Ben Tre province is located in the Mekong Delta region of southern Vietnam. It has a population of 1.3 million people and a largely agricultural economy. This coastal province is about 1.25 meters above sea level and is almost surrounded by water (Fig 1). Due to low piped-water coverage, river water is generally used as one of the main sources of water among the population. The last mass oral cholera vaccine program was implemented in 2002, and no cholera cases had been reported since 2005 [10, 16]. However, in 2010, an outbreak of cholera occurred in the province, starting in Mo Cay town on May 9. A total of 71 cases were identified in the outbreak (Fig 1).

We employed a purposive sampling technique to collect water samples from several types of water sources in Mo Cay and Giong Trom towns, where the first cholera cases were reported in this outbreak. They included 25 samples of river water obtained from the rivers closest to case-patients’ houses; six samples of indoor water, six wastewater samples, and two samples of drinking water in case-patients’ houses. In addition, 27 samples of fresh seafood were obtained from local markets where the case-patients lived.

From June 4 through July 2, 2010, we conducted a matched case-control study with a target sample size of 60 case-patients and 240 controls. Because the outbreak occurred in a not well-defined population and required rapid investigation, the case-control study would be the most appropriate choice of study design to identify risk factors for cholera infection [17]. In this study, the matching by age groups, sex, and living areas was used [18]. The sample size of this study for 80% power at 5% level of significance was calculated on the basis of the estimated percentage of controls exposed to unsafe water (30%), the ratio of controls to cases (4), and an odds ratio (OR, 2.25) as per the method of Kelsey and colleagues [19]. The estimated magnitude of OR that we used to calculate the sample size came from the previous work of Hoge and his colleagues [8]. The smallest reported OR was selected to ensure that the study was large enough and had sufficient statistical power in identifying risk factors with an expected OR of 2.25 or higher.

During this outbreak, all patients with acute watery diarrhea who sought diagnosis and treatment for their disease at local health facilities were immediately transferred to the nearest cholera treatment center. In all, there were four centers, which were established soon after the outbreak was confirmed and located in three distant local hospitals and one neighboring hospital (Fig 1). In these centers, case-patients were identified, quarantined, and treated, and they were considered eligible for inclusion if they had laboratory-confirmed V. cholerae identified through conditional culture of rectal swabs and resided permanently in Ben Tre province. For this study, case-patients identified since the beginning of the outbreak were consecutively recruited until the target sample size was reached. Case-patients who were under cholera treatment were prospectively recruited through the four cholera treatment centers, while case-patients who had already been discharged from these health facilities were contacted and recruited at their homes.

For each case, four community-based controls [20] were selected and matched by commune, sex and 5-year age group to control for the potential confounding effects of these three factors. They were recruited at their homes on the same day as the interview of their matched case. Specifically, the control search began with random selection of four sub-communes in the area where the case resided by using Microsoft Excel. In each of these sub-communes, currently registered houses were numbered and one house was then chosen by hand drawing from the Vietnamese bingo game, a set containing 90 balls numbered 1–90. Interviewers subsequently accessed the selected house, in which all household members were primarily screened by sex and age to identify a potential control. Controls were further screened by interviewers to ensure that they had not had acute diarrhea in the month prior to the interview. If no person in that house met the inclusion criteria, the interviewers went to the next house to the left until a matched control was found. No control persons that we approached refused to participate in the present study and be interviewed.

Trained health-workers used a structured questionnaire to collect the study information during face-to-face interviews. This questionnaire was adapted from a cholera case investigation form previously developed and used in Ho Chi Minh City in 2008. For the present study, it was extended and included a section for controls. Risk variables included were based on reports in the literature [8, 21] and primary assessments conducted at the beginning of the outbreak. The questionnaire contained questions to elicit information on socio-demographic characteristics, exposure to people with diarrheal diseases, recent travel history, and detailed consumption of food and water. The logic and language of questions were tested locally in Ben Tre before being used to collect data. For children aged <6 years, interviews were conducted with their parents or guardians. Potential contaminated food sources were identified through dichotomous questions about the different types of food eaten, eating places, and cooking methods (e.g., well-cooked, stir fried, cooked rare, and raw) within five days before the onset of diarrhea for case-patients or five days before being interviewed for control persons. Participants were also asked about their sources and characteristics of water used for drinking, cooking, and bathing, and their drinking habits (e.g., the use of boiled water and drinking water with ice) via several dichotomous questions in the previous seven days. Clinical symptoms and health-seeking behaviors since the onset of diarrheal symptoms were also collected. Rectal swabs were collected from controls after completion of the interview.

