Assessing malaria risk at night-time venues in a low-transmission setting: a time-location sampling study in Zambezi, Namibia

This study was conducted in six randomly selected health facility catchment areas (HFCAs), out of 13 total HFCAs in the western Zambezi Region in northeastern Namibia. The Zambezi Region is arid and sparsely populated and borders Zambia and Angola to the north and Botswana to the south. Malaria transmission is seasonal, peaks between December and April and is highly dependent on rainfall patterns. This study was conducted in the western Zambezi Region, where prevalence is highest and access to services lowest. The population was 35,381 based on 2011 census data [10] and the annual parasite index was 26 per 1000 persons in 2014, according to health information system data and a census conducted by the project partners.

The study was conducted simultaneously with a representative household survey in the same catchment areas [11], a facility-based case–control study (Smith et al. [12], pers.commun.), and a peer-referral study to examine clustering of malaria risk within social networks.

A formative assessment was conducted in March, 2015 to identify types of venues that were active between dusk and dawn, to identify their peak attendance times, and to engage with the community and venue owners regarding the potential acceptability and feasibility of the venue-based study design.

In each HFCA, one focus group discussion (FGD) was held with community leaders (“headmen”). Additionally, one FGD was held with community health volunteers in the combined study area. All FGDs were conducted in the Siliozi language and followed a semi-structured guide. Participants were asked to discuss the types of venues that would be likely to meet a set of pre-defined venue eligibility criteria, to identify potential challenges to recruitment, and to develop a map of such venues and a list of peak attendance times. The venue eligibility criteria included that the venues be open at least once weekly between sundown and sunrise, have clientele mostly older than age 14 years, have a peak attendance of at least 10 people, have clientele deemed likely to be willing to participate, and that carrying out the study would not be expected to pose a risk to survey staff. Based on findings from the FGDs, a limited number of types of venues that appeared to have the largest number of venue-goers, greatest duration of time outdoors during mosquito-biting hours, and accessibility, were selected for inclusion in the survey. Formative study staff then visited the venues identified to confirm peak attendance times, request permission to conduct the study, and identify locations within or outside of each venue to conduct malaria testing and confidential interviews.

TLS was used to select venues and times to recruit survey participants. A sampling frame of VDTs was first constructed. VDTs were defined as 2- to 4-h intervals of peak attendance at a specific venue (e.g., Bar #1 from 7 to 9 pm on Fridays). At the start of each month, VDTs were randomly selected from the sampling frame and scheduled onto a calendar of recruitment events. As possible, four events (two per each of two field teams) were scheduled each Wednesday, Friday and Saturday (days when peak venue attendance was identified) during the approximately 10-week data collection period. VDTs were selected by simple random sampling subject to logistical constraints: to limit travel time, a field team could travel to only one HFCA per evening; to limit burden on venue-goers, a venue could be visited no more than once weekly. An alternate VDT beginning on or after the start of each of these primary VDTs was also randomly selected, in case the primary venue was found to be closed, inaccessible, or had low attendance. Sampling quota were a minimum of 11 and a maximum of 25 participants per event, to meet a planned total sample size of 800 participants. The sample size was chosen to provide 80% power to detect a difference in polymerase chain reaction (PCR) parasite prevalence of 4 percentage points between households (anticipated to be 2%) and venues (anticipated to be 6%), assuming a design effect of 2 due to clustering by sampling event. Each month, the sampling frame was updated to reflect new venues or venue closures.

Field teams included an enumerator, a nurse and two interviewers. Enumerators designated an intercept line to count venue-goers present at any time during the event. When not conducting interviews, interviewers approached all venue-goers who crossed the line and verbally assessed eligibility.

Males and females ages 15 years and older who lived or worked in the Zambezi Region, had not participated previously and did not appear too intoxicated to provide informed consent were assessed for eligibility using a screening form and, if deemed eligible, invited to a confidential area outside and nearby the venue entrance. Those willing to provide written informed consent then completed a 15- to 30-minute interviewer-administered questionnaire via electronic tablet, which included demographics, outdoor activities between dusk and dawn, protective measures against mosquitos at the venue and place of sleeping, fever and malaria history, treatment-seeking for fever, and venue attendance patterns. The nurse then collected a finger-prick blood sample, conducted a malaria RDT and prepared a dried blood spot for molecular testing.

