Evaluating Hospital-Based Surveillance for Outbreak Detection in Bangladesh: Analysis of Healthcare Utilization Data

Background The International Health Regulations outline core requirements to ensure the detection of public health threats of international concern. Assessing the capacity of surveillance systems to detect these threats is crucial for evaluating a country’s ability to meet these requirements. Methods and Findings We propose a framework to evaluate the sensitivity and representativeness of hospital-based surveillance and apply it to severe neurological infectious diseases and fatal respiratory infectious diseases in Bangladesh. We identified cases in selected communities within surveillance hospital catchment areas using key informant and house-to-house surveys and ascertained where cases had sought care. We estimated the probability of surveillance detecting different sized outbreaks by distance from the surveillance hospital and compared characteristics of cases identified in the community and cases attending surveillance hospitals. We estimated that surveillance detected 26% (95% CI 18%–33%) of severe neurological disease cases and 18% (95% CI 16%–21%) of fatal respiratory disease cases residing at 10 km distance from a surveillance hospital. Detection probabilities decreased markedly with distance. The probability of detecting small outbreaks (three cases) dropped below 50% at distances greater than 26 km for severe neurological disease and at distances greater than 7 km for fatal respiratory disease. Characteristics of cases attending surveillance hospitals were largely representative of all cases; however, neurological disease cases aged <5 y or from the lowest socioeconomic group and fatal respiratory disease cases aged ≥60 y were underrepresented. Our estimates of outbreak detection rely on suspected cases that attend a surveillance hospital receiving laboratory confirmation of disease and being reported to the surveillance system. The extent to which this occurs will depend on disease characteristics (e.g., severity and symptom specificity) and surveillance resources. Conclusion We present a new approach to evaluating the sensitivity and representativeness of hospital-based surveillance, making it possible to predict its ability to detect emerging threats.


Estimating case and outbreak detection probability by distance -alternative models
We investigated more complex functional forms of distance in log-binomial regression models, such as polynomial terms up to the 5th degree or basic splines with knots at various positions (between 20 and 50 km distance). Model fit was compared based on the Akaike information criterion (AIC) and models with lowest AIC were selected. The results of the model selection procedure are summarized in Table A. While including a squared distance term led to better optical fit for severe neurological case detection probabilities (Fig B), no improvement of model fit was observed based on the AIC (AIC of 394 for model 1-a log-binomial regression model including only distance as explanatory variable vs. 396 for model 2-the model including additionally a squared distance term). We further explored if outbreak detection probabilities were influenced by model choice and changes in detection probabilities were only minimal ( Fig C). Based on model 2 we estimated that the probability of detecting outbreaks of three severe neurological cases was 60% at 10 km distance (compared to 59% with model 1) and 46% at 30 km distance from surveillance hospitals (compared to 47% with model 1). Outbreak sizes detected with ≥90% probability were of 8 at 10 km distance (8 with model 1) and 12 at 30 km distance (11 with model 1). We further explored the use of basic splines with knots at various locations, however did not detect model improvement based on AIC (Table A) Table A. Comparison of models to predict case detection probabilities by distance. Model selection was based on the lowest AIC, which for both disease types was the log-binomial regression model including distance as linear term (model 1).

Fig B.
Case detection probability for severe neurological disease estimated based on a log-binomial regression model including distance and squared distance as explanatory variables.

Fig C.
Probability of detecting an outbreak of three cases of severe neurological disease (outbreak threshold of at least one detected case) (A) and outbreak size necessary for 90% detection probability (B) based on case detection probabilities estimated by log-binomial regression models including distance and squared distance as explanatory variables.

Case detection bias by age
We investigated reporting probabilities and detection bias by age. Fig D shows detection probabilities of severe neurological and fatal respiratory infections by age estimated by basic spline regression (age basic splines of degree 4 for severe neurological and degree 2 for fatal respiratory disease in logistic regression models) and the proportion of all community and surveillance cases within a moving 5-year age window (at the midpoint of the window). While the youngest severe neurological cases were underrepresented among surveillance cases, the oldest fatal respiratory cases were underrepresented among surveillance cases. For presentational purpose, results are summarized for age categories (<5, 5-14, 15-59, and ≥60 years) in the main text.

Case detection bias by socioeconomic status quintiles
To investigate if classification of cases into socioeconomic status tertiles has influenced the results presented in the main text, we further investigated case detection by quintiles (Fig E). Results are consistent with the tertile analysis, where severe neurological cases of low socioeconomic status were underrepresented among surveillance cases while fatal respiratory cases of high socioeconomic status were overrepresented. Highest 13 17 4 (-3; 12) 0.232 37 43 6 (0; 12) 0.046 Table B. Characteristics of all cases identified in the community and identified cases who attended a surveillance hospital. Ninety-five percent confidence intervals and p-values were obtained using bootstrap.

Outbreak detection probabilities with alternative healthcare providers
To investigate the effects of integrating additional healthcare providers in the surveillance system on the capacity to detect outbreaks, case detection probabilities by distance from the original surveillance hospitals were estimated by combining surveillance hospitals with (i) other hospitals (all other hospitals attended by cases), (ii) healthcare providers of the local formal sector, and (iii) informal healthcare providers. These case detection probabilities were used to estimate the outbreak size required to reach a 90% outbreak detection probability by distance from surveillance hospitals. Including other hospitals that were attended by cases in the surveillance system would allow detecting outbreaks (defined as at least one detected case) of four severe neurological and eight fatal respiratory cases with ≥90% probability at any distance in the range of 0-40 km from the surveillance hospital (Fig F).

Fig F. Outbreak detection including alternative healthcare providers (outbreak threshold of at least one detected case).
Sizes of (A) severe neurological and (B) fatal respiratory disease outbreaks by distance from the original surveillance hospitals, achieving a ≥90% detection probability if alternative healthcare providers were included in the surveillance system.

Representativeness of cases attending alternative healthcare providers
To assess whether including other healthcare provider classes in the surveillance system would improve the representativeness of the surveillance system, we estimated the difference between case statistics based on community cases and those based on cases attending each of the healthcare provider types (Fig G).

Fig G. Representativeness of cases attending other healthcare providers. Absolute difference between (A)
severe neurological and (B) fatal respiratory case statistics (proportions of case characteristics) estimates based on community cases and those estimates based on cases attending each of the healthcare provider types. A negative difference indicates that proportions among cases attending a surveillance hospital are lower than among all cases in the community. Significant differences (bootstrap p≤0.05) are indicated with an asterisk (*).