The Journal of Infectious Diseases

Supplement Article

Geographic Pattern of Typhoid Fever in India: A Model-

Based Estimate of Cohort and Surveillance Data
Yanjia Cao,1, Arun S. Karthikeyan,2 Karthikeyan Ramanujam,2 Reshma Raju,2, Swathi Krishna,2, Dilesh Kumar,2 Theresa Ryckman,4
Venkata Raghava Mohan,3 Gagandeep Kang,2, Jacob John,3, Jason R. Andrews,1, and Nathan C. Lo5,
1
Division of Infectious Diseases and Geographic Medicine, Stanford University School of Medicine, Stanford, California, USA, 2Wellcome Research Unit, Christian Medical College, Vellore, India,
3
Department of Community Health, Christian Medical College, Vellore, India, 4Center for Health Policy and the Center for Primary Care and Outcomes Research, Stanford University School of
Medicine, Stanford, California, USA, 5Deparment of Medicine, University of California, San Francisco, San Francisco, California, USA

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

Background.  Typhoid fever remains a major public health problem in India. Recently, the Surveillance for Enteric Fever in India
program completed a multisite surveillance study. However, data on subnational variation in typhoid fever are needed to guide the
introduction of the new typhoid conjugate vaccine in India.
Methods.  We applied a geospatial statistical model to estimate typhoid fever incidence across India, using data from 4 cohort studies
and 6 hybrid surveillance sites from October 2017 to March 2020. We collected geocoded data from the Demographic and Health Survey
in India as predictors of typhoid fever incidence. We used a log linear regression model to predict a primary outcome of typhoid incidence.
Results.  We estimated a national incidence of typhoid fever in India of 360 cases (95% confidence interval [CI], 297–494) per
100 000 person-years, with an annual estimate of 4.5 million cases (95% CI, 3.7–6.1 million) and 8930 deaths (95% CI, 7360–12 260),
assuming a 0.2% case-fatality rate. We found substantial geographic variation of typhoid incidence across the country, with higher
incidence in southwestern states and urban centers in the north.
Conclusions.  There is a large burden of typhoid fever in India with substantial heterogeneity across the country, with higher
burden in urban centers.
Keywords.  enteric fever; geospatial model India; salmonella; typhoid fever; public health; vaccination.

Typhoid fever has an estimated global incidence of 11–21 mil- estimated an incidence of 265 (95% CI, 217–324) cases per
lion cases annually, resulting in 120 000–160 000 deaths [1–4]. 100 000 person-years [16, 17]. There have been no population-
Enteric fever is an acute febrile illness caused by ingestion of the based typhoid incidence studies for more than a decade, and
bacterium Salmonella enterica serotype Typhi (S Typhi) or sero- there have been no prior population-based studies from rural
type Paratyphi A, B or C, often through food or water contam- areas, where the majority of people reside. A  recent meta-
inated with human feces [5, 6]. The severe clinical presentations analysis, largely of facility-based studies, revealed that the pro-
of typhoid fever includes the development of sepsis, gastrointes- portion of individuals with positive blood cultures for S Typhi
tinal bleeding, intestinal perforation, and death [7, 8]. has been declining [10]. However, there remains scarce recent
A large proportion of the global burden of typhoid fever data on the incidence and geographic distribution of typhoid
is concentrated in South Asia, with a high incidence in India fever in India. This paucity of data is further complicated be-
[2–4, 9–14]. The Global Burden of Disease Study in 2017 esti- cause the geographic pattern of typhoid fever is expected to be
mated typhoid/paratyphoid incidence in India of 586 cases per highly heterogenous within the country [10, 18, 19].
100 000 person-years [4, 15]; however, these estimates extrapo- Accurate and recent estimates of typhoid fever incidence and
lated largely from regional data, because there have been few the spatial distribution in India are essential for public health de-
population-based studies in India. A 1996 study in Delhi found cision making such as vaccination strategies. The World Health
an incidence of 976 (95% confidence interval [CI], 763–1250) Organization (WHO) recently approved new Vi conjugate vac-
cases per 100 000 person-years, whereas a 2006 study in Kolkata cines against typhoid fever that provide high efficacy and dura-
tion of protection [20]. To address the need for locally relevant
data for typhoid fever burden in India to guide policy on use
of the conjugate vaccines in India, the Surveillance for Enteric
 

Correspondence: Nathan C.  Lo, MD PhD, Division of HIV, Infectious Diseases, and Global
Medicine, Department of Medicine, University of California, San Francisco, 1001 Potrero Ave, Fever in India (SEFI) study was conducted [21, 22]. The SEFI is
SFGH Bldg 3, Room 523, San Francisco, CA 94110, USA ([email protected]).
a multisite study that used both prospective cohorts and hybrid
The Journal of Infectious Diseases®  2021;224(S5):S475–83
© The Author(s) 2021. Published by Oxford University Press for the Infectious Diseases Society
surveillance designs in 10 urban and rural locations to provide
of America. This is an Open Access article distributed under the terms of the Creative Commons estimates on typhoid fever incidence. Although the SEFI study
Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted
reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
sites provide high-quality typhoid fever incidence data for these
DOI: 10.1093/infdis/jiab187 sites, there is a need for broader estimates of typhoid burden

