential STI risk factors at
the community level may increase individual-level risks for
HSV-2. Future studies should be designed to carefully measure drug use at the individual and community levels to determine whether the independent effects identified in this
study are valid or in fact result from misspecification of the
community-level variables.
Despite these limitations, this study obtained data which
provide a stepping-stone for future studies attempting to
identify community- or structural-level determinants of
HSV-2 disparities among population subgroups and communities. The identification of these structural determinants
may yield important prevention and intervention opportunities for limiting HSV-2 and HIV-1 transmission.
ACKNOWLEDGMENTS
Conflict of interest: none declared.
REFERENCES
1. Nahmias AJ, Lee FK, Beckman-Nahmias S. Seroepidemiological and -sociological patterns of herpes simplex
virus infection in the world. Scand J Infect Dis 1990;69:19–36.
2. Smith JS, Robinson NJ. Age-specific prevalence of infection
with herpes simplex virus types 2 and 1: a global review.
J Infect Dis 2002;186(suppl):S3–28.
3. Corey L, Langenberg AGM, Ashley R, et al. Recombinant
glycoprotein vaccine for the prevention of genital HSV-2
infection—two randomized controlled trials. JAMA 1999;282:
331–40.
4. Corey L. The current trend in genital herpes. Progress in
prevention. Sex Transm Dis 1994;21(suppl):S38–44.
5. Wald A, Link K. Risk of human immunodeficiency virus
infection in herpes simplex virus type 2-seropositive persons:
a meta-analysis. J Infect Dis 2002;185:45–52.
6. Serwadda D, Gray RH, Sewankambo NK, et al. Human immunodeficiency virus acquisition associated with genital ulcer
disease and herpes simplex virus type 2 infection: a nested casecontrol study in Rakai, Uganda. J Infect Dis 2003;188:1492–7.
7. Holmberg SD, Gerber AR, Stewart JA, et al. Herpesviruses
as cofactors in AIDS. Lancet 1988;2:746–7.
8. Reynolds SJ, Risbud AR, Shepherd ME, et al. Recent herpes
simplex virus type 2 infection and the risk of human immunodeficiency virus type 1 acquisition in India. J Infect Dis
2003;187:1513–21.
9. Xu F, Sternberg MR, Kottiri BJ, et al. Trends in herpes simplex
virus type 1 and type 2 seroprevalence in the United States.
JAMA 2006;296:964–73.
10. Corey L. Synergistic copathogens—HIV-1 and HSV-2.
N Engl J Med 2007;356:854–6.
11. Bünzli D, Wietlisbach V, Barazzoni F, et al. Seroepidemiology
of herpes simplex virus type 1 and 2 in western and southern
Switzerland in adults aged 25–74 in 1992–93: a populationbased study. BMC Infect Dis 2004;4:10. (Electronic article).
12. Garcı́a-Corbeira P, Dal-Ré R, Aguilar L, et al. Is sexual
transmission an important pattern for herpes simplex type 2
virus seroconversion in the Spanish general population? J Med
Virol 1999;59:194–7.
13. Hellenbrand W, Thierfelder W, Muller-Pebody B, et al.
Seroprevalence of herpes simplex virus type 1 (HSV-1) and
type 2 (HSV-2) in former East and West Germany, 1997–1998.
Eur J Clin Microbiol Infect Dis 2005;24:131–5.
14. Obasi A, Mosha F, Quigley M, et al. Antibody to herpes
simplex virus type 2 as a marker of sexual risk behavior in
rural Tanzania. J Infect Dis 1999;179:16–24.
15. Levett PN. Seroprevalence of HSV-1 and HSV-2 in Barbados.
Med Microbiol Immunol 2005;194:105–7.
16. Cowan FM, French RS, Mayaud P, et al. Seroepidemiological
study of herpes simplex virus types 1 and 2 in Brazil, Estonia,
India, Morocco, and Sri Lanka. BMJ 2003;79:286–90.
