Micro-epidemiology of malaria in an elimination setting in Central Vietnam

The field research team comprised the first author, the local principal investigator (NXX) and research assistants from the National Institute of Malariology, Parasitology and Entomology (NIMPE), Hanoi, previously trained in qualitative ethnographic research methods. The study team resided in Tra Cang for 2 months (April to June 2015) supported by the health staff at the Tra Cang CHC. The research team from NIMPE had extensive field experience in similar settings, and had previously visited the study setting on several occasions during the MAPARES study.

An open-ended research design centred on ethnographic methods guided the fieldwork. A purposive and adaptive sampling strategy was used to select informants, based on criteria such as history of malaria infection, mobility patterns, age, sex, and occupation. Emerging findings led to refinement of the sampling strategy and topic guides in order to maximize both variation and depth of information in the sample.

Data collection methods included audio-recorded in-depth interviews, informal conversations (not audio-recorded but recalled and summarized in field notes), informal group discussions, and participant observation. Participants were usually first visited at their hamlet houses, but the study team also accompanied participants to their farms and fields, especially those who slept overnight in their plot huts. Interviews were also conducted in the general store, schools and CHC located in the administrative centre of Village 5. All interviews were conducted in Vietnamese, with local Xe Dang guides assisting as translators for some interviews. Field diaries and memos were maintained to aid data interpretation.

Data were intermittently analysed in the field concurrent with data collection, leading to refinement of topic guides and further sampling. At the end of the fieldwork period, all data were imported into NVivo (NVivo for Mac v11.40, QSR International Pty Ltd) for data management. Descriptive open coding and analysis focused on agricultural and social activities occurring at times and locations that may be associated with exposure to local vectors’ bites, but also describing the general socio-cultural context, health-seeking behaviour and local understandings of malaria.

A simple random sample of 186 individuals aged 16 years and over was drawn from the population census of two hamlets (Xe Xua and Tu Nak), where malaria infections had been detected during the MAPARES study. The sample size calculation was based on detecting, with 80% power and at 5% significance, a minimum 20% difference in the odds of recent malaria exposure (defined by serology), for individuals staying overnight at forest farms and fields, compared to individuals who do not. As malaria sero-prevalence was unknown at the time of the cross-sectional survey, the maximum sample size (n = 186 at 50% exposed-group prevalence) was used.

The survey comprised a close-ended structured paper-based questionnaire. The questionnaire was piloted and revised, and the final version back-translated into English and checked for clarity. Trained surveyors from NIMPE administered the questionnaire at respondents’ households. Surveyors made up to three attempts to reach selected individuals. Data were double-entered and cleaned using Epi Info (Center for Disease Control and Prevention, Atlanta, USA).

The primary outcome was defined as recent exposure to malaria infection, estimated by serology and defined below. Serology was used because only four PCR-positive cases were detected during the six surveys, but also because serology outcomes may be more robust markers of exposure to malaria infection since it reflects cumulative changes in antibody titres over time, while PCR only detects malaria infections present at the time samples were taken [23‐28]. Individual mean percent-positive antibody levels (mean ‘PP’; mean value among duplicates of the sample optical density (OD) measured as a proportion of the strong positive control OD) against two P. falciparum antigens (AMA1, GLURP2) and two P. vivax antigens (AMA1, MSP1) were defined for all participants in the first and last (sixth) MAPARES surveys. Subsequently, these continuous variables were categorized into four groups indicating the extent of seropositivity (strong positive, weak positive, grey (transition) zone, negative) using the classification and regression tree (CART) method [29] and implemented using CART ^® software (Salford Systems, San Diego, USA). Cut-off points were defined using CART regression on each antibody distribution in the first survey, and applied to the sixth survey (Additional file 1). Regression trees were also computed for the changes in mean PPs from Survey 1 to Survey 6, to identify important minimum absolute increase and relative percent decrease that could be considered indicative of recent exposure and non-exposure, respectively. The following algorithm was used to define ‘recent exposure to malaria infection’:

The research focused on identifying activities that may increase exposure to vector bites as risk factors for recent exposure to malaria infection. Based on the qualitative strand, the questionnaire included items about time spent in locations of presumed differential exposure to biting vectors (e.g. deep forest, forest farms and fields, river, hamlet), as well as evening and night time activities in these different locations that delay usual sleeping time and/or disrupt bed net use (e.g. drinking, outdoor TV watching, hunting, fishing). Several household-level variables were also included, such as household size [30], number of children, household ITN/LLIN coverage, housing construction, and presence of another recently exposed individual. The assumed causal relationships between these variables is depicted as a directed acyclic graph [31] in Fig. 1, and formed the basis of the modelling strategy.

Logistic regression models with robust standard errors to account for clustering by household were used to calculate crude and adjusted odds ratios for the association between exposure variables and recent malaria exposure, separately for P. falciparum and P. vivax. In line with the causal modelling approach, the emphasis was on appropriate adjustment for confounding rather than statistical significance alone, with the number of exposure variables balanced against data scarcity, which was assessed by inflation of standard errors upon inclusion of additional variables. All exposure variables with p values less than 0.2 in the unadjusted analyses were included in the multi-variable models. Given the small sample size relative to the number of exposure variables, individual-level variables with p value exceeding 0.2 were removed from the multi-variable models, before household-level variables were included. Household-level variables were then retained if their inclusion meaningfully shifted the effect estimates of exposure variables without substantially increasing standard errors. Age and sex were forced (confounding) variables in all models. Sub-group analyses were conducted for respondents who slept overnight at their forest farms and fields, as their overall exposure profile was considered to differ from respondents who always slept in their hamlet. Sensitivity analyses restricted to individuals with serological data in both Survey 1 and Survey 6 were run. All regression analyses were conducted in Stata/IC 13 (StataCorp LLC, College Station, Texas, USA).