Geographic Pattern of Typhoid Fever in India  •  jid 2021:224 (Suppl 5) • S475

across India. The use of spatial modeling approaches has be- The Tier 2 SEFI sites (hybrid surveillance study) measured
come increasingly common to predict epidemiologic meas- hospitalized typhoid fever cases, and included 6 hybrid sur-
ures (eg, incidence and prevalence) in infectious diseases (eg, veillance sites. Each site measured the number of typhoid fever
malaria and schistosomiasis) in the absence of primary data cases identified in the hospital in persons 6 months and older
[23–26]. This modeling approach aims to leverage variables (including adolescents and adults) over a 2-year observation
from secondary datasets to predict incidence in areas without period (between February 2018 and March 2020). In each hy-
primary data on incidence, by calibrating the relationship of brid surveillance study site, healthcare utilization surveys were
these variables with incidence in areas with primary  data  on conducted to estimate the person-years of observation to adjust
incidence. the catchment population denominator when computing inci-
Although typhoid fever is common in India, the exact burden dence, with the methods as previously described [22, 28]. The
of disease and spatial heterogeneity are important to understand to incidence estimate was adjusted for the 60% sensitivity of blood
guide policy decision on national vaccination with the Vi typhoid cultures. The Tier 2 study sites were located in Chandigarh,

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

conjugate vaccines. To address this need, we applied a geospatial Nandurbar (Maharashtra), Kullu (Himachal Pradesh),
statistical model to estimate typhoid fever incidence across India, Karimganj (Assam), Anantapur (Andhra Pradesh), and East
using data from 4 cohort studies and 6 hybrid surveillance from Champaran (Bihar). Among the 10 study sites, Delhi, Kolkata
the SEFI study combined with a national household survey data. and Vellore in Tier 1 and Chandigarh in Tier 2 are urban areas,
whereas the rest are rural locations. Details of the computation
METHODS process for the incidence in these 10 study sites for Tier 1 and 2
Methods Overview are further described in the Appendix.
We used a statistical model to predict typhoid fever incidence
Study Data on Model Predictors
across India. We performed spatial data processing and inter-
The data on model predictors for typhoid fever incidence were
polation to match health and demographic variables geograph-
drawn from the Demographic and Health Survey (DHS) con-
ically to observed data on typhoid incidence from SEFI study
ducted from 2015 to 2016 in India [29]. The DHS are nationally
sites. The model calibration and prediction both utilized regres-
representative cross-sectional surveys on health and demo-
sion analysis to estimate typhoid incidence, which was reported
graphic variables that occur in many low- and middle-income
at a state and national level.
countries approximately every 5 years [30]. The DHS have also
Study Data on Typhoid Incidence been widely used by researchers and policymakers [31–33]. We
We used data on typhoid fever incidence from the SEFI study, extracted the following prespecified variables from the DHS
which was a multisite cohort (named Tier 1)  and hybrid sur- that were chosen based on their potential to predict incidence
veillance (named Tier 2) study of typhoid incidence. The SEFI of typhoid fever from prior literature: urbanicity (urban vs
study had 10 sites that each provided a site-specific typhoid in- rural in a cluster, defined by the Indian national government),
cidence estimate [21, 22]. In all study sites, typhoid cases were household wealth (quintile), household maternal education,
defined as blood culture-confirmed S Typhi cases over the du- household access to improved water and toilet, household size,
ration of the study. The Widal test was not used for diagnosis. household receipt of a third dose of the diphtheria-pertussis-
The spatial data for catchment areas at a village level for the 10 tetanus vaccination (a marker of healthcare access), and an-
study sites in India were provided by the SEFI study. We used thropometric measurements (stunting and underweight). We
spatial information on each study site using ArcGIS 10.7.1. simplified the definition of access to improved water and toilet
The Tier 1 SEFI sites (cohort study) measured clinical typhoid following the Joint Monitoring Program for Water Supply,
fever cases and included 4 cohorts. Each cohort enrolled 6000 Sanitation and Hygiene [34] guidance. We defined stunting and
children ages 6 months to 15 years that were observed over a 2-year being underweight based on the WHO Child Growth Standards
observation period (October 2017–February 2019). The study sites using height-for-age and weight-for-age more than 2 standard
were located in Delhi, Kolkata (West Bengal), Vadu (Maharashtra), deviations below a reference median, respectively. Population
and Vellore (Tamil Nadu). The incidence of typhoid fever was com- data were obtained from WorldPop [35]. Missing data were ex-
puted as the  number of blood culture-confirmed cases for each cluded. The variables were recoded and computed hierarchically
site, with the denominator being the number of person-years of as described in Supplementary Tables S1–S3 in the Appendix.
observation in the defined age group, with follow up censored for We performed a spatial interpolation of each DHS var-
15th birthday, withdrawal of consent/assent, febrile period, death, iable over India. Study variables were available at different
and completion of study. The incidence estimate was reported as levels including for each child, household, and cluster (see
number of cases per 100 000 person-years. We adjusted our esti- Supplementary Tables S2 and S3) and were each aggregated to
mate to account for consent/assent to obtain blood cultures and a mean at the cluster level (ie, the primary sampling unit where
blood culture sensitivity [21]. Further explanation of study method- the preexisting geographic area is known as census enumeration
ology for Tier 1 has been previously described [27]. areas). For each cluster in the DHS survey, the GPS location (ie,