Participants with positive RDT results were provided free treatment on-site. Blood samples were frozen before being sent to a laboratory in Windhoek for loop-mediated isothermal amplification (LAMP), and follow-up PCR following all positive LAMP results, and on 10% of negatives. No monetary or other incentives were provided.

Participant ID codes linked questionnaires, specimens and test results. Names were not recorded. The study protocol was approved by The University of California, San Francisco Committee on Human Research, Institutional Review Board, the University of Namibia Research Ethics Committee, and the Research Management Committee of the Ministry of Health and Social Services of Namibia.

Estimates of characteristics of the population reached by TLS were compared to results from a cross-sectional household survey covering the same geographic region and conducted between April and June, 2015. In this survey, 684 households were randomly sampled, 529 households enrolled, and 1918 household members interviewed and tested for malaria [11]. Sampling in the household survey was with probability proportional to number of households across HFCAs.

Respondents were asked about protective measures against malaria in their home including IRS and LLIN ownership, as well as whether they had been ill with fever anytime in the past 3 months, the last 2 weeks, and whether they sought treatment.

For categorical measures, proportions and 95% confidence intervals (CIs) were estimated; medians and interquartile ranges were estimated for continuous measures. Differences by venue type were assessed using Wald tests. Differences between the TLS and household surveys were assessed using two-sample Z-tests.

Risk factors for malaria infection by RDT, PCR or LAMP could not be identified due to the small number of positives detected. Associations were identified between self-reported fever in the past 3 months using bivariate and multivariable logistic regression, examining variables related to demographics, potential exposure at the venue and at the place of sleeping. Response categories were combined in some cases where bivariate associations were similar. The multivariable model included all variables with bivariate associations at the 10% level. Analysis was conducted in Stata version 12.0 using procedures for complex surveys.

Estimates of prevalence and tests for associations from the TLS sample were weighted and standard errors were linearized to account for clustering by recruitment event [13‐15]. Sampling weights were calculated as \(\frac{1}{{F}_{e}{A}_{i}}\), where F_e was the sampling fraction from respondent (i)’s sampling event and A_i was the frequency of attendance, defined as the number of days the respondent went to the same type of venue (i.e., bars or churches) in the past 30 days between sundown and sunrise [16]. Frequency of attendance was imputed at the sample mean, by catchment area and venue type for 14 (2.9%) participants due to non-response. A 5% trim was applied to the weights due to outliers ( > 3 standard deviations from the mean). Differences at the 5% level were considered statistically significant.

Several types of locations where individuals gathered between sundown and sunrise were identified during FGDs, including bars, churches, construction work sites, and cattle, fishing and police camps. The TLS study was limited to bars and churches as they were considered the most well-attended and accessible locations, often have windows or doors that are left open, and in some cases incomplete walls or open eaves. Twenty-one bars and four churches with evening hours were identified within the 6 catchment areas. VDTs, defined based on reported peak times, were generally between 3 and 8 pm at churches and 6 pm and 12am at bars. Selected HFCAs, venues, and households in the study area are shown in Fig. 1.

From March 27 to May 30, 2015, 480 study participants were recruited from 67 events held at 43 randomly selected VDTs, including 390 (81%) participants at bars and 90 (19%) at churches. Between 1 to 5 (median 3) sampling events were held per venue. Each venue contributed from 2 (0.4%) to 45 (9.4%) participants. Six additional events during the final month of data collection produced no new participants despite venue-goers in attendance. Of venue-goers approached, 80% at churches and 67% at bars were eligible and agreed to participate; this declined from 76% overall during the first 30 days to 51% during the latter half of the study period. The main reasons cited by field teams for non-participation were previous participation and lack of incentives.

Based on weighted analysis, an estimated 62.8% of individuals frequenting bars and evening church services were male and 37.2% were female (Table 1). The median age was 25 years and 26.1% were 35 years or older. Most had completed primary education (79.0%). Most venue-goers were Namibian citizens (97.9%), while 1.4% were Zambian. About one quarter (27.5%) worked in agriculture or fishing and 1 in 10 (11.0%) in construction or as security guards; 37% earned income from any of these activities. Bar patrons were more likely to be younger (median 25 vs. 36 years; P = 0.001), male (66.1% vs. 36.1%; P = 0.0281), and less likely to work in agriculture or fishing (17.8% vs. 52.9%; P = 0.014) than church-goers.

