Study area
Turkana County is located in the northwestern part of Kenya. The county shares borders internationally with Ethiopia to the north, Sudan to the northwest and Uganda to the west. Internally, the county borders Marsabit, Samburu, and West Pokot and Baringo Counties (Figure 1). The county is characterized by arid and semi-arid lands covered with sparse thorny shrubs. A large proportion of the county’s area consists of low-lying plains with isolated rocky mountainous, hilly ranges and several seasonal rivers. The rainfall pattern and distribution are unreliable and erratic over time ranging annually between 120 mm and 430 mm. Temperatures range annually from a low of 24°C to a high of 38°C with a mean of 30°C [17]. Administratively, Turkana County is divided into 17 divisions and 67 sub-locations [14]. Six administrative divisions namely, Loima, Oropoi, Kakuma, Lokichogio, Kibish and Kaaleng which served as the international frontier bordering divisions that reported initial PPR outbreaks in 2006 were purposively selected for this study. These divisions were perceived to be the foci of disease introduction into the county.
Study design, sampling unit, sample size calculation and sampling process
The study design was based on a proportionate stratified random sampling design while the sample frame was based on sheep and goat populations in the six administrative divisions that formed the study area. The sampling unit was an individual animal of specific age and vaccination status belonging to a village herd known locally as an adakar. In the Turkana community, an adakar entails a cluster of often-related households that pursue similar socio-economic activities such as search for pasture, water and security, under a trusted leader [15]. An adakar is, therefore, more or less synonymous to a village flock.
Since there is no serological test available that could differentiate animals vaccinated with homologous PPR vaccine from animals that had recovered from a natural PPR infection, the Turkana pastoral community, through focus group discussions (FGD), was deemed the best source of information regarding vaccination status of sheep and goats to aid in sampling. Together with the age structure, also sourced from FGD, these variables were subsequently used in the sample stratification. Five strata (young kids and lambs <6 months of age; middle-aged >6 months and <24 months of age vaccinated and unvaccinated groups; adults >24 months of age vaccinated and unvaccinated groups) were considered in this study for each of the two species (sheep and goats) investigated. Strata populations for each species were determined from the population of sheep and goats in the county, herd structure in Turkana herds established through participatory epidemiology approaches [19,20] and estimated vaccination prevalence of 14% in Turkana reported in unpublished data of Director of Veterinary services of the Government of Kenya.
For each species (sheep and goat), the stratum sample sizes determination was carried out using the formula by Bennett et al. [21] implemented within the ProMESA software program for statistical sampling in animal populations [22]. In determining the sample size, we ignored the sensitivity and specificity of the diagnostic test given their high values of 100% as reported by manufacturers in c-ELISA diagnostic test data control sheet. We assumed the prevalence of PPRV seropositivity was 50% with a relative error of 10%. We chose the 50% sero-prevalence because it provides the largest sample size (for given values of absolute error). The sample size was determined as 384 samples per each species and was then proportionately allocated to each of the strata based on sheep and goat population in each stratum. The strata sample sizes were determined as detailed in the online supplementary file.
The number of households in each adakar varies from 40 to 100 with an average of 70 [23]. The average number of sheep and goats per household were estimated at 34 (ranging between 3 and 100) and 54 (ranging between 7 and 167) respectively. We used this information to estimate the number of adakars in a sub-location and the population of sheep and goats in an adakars. A total population of 535 adakars was estimated in all the sub-locations within the six selected administrative divisions. The sheep and goat population for each adakars was estimated by dividing the population of sheep and goats in a sub location with number of adakars estimated in that sub-location. In this instance, we assumed equal herd sizes in adakars in any one sub-location.
All adakars in all six study divisions were allocated sequential numbers from 1 to 535. We arbitrarily listed the divisions beginning with Loima, Oropoi, Kakuma, Lokichoggio, Kibish and then Kaaleng divisions. For each division, the five animal strata populations were listed alongside each adakar. Cumulative population estimate per stratum for all adakars was calculated with the first animal in the stratum being from Loima and last being from Kaaleng. An individual animal was subsequently selected using simple random sampling using the random number function in Microsoft Excel®. Out of the 535 adakars estimated in the study area, selected animals fell in 155. Some animals selected were located in inaccessible adakars experiencing insecurity from livestock rustling, high mobility of the Turkana pastoralists and impassable roads. The inaccessible areas were in:
-
1)
whole of Oropoi division except Kalobeyei location,
-
2)
Lokichoggio division in such areas as Lorao location and sub locations of Songot and Lokudule and
-
3)
Kaleng division in Nadunga, Kangakipur, Kakelae and Loruth Esekon sub locations.
To compensate, additional random numbers were generated while keeping the stratum proportion rule. Animals were then selected if they fell in safe and accessible adakars. The final number of samples collected for each species was slightly higher (431 and 538 sheep and goat samples respectively).