S476  • jid 2021:224 (Suppl 5) • Cao et al

points with latitude and longitude) at the center of each sample per 100  000 person-years) and predictors (independent vari-
cluster was collected during field work or survey with varia- ables). We estimated the regression on the level of the study
tion for confidentiality inside the targeted administrative units site, including the sample size of the 10 SEFI sites. For each pre-
(by up to 2 kilometers for urban locations and 10 kilometers dictor, the variable population mean of the DHS clusters over-
for rural locations) [36]. The DHS survey for India in 2015 had lying each SEFI study site was estimated. We selected variables
a total of 28  395 clusters. We performed spatial interpolation by identifying the lowest Akaike Information Criterion (AIC)
using inverse distance weighting methodology for all variables value, with a goal of limiting to a single variable to prevent
from DHS (ie, cluster points) at 5 × 5-km resolution. The spatial overfitting given the limited sample size; in sensitivity analysis,
resolution of 5 km was (1) chosen to line up with the smallest we evaluated a 2–3 variable model. The final model was cali-
size of catchment area among the 10 study sites and (2) based brated to the selected variable(s) at the level of the 10 SEFI sites.
on resolution of available datasets. The interpolation process
was weighted by the number of households (or the number of Model Prediction

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

children for certain variables) in each cluster. The spatial weights We utilized the calibrated log linear regression to predict the
applied inverse distance with the power of 2 for the exponent of primary study outcome of typhoid incidence (cases per 100 000
distance that controlled the significance of surrounding points person-years) at the level of a 5 × 5-km grid using the selected
on the interpolated value (ie, the weight of known points on variable. There was a total of 160  800 grids in India. The ty-
unknown locations diminished with distance). The interpo- phoid incidence estimate was aggregated to a state level, with
lation results were evaluated through cross-validation. The population weighting for each grid. Statistical analyses were
cross-validation process separated the data into a training set to implemented in R 3.6.1. The methodology of this study is pre-
calibrate the model and a test set to evaluate the predictive per- sented in Figure 1.
formance on data that were withheld from the model during the We computed the 95% uncertainty interval (UI) for the pre-
calibration. The raster output from inverse distance weighting diction of typhoid fever incidence using a resampling process
methodology was then converted to 5 × 5-km polygon vector that accounted for the uncertainty in the original SEFI site es-
data to spatially intersect with study sites. The spatial data proc- timates of typhoid incidence. The process first sampled from
essing was implemented in ArcGIS 10.7.1. the 95% interval on typhoid incidence for each SEFI site, which
was bounded by a beta distribution. We then recalibrated the
Model Calibration model using the 10 SEFI study site incidence estimates, re-
We utilized a log linear regression model to estimate the rela- peated the typhoid incidence prediction for India, and stored
tionship between typhoid incidence (dependent variable, cases the mean estimate. This process was repeated 1000 times to

Typhoid
incidence
Weighted by
data
cluster size
10 SEFI sites

DHS data* Spatial Variable

Model
interpolation Spatial selection
Cluster calibration
intersect

Model
Cluster average prediction
Remove
missing data

Typhoid
incidence
in India

Figure 1.  Summary of the study design for prediction of typhoid incidence in India. The study design followed the outlined process in the figure. We used Demographic
and Health Survey (DHS) data on model variables to serve as predictors of typhoid incidence. The DHS variable data were averaged at a cluster level and then interpolated
on a 5 × 5-km grid. We geographically intersected the DHS model variable data with the Surveillance for Enteric Fever in India (SEFI) data on observed typhoid incidence. We
calibrated a model to estimate the relationship between each DHS model variable and typhoid incidence, and then we utilized a backward selection algorithm for variable
selection. When the Akaike Information Criterion was minimized, we used the selected variable(s) as the predictor of typhoid incidence for the model. The rectangles refer to
input/output data. The rhomboid shape refers to data processing. The gray shaded color indicates that additional data/processing steps.

Geographic Pattern of Typhoid Fever in India  •  jid 2021:224 (Suppl 5) • S477

ensure convergence of the estimate. The final 95% UI in this is summarized in Table 1. The calibrated model was able to
study was based on the 2.5% and 97.5% percentile of this range broadly reproduce the estimated pattern of typhoid incidence
of values. We also computed secondary study outcomes in- in many observed settings (Table 3).
cluding (1) total number of national annual typhoid fever cases
based on a population weighted incidence and (2) annual mor- Model Prediction of Typhoid Incidence

tality based on a 0.2% case-fatality rate [12]. The national incidence of typhoid fever was estimated to be 360
To test robustness in this analysis, we performed multiple cases per 100 000 person-years (95% UI, 297–494), adjusted for
sensitivity analyses. These included alternative variable selec- blood culture sensitivity. Based on this incidence, we estimated
tions and a leave-one-out analysis to determine the effect of re- 4.5 million (95% UI, 3.7–6.1 million) annual cases of typhoid
moving 1 study site on the overall typhoid incidence prediction. fever with approximately 8930 deaths (95% UI, 7360–12 260),
assuming a 0.2% case-fatality rate. The mean typhoid incidence
Ethic Statement in urban settings was 770 cases per 100 000 person-years (95%

