Study design and setting
We conducted a hospital-based case-case study using the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b) Diarrhoeal Diseases Surveillance System (DDSS) in Matlab and Dhaka hospitals.
We selected a case-case study design for the reasons outlined in the introduction above.
The DDSS employs icddr,b hospital staff to systematically record clinical, socioeconomic, and demographic data from diarrheal patients presenting to icddr,b hospitals prior to the patients receiving their diagnoses. All DDSS patients have their stool tested for enteric pathogens, including V. cholerae, Shigella, Salmonella, rotavirus, amoeba, and Giardia species. Regardless of etiology, we considered any case of diarrhea that required hospital treatment to be severe, and focused on those cases because they are of greatest clinical importance.
The Matlab hospital DDSS is part of a larger Health and Demographic Surveillance System (HDSS) created in Matlab sub-district, a rural area in east-central Bangladesh, in 1966. The HDSS employs Community Health Research Workers to record demographic, mortality, migration, and other relevant data through bimonthly visits to each household. The Matlab HDSS catchment area covers more than 200,000 residents, with all HDSS diarrheal patients treated at the icddr,b enrolled into the DDSS. Due to river and road access, and icddr,b’s well-established relationship the with the community, use of its facilities is assumed to be homogenous throughout the study area
Dhaka is the largest city in Bangladesh, with large numbers of residents living in substandard housing (slums). Since 1996, two percent of patients at the Dhaka Hospital have been systematically enrolled in the DDSS
. Due to the hospital’s location within the city limits, its services are considered to be accessible to all city residents. The administrative and clinical staff in Matlab and Dhaka received equivalent training to ensure comparability of care and DDSS data quality. In both settings, icddr,b hospitals provide free, high-quality diarrhea treatment.
In this case-case analysis, rural dwellers were defined as Matlab patients who were registered with the Matlab HDSS and self-reported currently living in a village. Urban dwellers were defined as Dhaka patients who self-reported currently living in slums or high-density mixed-use and residential areas.
We analyzed in-patient and out-patient data from patients who entered icddr,b hospitals for diarrhea treatment between January 1, 2000 and December 31, 2008. Since risk factors and necessary disease control measures among children under five may be different from older individuals, their risk factors were analyzed separately
. We also excluded those with missing age data, non-rural Matlab patients, non-urban Dhaka patients, those with neither cholera nor shigellosis, and those with enteric co-infections. Use of anonymized data prevented us from assessing if there were multiple admissions of the same patient.
Cholera and Shigellosis definitions
V. cholerae positivity was defined by the detection of V. cholerae O139 (Bengal), V. cholerae O1 El Tor Ogawa, V. cholerae O1 El Tor Inaba, V. cholerae O1 Classical Ogawa, or V. cholerae O1 Classical Inaba. Shigella spp. infection was defined by the detection of S. dysenteriae, S. flexneri, S. boydii, or S. sonnei. There were no changes in laboratory testing methods for V. cholerae or Shigella spp. during the study period.
The prevalence of potential correlates of diarrhea among hospitalized patients with cholera and shigellosis at icddr,b hospitals were compared. Self-reported sociodemographic characteristics included age, sex, number of household members, education, household income, urban residence, residence in a slum community, homeownership, and presence of concrete floors in the home. Education was defined as the patient’s education (for those ≥15 years old) or the mother’s education (for those <15 years old). Self-reported water and sanitation characteristics included the patient’s household having improved toilet facilities
, distance from the kitchen to drinking water (reported in feet and converted to meters for analysis), source of water, and drinking water treatment. Source of water was constructed by combining drinking and bathing water variables; if these were different, the least safe source was used for the analysis. Surface water was defined as that from a pond, river, or ditch. “Other” water treatment included use of tablets, filters, and sieves. Data regarding the source of water used for food preparation was unavailable for this analysis. Other potential correlates included the distance to the hospital (self-reported in miles and converted to km for analysis), the presence of a family member with diarrhea in the past week, and the season.
Clinical characteristics included general physical condition and clinical dehydration on admission as assessed by medical staff, self-reported days with diarrhea prior to admission, and the number of stools and history of vomiting in the 24 hours prior to admission. Data regarding patient deaths, if any, were not available.
Risk factor analysis
Assessed sociodemographic risk factors included age, sex, the number of household members, years of education, monthly household income (converted from Taka using the rate of exchange at the study period’s midpoint
), residence in a slum community, homeownership, and the presence of concrete floors in the home. Risk factors related to sanitation and water included improved toilet facilities, distance from the kitchen to the drinking water source (10-m increments), water source, and drinking water treatment. The distance from the home to the hospital (km) and the presence of a family member with diarrhea in the past seven days were also assessed.
We used Poisson regression with robust variance estimates to calculate risk ratios (RR) and 95% confidence intervals (95% CI) for cholera hospitalization risk factors
. The dependent variable in the regression model was cholera hospitalization (vs. shigellosis hospitalization) and the independent variables are listed under “Risk factor analysis” above. Due to substantial differences between Dhaka and Matlab, all regression analyses were stratified by urban or rural status. Only potential risk factors with less than 5% missing data were evaluated. Stata/IC 13.1 (StataCorp LP, College Station, TX) was used for all analyses. All P-values are two-sided.
(Statistically significant univariable predictors (p < 0.10) were considered candidates for the multivariable model. Predictors with a RR between 0.9 and 1.1 were excluded from consideration for the multivariable model due to small effect sizes. Strata with less than ten observations were also excluded from consideration for the multivariable model. We used variance inflation factors (VIF) to assess collinearity among the multivariable candidates. In the event of collinearity (VIF ≥10), we considered only the more biologically plausible predictor.
We built a multivariable regression model by sequentially adding and testing statistically significant candidates from the univariable analysis, in order of effect size. Continuous variables were retained in the model if the Wald test was significant (p < 0.05). A categorical variable was retained if the composite linear Wald test of all the variable’s strata and the Wald test for at least one individual stratum indicator variable was significant (p < 0.05).
 and age were included in the multivariable model as a priori adjustment variables. The seasonality adjustment variable was comprised of a restricted cubic spline of the day of the year on the date of visit (1–366). The spline had seven knots and was created to prevent the imposition of artificial categories or parameters on the data
The Research Review Committee (RRC) and Ethical Review Committee (ERC) of the icddr,b approved the hospital surveillance activities. Due to the high proportion of illiterate patients, the icddr,b RRC and ERC waived the need for written informed consent and approved the use of oral informed consent for all participants. Parents, guardians, next of kin, or caretakers provided oral informed consent for minors. icddr,b staff documented consent in the surveillance database. All data analyses were performed using anonymized patient medical records. The University of Washington Human Subjects Division/Institutional Review Board determined this research to be exempt from human subjects review because it did not fall under the definition of human subjects research under 45CFR46.