During the outbreak, rectal swabs of people with acute diarrhea were directly placed in Cary Blair Transport Medium and transported daily to the Pasteur Institute, Ho Chi Minh City (PI-HCMC) for identification of V. cholerae using a standard testing protocol [22, 23]. All controls’ rectal swabs and environmental samples were stored at the Ben Tre Provincial Preventive Centre and transported fortnightly to the PI-HCMC for testing. Specifically, swab specimens were enriched in alkaline peptone water (Oxoid) and inoculated for 6–8 hours. We then cultured the bacterial solution obtained on thiosulfate citrate bile salts sucrose agar (Merck) and carried out biochemical and agglutination tests (Denka Seiken). We further used polymerase chain reaction (PCR) to amplify the V. cholera isolations from the swab culture, environmental samples, to identify the genomic region encoding the O1/O139 genotypes, and to detect cholera toxins and other toxin-specific genes, specifically ctxA, ctxB, rtxC, and rstC, using previously published primers [24–28] according to the PCR method described by Nguyen et al.[29]. A 9 μl of amplicon was separated by electrophoresis on a 1.5% agarose gel and visualized with an ultraviolet light on the Gel Doc System (Bio-Rad Laboratories). Test results were immediately reported to the attending physicians and the local preventive centers for personal treatment and prevention purposes, respectively.

Statistical analysis was carried out using Stata version 14 (StataCrop LP, College Station, TX). Information about the participants in the case-control study were initially explored using descriptive statistics (including frequency and proportion for categorical variables and median and range for continuous variables), with comparisons between case-patients and controls made by McNemar’s chi-square tests. Children aged <6 years were classified as not having consumed a certain food or water if their parents or guardians reported that they did not consume it over the study period. To identify risk factors for cholera, conditional logistic regression with forward selection was used to estimate matched OR and 95% confidence intervals (CI) as per the method of Hosmer and his colleagues [30]. Initially, univariate analyses were employed to identify variables that were potential risk factors for cholera infection. Those variables with p<0.25, along with those which have been acknowledged to have biologic plausibility for increasing the risk of cholera, were considered for inclusion in the multivariable model. The analysis began by fitting the base model, which included cholera variable, participants’ age, and a variable that yielded the lowest corresponding p-value in univariate analyses. The variable with the next lowest p-value was progressively added to the model, and its contribution to model fit was tested by the log likelihood ratio test. If the test yielded a p-value less than 0.05, it was kept in the model. This process continued until the added variable made no significant improvement in model fit. As the level of education may behaviorally influence personal eating and drinking habits, we investigated potential effect modification by creating interaction terms between it and the other variables retained in the final model. In a sensitivity analysis, to accommodate missing data, we applied multiple imputation by chained equations, with an assumption of missingness at random. Ten imputed datasets were used with 1,000 iterations. We compared the risk factors found between the original and imputed data sets.

From May 9 (epidemiologic week 19) through August 3 (epidemiologic week 31), 2010, 71 cases of cholera were identified in all towns of Ben Tre province. The area with the highest number of cases was Mo Cay (46 cases), followed by Giong Trom (12 cases) and Ba Tri (7 cases) (Fig 1). The number of cases continued to increase, peaked during May 30 to June 5 (epidemiologic week 22), and then gradually decreased over time (Fig 2). No case-patient died during the outbreak.

After confirmation of the outbreak on May 12, 2010, Ben Tre rapidly rolled-out its control activities in the entire province. It included measures to isolate and treat case-patients at four cholera treatment centers, clean and disinfect their houses with chloramine B, and provide chloramine B to enhance the use of chloramine-treated water among people living in the areas where the case-patients emerged. Case-patient’s close contacts were traced, given a single dose of prophylactic ciprofloxacin, and monitored their health. Education campaigns were additionally made to encourage people in the entire province to practice safe water, proper sanitation, and food safety. As reports of primary risk assessments suggested a potential link between drinking iced water and the illness, on June 11, 2010, the government of Ben Tre prohibited the manufacture, transportation, sale, or supply of commercial ice in the two most affected towns of Mo Cay and Giong Trom.

In total, only V. cholerae O1 Ogawa, El Tor biotype was isolated from 71cholera cases. The ctxA, ctxB, rtxC, and rstC genes were exhibited in 97%, 100%, 100%, and 97% of all isolates, suggesting a presence of altered El Tor strains that combine characteristics of classical and El Tor strains. Due to resource constraints, only 193 rectal swabs taken from 240 controls were tested for cholera. V. cholerae was not detected in any controls’ rectal swabs.

