In total, 2100 unselected patients were recruited from June 2015 to September 2016. Enrolment in Rangamati district commenced from April 2016 onwards. Ten patients were excluded as they did not have confirmed malaria leaving 2090 enrolled patients with malaria. Enrolment broadly followed the pattern of seasonal peaks in incidence reported to NMEP with a slower uptake at the beginning of the study period (Additional file 1: Figure S1).
Males comprised 67% of the study population compared to 51% in the 2011 census for the 5 study districts in Chittagong Division (Additional file 1: Figure S3). The proportion who were male aged between 15 and 39 years was higher at 72% compared to 49% in the census (P < 0.001). When correcting for multiple comparisons across 5-year age bands, there was a higher proportion of males in age bands from 15 to 39 years in the study population than in the census (Bonferroni method, 17 age bands, significance p < 0.003). Children under 15 years comprised 34% of the study population, of which 55% were male. There were no differences in the proportion of adult females or children of either gender in the study population compared to the census. There were also no substantial differences in age or gender distribution in the study population compared to the NMEP surveillance data from the whole of Chittagong Division during the study period.
The five most reported occupations were farmer (21%), student (19%), forestry (16%), child (16%) and housewife (11%). Full details of occupations reported are shown in Additional file 1: Table S4. The majority of farmers (72%) were engaged in paddy farming. Of the people who worked in the forest, 21% worked in jhum cultivation (slash and burn farming), especially in the hilly areas, and another 21% worked in plantations close to the forest. The top three reported occupations for males were farming (28%), forest-related (21%), and student (16%) and for females, housewife (32%), student (24%) and child (22%). Of the 16 cases who reported their residence as outside the study catchment area, 14 were in the military.
Routine diagnostic testing found Plasmodium falciparum, P. vivax and mixed infection in 74% (N = 1543), 16% (N = 332) and 10% (N = 215) of enrolled individuals compared to 64%, 11% and 25% in the NMEP surveillance data for Chittagong Division during the same period. Only the proportion with mixed species infection was higher in the surveillance data (p < 0.001). Details of district, age, gender, occupation and forest status distribution by species are shown in Additional file 1: Tables S2-S5 and Figure S3. The proportion of children under 15 years of age was higher for P. vivax (45%) compared to P. falciparum (34%) (P < 0.001). No other significant differences were noted.
Spatial distribution of cases
Bandarban district in the south of Chittagong Division contributed 1033 (49%) of the enrolled cases (Fig. 2a) followed by Khagrachhari district in the north with 416 (20%) cases. The geographic distribution of enrolled cases broadly followed the spatial distribution of total cases reported to the NMEP (Fig. 2b). There was under-recruitment in some of the remote forested border areas in Bandarban (Additional file 1: Figure S2). Lama in Bandarban was the subdistrict with the highest recruitment of cases with 574 (27%) followed by Ramu with 254 (12%). There were no cases resident in the city of Chittagong, with Chittagong district having cases mainly in the forest fringe areas, such as Fatikchhari (11 cases), Lohagara (8 cases) and Banshkhali (6 cases) subdistricts.
Of the 2090 enrolled patients, 1631 (78%) reported travel within the previous 2 months. Of the patients who travelled, 729 (45%) reported travelling only for work, 178 (11%) reported travel for purposes other than work during the day and 724 (44%) reported overnight stays.
Travel from residence to the study site
The overall geographic pattern of travel from residence at the union level to the study site is shown in Fig. 3. It can be seen in 3a and 3b that these patterns are a combination of health facility catchment areas for people living locally plus some people travelling long distances across the country from their residence before presenting to a health facility at their destination. Panels c–g show the top 5 study sites which recruited the most cases (39%) in descending order of enrolled cases: (1) Ramu Upazila Health Complex, (2) Lama Ekata Laboratory Office, (3) BRAC Dighinala Laboratory, (4) Alikadam Upazila Health Complex, (5) Chittagong Medical College Hospital (CMCH). Ten percent of all cases from the study were enrolled at Ramu Upazila Health Complex. This is located in Cox’s Bazar, a coastal district in the south of Chittagong Division, itself an area with relatively low incidence but situated adjacent to the high-incidence areas in Bandarban district. There was a separate ongoing malaria research project at this site during the study period which may have accounted for the high enrolment. CMCH, the main tertiary referral hospital in Chittagong city (otherwise referred to as Chittagong city corporation), had the widest catchment area with zero cases resident in the city (where there is thought to be no malaria transmission), 51% resident in other subdistricts in Chittagong district, 47% resident in the Chittagong Hill Tracts and 2% resident in another division. The sites in Lama, Dighinala and Alikadam had smaller catchment areas and relatively high malaria incidence rates.