Nearly a third (28.8%) of participants frequented bars or churches between sundown and sunrise at least weekly. Most (79.3%) spent more than 1 h during a typical visit to the venue where they were surveyed and 44.5% spent 3 h or more. About half (53.8%) had been bothered by mosquitoes at the venue previously; 2.5% were using mosquito repellent at the time of the interview. These estimates of potential exposure to mosquito bites were similar at bars and churches.

About half (50.8%) of venue-goers slept in structures the previous night where IRS had been conducted in the past year, 43.7% owned a mosquito net and 26.9% had slept under a net the previous night. Fifty participants (10.5%) had slept outdoors in the past 2 weeks and 82 (17.3%) had spent the night outside of their home village in the past 4 weeks. Of the 82 participants reporting overnight travel, 78 travelled within Namibia, 2 to Zambia and 2 did not report their destination. These estimates were also similar at bars and churches.

The estimated prevalence of malaria infection was 4.7% by LAMP and 0.5% by RDT, with no differences by type of venue. Of the 17 LAMP-positive participants, 13 were recruited from bars (11 males and 2 females) and 4 from churches (3 males and 1 female); LAMP-positive participants’ ages ranged from 17 to 42 years (median 24 years) and were similar by sex and venue type. Only 2 of the 17 LAMP-positive participants tested positive by RDT. Two of the three RDT-positive participants tested positive by LAMP.

An estimated 52.1% of venue-goers had self-reported fever in the 2 weeks prior to the survey and 64.2% in the prior 3 months. Of the latter, 37.9% (105) sought treatment or advice for the most recent fever episode. None of those who reported fever in the past 3 months reported receiving chloroquine or quinine: five received Coartem®; one received SP/Fansidar®; 61 received Panadol®; and 26 received other unspecified treatments. Of the six participants with fever in the past 3 months who took Coartem or SP/Fansidar anti-malarials at last fever, none tested positive for malaria by RDT, PCR or LAMP. Of the 88 who received other or no treatments, five (5.7%) tested positive for malaria: two by RDT, one by PCR and five by LAMP; all five reported fever in the past 2 weeks.

Self-reported fever in the past 3 months was associated with engaging in livelihood activities that included agriculture, fishing, construction and security work (OR = 1.7; P = 0.030), ever being bothered by mosquitos at the venue (OR = 2.8; P < 0.001), not wearing personal protective measures when interviewed (OR = 6.5; P = 0.043), and Zambian citizenship (OR = 12.4; P = 0.009) (Table 2). Malaria infection by LAMP was not significantly associated with self-reported fever in the past 2 weeks or 3 months (P ≥ 0.4).

Relative to the household population aged 15 years and older, venue-goers were more likely to be male (62.8% vs. 41.5%; P < 0.001), aged 15 to 24 years (46.8% vs. 34.0%; P = 0.013), and less likely to be age 45 years and older (10.9% vs. 29.0%; P < 0.001) (Table 3). Venue-goers were more likely to have spent the night away from their home village in the past month (17.3% vs. 8.9%; P = 0.008), to have experienced fever in the past 2 weeks (52.1% vs. 33.2%; P < 0.001) and 3 months (65.2% vs. 36.9%; P < 0.001), less likely to have sought treatment at last fever (37.9% vs. 52.1%; P = 0.011), and moderately less likely to reside in a structure where IRS had been conducted in the past year (50.8% vs. 60.0%; P = 0.063). Prevalence of malaria infection by LAMP between the venue-going population and the general population was not significantly different (4.7% vs. 2.8%; P = 0.740). Bar and church sub-samples both differed from the general population with respect to prevalence of fever and spending the night away from home; bar-goers differed from the general population with respect to age (P < 0.001), sex (P < 0.001), IRS in past year (P = 0.031), and treatment-seeking for fever (P = 0.011) (Fig. 2).