Burden of typhoid and paratyphoid fever in India

Background: In 2017, over half the global burden of typhoid fever was projected to have occurred in India. In the absence of contemporary population-based data, it is unclear whether declining trends of hospitalization for typhoid in India reflect increased antibiotic treatment or a true reduction in infection. Methods: We conducted weekly surveillance for acute febrile illness and measured the incidence of blood culture-confirmed typhoid fever in a prospective cohort of children 6 months to 14 years old at three urban and one rural site in India between 2017 and 2020. At an additional urban and five rural sites, we combined blood culture testing of hospitalized patients with fever with health care utilization surveys to estimate incidence in the community. Results: 24,062 children were enrolled across four cohorts, contributing 46,959 child years of observation (CYO). 299 culture-confirmed typhoid cases were recorded, with incidence per 100,000 CYO of between 576 and 1173 in urban sites, and 35 in rural Pune. The estimated incidence of typhoid fever from hospital surveillance ranged between 12 and 1622 per 100,000 CYO in children 6 months to 15 years, and between 108 and 970 per 100,000 person-years among those above 15 years, although there was more uncertainty in these estimates. S. paratyphi was isolated from 33 children, overall incidence of 68 per 100,000 CYO after adjusting for age Conclusions: The incidence of typhoid fever in urban India remains high.


Burden of typhoid and paratyphoid fever in India
permitted in the analysis. Age-group specific incidence was estimated using the 'stsplit' function in STATA and the overall incidence calculated as an average across age-groups, weighted to reflect the age structure of the underlying population at each site. Confidence intervals were calculated using the quadratic approximation to the Poisson log-likelihood for the log-rate parameter. We investigated the association of typhoid incidence with baseline household characteristics and location using Andersen-Gill's proportional hazards model. 1

Tier 1 notification
We did not explicitly measure the duration of time between notification and visit in the cohort study. However, since most of the weekly surveillance visits were in-person, the bulk of the notifications for cases were at the time of the first visit. When notifications were identified during the telephonic contact, visits were initiated within one day. We found that the mean interval between the onset of fever and the first visit was 1.89 days (sd 1.31days).

Tier 2 Hospital based surveillance sites
The surveillance sites were chosen to represent different geographic and risk settings such as climate and population density. We considered the site's ability to conduct research and the proportion of the catchment population accessing healthcare at the study hospital. Since all sites were required to conduct blood cultures on all febrile admissions, this necessitated the establishment of culture facilities at some facilities, with quality control procedures including proficiency testing at all facilities. All sites provided data on febrile hospitalizations during 2016 and 2017, and geographically contiguous administrative areas from which 80% of the hospitalized febrile patients came to the study facility constituted the catchment for each site.
A brief description of surveillance sites follows.
1) The Anantapur site in the southern state of Andhra Pradesh had a catchment population of 487,000 served by the 350-bed Rural Development Trust Hospital (RDT) at Bathallapali 2) The Nandurbar site in the western state of Maharashtra had a population of 311,000 with a predominantly tribal population was served by Chinchpada Christian Hospital, a 50-bed charitable hospital.
3) The East Champaran site in Bihar is one of the most backward districts of the country. This site has a catchment population of 530,000 in the Raxaul and Ramgarhwa blocks and is served by the Duncan Hospital, Raxaul, a 200-bedded surveillance facility.
4) The Karimganj site in the north-eastern state of Assam is located at the tri-junction of Assam, Tripura, and Mizoram. The Makunda Christian Leprosy and General Hospital is a 160-bed charitable facility that serves Lowairpoa and Patharkandi blocks with a population of 384,000.
5) The Kullu site in the Himalayan ranges of Himachal Pradesh, with a catchment population of 123,000, is served by the 55-bed Lady Willingdon Hospital in Manali. 6) Chandigarh, the only urban site, identified a population of 143,000 in Sectors 45 and 52 as its catchment population. The Civil Hospital at Sector 45 Chandigarh, a 50-bedded government hospital, was the primary inpatient healthcare facility for this population.

Establishment of the hospital surveillance system
Physicians recruited patients older than six months hospitalized at each study facility with presenting complaint of fever, irrespective of the duration or recorded temperature. At admission, study personnel took informed consent and collected age-appropriate blood samples for blood culture. The patients were managed using sitespecific treatment protocols. The study protocol required no intervention other than blood culture and data collection. We collected sociodemographic information, history of prior treatment, clinical and laboratory data on electronic case report forms (CRF). The data was stored on a secured, cloud-based, custom-built data management system. Patients with blood-culture confirmed enteric fever were monitored daily to document temperature trends, antibiotic therapy, and complications if any.

Quality control of data and monitoring
The study protocol was harmonised by the coordination unit at the Christian Medical College, Vellore (CMCV), in consultation with the Scientific Advisory Process for Optimal Research on Typhoid (SAPORT), an advisory group established by the Bill and Melinda Gates Foundation, with participation from the World Health Organisation. The coordination unit monitored sites through frequent site visits and data validation checks.
Deviations from the protocol were addressed by joint review and re-training. Laboratories participated in an external quality assurance system.

Laboratory methods
At admission, age-appropriate blood volumes (3 ml for infants; 5 ml for those between 1 and 15 years; 8 -10 ml for those older than 15 years) were collected and cultured using an automated system (BD BACTEC TM blood culture system). All the S. Typhi and S. Paratyphi isolates were reconfirmed and characterised for antimicrobial susceptibility at the central laboratory at CMCV.

