Population

In late 2004 there were an estimated 526,000 internally displaced persons (IDPs) in eastern Burma, and at least 240 villages had been destroyed, forcibly displaced or abandoned in the prior two years [14]. Conservative estimates of ongoing displacement suggest that an additional 167,000 people and 300 villages were forced to move in the two years subsequent to the 2004 report [15].

Data in this study were collected from the so-called "black zones" in eastern Burma where health services are unavailable from either the military regime or international organizations. Services for a population of approximately 250,000 are provided primarily by ethnic health organizations of the Karen Department of Health and Welfare (KDHW) and the Backpack Health Worker Team (BPHWT), whose broad geographic target area extends from Mergui-Tavoy in the South to Karenni (Kayah) area in the North, and from the Thai-Burma border to slightly west of the Sittang River in eastern Pegu (Bago) Division. (Figure 1) For the purposes of service delivery and health information the two populations are mutually exclusive, in that BPHWT was designed to serve populations unable to access ethnic health clinics due to distance and/or security.

KDHW administers 33 clinics to provide primary health care to approximately 95,000 persons. These semi-permanent clinics are located in relatively stable areas of Karen State, but are designed for rapid relocation in the case of threats to population security. Eleven clinics have been forced to relocate since 1998, five from October 2006 to April 2007. BPHWT is comprised of over 300 health workers divided into 76 teams designed to reach an additional 152,000 persons in less stable areas. Since inception of the program in 1998, seven BPHWT health workers have died while carrying out their health care provision responsibilities.

This report summarizes and compares Pf prevalence estimates derived from two types of data sources: cross-sectional screenings conducted as part of the KDHW malaria program from 2003 to 2006, and retrospective cluster surveys designed to estimate infant mortality rates in the entire BPWHT and KDHW populations in 2004 and 2006. Both the malaria program and the cluster surveys identified Pf parasitemia with a rapid diagnostic device (RDT; Paracheck-Pf^® Orchid Biomedical Systems, Goa, India).

Integrated Malaria Control Program

In 2003 the KDHW initiated an integrated malaria control program in four villages with a total population of 1,819. By 2006 the program reached 28,498 persons in 49 villages (village population size range: 162 – 1,824). This population is a subset of the entire KDHW population of 95,000. The pilot program included distribution of long-lasting insecticide treated nets (LLITNs), malaria education messages, and early detection with the Paracheck-Pf^® device and therapy with mefloquine-artesunate for three days (MAS3). Baseline screenings were conducted prior to initiation of malaria control activities, permitting the estimation of malaria prevalence among new villages in each year.

The decision to actively screen a population living in an area of unstable transmission was based on the dramatic success of a similar strategy in Vietnam [16, 17] and later in Brazil [18, 19] and Cambodia [20]; and on growing, albeit inconsistent, evidence for asymptomatic infections in areas of unstable transmission [21–23] including Burma [24, 25].

Screening was universal in the first phase of the program (2003–2004). However, in order to reduce costs, limited screening was conducted in 10 of 14 new villages in 2005, and all new villages in 2006 (N = 27). Limited screenings included a systematic sample of 100 heads of household. Females were preferentially sampled during limited screenings in order to minimize the workload of health workers operating in a conflict zone, and to maximize the likelihood of identifying parasitemia in women of reproductive age. Villages with fewer than 100 households in 2006 (N = 11) screened only one person per household. All participants with a positive test result in either the malaria program or the cluster surveys (described below) received MAS3, as recommended by regional guidelines [26].

Parasitemia prevalence is reported as the proportion of the screened population with a positive Paracheck-Pf^® test result [(number Pf positive)/(total number screened)]. Estimates in 2005 and 2006 were weight-adjusted by village population size. Confidence intervals for prevalence estimates were calculated for finite populations to account for near-complete sampling by multiplying the standard error by the square root of (1 - p), where p is the proportion of the population that is sampled [CI = +/- 1.96 * SE$\sqrt{(1-p)}$].

Prevalence estimates from the eleven villages conducting universal screening were stratified by sex and age (<5, 5–14, and 15+ years), and crude odds ratios and their 95% confidence intervals were calculated. The rainy season was defined as the 5 months from June through October to account for parasite development in the mosquito following the onset of the rainy season between May and early October.

Cluster Survey Design

This report included results from four retrospective mortality cluster surveys conducted in two different years in the two mutually exclusive target populations of the BPHWT and KDHW. Between October and December in 2004 and 2006, BPHWT and KDHW health workers conducted retrospective household surveys of vital events and human rights violations occurring in the 12 months prior to the interview. The design, implementation, and operational method of the surveys have been described previously [13, 27]. Briefly, in 2004 and 2006 annual village census information was used to construct a sampling frame for the target population (~130,000) and spanning eight administrative areas (Figure 1). In 2004, one hundred village-based clusters (200 in 2006) were selected proportionate to population size and twenty (10 in 2006) households within each cluster were selected using systematic interval sampling. Design and implementation of surveys in KDHW areas differed only in the size of the sampling frame (~95,000).