Ethics statement
This field serological study was conducted in manner to ensure quality and integrity of the research. The ethical approval as well as consent of this study was sought from Directorate of Veterinary Services who granted the approval and permission for collection of field laboratory samples on Peste des petits ruminants vide letter referenced “Ref.Meat/Vol.XIV/42 dated 1st July 2011. The Directorate of Veterinary Service belongs to the State department of Livestock development in the Ministry of Agriculture, Livestock and fisheries development of the Government of Kenya. Consent was also sought from Turkana herders for voluntary presentation of their small stock for collection of blood samples which they granted and facilitated the exercise.
Serum collection and storage
During serum collection activity, the pastoral herders were asked to recall and provide information on vaccination status of each of the animals selected for sampling. Blood was collected by jugular-vein puncture using venoject needles and vacutainer tubes (Venoject, UK). The blood was transported to the field laboratories where it was left to clot overnight. The serum was decanted into sterile tubes and centrifuged to remove the remaining red blood cells before being transferred to 2-ml cryovials and stored at -20°C.
Competitive Enzyme Linked Immuno-sorbent Assay (c-ELISA) for antibody detection
The peste des petits ruminants c-ELISA test kit ID Screen® PPRC, product code PPRC 1209, Lot 320 from IDVET innovative Diagnostic, Montpellier, France with an expiry date of July 2013 and assay protocol was supplied by the manufacturer. The test kit was used as per manufacturers recommended protocol to determine the presence of antibodies against PPRV in the samples of sheep and goats sera following the protocol supplied [24].
Statistical analysis
Ascent® Software version 2.6 (Thermo Electron Corporation, Theorem Electron Oy, Vantaa, Finland), a Windows-based Software designed to power all Thermo’s Ascent® microplate research instruments, was used to control the Thermo Scientific Multiskan® EX microplate reader used for the c-ELISA. The software’s spreadsheet function was used to generate results data that were subsequently exported to Microsoft Excel®, (Microsoft Inc. USA) and frequency plots generated. SPSS statistical software version 17.0 (IBM Corp., Armonk, NY) was used to generate descriptive statistics based on variables investigated.
For each species, the prevalence was estimated as: p = y/n, where y denoted the total number of animals positive for PPRV antibodies out of the sample size, n. This formula was used to compute not only the overall sero-prevalence for a species but also divisional-specific sero-prevalence by replacing the numerator and denominator to the relevant number of animals in the respective administrative unit. Differences in the sero-prevalences were tested using the chi-square test.
Univariable models were first run to assess the relationship between PPRV antibody sero-prevalence and individual risk factors for PPRV sero-positivity. The risk factors assessed included sex, age group, vaccination status and administrative division. The significance level was set at P ≤ 0.1. A multivariable logistic regression model was subsequently built using significant variables in the univariable analysis by extending the univariable model to include other risk factors. In the latter analysis, all the significant risk factors were initially offered to the model. Model building used backwards elimination method to decide on the factors to exclude from the model using the likelihood ratio test (P < 0.05). The strength of association between the risk factor and PPRV sero-positivity was estimated using the odds ratios (OR) which were directly derived from the coefficient estimates from the logistic regression models. The odds ratio is a relative measure of risk that describes how much more likely it is that an animal which is exposed to the risk factor under analysis will develop the outcome as compared to an animal which is not exposed. If the odds ratio is 1, the risk factor is unlikely to be associated with the risk of PPRV sero-positivity. For an odds ratio greater or less than 1, the likelihood that the risk factor is associated with risk of sero-positivity increases, and the stronger the association. A plausible interaction – between sex and age - was tested for both species.
The relationship between PPRV infection sero-status and the significant risk variables was finally evaluated by fitting mixed-effect models with the sub-location as a random effect. The latter step was carried out to provide, as much as possible, statistically unbiased estimates of sero-prevalence with associated uncertainty adjusted for clustering of PPRV sero-positivity responses within sub-locations. The intra-cluster correlation coefficient (ρ) is a measure of correlation of observations in a cluster e.g., herds, villages, agro-ecological zones or administrative units. In this study, for each species, ρ for each division were computed indirectly through accounting for heterogeneity of data in sub-locations via the random effect variance. In this instance, the error variance was fixed at π2/3 to substitute for the level 1 (animal-level) variance (ε
i
) [25]. Thus, for each species and for each division, ρ was calculated as:
$$ {\sigma^2}_{\mathrm{sub}-\mathrm{location}}/\left({\sigma^2}_{\mathrm{sub}-\mathrm{location}}+{\pi}^2/3\right) $$
where σ2
sub-location is the variance due to sub-location-specific random effects whereas the sum of σ2
sub-location and π2/3 is the total variance in the data for each division. Assuming the data is organized as a 2-level hierarchy, the intra-divisional correlation coefficient is the proportion of division-level variance out of the total variance for that division [25]. Coefficients close to zero indicate that responses (in our case PPRV sero-positivity) within clusters are no more similar to each other than responses from different clusters (implying that the response is randomly distributed among clusters) and vice versa. To evaluate whether ρ was associated with the magnitude of the serological response of the animals, non-parametric correlations (Spearman correlation coefficient) between ρ and the sero-prevalence was computed.
The sero-prevalence maps were produced using ArcGIS version 9.1 (ESRI, Redlands, California).