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

This project did not meet the definition of human subjects re- UI, 620–1040), whereas the mean incidence was 150 cases per
search at Stanford University given use of aggregated estimates 100 000 person-years (95% UI, 130–210) in rural settings. We
of typhoid fever incidence without identifiable or person-level noted there was substantial variation in predicted typhoid in-
data. In the  SEFI study, all participants provided informed cidence across the country in Figure 3. The incidence ranged
consent with institutional review board approval at Christian from 149 cases per 100 000 person-years (95% UI, 130–213) for
Medical College, Vellore, as well as approval at each study site. Himachal Pradesh to 1245 cases per 100 000 person-years (95%
UI, 963–1702) for Delhi. Table 2 summarized the predicted inci-
RESULTS dence for each state as well as the prevalence of state urbanicity.
Surveillance for Enteric Fever in India Data on Typhoid Incidence In general, there was higher incidence in southern and western
The overall mean incidence of typhoid fever in the 10 SEFI states (eg, Maharashtra and Tamil Nadu) and urban centers in
sites was 615 cases per 100  000 person-years. The location of the north (eg, Delhi and Chandigarh), whereas there was lower
the 10 sites and corresponding typhoid incidence is displayed incidence in rural northern states (eg, Arunachal Pradesh and
in Figure 2. Typhoid incidence was higher in the cohort studies Himachal Pradesh). Approximately 50% of the geographic area
(Tier 1) with an average of 1080 cases per 100 000 person-years in the country where over 70% population reside had incidence
compared with the hybrid surveillance study (Tier 2) with an over 100 cases per 100 000 person-years; we found that less than
average of 304 cases per 100 000 person-years. The study sites in 10% of the geographic area of the country with approximately
urban settings had a higher mean incidence of typhoid fever of 25% of the country population had incidence over 500 cases per
1310 cases per 100 000 person-years, compared with rural sites, 100 000 person-years.
which had a mean incidence of 151 cases per 100 000 person-
Sensitivity Analysis
years. Among the 10 sites, 3 had typhoid incidence below 100
We performed multiple sensitivity analyses to determine robust-
cases per 100 000 person-years: 61 (95% UI, 24–125) in Vadu, 72
ness of the model prediction. We tested the effect of removing 1
(95% UI, 50–133) in East Champaran, and 79 (95% UI, 59–133)
study site on the overall typhoid incidence prediction, which had
in Karimganj. In addition, 3 sites had an incidence between 100
modest effect on the national typhoid incidence estimate. The re-
and 500 cases per 100 000 person-years: 154 (95% UI, 98–280)
sult of sensitivity analysis was presented in Supplementary Table
in Nandurbar, 266 (95% UI, 176–412) in Anantapur, and 274
S4 in the Appendix. In this analysis, we found that removal of
(95% UI, 179–443) in Kullu. Finally, 4 sites had an incidence
Vellore had the largest effect on the estimate. We also tested al-
of over 500 cases per 100 000 person-years: 981 (95% UI, 717–
ternative variables in the prediction model for typhoid incidence.
1416) in Chandigarh, 1095 (95% UI, 913–1302) in Delhi, 1187
Using a 2-variable model (ie, urban prevalence and improved
(95% UI, 998–1400) in Kolkata, and 1977 (95% UI, 1740–2236)
toilet access), we estimated a national incidence of 364 cases per
in Vellore.
100 000 person-years (95% CI, 287–530). A map of state variation
Variable Selection and Model Calibration is available in Supplementary Figure S2 in the Appendix.
The spatial interpolation of each DHS  variable is available in
DISCUSSION
Supplementary Figure S1 in the Appendix. The prediction error
statistic (root mean squared error) from the inverse distance In this study, we used a geospatial statistical model to estimate
weighting interpolation for each DHS  variable was also dis- the incidence of typhoid fever in India. We computed a national
played in Supplementary Figure S1. We performed a variable incidence of approximately 360 cases per 100 000 person-years
selection process of DHS variables to predict typhoid fever in- with higher burden in urban centers, corresponding to 4.5 mil-
cidence, and identified that urban prevalence minimized the lion cases and 8900 deaths annually in India. This study utilized
AIC. The result from the model calibration using DHS variables statistical modeling of incidence data on typhoid fever from 4

S478  • jid 2021:224 (Suppl 5) • Cao et al

 Kullu

Chandigarh

Delhi

East Champaran

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

Karimganj

Kolkata

Nandurbar

N Vadu

W E

Typhoid incidence Anantapur

< 200
201 – 1000
Vellore
1001 – 2400
Study tiers
Tier 1
Tier 2
610 305 0 610 Kilometers

Figure 2.  Spatial distribution of the 10 Surveillance for Enteric Fever in India (SEFI) study sites in India. The circles indicate the location of the 10 SEFI sites. The pink circles
refer to 4 cohort study sites in Tier 1, and orange circles refer to 6 hybrid surveillance surveillance study sites in Tier 2. The size of the circles were categorized in 3 levels
of incidence: fewer than 200 cases per 100 000 person-years (small circle), 201–1000 cases per 100 000 person-years (medium circle), and over 1000 cases per 100 000
person-years (large circle).

