Passive case detection of malaria in Ratanakiri Province (Cambodia) to detect villages at higher risk for malaria

Ratanakiri Province, located in the remote northeast of Cambodia, is a mountainous and hilly area covered by tropical forest, rubber plantation or fallow land. The province borders the provinces of Mondulkiri to the south and Stung Treng to the west and the countries of Laos and Vietnam to the north and east, respectively. It comprises nine administrative districts, 49 communes and 240 main-villages [26]. Some main-villages have annex-villages or satellite-villages, which are informal settlements, often remote or geographically isolated and they are established mainly because of population movement and growth [27]. In 2013 there were 110 annex-villages in Ratanakiri (personal communication with the malaria supervisor of the provincial health district). The province houses a population of 183,699 inhabitants (2013) [28] belonging to ten different minority groups that are mostly subsistence farmers cultivating seasonal crops such as dry rice, cassava and beans on slash-and-burn forest fields [29]. A large number of families possess farms far from their villages and own two houses, one in the village and the other in the farm, often located in the forest where they usually stay during planting and harvesting season [29]. During the rainy season (June–October) some areas are not accessible because of poor road conditions and some places are only accessible by boat and or on foot. In 2013 this province had 11 health centres, 19 health posts and a provincial hospital [26]. These health facilities are under the management of Ratanakiri operational district (OD), which is overviewed by Ratanakiri provincial health department (PHD). Some 136 remote malaria-endemic villages have village malaria workers (VMWs) in place. To install a VMW in a village, the village must be located more than 5 km or a 1-h walk from the nearest public health facility (HF). Two villagers, preferably one male and one female, per village are selected as VMWs through community consensus [30]. The VMWs are trained to diagnose malaria using rapid diagnostic tests (RDT) and to treat uncomplicated cases [5]. Malaria endemicity of this province is among the highest in the country with a peak between August and November) and people who stayed overnight in plot-huts at farms were found to be more at risk of malaria infections [11]. LLINs are the main malaria prevention tool used in the province and a combination of dihydroartemisinin and piperaquine is the anti-malarial used for the treatment of all Plasmodium spp.

In 2010 a Malaria Information System (MIS) database was created to collect information from all areas at risk [31]. When required, additional support was provided by the MalaResT project [32], such as per diem and travel costs for OD staff to mentor health centres how to record data. MIS captured all individual malaria cases from both VMWs and HFs (in- and outpatients) obtained through PCD [33]. VMWs report the number of cases tested, both having a negative and positive result, while HF reports contain positive cases only. No information from the private sector was captured by the system. Monthly reports of VMWs and HFs are collected by operational districts. They record information on name and code of district, HF and village of origin, year, month, laboratory test results related to malaria parasite species, death, age group (up to 4, 5–14, 15–49 and 50 years and above), and gender. The data entry was done at operational district level using separate databases for VMWs and HFs and monthly updated data extracts were sent to the National Centre for Malaria Control (CNM) by email where they were automatically entered into the national database. The MIS also recorded number of reports received from each VMW village and HF per year. The person at operational district level was responsible for supervising the HFs and VMWs, checking the quality of data and following up on missing reports.

Only RDT, CareStart™ Malaria, pLDH/HRP2 COMBO (PAN/Pf), was used by VMWs to confirm malaria. This test was also used by health staff at HFs without microscopes or when a microscopist was not available, e.g., at night-time or during holidays. However, the proportion of each tool used at HFs was unknown since it was not recorded in the MIS. The national guidelines for interpretation of RDTs were as follows: tests with no band for control are considered invalid (to be repeated); presence of control band alone is negative for Plasmodium; presence of control and Pf bands is positive for Plasmodium falciparum; a control and PAN bands is Plasmodium vivax; and, control, Pf and PAN are classified as mixed infections. However, the last case (presence of control, Pf and PAN bands) can also be P. falciparum alone [34‐36].

The two datasets (VMW and HF) were combined into one to get monthly incidence cases by village/annex-village every year, taking into account the frequency of monthly reporting. Cases that were referred to a HF by a VMW and notified as such were removed from the VMW dataset to avoid double counting.

The average number of monthly reports (to MIS) is calculated for VMW and HF systems and by year. The total number of malaria cases per year and per village was further reported on a map and this regardless of frequency of reporting. Taking into account that not all localities had VMWs, and that HFs might deal with patients coming from localities not covered by them, the total number of positive cases per locality per year is provided if at least one report from a local VMW is available or/and at least one report from any HFs in the province reporting a case from that village. Data were considered to be missing for a locality and a specific year if no report from VMW was available and no report for that specific locality from any HFs in the province. According to annual number of positive cases the reporting villages were classified into eight categories (i.e., villages with number of cases of 0, 1, 2–5, 6–12, 21–50, 51–100, >100, and missing data).

