The authors have declared that no competing interests exist.
Conceived and designed the experiments: EL BL. Performed the experiments: EL. Analyzed the data: EL TML. Contributed reagents/materials/analysis tools: EL TML BL. Wrote the paper: EL BL.
Understanding the spatio-temporal pattern of malaria transmission where prevention and control measures are in place will help to fine-tune strategies. The objective of this study was to assess the effect of mass distribution of bednets and indoor residual spraying (IRS) with insecticides on the spatio-temporal clustering of malaria in one malaria endemic village in south Ethiopia.
A longitudinal study was conducted from April 2009 to April 2011. The average population was 6631 in 1346 locations. We used active and passive searches for malaria cases for 101 weeks. SatScan v9.1.1 was used to identify statistically significant retrospective space–time clusters. A discrete Poisson based model was applied with the aim of identifying areas with high rates. PASW Statistics 18 was used to build generalized Poisson loglinear model.
The total number of both types of malaria episodes was 622, giving 45.1 episodes per 1000 persons per year; among these, episodes of
The risk of getting malaria infection varied significantly within one village. Free mass distribution of ITNs did not influence the spatio-temporal clustering of malaria, but IRS might have eliminated malaria clustering.
Malaria is a leading health problem in Ethiopia, where 67% of the 82 million people are estimated to be at risk. There were 1 036 316 confirmed cases of malaria in 2009. The dominant plasmodium species are
Nationally, the number of cases has declined since 2005 due to an expansion in the malaria control programmes. However, malaria admissions increased in 2009
To implement malaria prevention and control measures, and to understand risk dynamics, application of Geographic Information System (GIS) has been emphasized, in particular to provide a precise definition of the time and location of epidemics
Spatial epidemiological studies at a finer geographic scale, such as households, help to increase understanding of the varied pattern of malaria infection and transmission
Chano Mille Kebele (Kebele is the lowest administrative unit in Ethiopia) is 492 km south of Addis Ababa (
Chano Mille Kebele was selected purposely for the study of malaria epidemiology. Three main irrigation ditches run from the neighbouring Kebele in the west, cross the Kebele and may also end within the Kebele. There are two adjacent Kebeles, Chano Dorga to the north-west and Chano Chalba to the south-west. The area to the east and south-east sides of the Kebele, extending to Lake Abaya, is used for agricultural purpose. Most of the households grow mango trees within their compounds.
There was one health post in the village staffed by a health extension worker. A health post provides basic health services, including malaria diagnosis using rapid diagnostic test (RDT) kits and treatment with Artemether–Lumefantrine.
The cohort study was carried out from April 2009 to April 2011. Both active and passive surveillance schemes were used. Each household was visited every week for 101 weeks looking for cases of fever (temperature ≥37.5 degrees Celsius); if needed patients were referred to the health post for diagnosis and treatment of malaria (active case finding). Each day, we checked whether the referred cases had visited the health post. During the days between the visits, the residents were advised to self-report to the health post if they became febrile (passive case-finding).
We gave a unique household number to each household before the first census. The geographic coordinates of all households were recorded using GPS during the first census. The GPS reading for the new households was performed during the midway census. We also recorded GPS coordinates for the main vector breeding sites. These vector breeding sites are swampy areas close to Lake Abaya, with many hoof prints of cattle and hippopotami. Such small water bodies are formed mainly after flooding of the lake during the rainy season. We did not find larvae of
The laboratory technician used a single finger prick to collect blood samples for RDT and prepared thick and thin blood films for microscopic evaluation. Based on the results of the RDT, the patients were treated with Artemether–Lumefantrine (
We used SatScan v9.1.1 (
A maximum spatial cluster size of 50% is recommended because it should capture all clustering. In our case, it captured all the smaller clusters (most likely and secondary) that we observed while running the analysis with smaller maximum spatial cluster size within one most likely cluster, and it provided no secondary cluster. The greater portion of the cluster circle had no households in it. The clusters with maximum cluster size smaller than 50% yielded smaller clusters with varying relative risks. A maximum spatial cluster size of 35% yielded a very small secondary cluster with only 34 people in it, and the greater portion of the cluster circle did not hold households. A maximum spatial cluster size of 15%, to accommodate an additional secondary cluster, pushed the most likely cluster to the edge of the village and again the greater portion of the cluster circle contained no households. This revealed three secondary clusters, of which one contained only 21 people and the other was the same as the secondary cluster detected using the 25% maximum spatial cluster size restriction. Therefore, we decided that the 25% maximum cluster size restriction was most appropriate to show the malaria clustering activity of the study area because the portion of the most likely cluster circle with no households was relatively small and this specification yielded a secondary cluster. All the results reported here were obtained with this maximum spatial window size. The same maximum spatial cluster size was used to investigate the differences in clustering by the type of malaria and years of study. (Supplemental
The PASW Statistics 18 program (Chicago, IL, USA) was used to fit a generalized Poisson loglinear model. The dependent variables were episodes of vivax and falciparum malaria. The potential predictors considered were: distance to the vector breeding site, number of households located between each household and the vector breeding site (household count), sex, age, wealth index and total number of nights spent under insecticide-treated nets (ITNs). Every week, residents were asked whether they slept under ITNs the night before the interview and the names of the household member who had slept under ITNs were recorded. To get the total number of nights spent under ITNs, we summed-up the weekly data for each individual. The number of weeks for which each individual had been observed was used as a scale weight variable. The scale parameter method was Pearson chi-square and a robust estimator was used for the covariance matrix. The log-likelihood function was kernel. The ratio of the Pearson chi-square value to its degrees of freedom was used to rule out over-dispersion. This value was 1.34 and 1.28 for vivax and falciparum episodes, respectively. Given that these values did not deviate significantly from 1, we assumed that the Poisson distribution was a good fit for the data. Statistically significant (P-value <0.05) variables detected during bivariate analyses were considered for the multivariate model. The exponential form of the estimates was interpreted as the incidence rate ratio (IRR). We reported the IRR with 95% confidence intervals (CI).