cohort and 6 hybrid surveillance studies, while incorporating in larger urban centers that may be related to living conditions
data from a national health survey to use as predictors of ty- such as density, sanitation, and other environmental factors, al-
phoid incidence. The key study limitation was a modest sample though there was still typhoid fever found in rural areas. Prior
size to calibrate our model, which mainly relied on urbanicity epidemiologic work has demonstrated the risk of typhoid fever
alone to predict typhoid fever incidence. Our national estimate in urban centers [9, 19]. This suggests a higher risk in these
on incidence of typhoid fever in India is generally consistent areas, which may support prioritization of vaccination in these
with previous studies and supports that there is a large burden settings [37]. We also found a relationship between typhoid in-
of typhoid fever in India that would benefit from national cidence and growth metrics, vaccination, and improved toilet/
vaccination. water; some of these variables may have a causal relationship
Our findings suggest substantial variation of typhoid inci- with typhoid fever infection, but they also may only be correl-
dence across the country. We found higher incidence in urban ated with risk factors for typhoid fever, and this analysis is lim-
centers in the north and southwestern states and lower inci- ited by a small sample size.
dence in northern rural regions driven largely by the single WHO now recommends introducing typhoid conjugate
model variable of urbanicity. This urban-rural disparity high- vaccines in areas with high typhoid incidence [38]. Prior
lights that the burden of typhoid fever in India is predominately cost-effectiveness models have found that typhoid conjugate

Geographic Pattern of Typhoid Fever in India  •  jid 2021:224 (Suppl 5) • S479

Table 1.  Univariate Regression on Predictors of Typhoid Fever Incidence about whether and how to introduce typhoid conjugate vac-
cines. Data on the burden of typhoid fever across the country
Univariate Regression
are important to support  these decisions. Despite substantial
Variables Coefficient 95% CI AIC spatial variation, we estimated that the incidence of typhoid in
Urban prevalencea 2.83 (2.47–3.18) 4.24 all states were likely above 100 per 100 000 person-years, which
Improved toilet access (binary) 2.72 (0.15–5.29) 34.35 has been defined as “high burden” for typhoid [3]. These find-
Education level (tertile) 1.45 (−0.76 to 3.55) 36.77 ings suggest that a nationwide vaccine introduction, rather than
Access to vaccination (3rd dose, 6.54 (1.98–11.11) 31.79
diphtheria, tetanus, and
a geographically targeted one, may be required for control of
­pertussis) typhoid and mitigation of its health impacts in India.
Wealth (quintile) 0.78 (0.19–1.37) 32.61 The study findings should be interpreted within the limita-
Household size 0.02 (−0.96 to 1.00) 38.64
tions of the data and analysis. The study had a key limitation in
Improved water access (binary) 0.62 (−4.99 to 6.24) 38.59
sample size; we used 10 sites to calibrate the model to predict

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

Stunting prevalence −7.68 (−11.96 to −3.39) 29.31
Underweight prevalence −3.36 (−8.15 to 1.41) 36.51 typhoid fever incidence across India, which limited the accu-
Abbreviations: AIC, Akaike Information Criterion; CI, confidence interval.
racy and validation of the model prediction. Measurement of
The model in the univariate regression was a log linear regression using the 10 study sites typhoid fever incidence at a given location requires a resource-
(N = 10). In this regression test, the dependent variable was typhoid incidence at each study
site (cases per 100 000 person-years). The coefficient represents a log transformation.
intensive methodology (eg, multiyear cohort study or hybrid
a
Urban prevalence was computed as the average of a binary urban/rural household variable surveillance), which limits the number of locations where ty-
at the cluster level.
phoid incidence can be reliably estimated. We adjusted our
case data for an estimated 60% sensitivity of blood cultures for
vaccines  would be cost-effective in routine immunization diagnosis of typhoid fever. However, this adjustment could be
programs for countries with high typhoid incidence [39, 40, limited because blood culture sensitivity varies across locations,
41]. The Government of India is preparing to make a decision and it could further result in underestimation or overestimation

B Predicted incidence

40 000

30 000
Frequency

20 000

10 000

Predicted incidence
per 100,000 person-years
< 200 0
200 – 299 N
300 – 499 0 500 1000 1500
0 240 480 960 Kilometers
500 – 699
Incidence (typhoid fever cases per 100,000 person-years)
≥700

Figure 3.  Predicted incidence of typhoid fever in India. We calibrated a statistical model to predict typhoid fever incidence in India using data from 10 Surveillance for
Enteric Fever in India study sites. We used a log linear regression model to predict typhoid incidence across the country using secondary data obtained from Demographic
and Health Survey in India. The estimated incidence was at 5 × 5-km grid level and was aggregated at state level and mapped in (a). The histogram of incidence at original
grid level was visualized in (b).