At each household, surveyors explained the objectives and obtained verbal consent for participation. The survey included a listing of all household members by age and sex, and documented falciparum malaria parasitemia for the respondent using the Paracheck-Pf^® device.

Sample Size and Analysis of Cluster Surveys

The proposed sample size for each survey was based on a balance of operational feasibility and resource constraints and the goal of continued monitoring of the infant mortality rate. Population proportions were estimated for several morbidity outcomes, including the proportion of respondents testing positive for Plasmodium falciparum. All confidence intervals were adjusted for the cluster sampling. The sample size allows for the estimation of parasitemia prevalence to within 2%, assuming baseline prevalence = 10%, overall survey completion rate = 85%, and design effect = 2.0.

Ethical Approval

Data were collected as part of routine program monitoring and evaluation. Data forms were brought from the field to Mae Sot, Thailand where they were entered into a computerized database (Microsoft ACCESS) and were cleaned using range and internal consistency checks. The survey protocol and malaria program data collection instruments were approved by local leaders of the Burma Medical Association. The Johns Hopkins University Committee on Human Research approved the secondary analysis of the cluster survey data. The authors of this paper were responsible for the secondary analysis, conducted with Stata 8.2 (Stata Corp., College Station, TX, USA).

Malaria Program Screenings

Between 2003 and 2006 a total of 9,796 RDTs were performed among 28,410 villagers participating in 11 universal (n = 5,872) and 36 limited (n = 3,924) baseline screenings. Each baseline screening was completed in approximately 3 (median) days, (range 1–7). Overall participation in universal screenings was 98.1% (village range 87–100%) of the expected population. Overall 800 RDTs were positive for Pf, representing a weighted-mean prevalence of 9.5%, 95% CI: 8.7 – 10.2.

Overall prevalence estimates derived from baseline universal and limited screenings in each year from 2003 to 2006 are presented in Figure 2. Prevalence in malaria program areas was similar over four years (2003: 8.4%, 95% CI: 8.3 – 8.6; 2004: 7.1%, 95% CI: 6.9 – 7.3; 2005:10.5%, 95% CI: 9.3 – 11.8 and 2006: 9.3%, 95% CI: 8.2 – 10.6).

There was substantial inter-village variation of Pf prevalence among villages (range 0% – 28.6%). In 2005, the only year that included both universal and limited screenings, combined prevalence in ten villages conducting limited screening (12.5%, 95% CI: 10.6 – 14.4%) was higher than in four universally screened villages (6.4%, 95% CI: 6.3 – 6.5).

Age, Sex & Season

Cluster Survey Results

Age

The higher Pf prevalence we observed in children compared to adults is consistent with population surveys in ecologically similar areas of Laos, [32] Cambodia [20, 23] and Burma. For example, Tun-lin et al. documented higher prevalence in children under ten (30%–50%) than in adults (10–27%) during four successive screenings (n = 146 – 168) in a single village in central Burma in 1992 – 1993 [28]. However, an HRP-II antigen assay might overestimate Pf prevalence in children relative to adults, as acquired immunity in adults might lead to lower parasitemia levels and may decrease the sensitivity of the antigen assay.

Sex

We did not observe large differences in Pf prevalence between males and females in either children or adults. These results differ from the observation of a four-fold higher Pf prevalence among male (9%) vs. female (2%) adults in four Burmese villages immediately across the Thai border with access to early detection and treatment (EDT), [31] as well as from other studies southeast Asia that have documented increased exposure of male adults to infected mosquitoes due to forest related activities [23, 28, 33, 34]. The discrepant observations may reflect a difference in forest-related behaviors or an influence of the location and/or stability of villages; but may also reflect the lack of access to EDT or other malaria control interventions prior to our surveys. The higher Pf prevalence in males noted in other studies may reflect the relative impact of malaria control programs in adult men and women, and may not reflect the sex distribution of asymptomatic Pf among adults prior to program implementation.

Season

Overall prevalence was higher during the rainy season in both 2005 and 2006. This seasonal variability, however, appears to be less than that observed in Pf incidence among Burmese migrants [10] and refugees [35] in Thailand. These data are similar to those from bi-annual screenings conducted in four Burmese villages in 2006 (Pf prevalence 3.9% in both the rainy and dry seasons) in the setting of ongoing malaria control [31].

Malaria Program Screenings vs. Cluster Surveys

In 2004, cluster surveys produced higher estimates of Pf prevalence (12.4% and 11.8%) than program areas (7.1%). There are several possible reasons for this discrepancy. In 2004, malaria program screenings included nearly the entire population, whereas cluster surveys screened only heads of households, who may be more likely to engage in behaviours with elevated malaria risk, such as forest-related activities. While we did not directly measure malaria risk behaviours, in universally screened malaria program areas we observed that adults were at significantly lower risk than children. Alternatively, the higher prevalence reported in the cluster surveys in 2004 may reflect differences in village location, stability and/or exposure to human rights violations. Studies have documented an increased risk of malaria among migrants [34, 36] and in the setting of complex emergencies [7, 37]. Results reported elsewhere [13] from the 2004 survey in BPHWT areas suggest that malaria prevalence may be associated at the household level with forced displacement, forced labour, and destruction of food supply, and that exposure to multiple human rights violations increases risk.