Travel from residence to reported destination
Looking at the distribution of travel from residence to reported destination by administrative unit, of the 1631 people who reported travel, 1064 (65%) travelled to another location only within the smallest spatial unit of analysis, the union and 261 (16%) travelled outside the union but only within their own district. For longer distance travel, 273 (17%) cases travelled to another district within Chittagong Division, 23 (1%) to another division in Bangladesh outside the study area and 14 (1%) to another country. International travel comprised 6 people visiting Myanmar from Bandarban district, 4 going to India from Rangamati district, 1 to the Kingdom of Saudi Arabia for the Haj, 1 to the Democratic Republic of the Congo for peacekeeping, and 1 to Mozambique for business. One person cited residence in Lunglei, India, and visited Rangamati district for treatment.
Most travel outside unions of residence was in the south of the study region between Cox’s Bazar district (coastal region) to forested areas in Bandarban (31% of days and 45% of nights of overall travel, rising to 64% and 72% respectively for inter-district travel). Accounting for total travel reported by residents in Cox’s Bazar district both within and outside Cox’s Bazar, travel to Bandarban amounted to 32% by day and 80% by night (Additional file 1: Figures S5A and S5B). Travel was clustered in the north and south of the Chittagong Hill Tracts, with less than 1% of travel days and nights between northern and southern districts (Additional file 1: Figures S5 and S6). Most of the inter-district travel from Chittagong district, which is the site of the division capital and an area of low malaria endemicity, was to the malaria-endemic district of Bandarban (95% of travel days and 83% of overnight stays).
The overall geographic pattern of total days away from place of residence over the previous 2 months at a reported destination is shown at the union level in Fig. 4. A total of 87,683 days of travel were reported by all cases over the 2 months. In total, 80% of travel days were within the same union (Additional file 1: Table S6). Of days travelled, 81% were for work, of which 87% were within the same union. 12% of travel days were to a different district from their district of residence. For people travelling to the forest, travel outside the district of residence comprised 35% of travel days. Lama union had the most travel days within the union with 6213 (7%) days, followed by Alikadam union at 5319 days (6%). Out of the top 5 within union travel destinations, 3 were within Lama subdistrict (Additional file 1: Table S8). Looking at travel outside the union, the most travelled route was between Kachhapia in Ramu, Cox’s Bazar district to a neighbouring union Docchari in Bandarban district at 2006 days (2%).
The overall geographic pattern of travel by nights spent away from home (Additional file 1: Figure S4) was similar to that for days. In total, 10,817 nights were spent away from home with Alikadam union in Bandarban district reporting the most amount of overnight travel (Additional file 1: Table S9), all within the same union (593 or 5%). The highest recorded travel by nights between unions was from Kachhapia union, in Ramu subdistrict (Cox’s Bazar district) to Docchari union in Naikhongchhari subdistrict (Bandarban district) at 544 nights (5%). Only 34% of cases travelled within the same union when spending nights away from home compared to 48% travelling for non-work purposes during the day (Additional file 1: Table S7). The majority of nights (81%) spent away were for travel to the forest. Of cases reporting overnight travel, 44% travelled outside their district of residence. Over half the travel from outside the study area (134 nights, 55%) was to Bandarban district.
Residents from Chittagong district tended to travel further than other residents in the study area (median (IQR) 39 (22–56) vs 18 (10–32), p = 0.0004), as did patients from outside the study area (p < 0.001), Additional file 1: Figure S9.
Malaria sources and sinks
Lama and Alikadam subdistricts in Bandarban district, Chakaria and Ramu in Cox’s Bazar and Dighinala in Khagrachhari had the highest numbers of travellers to other endemic subdistricts (Additional file 1: Figure S10A). Using method 1, adjusting for the number of enrolled cases by subdistrict of residence (Additional file 1: Figure S10B), the top 4 ranked source subdistricts were all in Cox Bazar’s district (Pekua, Chakaria, Ramu and Ukhia—forming a contiguous region), with travel mostly to Alikadam, and Naikhongchhari subdistricts in Bandarban district for farming and forestry work. The next highest ranked sources were Kaptai (but with only 3 outgoing visitors and 6 enrolled cases) in Rangamati, adjoining a large lake, followed by Panchhari and Dighinala in Khagrachhari district with most travel to Sajek union in Baghaichhari. Using method 2, adjusting for both enrolment and origin and destination malaria API, the top 6 ranked sources were Belaichhari in Khagrachhari district, Alikadam in Bandarban district, Lama in Bandarban district, Ramu in Cox’s Bazar district, Matiranga in Khagrachhari district and Chakaria in Cox’s Bazar district (Additional file 1: Figure S10C). Including API had a major influence on the results. For example, Belaichhari had the second highest API in the Chittagong Hill Tracts but only 10% of cases travelled out from this subdistrict. Being next to the Indian border, the main means of travel in this hilly area was through a waterway and people travelling out of this subdistrict were likely to stay overnight. Lama and Alikadam were higher ranked as sources using method 2, with higher API, and substantial numbers of travellers to nearby subdistricts, in contrast to ranked sources adjusted solely for enrolment.