S480  • jid 2021:224 (Suppl 5) • Cao et al

Table 2.  Predicted Incidence of Typhoid Fever at a State Level in India Table 3.  Comparison of observed and predicted incidence of typhoid
fever in SEFI study sites
State Incidence (95% UI) %Urban Population
Site Original Predicted
Andaman and Nicobar 232 (196–324) 31.5
Andhra Pradesh 390 (321–534) 38.4 Incidence 95 %UI Incidence 95 %UI
Arunachal Pradesh 204 (175–286) 19.2
Anantapur 266 (176–412) 400 (334–543)
Assam 166 (144–235) 15.2
Chandigarh 981 (717–1416) 941 (744–1280)
Bihar 162 (140–230) 13.8
Delhi 1095 (913–1302) 1313 (1010–1799)
Chandigarh 905 (719–1228) 90.7
East Champaran 72 (50–113) 80 (71–124)
Chhattisgarh 305 (253–421) 29
Karimganj 79 (59–133) 96 (86–144)
Dadra and Nagar Haveli 446 (369–607) 45.5
Kolkata 1187 (998–1400) 1313 (1010–1799)
Daman and Diu 564 (455–768) 72.3
Kullu 274 (179–443) 274 (239–371)
Delhi 1245 (963–1702) 97.2
Nandurbar 154 (98–280) 137 (120–198)

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

Goa 400 (339–540) 56.6
Vadu 61 (24–125) 192 (174–263)
Gujarat 457 (374–624) 44
Vellore 1977 (1740–2236) 1185 (926–1613)
Haryana 393 (326–536) 39.7
Himachal Pradesh 149 (130–213) 13.2 The original typhoid incidence was provided by SEFI. The predicted incidence for each site
was based on the model prediction. All incidence estimates are presented as cases per
Jammu and Kashmir 249 (210–346) 24.1 100 000 person-years.
Jharkhand 298 (248–411) 28.1 Abbreviations: SEFI, Surveillance for Enteric Fever in India; UI, uncertainty interval
Karnataka 441 (362–602) 43.3
Kerala 429 (356–582) 44.4
in the model to limit the risk of overfitting given limited data,
Madhya Pradesh 305 (255–419) 30.7
Maharashtra 515 (418–703) 48.4 both of which limited the model’s predictive accuracy (Table
Manipur 342 (283–470) 32.8 3). The model identified a negative association between ty-
Meghalaya 254 (211–354) 22.3 phoid incidence and being stunted or underweight (Table 1).
Mizoram 444 (359–609) 45.4 Our urban sites (eg, Chandigarh, Delhi, Kolkata, and Vellore)
Nagaland 264 (224–365) 26.7
had substantially higher typhoid incidence than rural sites, yet
Orissa 224 (190–313) 20.8
stunting and underweight are much less common in urban lo-
Puducherry 659 (521–904) 55.5
Punjab 427 (353–580) 43.9 cations, which likely explains this negative association. Due to
Rajasthan 307 (256–424) 30 the small sample size for the model calibration, we were unable
Sikkim 199 (174–276) 21.8 to perform a meaningful validation of the model prediction. In
Tamil Nadu 494 (407–669) 50.2 addition, the 4 study sites in Tier 1 included all children, and
Tripura 285 (237–394) 26.4
3 of these sites were in urban settings. Because typhoid fever
Uttar Pradesh 282 (235–390) 27
Uttaranchal 360 (302–490) 37.4
has greater risk for children, this site selection could bias our
West Bengal 395 (323–543) 36.2 estimate towards a higher incidence given the urban settings
Country average 360 (297–494) 34.3 and pediatric population included in this study. Estimation bias
Abbreviations: UI, uncertainty interval. could have also resulted from the differences in the time frame
All incidence estimates are presented as cases per 100 000 persons between sample collection for SEFI and DHS. Finally, due to a
limited sample size, we did not include age-specific incidence
estimates, although there is likely a strong age correlate of risk.
of typhoid incidence in certain areas. To address the concern
CONCLUSIONS
of the influence of a single site on the overall national typhoid
estimate given a small sample size, we performed sensitivity There is a substantial disease burden of typhoid fever across
analyses and found that 1 site did not disproportionately af- India, with higher typhoid incidence in urban centers. This
fect the estimate. We relied upon secondary data from the study supports immunization with the Vi conjugate typhoid
DHS to predict typhoid incidence; these data required some vaccine to address the disease burden from typhoid fever in
data processing, interpolation, and aggregation to a grid level India.
(5 × 5 km), which could introduce imprecision and bias into
the estimate. An independent assessment of DHS data was Supplementary Data

done through comparison to the state urban population data Supplementary materials are available at The Journal of Infectious
from the Indian Ministry of Home Affairs [42]. Supplementary Diseases online. Consisting of data provided by the authors to
Table S5 in the Appendix shows our computed urban pop- benefit the reader, the posted materials are not copyedited and
ulation and data from the national India Census, which were are the sole responsibility of the authors, so questions or com-
overall comparable. We constrained the number of predictors ments should be addressed to the corresponding author.

