Acta Tropica 103 (2007) 157–162
Descriptive spatial and spatio-temporal analysis of the 2000–2005 canine rabies endemic in Santa Cruz de la Sierra, Bolivia K. Suzuki a,∗ , J.A.C. Pereira b , R. L´opez b , G. Morales c , L. Rojas c , L.E. Mutinelli a , E.R. Pons a a
Facultad de Ciencias Veterinarias, Universidad Nacional de La Plata, Av. 60 y 118, La Plata, B1900AVW, Buenos Aires, Argentina b Facultad de Ciencias Veterinarias, Universidad Aut´ onoma Gabriel Ren´e Moreno, Av. 26 de Febrero entre Av. Busch y Centenario, Santa Cruz de la Sierra, Bolivia c Laboratorio de Investigaci´ on y Diagn´ostico Veterinario, Av. Ejercito N◦ 153 y Casilla N◦ 29, Santa Cruz de la Sierra, Bolivia Received 3 November 2006; received in revised form 25 May 2007; accepted 1 June 2007 Available online 14 June 2007
Abstract The authors analysed laboratory samples testing positive for rabies virus collected between 2000 and 2005 from Santa Cruz de la Sierra, Bolivia, to assess the spatial and spatio-temporal distribution of the endemic of canine rabies. According to the results of the spatial analysis, canine rabies was not evenly distributed in the city. Significant spatio-temporal clustering of canine rabies cases showed a trend towards clusters with high incidence rates in two particular areas compared with surrounding areas in the city. Benefits obtained from each of the analytical techniques utilised are described and practical difficulties in conducting and interpreting the results are discussed. © 2007 Elsevier B.V. All rights reserved. Keywords: Rabies; Bolivia; Spatial analysis; Spatio-temporal analysis
1. Introduction Rabies is an acute, progressive, and incurable viral encephalitis, and still an important public health problem in Latin America (Favi et al., 2003). Where rabies remains a major threat within Latin America, domestic dogs act as the main reservoir for the virus and primary source of human exposure through animal bite
∗ Corresponding author. Tel.: +54 221 425 3276; fax: +54 221 425 3276. E-mail address:
[email protected] (K. Suzuki).
0001-706X/$ – see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.actatropica.2007.06.003
(Wunner, 2005). Although most municipalities have experienced interrupted circulation of the rabies virus in dogs with appropriate control measures, endemic areas persist in several countries due to deficiencies in dog vaccination and social and environmental conditions that bring people into contact with dogs (Belotto et al., 2005). In Bolivia, 78% of the total cases of animal rabies (N = 8627) corresponded to canine rabies, and 92% of all human rabies cases (N = 103) were transmitted by dogs between 1990 and 1999. Between 1997 and January 2001, eight of the nine departments of Bolivia reported human and canine rabies cases, the most affected departments being Santa Cruz (which
158
K. Suzuki et al. / Acta Tropica 103 (2007) 157–162
is the department where Santa Cruz de la Sierra is located), Cochabamba, and La Paz (Ministerio de Salud y Prev´ısion Social, 2001, 2002). Cities in poorer departments such as Santa Cruz de la Sierra lag behind in control efforts because resources are scarce and programmes are poorly focussed (Widdowson et al., 2002). Precise knowledge of the distribution of endemic canine rabies in the city will help focus control measures in a resource-poor environment. In other words, given the present situation in Santa Cruz de la Sierra, an assessment of spatial and temporal patterns of canine rabies cases that might influence the success of the on-going vaccination campaign, in order to reduce risk of rabies in domestic dogs and humans, is needed. The objective of the study was to analyse the spatial and spatio-temporal distribution of the endemic of canine rabies in Santa Cruz de la Sierra from January 2000 to December 2005, to facilitate allocation of resources to potential disease interventions, such as vaccination. 2. Materials and methods 2.1. Study area Santa Cruz de la Sierra is the capital city of the Department of Santa Cruz, located in the eastern part of Bolivia (17◦ 45 S, 63◦ 14 W) at 416 m above sea level. The city has a sunny and semi-tropical climate, with an average temperature of 21 ◦ C in winter and 32 ◦ C in summer. The total land area is about 370 square kilometres, divided into 16 districts and an industrial area. The population is estimated at 1.4 million people (Ministerio de Salud y Prev´ısion Social, 2006). The Bolivian government issued a regulation of special sanitary measures for rabies control in November 2005, owing to the increased incidence of dog and human rabies cases in recent years (Ministerio de Salud y Deportes, 2005). The last mass vaccination campaign was implemented on 26 and 27 August 2006 (Gobierno Municipal Aut´onomo de Santa Cruz de la Sierra, 2006). Dogs aged ≥ 1 month are eligible for vaccination during the campaign and the owner receives a rabies vaccine certificate following vaccination. Nevertheless, vaccination coverage data are unreliable due to lack of information about the canine population in the study area. Canine rabies is a major problem in Santa Cruz de la Sierra, accounting for >90% of all animal rabies in the city (Widdowson et al., 2002). 2.2. Data The veterinary research and diagnostic laboratory (Laboratorio de Investigaci´on y Diagn´ostico Veterinario;
LIDIVET) receives samples of brain tissue for rabies diagnosis from suspected cases in animals and humans. Impression smears of brain tissue from the cerebellum, Ammon’s horn, and medulla are examined after staining by direct fluorescent antibody test. Since 2000, the location where the animal was found has been recorded at street level, as well as the date when the animal was found, species, age group, sex, bite history, and the results of the diagnostic test. Data collected are entered into the municipal rabies database “Registro de datos y resultados de rabia” using the Microsoft Access 97 (Microsoft Corporation, Redmond, USA) managed by LIDIVET. None of the authorities provided a basis for a relevant understanding of the recent canine population in the study area owing to nonexistent census data. 2.3. Analysis All the records of canine samples examined at LIDIVET for rabies from within the municipal boundaries of Santa Cruz de la Sierra between 1 January 2000 and 31 December 