17. Smith JS, Herrero R, Bosetti C, et al. Herpes simplex virus-2
as a human papillomavirus cofactor in the etiology of invasive
cervical cancer. J Natl Cancer Inst 2002;94:1604–13.
18. Diez Roux A, Aiello A. Multilevel analysis of infectious
disease. J Infect Dis 2005;191(suppl):S25–33.
19. Aral SO, Padian NS, Holmes KK. Advances in multilevel
approaches to understanding the epidemiology and prevention
of sexually transmitted infections and HIV: an overview.
J Infect Dis 2005;191(suppl 1):S1–6.
20. Steinbrook R. HIV in India—a complex epidemic. N Engl
J Med 2007;356:1089–93.
21. Go VF, SriKrishnan AK, Sivaram S, et al. High HIV
prevalence and risk behaviors in men who have sex with
men in Chennai, India. J Acquir Immune Defic Syndr
2004;35:314–19.
22. Chandrasekaran P, Dallabetta G, Loo V, et al. Containing
HIV/AIDS in India: the unfinished agenda. Lancet Infect
Dis 2006;6:508–21.
23. Marx MA, Murugavel KG, Tarwater PM, et al. Association of
hepatitis C virus infection with sexual exposure in southern
India. Clin Infect Dis 2003;37:514–20.
Am J Epidemiol 2008;167:1495–1503
Community-level Determinants of HSV-2 Infection
24. Kish L. A procedure for objective respondent selection
within the household. J Am Stat Assoc 1949;44:380–7.
25. Kumar K. Conducting key informant interviews in
developing countries. (USAID Program Design and Evaluation Methodology Report no. 13). Washington, DC: US
Agency for International Development, 1986.
26. Akaike H. Information theory and an extension of the
maximum likelihood principle. In: Petrov BN, Csaki
F, eds. Second international symposium on information
theory. Kiado, Budapest: Akademiai, 1973:267–81.
27. Schwartz G. Estimating the dimensions of a model. Annu
Stat 1978;6:461–4.
28. Ellingson S, Laumann EO, Paik A, et al. The theory of
sex markets. In: Laumann EO, Ellingson S, Mahay J, et al,
eds. The sexual organization of the city. Chicago, IL:
University of Chicago Press, 2004:3–38.
29. Potterat JJ, Rothenberg RB, Woodhouse DE, et al.
Gonorrhea as a social disease. Sex Transm Dis 1985;12:
25–32.
30. Go VF, Frangakis C, Nam IV, et al. High HIV sexual risk
behaviors and sexually transmitted disease prevalence among
injection drug users in northern Vietnam: implications for
Am J Epidemiol 2008;167:1495–1503
31.
32.
33.
34.
35.
36.
37.
1503
a generalized HIV epidemic. J Acquir Immune Defic Syndr
2006;42:108–15.
Khan S. Culture, sexualities, and identities: men who have
sex with men in India. J Homosex 2001;40:99–115.
Lau JT, Wong WS. HIV antibody testing among male
commercial sex networkers, men who have sex with men
and the lower-risk male general population in Hong Kong.
AIDS Care 2002;14:55–61.
Ross MW, Williams ML. Sexual behavior and illicit drug
use. Annu Rev Sex Res 2001;12:290–310.
Saxe L, Kadushin C, Beveridge A, et al. The visibility of
illicit drugs: implications for community-based drug control
strategies. Am J Public Health 2001;91:1987–94.
Bobashev GV, Anthony JC. Clusters of marijuana use in
the United States. Am J Epidemiol 1998;148:1168–74.
Petronis KR, Anthony JC. Perceived risk of cocaine use and
experience with cocaine: do they cluster within US neighborhoods and cities? Drug Alcohol Depend 2000;57:183–92.
Neaigus A, Friedman SR, Jose B, et al. High-risk personal
networks and syringe sharing as risk factors for HIV infection
among new drug injectors. J Acquir Immune Defic Syndr Hum
Retrovirol 1996;11:499–509.