Healthcare utilization survey
Two rounds of healthcare utilization surveys were conducted in 2018 and 2019 to identify the proportion of all febrile hospitalizations from the catchment population at the study hospital. A two-stage cluster sampling strategy was used for the selection of households. In the first stage, a random sample of 100 geographical clusters was selected in each site using probability proportional to size method. In the second stage, 50 households were selected from each cluster by systematic random sampling. The methods of the healthcare utilization survey are detailed elsewhere. 2

Incidence calculation and statistical methods
Crude incidence of severe (hospitalized) enteric fever was calculated by dividing the number of culture-confirmed typhoid and paratyphoid fevers by the catchment population and adjusting for the period of surveillance. The catchment area population was estimated based on projections for the year 2019 from the 2011 census. Only culture-positive enteric fever cases at the study facility from the catchment area contributed to the numerator.
Given the many factors that affect enteric fever incidence measurement through hospital-based surveillance, we performed adjustments (Supplementary Figure S1). Since only a subset of all febrile hospitalizations occurred at the surveillance facility, we adjusted for the proportion of illness with similar presentations treated elsewhere based on the healthcare utilization survey (A2). This ranged from 0.10 in Nandurbar to 0.38 in Chandigarh and Karimganj. 2 A subset of participants admitted with fever failed to receive a blood culture either because of nonconsent or operational reasons (ranging from 0-40% among sites), and they were assumed to have a risk of typhoid like those who received a blood culture at the same age and site (A3). We adjusted for the poor sensitivity of blood culture using an adjustment based on blood volume inoculated in culture. (A4). The correction for blood culture sensitivity was assumed to be 60% with an uncertainty range from 50-70%. 3 These adjustments were incorporated in a probabilistic multiplier model. A Monte Carlo approach using a beta distribution generated 1,000 randomly sampled iterations to account for uncertainty in the multiplication parameters. We used the simulation results to obtain the median and 95% uncertainty intervals for the number of total cases of febrile hospitalizations in the catchment area (Supplementary Table S7, S8).
As an additional analysis to permit comparison with studies that estimate the incidence of enteric fever of all severity, we adjusted our estimates of severe enteric fever for the proportion of enteric fevers that require hospitalization (0.154) in cohort studies that formed the tier 1 surveillance (A1). We performed a sensitivity analysis using a lower blood culture sensitivity of 40% (varied 30-50%) for the site-and age-specific fraction of study participants who received antibiotics prior to blood culture collection. Statistical analysis was performed using STATA 15.0 and R 3.6.1 Figure S1: Typhoid disease hybrid surveillance pyramid. The apex represents culture-confirmed typhoid hospitalizations identified in the study hospital. The base of the triangle represents the true burden of typhoid fever in the community. To estimate the true incidence in the community, the crude incidence rate obtained from the hospitalbased surveillance is adjusted for A1 (severity of typhoid requiring hospitalization), A2 (proportion seeking healthcare at study facility), A3 (Proportion of hospitalized fevers receiving blood culture), & A4 (sensitivity of blood culture).

Figure S2: Incidence of culture-confirmed typhoid and paratyphoid fever by Tier 1 cohort site, age, and time.
A) Incidence of typhoid fever and B) incidence of paratyphoid fever by study site and age group. Error bars indicate 95% confidence intervals. C) Monthly incidence (unadjusted) of typhoid and paratyphoid fever at each site with 95% confidence intervals given by the shaded region and periods of high rainfall (monsoon) indicated by the grey rectangles. Incidence in the 0.5-14 years age group is adjusted to match the underlying age-distribution in panel A and B.   Data are n (%) except where indicated. *Adjusted to match the underlying age distribution of the populations. ¶An episode of fever was counted as a fever from day one to the last day with elevated temperature that was followed by three fever-free days ‡A child with a fever episode of three or more consecutive days was categorized as potential enteric fever (PEF). ∑Children with PEF were eligible for blood culture. $A child who satisfied the PEF criteria but was afebrile over the last 12 hours prior to the time of performing the blood culture. CYO=child-years of observation. IQR=inter-quartile range.

Highest education Mean (SD)
No of years 10.9 (3.8) 9.3 (4.8) 10.9 (4.7) 9.4 (4.9) 9.3 (3.2) 11.7 (3.9) 9.8 (4.3)   * per 100,000 Person Years. # Estimated using Monte Carlo simulation The 95% range for the A1 and A4 adjustment factors were sampled from beta distributions with shape parameters (A1: 46, 253) and (A4: 57, 38), respectively, based on certainty of these estimates. The 95% range for the A2 adjustment factor was sampled from a normal distribution with a logistic transformation. * per 100,000 Person Years. # Estimated using Monte Carlo simulation The 95% range for the A1 and A4 adjustment factors were sampled from beta distributions with shape parameters (A1: 46, 253) and (A4: 57, 38), respectively, based on certainty of these estimates. The 95% range for the A2 adjustment factor was sampled from a normal distribution with a logistic transformation. * per 100,000 Person Years. # Estimated using Monte Carlo simulation This sensitivity analysis used data on the proportion of participants with antibiotics prior to blood culture collection by age group and site, incorporating a lower sensitivity estimate (40%) for blood cultures taken from these study participants. * per 100,000 Person Years. # Estimated using Monte Carlo simulation This sensitivity analysis used data on the proportion of participants with antibiotics prior to blood culture collection by age group and site, incorporating a lower sensitivity estimate (40%) for blood cultures taken from these study participants.