Location and stability may also have contributed to the higher prevalence observed in 2005 among malaria program villages conducting limited screening among heads of household (population weighted prevalence 12.5%, 95% CI: 7.5 – 17.5) compared to universally screened villages (6.4%, 95% CI: 6.3 – 6.5), which tended to be located among more stable populations. The villages with the highest prevalence in both 2004 (Mae Ngaw, 17%) and 2006 (Ei Tu Hta, 29%) were the least stable villages in those years. Mae Ngaw subsequently was destroyed by the military in early 2005 and Ei Tu Hta was a newly formed encampment for persons internally displaced in early 2006 by an escalation of violence near the new Burmese capital of Pyinmana (Naypyidaw).

Limitations

Sequential estimates of Pf prevalence in new malaria program areas perforce relied on screening different villages in each term, which likely resulted in substantial bias by area and other unmeasured factors. An alternative approach to include longitudinal measurements in intervention-naïve villages would have minimized this bias, but was not possible in this setting, as implementing partners felt it would be unethical to withhold effective interventions from vulnerable populations. Furthermore, the increasingly large number of areas included in screenings, as well as triangulation with estimates from cluster sample surveys, enhances the external validity of our findings to other villages in "black zones" of eastern Burma.

We did not conduct universal screenings in all villages. However, the number of RDTs from universal screenings performed from 2003–2005 (n = 5,871) permitted evaluation of associations with age and sex; and limited screening resulted in significant cost savings to facilitate program expansion to additional villages. In universally screened villages, the Pf prevalence overall was higher (7.2%) than the estimate among adult women (4.4%) and this relationship was consistent for each year in which universal screening was conducted. This suggests that the population prevalences in cluster surveys and in program areas conducting limited screening, where adult women were over-sampled, likely underestimate the true population-based burden of parasitemia.

The use of a rapid diagnostic test may have limited our ability to detect low level parasitemia [38–40]. However, Paracheck-Pf^® has demonstrated impressive sensitivity and specificity under field conditions during asymptomatic screening of children in India (sensitivity/specificity 94.4 & 89.0%, respectively), [41] and in Tanzanian villages with either high (40.1%), low (4.3%), or very low (1.9%) P. Falciparum prevalence (sensitivity 83.6, 100%, n/a; specificity 94.1%, 99.5%, 98.4% respectively) [42, 43]. Furthermore, the alternative diagnostic strategy in areas where PCR is unavailable – field microscopy – has shown poor sensitivity (~10%) for asymptomatic P. Falciparum parasitemia [22] when compared to expert microscopy in western Thailand, [44] suggesting Paracheck-Pf^® may be at least as accurate as field microscopy in this setting. The accuracy of RDTs may be compromised by high-temperatures or prolonged storage under field conditions, [42] but storage in thatched huts likely minimized extreme temperatures in our case. Low RDT sensitivity would have resulted in an underestimation of parasitemia prevalence. It is unlikely that false positives (due to low specificity) accounted for a high proportion of prevalent cases during baseline screenings, given the consistently low prevalence (<2%) recorded during follow-up in most malaria program areas during program implementation [45].

Although most participants were asymptomatic at the time of testing, those with a positive RDT were treated immediately; therefore we are unable to distinguish between pre-clinical and chronic asymptomatic infection. Other studies from Burma, [24, 25] Cambodia, [23] Tanzania [43] and South America [18, 21, 46] suggest that protective immunity (premunition) is not uncommon in the setting of unstable transmission; and that asymptomatic infection is infectious to mosquitoes despite a low asexual parasite burden [46–48]. The relatively high prevalence of predominantly asymptomatic parasitemia in this report adds to the growing body of evidence supporting the presence of asymptomatic infection in areas of unstable malaria transmission. Additional studies are needed to estimate the prevalence of asymptomatic carriers in eastern Burma and to evaluate the role of active case detection in reducing malaria transmission.

We did not estimate the prevalence of plasmodium vivax (Pv), though Pv appears to account for no more than 20% of malaria infections in Burma [6] and almost certainly represents an even smaller fraction of malaria related deaths. We did not directly measure rainfall, [49] migration, [36] forest-related activity, proximity to water [50] or other risk factors for malaria [51] that may have confounded the associations we observed.

Rapid testing with accurate RDTs is easily integrated into malaria control programs and cluster surveys designed to estimate other health indicators, and provides a simple and cost-effective means to estimate cross-sectional prevalence of parasitemia. Triangulation of data from different sources enhances the validity of parameter estimates. Additional studies are necessary to quantify malaria risk in eastern Burma, including the role of age and sex, elevation, season, migration, forest-related activities and civil conflict. Increasing capacity for EDT offers an opportunity to directly monitor the more clinically relevant incidence of symptomatic Pf, and to improve our understanding of the relationship between Pf incidence and prevalence in this setting. In a region with highly drug resistant Pf [2, 52] and ubiquitous fake antimalarials [3, 53] efforts to track treatment failures and to monitor in vitro drug susceptibility and antimalarial quality should be expanded in unstable areas in eastern Burma.