We used ESRI ®ArcMap™ 9.3(CA, USA) to calculate the distance of each household from the vector breeding site (in km) and to produce the maps.
To get household count
A recent paper by the authors has reported the predictors of falciparum malaria episodes, which included the effect of meteorological covariates (total rainfall, temperature and relative humidity), ITN utilization rate, efficacy of the insecticides used for IRS, and other factors. The statistical models employed were auto-regressive integrated moving average models with a transfer function model, generalized Poisson loglinear model and generalized estimating equation with logit link function. Principal Component Analysis was used to construct wealth index. The variables included were presence of electricity, watch, radio, TV, mobile telephone, refrigerator, separate room used for kitchen, bicycle, any land used for agriculture, livestock, account in bank or credit association and latrine facility. In addition the main materials of the floor, the roof and the wall were included. The technical details of the construction of the wealth index were reported elsewhere
The Regional Health Research Ethics Review Committee of the Southern Nations, Nationalities and People’s Regional Health Bureau has approved this research project. Informed verbal consent was obtained from all study participants and recorded by the research team on the ethical consent form (with prior approval from our ethics review committee). For minors, consent was obtained from their caregivers or legal guardians. All cases of malaria were treated immediately. Given that the blood samples were collected only for the purpose of malaria diagnosis and treatment, written consent was not suggested by the Regional Health Research Ethics Review Committee.
The study population (average of the three censuses) was 6631 in 1388 households (1346 locations). Within the study period, the total number of both types (
Types of malaria | ||||||
|
|
|||||
Number | Percent | Number | Percent | |||
Number of episodes | Sex | Male | 208 | 65.8 | 196 | 64.1 |
Female | 108 | 34.2 | 110 | 35.9 | ||
Age in years | <5 | 45 | 14.2 | 71 | 23.2 | |
5–14 | 146 | 46.2 | 130 | 42.5 | ||
15–24 | 85 | 26.9 | 71 | 23.2 | ||
>24 | 40 | 12.7 | 34 | 11.1 | ||
Total | 316 | 306 | ||||
Annual number of episodes per 1000 | 22.9 | 22.2 | ||||
Number of locations | 226 | 199 | ||||
Mean (SD) distance of locations fromthe vector breeding site (km) | With episodes | 2.28 (0.36) | 2.36 (0.34) | |||
Without episodes | 2.53 (0.33) |
2.51 (0.34) |
Number of locations: 1120.
Number of locations: 1147.
SD: standard deviation.
