Study design and implementation
The design of this study differed from conventional epidemiological observational studies in that it did not aim to obtain a representative sample of the population. Rather, the focus was on active detection of field virus and disease, elucidation of contact patterns and increasing understanding of potential high-risk points and practices for transmission of infection within and between different avian populations, production systems as well as enterprise types (chicken production Sectors and ducks). Consequently, this research consisted of a combination of complementary research activities, rather than a single study. Such a mixed methods research approach, which integrates quantitative and qualitative research methods, may help to improve understanding of potential biases of the observed data (Kristensen et al., 2008), and allow for a deeper exploration of the topic under investigation than can be achieved by either approach alone (Limon et al., 2014). It also provides a framework for integrating data from disparate sources (Vink et al., 2016).
Figure 2 provides a schematic overview of the study activities. An initial expert consultation meeting bringing together international experts in different fields of relevance to H5N1 HPAI was convened in November 2015 to elicit hypotheses related to reservoirs of persistence of the virus, risk factors for transmission of infection, production systems and linkages between systems. Potential methods to test and investigate these hypotheses were subsequently discussed. A subset of eight hypotheses which could feasibly be investigated in Indonesia (Figure 3) were utilised to inform the design of this endemicity study.
A stratification of enterprise types was applied, grouping Sector 1-3 commercial poultry farms, collector yards and live bird markets on the one hand, and Sector 4 backyard and non-commercial producers and nomadic duck flocks on the other. The rationale for this was that the first group was less numerous, so that a census of all properties in Purbalingga District could be included, whereas a sample needed to be taken of the second group. The sampling designs for the subsequent active sampling differed between these two strata, too.
The first part of the study aimed to characterise poultry populations and trade and contact networks; it was conducted from April to July 2016. All Sector 1-3 commercial poultry farms (n=204), collector yards (n=129) and LBMs (n=18) were visited and questionnaires were conducted by 22 trained enumerators and 25 Government of Indonesia District Animal Health Officers of Purbalingga District. If information arose about locations that had not been included, these were also visited. In addition, visits to all 239 villages in Purbalingga District were performed and questionnaire surveys were conducted in a minimum of 54 households. As there was no list of households, these were randomly selected. A total of 14,368 households were interviewed. The questionnaires included signalment information, flock location, details of numbers and species of poultry kept, information on biosecurity practices and flock management (including vaccination), purchase and sales of live poultry over the previous 12 months, information on the incidence of disease. All data were entered into a Microsoft Access database.
The second part of the study aimed to detect AIV in the different enterprise types and populations by implementing a series of surveys in two sampling rounds (October 2016 – February 2017 and August – November 2017). For Sector 1-3 commercial farms, collector yards and LBMs, risk-based, stratified sampling was performed. Following geographical stratification into high, medium and low backyard poultry population density areas, further stratification into Sector 1-3 commercial farms, collector yards and live bird markets was done. Subsequently, random selection was carried out to identify properties, assuming a design prevalence of 2% with a confidence level of 95% that at least one HPAI H5 virus would be detected. All sampling was performed by 25 District Animal Health Officers.
Thirty-two commercial farms were selected. No broiler farms were included in the study, but all commercial duck farms (n=6) were included in this selection, with the remainder consisting of layer farms. Oropharyngeal swabs were taken from a fixed number of 30 birds per property, pooling six swabs into one viral transport medium (VTM) sample. The sampling was risk-based insofar as possible, that is, any birds that showed suspect clinical signs as well as dead birds were preferentially selected. Farms were sampled once during the first round (high season) and twice during the second round (low season).
A selection of 41 collector yards was made. Twelve environmental swab samples were pooled in two VTM samples. If any birds were observed with suspect clinical signs, rapid testing and individual sampling were performed. The collector yards were sampled three times during the first round and four times during the second round.
All 18 LBMs in the District were sampled. The same sampling protocol was based on a protocol developed for previous studies in markets in Indonesia (Indriani et al., 2010). LBMs were also sampled three times during the first round and four times during the second round.
An enhanced passive surveillance scheme was initiated by offering incentive payments to District Animal Health Officers for reporting sudden mortality events in poultry, with at least one positive AIV rapid test result of sick and dead birds. If these birds had died or been removed, environmental samples were taken.
In addition to the enhanced passive surveillance in Sector 4 poultry, it was decided to expand the sampling to include nomadic duck flocks. Twelve ducks per flock were sampled and the samples were pooled into two VTMs. In the first round, all 20 flocks that were included in the questionnaire survey were sampled. In the second round, 196 flocks were sampled and the number of sampled ducks per flock was increased to 30, pooled into 5 VTM samples.
As only four reports were received using the incentive scheme, it was decided to actively sample backyard poultry in the second sampling round. As the frequency of mortality events in backyard poultry during the low season was likely to be low, in addition to which such outbreaks were unlikely to be reported by owners of such birds, a purposive sampling technique was used. Snowball sampling (also known as chain-referral sampling) is a non-probability sampling technique where initial study subjects are asked to identify future subjects from among their acquaintances; the technique is utilised for the investigation of ‘hidden’ populations which are difficult for researchers to access, or for rare conditions or events which are unlikely to be detected through randomised sampling techniques (using community knowledge or ‘rumour’). Following a sample size calculation specifying a design prevalence of 5% and a confidence level of 95%, 187 villages in Purbalingga were visited. An initial household keeping backyard poultry was randomly selected, and a questionnaire interview was conducted. In the interview, the respondents were asked if they experienced any mortality in their poultry during the previous week and if not, if they could identify any acquaintances who may have had. If they could, this acquaintance was requested to be interviewed. If not, the respondents’ closest neighbour owning backyard poultry was asked to be interviewed. Using this methodology, 54 households keeping chickens were sampled in each village. In the interviews, numbers and type of poultry were also recorded. If sick or dead bird were identified, individual swab samples were taken; if the carcasses could not be identified, environmental swabs were collected. Following this methodology, 127 households keeping backyard chickens were sampled.