Subdistricts with the highest numbers of incoming visitors from endemic areas were as follows: Naikhongchhari in Cox’s Bazar district and Alikadam in Bandarban district with visitors from Ramu, followed by Baghaichhari in Khagrachhari district, Bandarban Sadar, Lama in Bandarban district, Cox’s Bazar Sadar in Cox’s Bazar district and Thanchi (remote forested area) in Bandarban district (Additional file 1: Figure S10D). The top ranking sinks, when adjusting for enrolled cases at origin (method 1), were Bandarban Sadar, Khagrachhari Sadar, Alikadam, Thanchi, Dighinala and Rowangchhari (another remote forested area in Bandarban district) (Additional file 1: Figure S10E). Method 2 gave similar results, but subdistricts with lower APIs such as Cox’s Bazar Sadar, Chakaria, Lohagara (forest fringe area near Chittagong) and Khagrachhari Sadar were ranked higher than Bandarban Sadar. Thanchi and Alikadam had the highest change in ranking from top 25% to the bottom 25% due to their higher API. Similar results were noted when considering days travelled and overnight stays (results are included in Additional file 1: Figure S11 for sources and S12 for sinks), with higher numbers of travellers and longer travel duration moving an area up the ranking.
Travel and demographics
Age and gender
Travel patterns varied substantially between different demographic groups. For instance, men travelled further and more compared to women (Fig. 5, Additional file 1: Figure S13, and Table S12). A total of 100% of men aged 15–39 years travelled compared to 65% for the rest of the study population. In total, 37% of men aged 15–39 travelled outside their subdistrict of residence compared to 15% for all other men (p < 0.001).
The geographic pattern of days of travel also differed greatly by occupation (Additional file 1: Figure S14). Farmers travelled the most (27,409 or 31% of days travelled). Children were reported to be either a student or a child when asked about occupation. Of the those who reported travel, only 14 children (28%), and 87 students (22% of the subgroup) travelled outside their union of residence. The distance travelled by military was the furthest, with 23% of their reported travel days greater than 100 km. Those with forestry-related occupations contributed the most amount of travel when looking at nights spent away from residence (3806 nights or 35%), forming two clusters of travel in the north and south of the Hill Tracts.
Odds of demographic factors and travel to another area
A univariate analysis was done by whether a group travelled or not, and the extent of their travel, i.e. outside the union, subdistrict or district for the following demographic factors—age, occupation, travel to forest and gender. Women, housewives, children aged under 5 years, people who lived in the forest and who did not visit the forest were less likely to travel (see Additional file 1: Figure S16 for results).
Significant factors from the univariate analysis were then included in a multivariate logistic regression to determine which groups travelled the most outside the administrative boundary of their place of residence (Table 1). The following groups were not included due to possible confounding: occupation listed as child (defined as age < 15 years), and housewife. Forest visits were not included as a separate factor due to overlap with forest-going occupations.
It was found that children under 15 years of age and women were least likely to travel beyond the union of residence (Table 1). Children and women also comprised the largest group amongst forest dwellers (63%). Students were significant for travel only beyond the union, but not further. Of those who reported travel, 44% of students aged 15–19 reported travel beyond the union, compared to 20% beyond the subdistrict. Males aged 25–49 were significant for travel at all administrative levels and consisted mainly of forest workers, farmers who reported travel to forest and labourers. Two thirds (63%) of labourers travelled to the forest from Cox’s Bazar to Bandarban. Military and businessmen travelled the furthest (p < 0.001).
This analysis was then restricted to subsets of travellers from the top quartile of predicted source subdistricts who travelled to another subdistrict derived using method 1 (Additional file 1: Table S13). The majority of the enrolled cases in source subdistricts were resident in Cox’s Bazar district (295 or 14% of total enrolled), followed by Khagrachhari district (118 or 8%). The odds ratios were different from the overall pattern when considering the whole study region. Military were less likely to travel if from a source subdistrict than from a non-source subdistrict, i.e. for cases resident in Cox’s Bazar or Khagrachhari district more than 90% travelled within the same district compared to cases originating elsewhere (Additional file 1: Figure S17A). Similarly, for forestry and farming occupations, of those residing in Cox’s Bazar district, 57% and 80% travelled to neighbouring subdistricts in a different district (i.e. Bandarban) rather than within Cox’s Bazar district. Heatmaps of demographic factors illustrating the proportions of travel between different districts are provided in the Additional file 1: Figure S17: A-I.