Both the most likely and the secondary clusters were found on the south-east edge of the village, facing the vector breeding site (
Meanwhile, green dots refer to
The most likely space–time cluster lasted for 9 months (of the 25 months of the study) for both
Both types of malaria |
|
|
|
Number of locations included | 326 | 326 | 332 |
Coordinates | 6.1100N, 37.6005E | 6.1100N, 37.6005E | 6.1105N, 37.5986E |
Radius (km) | 0.43 | 0.43 | 0.33 |
Time frame | Dec. 2009 to Aug. 2010 | Dec. 2009 to Aug. 2010 | Nov. 2009 to Jul. 2010 |
Population | 1626 | 1626 | 1653 |
Number of episodes | 230 | 133 | 106 |
Expected episodes | 54.99 | 27.94 | 27.4 |
Annual episodes/1000 | 188.6 | 109.0 | 85.8 |
Observed/expected | 4.18 | 4.76 | 3.87 |
Relative risk | 6.05 | 7.49 | 5.39 |
Log likelihood ratio | 184.43 | 124.51 | 77.11 |
P-value | <0.001 | <0.001 | <0.001 |
Both types of malaria |
|
|
Number of locations included | 81 | 102 |
Coordinates | 6.1077N, 37.5947E | 6.1071N, 37.5946E |
Radius (km) | 0.17 | 0.22 |
Time frame | Dec. 2009 to Aug. 2010 | Dec. 2009 to Apr. 2010 |
Population | 407 | 489 |
Number of episodes | 43 | 23 |
Expected episodes | 13.76 | 4.48 |
Annual episodes/1000 | 140.8 | 113.8 |
Observed/expected | 3.12 | 5.13 |
Relative risk | 3.28 | 5.47 |
Log likelihood ratio | 20.46 | 19.67 |
P-value | <0.001 | <0.001 |
The space
Year I(Apr. 2009 to Mar.2010) | Year II(Apr. 2010 to Apr.2011) | ||
Most-likely cluster | Secondary cluster | Most-likely cluster | |
Number of locations included | 264 | 79 | 322 |
Coordinates | 6.1105N, 37.6009E | 6.1077N, 37.5947E | 6.1099N, 37.6009E |
Radius (km) | 0.39 | 0.17 | 0.47 |
Time frame | Dec. 2009 to Mar. 2010 | Jan. 2010 to Mar. 2010 | Apr. 2010 to Aug. 2010 |
Population | 1335 | 399 | 1604 |
Number of episodes | 107 | 22 | 115 |
Expected episodes | 21.9 | 4.9 | 27.6 |
Annual episodes/1000 | 241.9 | 223.8 | 171.2 |
Observed/expected | 4.89 | 4.52 | 4.2 |
Relative risk | 6.77 | 4.77 | 6.21 |
Log likelihood ratio | 97.79 | 16.52 | 93.13 |
P-value | <0.001 | 0.002 | <0.001 |
The three major preventive measures applied by the government during the study period included: IRS with Dichlorodiphenyltrichloroethane (DDT): 91% of the houses were sprayed in June 2009, free mass ITN distribution (2.3 ITNs per household) in March 2010, and IRS with Deltamethrin: 97.5% of the houses were sprayed in July 2010. The spatial coverage of IRS and ITNs are presented in
The shaded part indicates the time-span of the most likely space–time cluster.
Of the total number of falciparum and vivax episodes, 224 and 203 episodes, respectively, occurred among permanent residents for whom we had follow-up data.
Living nearer to the vector breeding site increased the risk of acquiring falciparum malaria, that is, each 1 km closer to the vector breeding site added 4.93 (95% CI: 2.59–9.35) times more risk. Household count was negatively associated with both types of malaria episodes with all vector’s search angles considered during bivariate analyses. However, the household counts of the first three lower search angles (1, 5 and 10 degrees) were found to be statistically significant in the multivariate model for falciparum episodes. As the vector’s search angle decreases, the effect of household count increases. With a search angle of 1 degree, for each additional household between a household of interest and the vector breeding site, the risk of getting falciparum malaria decreases by 2%. Male participants had 1.63 (95% CI: 1.22–2.18) times more risk of acquiring falciparum malaria. When compared with adults aged >24 years, children aged 5–14 years had 3.82 (95% CI: 2.52–5.78) times more risk. Having higher wealth index was marginally failed to be protective against falciparum malaria (P-value: 0.051) in the model containing household count. Nevertheless, for search angles above 15 degrees, the wealth index regains statistical significance as the household count becomes no more statistically significant. The total number of nights spent under ITNs was not associated with the total number of falciparum episodes. Regarding vivax malaria, household count, sex, wealth index and the total number of nights spent under ITNs were not significant predictors. Meanwhile, when compared with adults aged >24 years, children <5 years old had 7.6 (95% CI: 4.2–13.74) times more risk, and living 1 km closer to the vector breeding site conferred 2.9 (95% CI: 1.2–6.99) times more risk of acquiring vivax malaria (
Variable (n = 8121) | ||||||
Crude IRR (95% CI) | Adjusted IRR (95% CI) | Crude IRR (95% CI) | Adjusted IRR (95% CI) | |||
Distance (km) fromvector breeding site |
11.11(6.67–20.0) | 4.93(2.59–9.35) |
4.55(2.7–7.69) | 2.9(1.2–6.99) |
||
Household count with a searchangle of 1 degree |
0.98(0.96–0.99) | 0.98(0.96–0.99) |
0.97(0.96–0.98) | 0.99(0.97–1.01) | ||
Sex : Male | 1.65(1.23–2.23) | 1.63(1.22–2.18) |
1.26(0.87–1.83) | NA | ||
Age in years |
<5 | 3.37(2.02–5.62) | 3.07(1.87–5.03) |
7.82(4.31–14.2) | 7.6(4.2–13.74) |
|
5–14 | 4.32(2.83–6.58) | 3.82(2.52–5.78) |
6.91(3.95–12.12) | 6.53(3.74–11.41) |
||
15–24 | 2.1(1.28–3.44) | 2.14(1.31–3.48) |
1.84(0.91–3.74) | 1.84(0.91–3.75) | ||
Wealth index | 0.76(0.65–0.89) | 0.91(0.83–1.0) |
0.95(0.78–1.16) | NA | ||
Total number of nightsspent under ITNs | 1(0.99–1.01) | NA | 1(0.99–1.01) | NA |
Reference category: >24 years.
