A Bayesian approach for estimating typhoid fever incidence from large-scale facility-based passive surveillance data

Abstract

Background. Decisions about typhoid fever prevention and control are based on current estimates of typhoid incidence and their uncertainty, which can be difficult to measure. Limits of using facility-based estimates alone- the lack of specific clinical diagnostic criteria, poorly sensitive and specific diagnostic tests, and scarcity of accurate and complete datasets- contribute to difficulties in calculating population-level incidence of typhoid. Methods. Using data from the Strategic Alliance across Africa & Asia (STRATAA) programme, we integrated information from demographic censuses, healthcare utilization surveys, facility-based passive surveillance, and serological surveillance from sites in Malawi, Nepal, and Bangladesh in order to adjust crude incidence estimates to account for under-detection. We developed an approach using a Bayesian framework that adjusts the count of reported blood-culture-positive cases of typhoid for each of the following phases: healthcare seeking, blood culture collection, and blood culture detection. We estimated the proportion of "true" typhoid cases occurring in the population under surveillance captured at each phase by combining information from the observed cases from the STRATAA datasets and estimates from prior published studies. We confirmed that the model was correctly formulated by comparing to simulated data. Results. The ratio between the observed and adjusted incidence rates was 8.2 (95% CI: 6.4-13.3) in Malawi, 13.8 (95% CI: 8.8-23.0) in Nepal, and 7.0 (95% CI: 5.5-9.1) in Bangladesh, and varied by age across the three sites. The probability of having blood drawn for culture led to the largest adjustment in Malawi, while the probability of seeking healthcare contributed the most to adjustment factors in Nepal and Bangladesh. Adjusted incidence rates were mostly within the limits of the seroincidence rate of typhoid infection determined by serological data. Conclusion. Passive surveillance of blood culture-confirmed typhoid fever without adjustment for case ascertainment, sample collection and diagnostic sensitivity results in considerable underestimation of the true incidence of typhoid in the population. Our approach allows each phase of the typhoid reporting process to be synthesized to estimate the adjusted incidence of typhoid fever while correctly characterizing uncertainty in this estimate, which can inform decision-making for typhoid prevention and control.

Burden of enteric fever at three urban sites in Africa and Asia: a multicentre population-based study

“…Use of alternative health-care facilities and pharmacies was higher at Asian sites than the African site, leading to higher inflation of incidence once adjusted, whereas in Blantyre, fewer individuals who sought health care at study facilities received a blood culture due to high incidence of febrile illness, inadequate health-care infrastructure, and parental refusal. 18 In Blantyre, individuals with typhoid presented with more severe fever and a longer duration of fever, and hospitalisation was two times higher than at the other two sites. This difference might indicate that despite the adjustments for sampling and health-care seeking behaviour, undetected mild or subclinical disease remains an ongoing burden, consistent with the comparatively high rates of seroconversion observed at the Blantyre site.…”

“…The largest adjustments at each site were for the the probability of being enrolled and receiving a blood culture in Blantyre (2•9-fold increase) and for the probability of seeking health care in Kathmandu (6•7-fold increase) and Dhaka (3•7-fold increase; table 2, figure 2). 18 The antimicrobial susceptibility profiles for both S Typhi and S Paratyphi A differed at each site (appendix p 9 The seroconversion data showed different rates of presumed exposure to and infection with S Typhi in the three sites (figure 3; appendix p 15). Overall seroincidence was highest in Kathmandu (7631 episodes [95% CI 5913-9691] per 100 000 person-years of observation), was more than seven times higher than the estimated adjusted incidence, and more than 100 times higher than the observed blood-culture-confirmed incidence.…”

“…Adjustments factors were site-specific and varied by age group. 18 We calculated median values and 95% Bayesian credible intervals for the adjusted incidence estimates ( appendix p 2 ).…”

Environmental surveillance for Salmonella Typhi as a tool to estimate the incidence of typhoid fever in low-income populations

“…The estimated incidence of typhoid fever among children in Vellore during 2017-2020 was about twice that for Blantyre during a similar time period 2016-2018 (approximately 2% vs 1% per year of observation for children <15 years old) [29]. Pilot ES surveillance data from Vellore based on direct molecular detection methods using either grab samples (bag-mediated filtration) or trap samples (Moore swabs), detected S .…”

bootComb—an R package to derive confidence intervals for combinations of independent parameter estimates

“…The impact of study or facility-based limitations on parameter estimates is well-known 1 and common adjustment factors include the probability of seeking healthcare or of receiving a diagnostic test (both in the case of facility-based estimates), the incidence or prevalence of a related condition (in the case of a conditional disease prevalence/incidence), or the operational characteristics of the diagnostic test (in the case of imperfect diagnostic tests). A recent example includes the estimation of typhoid incidence 1 where a Bayesian model was used to derive adjustment factors. While usually easy to combine point estimates, it is often difficult to obtain a valid confidence interval (CI) for the combined parameter.…”