Integrating Facility-Based Surveillance With Healthcare Utilization Surveys to Estimate Enteric Fever Incidence: Methods and Challenges

Abstract

Cohort studies and facility-based sentinel surveillance are common approaches to characterizing infectious disease burden, but present trade-offs; cohort studies are resource-intensive and may alter disease natural history, while sentinel surveillance underestimates incidence in the population. Hybrid surveillance, whereby facility-based surveillance is paired with a community-based healthcare utilization assessment, represents an alternative approach to generating population-based disease incidence estimates with moderate resource investments. Here, we discuss this method in the context of the Surveillance for Enteric Fever in Asia Project (SEAP) study. We describe how data are collected and utilized to adjust enteric fever incidence for blood culture sensitivity, facility-based enrollment, and healthcare seeking, incorporating uncertainty in these parameters in the uncertainty around incidence estimates. We illustrate how selection of surveillance sites and their coverage may influence precision and bias, and we identify approaches in the study design and analysis to minimize and control for these biases. Rigorously designed hybrid surveillance systems can be an efficient approach to generating population-based incidence estimates for infectious diseases.

Figure 1.

Typhoid disease pyramid, where the base represents all typhoid patients in a catchment area and the apex represents culture-positive cases detected at study sites. Culture sensitivity is estimated from the literature, enrollment capture is estimated at facilities, and facility coverage is estimated by a household survey. These factors are utilized to adjust the crude incidence.

Figure 2.

Relative half-width of the 95% confidence interval (CI) achieved by hybrid surveillance as a function of effect sample size (ie, number of households with an individual meeting study definition for febrile illness) and fraction of patients captured at study sites. Here, incidence is 500 per 100000 and the population size is 500000. A higher proportion of participants seeking care at the facility improves the precision of the adjustment factor estimate as well as the precision in the numerator, as more typhoid cases present to the facility. A larger healthcare utilization sample size can improve precision in the adjustment factor estimate, but greater uncertainty in the incidence estimate remains due to uncertainty in the case number.

Figure 3.

A, Log odds ratio for typhoid and care seeking, by age, in Surveillance for Enteric Fever in Asia Project (SEAP) data from Nepal. The reference group is age 0. B, Log odds of typhoid and care seeking by demographic and severity indicators in SEAP data from Nepal. Abbreviation: CI, confidence interval.

Figure 4.

Percentage bias in incidence estimate as a function of the odds ratio for care seeking at study facilities among typhoid patients compared with nontyphoid patients, and the percentage adjusted for in the analysis. The proportion seeking care at the study site is 0.1 (A) and 0.5 (B).

Figure 5.

Incidence estimates and uncertainty intervals according to adjustment method and proportion captured at the surveillance site. We assume 1 year of surveillance in a population of 100000 individuals with typhoid incidence of 500 per 100000, a healthcare utilization assessment of 5000 households, and odds ratio of care seeking at study sites for typhoid patients (bias) of 1.5. Culture sensitivity (assumed 59% [95% confidence interval, 54%–64%]); proportion of eligible individuals at study site enrolled (assumed 80%); adjustment for care seeking (varied, x-axis); bias: adjustment for biases in healthcare seeking (correction of 67% of bias). The dashed line represents the true prevalence.