Travel to forest
Patients were asked about travel to the forest and the reason for this travel. A third of cases reported that they live in the forest (“dweller”, 31%), a third reported having visited the forest (“visitor”, 32%) and just over a third (37%) reported not staying in or visiting the forest (“no visits”). There were no differences in forest residence or visit status between those with different malaria species (P. falciparum vs. P. vivax vs mixed).
Forest travel and demographics
Forest dwellers had proportionally more children, 353 (54%) and women, 299 (46%) compared to the groups who did not visit the forest, 346 children (45%) and 294 women (38%), and also for those who reported forest travel, 62 children (9%) and 106 women (16%) (Additional file 1: Table S15).
When looking at reported forest travel (“forest visitor”) by age (5-year bands) and gender (Additional file 1: Figure S18), males aged 25–49 years were over-represented, 303 (54%) compared to the group not visiting the forest (“No forest visits”) (137 (29%), p < 0.0001). This group was also over-represented compared to the study area census population, where males aged 25–49 constituted 32% of the census population (p < 0.001). Similarly, young males age 0–14 were under-represented, 27 (5%) in those who visited the forest compared to those who did not (184 (39%), p < 0.001). These differences were also found when correcting for multiple comparisons, as described earlier. There were no significant differences in the age or gender of the “forest dweller” or “no forest visit” groups compared to the census data for those living in Chittagong Division. Forest dwellers travelled shorter distances overall (11 km, 7–23), compared to both forest visitors (21 km, 11–38, p < 0.001) and “no forest visit” group (23 km, 11–39, p < 0.001). There was no significant difference in overall distances travelled between forest visitors and the no forest visit group (p = 0.4939). The forest visit group travelled more by days and nights compared to the no forest visit group (p < 0.001) (Additional file 1: Table S14, Figures S19, and S20).
Reasons for forest travel
Figure 6 shows reported reasons for travel to forest. For jhum farming, people reported various reasons such as farming paddy, vegetables and turmeric. A quarter of the people travelling to plantations specified working in a rubber plantation (17 cases or 23%). Only 1 person reported hunting as a reason for visiting the forest. The median distance travelled to the forest was 21 km (IQR 11–34). Again, people who travelled for military (27 km, IQR 20–177) and government reasons travelled the furthest (293 km, IQR 247–352)
Study subjects were also asked about reasons for non-work travel which were reported by 316 cases. Of this subset, 37% reported travel for social reasons, 28% reported going to the market and a further 8% reported travel for forest-related activities. The median distance travelled for non-work travel was 17 km (IQR 8–33 km). Additional file 1: Figure S22 summarises reasons for non-work travel with distances travelled. People travelled the furthest for forest activities (reported as part of frequent non-work activity) with a median distance of 61 km and the shortest distances when going to the market at 7 km median distance. See Additional file 1 for a detailed breakdown of reasons for travel, including the “miscellaneous” group.
Forest travel and forest cover
Of the group who worked in the forest (as part of their occupation), 12% did not report visiting the forest in the preceding 2 months. Informal feedback from study staff revealed that reported forest travel in the study may have been affected by subjectivity regarding the definition of forest, in that different people had different ideas about what is meant by forest.
The median (range) forest cover by union for unions of residence in the study area was 24% and in the 3 Chittagong Hill Tracts (CHT) districts 67.5%. The overall per-person forest cover was 68% in the study area. There was no significant difference (Additional file 1: Figure S23A and B) in per-person % forest cover between unions of residence of people who reported living in the forest (median 67% forest cover, n = 657) and those who did not (median 68% forest cover, n = 1433). Unions of residence for enrolled patients in Cox’s Bazar and Chittagong districts had median forest cover of 4.5% and 9% respectively, and a per-person forest cover of 23 and 11% respectively. The 3 Chittagong Hill Tract districts had a median per-person forest cover of 68%. The border areas near India and Myanmar were heavily forested with 75% to 100% cover Additional file 1: Figure S23A. Looking at the travel destination of people who reported having travelled to the forest, especially in Bandarban district, they mostly visited heavily forested regions near the border (56% visited unions with ≥ 75% forest cover), compared to people who did not report visiting the forest (39%, p = < 0.001, Additional file 1: Figure S23B). Forest cover in destination unions for people who reported visiting the forest was 77%, higher than in destination unions for those who did not report visiting the forest (67%, p value < 0.001). The highly forested areas in the study area also had higher malaria incidence rates (median API 50.4 vs 2.7, p < 0.001), and lower population density (median 58 people per square km vs 692 people per square km, p < 0.001, Additional file 1: Figure S24).