Annual incidence rates per 1000 inhabitants at commune level for all and each specific Plasmodium spp were calculated. This calculation used projected annual population by commune as denominator. The population projection was done by multiplying population from Cambodia National Census 2008 by annual growth rate taken from population projections for Cambodia, 2008–2030 for Ratanakiri province (2009: 2.54%, 2010: 2.14%, 2011: 1.96%, 2012: 1.97%, 2013: 1.94%, 2014: 1.90%) [37]. To estimate a general trend of malaria incidence in the province, only villages reporting every year (2010–2014) were considered. To estimate the decrease of malaria incidence, the model used was a mixed-effects negative binomial regression using village as a random effect and year as categorical fixed effect. Each model was verified versus a standard negative binomial regression. The analysis was done in Stata/MP 14.2 (StataCorp. 2015 Stata: Release 14. Statistical Software, College Station, TX, USA). The annual trends of total incidence rates were presented in a graph by Plasmodium spp. Each year, the communes were classified into seven categories according to their levels of annual incidence rate (i.e. 0, >0 to <1, 1–50, 51–100, 101–150, >150, and missing data). Data analyses were performed using program R v.3.2.0 [38].

SaTScan v9.4 64-bit, software for the Spatial Scan Statistics, was used to identify clusters with increased risk of malaria in each year. For this spatial analysis, the program identified clusters by gradually checking multiple circular windows with a variable circle size of maximum 50% of the population at risk [39] based on the census in 2008. The analysis estimated the risk of malaria outside and inside each window under a null hypothesis of equal risk using a discrete Poisson model. Using the default maximum number of replication of 999 Monte Carlo simulations, the window having the maximum likelihood was assigned to be a cluster and a p value was provided. A cluster with p value lower than 0.05 was considered as significant. Spatial clustering analysis of malaria cases (all and by species) was done by year.

The GPS coordinates of the villages were provided by the Cambodian national MIS. During a survey [11], 117 out of 240 villages were checked for their GPS coordinates in the field and compared to the MIS coordinates. Most villages (92%) were in a radius of fewer than 500 m, 2.6% between 0.5 and <1 km, and 4.2% between 1 and 4 km. For only one village, the MIS coordinates did not match all the coordinates observed during the survey. The correct GPS positions were then applied for all 117 villages but remained unchanged for the other 123 villages. For the present study, an overall accuracy of 95% for the GPS positions within a radius of less than 1 km was estimated.

Some 157 out of 277 (57%) villages reported every year and were considered for estimating the trend malaria incidence between 2010 and 2014. For these 157 villages the overall annual incidence rates per 1000 inhabitants of all Plasmodium spp. decreased dramatically from 85.9 in 2010 to 30.4 in 2014 (Fig. 2). Using a mixed-effects negative binomial regression this decrease was significant from 2012 up to 2014 compared to 2010 (2014 vs 2010: 65%; 95% CI 61–69%). Falciparum cases decreased even more (2014 vs 2010: 79%; 95% CI 76–82%) whereas the incidence for P. vivax (2014 vs 2010: 19%; 95% CI 5–32%) and mixed infections fluctuated almost every year (Additional file 3).

The number of reporting villages increased from 174 in 2010 to 226 in 2014 (Additional file 4). The number of malaria cases was further reported by village and the incidence was calculated per commune (Fig. 3). Villages with more than 50 malaria cases reported annually remained concentrated at peripheral areas of the province especially along the Sesan River, in the northwest and in the southeast (Fig. 3). This trend looked similar for P. falciparum and P. vivax separately (Additional files 5 and 6). Since 37 and 28% of villages belonging to 35 and 26 communes in 2010 and 2011, respectively, had no report, the incidence of those communes was not calculated (Fig. 3).

Each year several clusters with a significantly higher risk of malaria cases (all species combined) were found, indicating that malaria is not evenly distributed amongst the province. There were ten to 16 different, significant, spatial clusters each year. The number of significant clusters decreased steadily from 2010 (16) to 2013 (ten) before a slight increase in 2014 (12). Every year the minimum number of villages per significant cluster was one, whereas the maximum varied from 62 in 2010 to 41 in 2014. Detailed information of the significant clusters can be found in Additional file 7. Big clusters tended to extend to the east (along the Cambodian–Vietnamese border) and the north of the province (along the Sesan River). Figure 4 shows that cluster 1 was stable throughout all years and within this cluster there were 21 villages appearing every year and 20 other villages (in the group of village 1B) appeared from 2012 until 2014. Eleven villages of 1B did not report in 2010 and seven villages of this group did not report in 2011. Cluster 2 was shrinking progressively from 2012 to 2014 (2013: clusters 2 and 3; 2014: cluster 4). Cluster 5 observed in 2012 also remained stable from 2012 to 2014, but in 2013 and 2014 it was split into two smaller clusters (2013: clusters 7 and 7′; 2014: 3 and 3′) of three villages.

The distribution of the P. falciparum and P. vivax clusters were somewhat different although large overlaps were observed (see Additional files 8, 9, 10, 11). In 2010, the large P. vivax cluster (cluster 1) has a more western distribution compared to the large P. falciparum cluster (cluster 2). In 2013, P. falciparum and P. vivax clusters overlapped each other. In 2014, a new large P. vivax clusters (cluster 4) appeared at the western border of the province.