Geographic Pattern of Typhoid Fever in India  •  jid 2021:224 (Suppl 5) • S481

Notes 8. Parry CM, Hien TT, Dougan G, White NJ, Farrar JJ. Typhoid
Acknowledgments. We thank the study participants and re- fever. N Engl J Med 2002; 347:1770–82.
search team who assisted with the Surveillance for Enteric Fever 9. Corner RJ, Dewan AM, Hashizume M. Modelling typhoid
in India (SEFI) studies. risk in Dhaka metropolitan area of Bangladesh: the role of
Author contributions. Y.  C.  and N.  C. L.  had full access socio-economic and environmental factors. Int J Health
to all of the data in the study and take responsibility for the Geogr 2013; 12:13.
integrity of the data and the accuracy of the data analysis. 10. John J, Van Aart CJ, Grassly NC. The burden of typhoid and
G. K., J. J., J. R. A., and N. C. L. contributed to study concep- paratyphoid in india: systematic review and meta-analysis.
tion. All authors contributed to study design, data analysis, PLoS Negl Trop Dis 2016; 10:e0004616.
and data interpretation. Y.  C.  and N.  C. L.  wrote the first 11. Kanungo  S, Dutta  S, Sur  D. Epidemiology of typhoid
draft. All authors contributed to intellectual material and ap- and paratyphoid fever in India. J Infect Dev Ctries 2008;
2:454–60.

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

proved the final draft.
Disclaimer. The funding organizations had no role in the de- 12. Yu  AT, Amin  N, Rahman  MW, Gurley  ES, Rahman  KM,
sign and conduct of the study; collection, management, anal- Luby  SP. Case-fatality ratio of blood culture-confirmed
ysis, and interpretation of the data; and preparation, review, or typhoid fever in Dhaka, Bangladesh. J Infect Dis 2018;
approval of the manuscript; or the decision to submit the man- 218:222–6.
uscript for publication. 13. Sur D, Barkume C, Mukhopadhyay B, Date K, Ganguly NK,
Supplement sponsorship. This supplement is sponsored by Garrett  D. A retrospective review of hospital-based data
the Christian Medical College Vellore Association. on enteric fever in India, 2014–2015. J Infect Dis 2018;
Financial support. This study was funded by the Bill & Melinda 218:S206–13.
Gates Foundation (OPP1159351). N.  C. L.  is supported by the 14. Sur  D, Ali  M, von  Seidlein  L, et  al. Comparisons of pre-
University of California, San Francisco (Department of Medicine). dictors for typhoid and paratyphoid fever in Kolkata, India.
Potential conflicts of interest. N.  C. L.  reports personal fees BMC Public Health 2007; 7:289.
from the World Health Organization, unrelated to the present 15. Kim S, Lee KS, Pak GD, et al. Spatial and temporal patterns
work. All authors have submitted the ICMJE Form for Disclosure of typhoid and paratyphoid fever outbreaks: a worldwide
of Potential Conflicts of Interest. Conflicts that the editors consider review, 1990–2018. Clin Infect Dis 2019; 69:S499–509.
relevant to the content of the manuscript have been disclosed. 16. Sur  D, Ochiai  RL, Bhattacharya  SK, et  al. A cluster-
randomized effectiveness trial of Vi typhoid vaccine in
India. N Engl J Med 2009; 361:335–44.
References 17. Sinha A, Sazawal S, Kumar R, et al. Typhoid fever in children
1. World Health Organization. Typhoid. Available at: https:// aged less than 5 years. Lancet 1999; 354:734–7.
www.who.int/immunization/diseases/typhoid/en/. 18. Ochiai RL, Acosta CJ, Danovaro-Holliday MC, et al; Domi
Accessed 1 April 2020. Typhoid Study Group. A study of typhoid fever in five Asian
2. Mogasale V, Maskery B, Ochiai RL, et al. Burden of typhoid countries: disease burden and implications for controls.
fever in low-income and middle-income countries: a sys- Bull World Health Organ 2008; 86:260–8.
tematic, literature-based update with risk-factor adjust- 19. Bhunia  R, Hutin  Y, Ramakrishnan  R, Pal  N, Sen  T,
ment. Lancet Glob Health 2014; 2:e570–80. Murhekar M. A typhoid fever outbreak in a slum of South
3. Crump  JA, Luby  SP, Mintz  ED. The global burden of ty- Dumdum municipality, West Bengal, India, 2007: evidence
phoid fever. Bull World Health Organ 2004; 82:346–53. for foodborne and waterborne transmission. BMC Public
4. Stanaway JD, Reiner RC, Blacker BF, et al. The global burden Health 2009; 9:1–8.
of typhoid and paratyphoid fevers: a systematic analysis for 20. World Health Organization. Typhoid vaccine: WHO pos-
the global burden of disease study 2017. Lancet Infect Dis ition paper - March 2018. Wkly Epidemiol Rec 2018;
2019; 19:369–81. 13:153–72.
5. Bhan MK, Bahl R, Bhatnagar S. Typhoid and paratyphoid 21. Vellore CMC. National surveillance system for enteric fever
fever. Lancet 2005; 366:749–62. in India. 2018:1–21.
6. Hornick  RB, Greisman  SE, Woodward  TE, DuPont  HL, 22. Vellore CMC. National surveillance system for enteric fever
Dawkins AT, Snyder MJ. Typhoid fever: pathogenesis and in India hospital-based surveillance (Tier 2). 2018; (Tier
immunologic control. 2. N Engl J Med 1970; 283:739–46. 2):1–20.
7. Buckle  GC, Walker  CLF, Black  RE. Typhoid fever and 23. Li Y, Shetty AC, Lon C, et al. Detecting geospatial patterns
paratyphoid fever: systematic review to estimate global of Plasmodium falciparum parasite migration in Cambodia
morbidity and mortality for 2010. J Glob Health 2012; using optimized estimated effective migration surfaces. Int
2:1–9. J Health Geogr 2020; 19:13.