Significant at P-value <0.05.
The reciprocal of the IRR (95% CI) was presented to show the risk of being closer to the vector breeding site.
Household count refers to the number of households located between each household and the vector breeding site. For search angles of 5 and 10 degrees, the effect measures, adjusted IRR (95% CI), became 0.995 (0.991–0.999) and 0.997 (0.995–0.999), respectively.
P-value: 0.051.
NA: Not applicable.
There was a space–time clustering of malaria at household level. Free mass distribution of ITNs did not affect the spatio-temporal clustering of malaria, but IRS might have. Living nearer to the vector breeding site increased the risk of acquiring malaria infection. These differences in malaria risk within a population who live in one village reflect the complexity of the disease transmission dynamics.
Our study confirms the findings of other studies that have shown spatio-temporal clustering of malaria cases at varying geographical extents
Clustering varied by the type of malaria
An increase in risk of getting vivax malaria has occurred in two locations (vivax malaria had significant secondary cluster) and the relative risks within vivax malaria clusters were smaller than that of falciparum malaria cluster. This may suggest that a targeted intervention could be easier to apply for falciparum than for vivax malaria in the study area.
There is ample evidence that sleeping under ITNs protects against malaria infection
The time span (December 2009 to August 2010) for the spatio-temporal clustering of both types of malaria ended when the possible effect of IRS with Deltamethrin (sprayed in July 2010) started. Although it may not be possible to reach the conclusion that IRS alone eliminated the spatio-temporal clustering of malaria without considering the effects of other factors such as rainfall and temperature, it is also not possible to state that the timing of the possible effect of IRS and the end of the clustering coincided simply by chance. Thus, we suggest that IRS with Deltamethrin has possibly suppressed the transmission to the level where little power to identify clusters remains. Meanwhile, the location where we observed the clustering activities was almost perfectly covered with IRS where only fewer than 10 households did not receive the intervention. A recent paper by the authors has discussed the reasons for the differences in the risk of falciparum malaria with regard to sex, age, wealth index, ITN use (with a 2-week lag in the effect) and other factors. We also showed that, among the meteorological covariates, rainfall (with a lag of 6 weeks) was a significant predictor of falciparum malaria. When controlled for the effect of rainfall, IRS with Deltamethrin significantly reduced the incidence of falciparum malaria; however, utilization rate of ITNs did not
All the clusters observed were on the south-east side of the community, and near to the identified vector breeding site on the shore of Lake Abaya. This implies that the greater risk of malaria infection among these households served as a ‘barrier’ between the breeding site and households that lived to the north-west of the cluster. This was supported by the analysis showing the significant effect of number of households located between each household and the vector breeding site while using the vector’s search angle scenarios of 1, 5 and 10 degrees. Meanwhile, the clustering activity observed close to the vector breeding site in our study area could provide an example of what Bousema et al. described as “hotspots of malaria transmission in the dry and wet season”. These hotspots were referred to as groups of households that have an increased risk of malaria infection within a focus of malaria transmission
The edge effect is worth mentioning, because we observed clustering activity on the edge of the village. An edge effect may result in a biased estimate of risk at the edge of a study area provided that there are no data for the adjacent localities
The risk of malaria infection varied within one village, and there was spatio-temporal clustering of malaria episodes at household level. The vector breeding site identified may have played a role in the clustering of malaria. Mass distribution of ITNs did not influence the spatio-temporal clustering of malaria, but IRS with Deltamethrin might have eliminated the clustering activity. Local knowledge of malaria transmission and follow-up on ITN use, combined with targeted interventions, may improve the existing malaria prevention and control efforts.
(TIF)
(TIF)
(TIF)
(TIF)
(TIF)
(TIF)
(DOCX)
We thank the residents of Chano Mille Kebele for their cooperation throughout the study period.