Two out of twenty five river water samples and one out of six wastewater samples were culture positive for V. cholerae. PCR analysis revealed the presence of V. cholerae O1 Ogawa carrying ctxA in all of these three positive-culture samples. V. cholerae was absent in two samples of drinking water, six samples of indoor water, and twenty-seven samples of fresh seafood.

After excluding seven case-patients lost to follow-up and four case-patients whose disease occurred after the target sample size had been reached, a total of 60 case-patients occurring from May 9 through June 26 were consecutively included in a matched case-control study and matched to 240 controls. Forty case-patients were prospectively recruited through the cholera treatment centers and twenty case-patients were recruited at their homes. The case-patients were interviewed from June 4 through July 2, 2010. Nine case-patients and 32 control persons had missing data on at least one variable. Except for age, no significant discrepancies were recorded between the missing and non-missing subjects on variables (Table A in S1 File).

Demographic features and risk factors are listed in Table 1. Two thirds of subjects were females. Participants’ ages ranged from two months to 83 years for the case-patients and from one month to 87 years for the controls. For those aged six years or older, 73% of the case-patients and 57% of the controls reported having little education (primary education or illiterate).

The median number of case-patient’s stools within a day prior to hospital admission was eight (range: 2–20), 55% had watery stools, 40% had abdominal pain, and 53% had vomiting. The mean time of admission to local hospitals was 0.7 days after the onset of the disease, and the mean time lag between case onset and enrolment of controls was 12 days among the case-patients.

Fishpond or river toilets were frequent in both the case-patient’s and control’s houses (73% and 73%, respectively). A flush toilet in house was reported in 22% of the case-patients and 25% of the controls (p = 0.404). Thirty percent of the case-patients reported that their main sources of water for use were close to a toilet, whereas only 14% of the controls had a similar situation (p <0.001). A self-report of changes in the color, odor, appearance and taste of water used was seen in 15% of the case-patients and 8% of the controls (p = 0.029).

Among the case-patients surveyed, 22% reported drinking iced tea within one week before the disease onset, compared with 3% of the controls in the week before being interviewed (p <0.001). A higher percentage of the case-patients than the controls consumed water with ice (81% vs. 68%, p = 0.001), unboiled water (65% vs. 43%, p <0.001), and sedimented river water (12% vs. 5%, p = 0.014). Up to 52%, 33%, and 42% of the case-patients used sedimented river water as the source of water for bathing, brushing teeth/gargling, and cooking, respectively. These levels were nearly twice as high as those for the controls (30%, 16%, and 21%, respectively; all p-values <0.001). The case-patients were less likely to use stored rainwater for both drinking (70% vs. 89%, p <0.001) and cooking (23% vs. 41%, p <0.001), but they were more likely to drink bottled water (20% vs. 13%, p = 0.027) than the controls. The use of indoor tap water for both drinking and cooking was rarely seen in this sample.

The case-patients were less likely to eat cooked seafood (62% vs. 81%, p<0.001), steamed vegetables (12% vs. 31%, p <0.001) and raw vegetables (23% vs. 32%, p = 0.027) in comparison with the controls.

While only 2% of the controls were living with people who had acute diarrhea, 15% of the case-patients did so (p <0.001). The case-patients had a greater percentage of travel out of town in the week prior to the disease onset than the controls (37% vs. 20%, p <0.001).

Risk factors for cholera infection identified through the multivariable conditional logistic regression are summarized in Table 2. Individuals who reported drinking iced tea (adjusted OR (aOR) = 8.40, 95% CI: 1.84–39.25), not always boiling drinking water (aOR = 2.62, 95% CI: 1.03–6.67), living with people who had acute diarrhea (aOR = 13.72, 95% CI: 2.77–67.97), having the main household sources of water for use close to a toilet (aOR = 4.36, 95% CI: 1.37–13.88), and having little or no education (aOR = 4.89, 95% CI: 1.18–20.19) were significantly associated with an elevated risk of cholera. The risk of cholera was lower among persons who drank stored rainwater (aOR = 0.17, 95% CI: 0.04–0.63), ate cooked seafood (aOR = 0.27, 95% CI: 0.10–0.73), and ate steamed vegetables (aOR = 0.22, 95% CI: 0.07–0.70). No significant interactions were recorded throughout the analysis. Only the level of education become insignificant in another regression analysis based on a multiple imputation approach (Table B in S1 File).