S482  • jid 2021:224 (Suppl 5) • Cao et al

24. Weiss DJ, Lucas TCD, Nguyen M, et al. Mapping the global 34. WHO/UNICEF Joint Monitoring Programme (JMP) for
prevalence, incidence, and mortality of Plasmodium falcip- Water Supply and Sanitation. Available at: https://www.
arum, 2000–17: a spatial and temporal modelling study. unwater.org/publication_categories/whounicef-joint-
Lancet 2019; 394:322–31. monitoring-programme-for-water-supply-sanitation-
25. Lai YS, Biedermann P, Ekpo UF, et al. Spatial distribution of hygiene-jmp/. Accessed 1 December 2019.
schistosomiasis and treatment needs in sub-Saharan Africa: 35. WorldPop. Open spatial demographic data and research.
a systematic review and geostatistical analysis. Lancet Infect Available at: https://www.worldpop.org. Accessed 1 October
Dis 2015; 15:927–40. 2019.
26. Tamrakar D, Vaidya K, Yu AT, et al. Spatial heterogeneity of 36. Burgert  CR, Colston  J, Roy  T, Zachary  B. Geographic dis-
enteric fever in 2 diverse communities in Nepal. Clin Infect placement procedure and georeferenced data release policy
Dis 2020; 71:205–13. for the demographic and health surveys. DHS Spat. Anal.
27. John J, Bavdekar A, Rongsen-Chandola T, Dutta S, Kang G; Reports No. 7. ICF International Calverton, Maryland, 2013.

Downloaded from https://academic.oup.com/jid/article/224/Supplement_5/S475/6433809 by guest on 28 February 2023

NSSEFI Collaborators. Estimating the incidence of enteric 37. Andrews JR, Vaidya K, Bern C, et al. High rates of enteric fever
fever in children in India: a multi-site, active fever surveil- diagnosis and lower burden of culture-confirmed disease in
lance of pediatric cohorts. BMC Public Health 2018; 18:594. peri-urban and rural Nepal. J Infect Dis 2018; 218:214–21.
28. Andrews JR, Barkume C, Yu AT, et al. Integrating facility- 38. World Health Organization. WHO recommends use of first
based surveillance with healthcare utilization surveys to typhoid conjugate vaccine. Available at: http://www.who.int/
estimate enteric fever incidence: methods and challenges. J immunization/newsroom/press/who_recommends_use_
Infect Dis 2018; 218:268–76. first_typhoid_conjugate_vaccine/en. Accessed 1 November
29. USAID. Demographic and health surveys standard re- 2019.
code manual for DHS 7. 2018: pp 145. Available at: 39. Lo NC, Gupta R, Stanaway JD, et al. Comparison of strat-
https://dhsprogram.com/pubs/pdf/DHSG4/Recode7_ egies and incidence thresholds for vi conjugate vaccines
DHS_10Sep2018_DHSG4.pdf. Accessed 1 October 2019. against typhoid fever: a cost-effectiveness modeling study.
30. Demographic and Health Surveys. Available at: https:// J Infect Dis 2018; 218:232–42.
dhsprogram.com/Data. Accessed 1 October 2019. 40. Bilcke J, Antillón M, Pieters Z, et al. Cost-effectiveness of
31. Khare  S, Kavyashree  S, Gupta  D, Jyotishi  A. Investigation routine and campaign use of typhoid Vi-conjugate vaccine
of nutritional status of children based on machine learning in Gavi-eligible countries: a modelling study. Lancet Infect
techniques using Indian demographic and health survey Dis 2019; 19:728–39.
data. Procedia Comput Sci 2017; 115:338–49. 41. Ryckman  T, Karthikeyan  A, Kumar  D, Cao  Y, Kang  G,
32. Oxlade  O, Murray  M. Tuberculosis and poverty: why are Goldhaber-Fiebert J, John J, Lo N, Andrews J. Comparison
the poor at greater risk in India? PLoS One 2012; 7:e47533. of Strategies for Typhoid Conjugate Vaccine Introduction in
33. Sreeramareddy  CT, Shidhaye  RR, Sathiakumar  N. India: A Cost-Effectiveness Modeling Study.
Association between biomass fuel use and maternal re- 42. Government of India Ministry of Home Affairs. Available
port of child size at birth–an analysis of 2005-06 India at: https://censusindia.gov.in/2011-prov-results/paper2-
Demographic Health Survey data. BMC Public Health vol2/data_files/Meghalaya/CHAPTER_IV.pdf. Accessed 1
2011; 11:403. March 2021.

Geographic Pattern of Typhoid Fever in India  •  jid 2021:224 (Suppl 5) • S483