Access to a bed net in hamlet houses was reported to be very high (97.5%), and 75% reported always sleeping under a bed net (Table 1), however only 30% of households had a household bed net ratio of 1–2 people per bed net. Only 10 (6.25%) surveyed individuals reported sleeping in another house in the same hamlet, 8 of whom reported using bed nets when sleeping in a different house. Overall, neither individual bed net use when sleeping in the hamlet or bed net condition was associated with recent P. falciparum or P. vivax infection (Tables 2, 3).

Consumption of rice wine in the evenings in the hamlets was very common amongst men (84%) and common amongst women (44%), and 16% of men and 8% of women (12% overall) reported to drink every evening (Table 1). Rice wine was usually consumed outdoors in groups and had multiple functions, such as to aid sleep and soothe aches and pains after a day spent working in the fields, to facilitate social interactions (“first we invite each other for drinking, then we start talking”), transactionally (“paying for favours in wine”), and for a range of traditional ceremonies. Rice wine consumption delayed sleeping time; the median sleeping time was 8 pm if not drinking (IQR 7–9 pm), 9 pm after drinking (IQR 9–10 pm). Evening drinking hindered bed net use as, when asked generally, 88.5% of survey respondents reported to always sleep under a bed net, but only 65.4% reported to sleep under a bed net after drinking in the evening. There was a non-significant trend towards increased risk of P. falciparum exposure amongst daily drinkers in multivariable analyses (OR 2.15, 95% CI 0.71–6.51) (Table 4). However, when restricting analysis to individuals who never slept at the farms, daily drinking was their main risk factor for recent P. falciparum exposure (OR 9.52, 95% CI 1.93–46.87, p = 0.006) (Table 4).

Watching television/DVDs (hereafter ‘TV’) in the evenings was a common social activity. Most often, people watched TV together in a central house (26.5%), particularly in Tu Nak, many or all of whom sit outside as there is not enough space in the house. Only 17.5% watched TV at home (thus indoors), reflecting low TV ownership (Table 1). Watching TV in the evenings at another’s house, a proxy for outdoor TV watching, was associated with increased risk of recent P. vivax exposure (OR 6.29, 95% CI 1.49–26.58) (Table 5).

There were 72 (45%) survey respondents who slept overnight at their forest farms and fields. The longest duration of stay was mostly less than 1 week and was not associated with recent malaria, but 10% of overnight sleepers stayed for 1 week to more than 1 month continuously, which was associated with recent exposure to P. falciparum malaria amongst farm sleepers (adjusted OR 6.99, 95% CI 1.22–40.01, p = 0.029) (Table 4).

Over 90% of farm sleepers reported to sleep under a ITN or LLIN, and qualitative research suggested that bed net use at farms was consistently high owing to perceived high insect nuisance. Evening sleeping times at farms were earlier than in the hamlets (median sleep time 7 pm, IQR 6–8 pm). Not sleeping under a bed net at the farm was associated with recent P. vivax exposure (adjusted OR 17.57, 95% CI 1.91–160.80) (Table 5).

Evening drinking also occurs at the farms and fields (18.1% of men and 16.9% of women sometimes drink), but intermittently rather than regularly, and was not associated with P. falciparum or P. vivax exposure (Tables 2, 3).

46% of respondents reported spending evening hours in the forest, for collecting wild plant foods (24%), hunting (32.5%), collecting firewood (11.3%) and logging (3.8%) (Table 1). Different forest activities had different exposure patterns. Respondents reported to go hunting at night, but sleep at their plot hut or village house, whereas logging required multi-night stays in the forest. Logging was strongly associated with recent exposure, particularly for P. falciparum (OR 10.9, 95% CI 2.01–58.51, p = 0.006) but also for P. vivax (OR 5.45, 95% CI 1.26–22.76, p = 0.02). Insecticide-treated hammock nets were rarely taken to the forest (only two individuals in the survey).

Evening fishing was reported by 20 (12.50%) respondents, mostly men (75%) and mostly in Xe Xua hamlet where the river is located. Men went fishing in small groups of 4–6 people and returned late in the evening (median 10 pm, IQR 9 pm–12 am). Evening fishing was weakly associated with recent P. falciparum exposure amongst farm sleepers (OR 7.99, 95% CI 1.10–58.04, p = 0.04) (Table 4).

Increasing age was associated with increased risk of recent exposure to P. falciparum, less so for P. vivax. Females were at higher risk of recent malaria exposure in univariable analyses, an effect that strengthened after adjusting for evening and outdoor activities, which were more prevalent in males.

Higher numbers of children living in a household was associated with decreased risk of P. falciparum and P. vivax exposure. Household bed net ratio or other seropositive members were not associated with P. falciparum or P. vivax exposure in adjusted analyses. Having a hamlet house built on stilts rather than ground level was not associated with P. falciparum or P. vivax exposure overall, but was protective against P. falciparum exposure amongst non-farm sleepers (OR 0.12, 95% CI 0.03–0.60, p = 0.01) (Table 4).