2005 inclusive were first extracted from the entire database. The records of location and date on laboratory-confirmed rabid dogs were then used for the following analyses. Initially, the geographical point location datasets were plotted with ArcView 3.3 (ESRI, Redlands, CA, USA) and then smoothed by using Gaussian kernel methods (Kelsall and Diggle, 1995) implemented in CrimeStat III (Levine, 2004). A bandwidth of 0.6 kilometres was used, calculated by using the adaptive bandwidth selection method (Bailey and Gatrell, 1995) and a grid cell size of 0.01 square kilometres. Gaussian kernel density surfaces of the cumulative number of canine rabies cases in Santa Cruz de la Sierra up until cut-off dates at 31 December 2000, 2001, 2002, 2003, 2004, and 2005, were made by using the technique described above. A relief map, providing an estimate of the cumulative incidence of a canine population experiencing a case of rabies at each defined cut-off date, adjusted for the heterogeneity in the spatial distribution of the underlying population-atrisk, would be useful for further assessment. However, such further assessment was not considered in this instance as the canine population census information was unavailable. The space-time scan statistic was implemented in SaTScan ver. 7.0.1 (Kulldorff M. and Information Management Services, http://satscan.org/, 2006) to investigate the data for any potential clustering of canine rabies cases. The space-time scan statistic was defined by a cylindrical window with a circular geographic base
K. Suzuki et al. / Acta Tropica 103 (2007) 157–162
and with height correlated to time. The base was defined precisely as for the spatial scan statistic, while the height expressed the time period of possible clusters. The cylindrical window was subsequently moved in space and time, so that for each potential geographical location and size, it also visited each possible time period. In effect, an uncountable number of overlapping cylinders of different size and shape was obtained, jointly covering the whole study area, where each cylinder reflected a possible cluster. The likelihood function was maximised over all window locations and sizes. The window with the maximum likelihood was the ‘most likely cluster’ or the cluster least likely to be owing to by chance. The software also determined ‘secondary most likely cluster’ in addition to the most likely cluster, according to their scan statistic. As only the case data were available, the space-time permutation model was used which adjusted for natural temporal and geographical variation, and dealed with the lack of suitable population-at-risk data (Kulldorff et al., 2005). The dataset was scanned for clusters in spatial and temporal dimensions ranging from 0 to 50% of the total population-at-risk (up to 50% of the study area and 50% of the study period). Statistical significance was evaluated using Monte Carlo hypothesis testing with 999 iterations. The number of iterations chosen was considered to provide moderate statistical power whilst minimising computing time. Other statisti-
159
cal calculations were performed with the R software ver. 2.4.0 (Ihaka and Gentleman, 1996). A value of P < 0.05 was used throughout the analyses to define statistical significance. 3. Results During 2000–2005, 184 to 872 canine samples were tested annually in the study area; 5–58% of them were rabies positive (972 confirmed rabid dogs). In the same period, human rabies cases occurred occasionally (0–6 cases annually). In Fig. 1, density maps for ‘2000’ to ‘2005’ show the cumulative number of canine rabies cases in the study area for the periods from 1 January 2000 up until cut-off dates at 31 December in each of the respective years. On visual inspection, the highest densities of canine rabies by 31 December 2005 were in areas around the central place of the study area. The space-time scan statistic confirmed a visual assessment of clustering of high incidence rates (Fig. 2, Table 1). The ‘most likely cluster’ was centred in District 2 and across the neighbouring districts, and the second most likely cluster was in District 5. These significant spacetime clusters involved 60% (95% CI: 36–80%, N = 15) and 15% (95% CI: 7–31%, N = 33) of canine rabies cases in a 6- and 5-month time period, respectively. There were no other statistically significant clusters.
Fig. 1. Gaussian kernel density surfaces of the cumulative number of canine rabies-positive cases in Santa Cruz de la Sierra for the periods from 1 January 2000 to 31 December in each of the respective years; 2000, 2001, 2002, 2003, 2004 and 2005. The legend indicates the relative percentile ranks of the kernel density values based on the outputs for the period from 1 January 2000 to 31 December 2005.
160
K. Suzuki et al. / Acta Tropica 103 (2007) 157–162
Fig. 2. Map of locations of canine rabies cases in Santa Cruz de la Sierra between 2000 and 2005, showing district boundaries with each district number, and clusters of disease identified using space-time permutation scan statistic. The most likely and the second most likely cluster are indicated by a thick and thin circle, respectively.
4. Discussion 2This study demonstrated the utility of an official electronic record database for analyses of spatial and spatio-temporal characteristics of canine rabies. The database provided information on the results of the diagnostic test performed on the suspected population for canine rabies, allowing analysis of individual-based data for the study. The modest-sized datasets posed a
number of computational challenges and, as a result, some analytical compromises had to be implemented. The determination of smoothing parameters using robust techniques such as cross-validated log-likelihood function methods (Diggle et al., 2005) was attempted but proved impracticable as a result of the need for a larger dataset. The application of edge correction techniques to the kernel density surfaces (Zheng et al., 2004) would have been desirable, but was not computationally pos-
Table 1 Locations and periods of the two clusters of higher proportion of canine rabies shown in Fig. 2 Cluster
Coordinates
Radius (km)
Time frame (number of month)
Scan statistic
P
Most likely
17.775296 S 63.164741 W 17.723519 S 63.163932 W
2.03
September 2002–February 2003 (6)
11.0
