2.3 Data collection and quality
2.3.2 Passive sampling
The logistics and expense associated with capturing wild animals and testing them for the presence of disease makes it difficult to apply direct capture methods over large geograph- ical areas (McKenzie 2004). As a result, passive sampling and monitoring techniques tend to be widely used to estimate population size and disease prevalence in wildlife popula- tions.
Monitoring natural mortality and morbidity events
Morbidity and mortality events that have been spontaneously reported by members of the general public to health agencies provide an important source of information regarding the disease status of wildlife populations. Using this approach, surveillance for diseases of wildlife can be undertaken at relatively low cost (Stitt et al. 2007). For example, surveil- lance targeting dead wild birds, in particular American crows Corvus brachyrhynchos, plays a critical role in WNV surveillance in the United States (Eidson et al. 2001).
In an attempt to use point locations for analysing the spatial process of reported dead birds from New York City in 2000 and 2001 to provide an indication of WNV activity, Mostashari et al. (2003) used information recorded through an interactive voice-response telephone system and the Internet that recorded details on the date, location and species of dead birds found by members of the public. While this approach has many advantages (particularly in terms of cost), recall bias among those contributing to the program is likely
to be a problem. Also, because it is unlikely that members of the public carry GPS units, little information on the precise locations of where the carcasses have been retrieved can be recorded. Therefore, data are most often aggregated at the public-health district level, limiting the level of spatial detail able to be derived from these systems (see Eidson et al. 2001, Tinline et al. 2002, Guerra et al. 2003, Blanton et al. 2006, Beroll et al. 2007, David et al. 2007, Recuenco et al. 2007 for examples of spatial analyses using aggregated data).
Geographical variations in detection and reporting capabilities may occur due to pub- lic awareness differences (which, in turn, depends on the intensity and the duration of campaigns to inform the public about a disease of interest), human and animal population density, and the species of interest. These factors have the potential to introduce selec- tion biases in computed mortality rates arising from natural mortality and morbidity event monitoring. For example, Ward et al. (2006) evaluated changes in both the detection and reporting probabilities of dead crows in the WNV surveillance system implemented in urban and rural areas of Georgia, USA, during July and September 2003. These authors found that the proportion of crow decoy surrogates detected in urban areas (61%, SE 2.4%) was approximately twice that of rural areas (29%, SE 2.3%), and the proportion of decoys reported in urban areas (27%, SE 2.3%) was approximately three times that of rural areas (10%, SE 2.8%). Ward et al. (2006) concluded that human population density and associated factors had the potential to influence dead crow detection and reporting and, therefore, the reported spatial distribution of WNV. As another example, acknowl- edging imperfect detection probability, Hoff et al. (1973) suggested that the recovery rate of deer carcasses in North Dakota was not more than 10%. These authors then multiplied the number of deer found dead by a correction factor of ten to obtain an estimate of total deer mortality.
In addition to the absence of information regarding the size and temporal variations of a population at risk, carcasses are assimilated into the environment at a rate that depends on environmental conditions. Ward et al. (2006) recorded that most monitored carcasses (82%) disappeared or were decayed within a 6 day period, with an average carcass per- sistence of 1.6 days in rural areas and 2.1 days in urban areas. It would be reasonable to assume that weather conditions and the presence (and abundance) of scavenger species may influence the rate of carcasses disappearance, which in turn introduces spatial and
temporal variations in reported disease frequency.
Monitoring human-related mortality and morbidity events
Because of convenience, many epidemiological studies of wild species target commercial or recreational hunters as a source of data (e.g. Couvillon et al. 1980, Kane & Litvaitis 1992, Lugton et al. 1998, Fuchs & Deutz 2002, Koehler & Pierce 2005). Hunters can be questioned on the location of each single kill and these locations can be plotted on a map of the area of interest (Muller et al. 1998, Tackmann et al. 1998). Involving hunters is not an easy task as many are reluctant to participate in surveys. This is because: (1) many believe that the outcome of the study would increase regulations, affecting their recreational activities and/or (2) mistrust and suspicion of government agencies (Mason & Fleming 1999). In addition, the data collected from hunters may be biased. Recall bias may occur when hunters are asked to precisely recall the location at which an animal was shot or captured (McKenzie 2004). Selection bias may occur because trophy hunting is targeted towards adult male animals. Furthermore, the disease status of an individual animal may influence its probability of capture (Wobeser 2007, p: 18). Bellrose (1959), for example, found that ducks that had ingested lead shot were more likely to be killed by hunters, over-representing lead-exposed ducks in their study. This particular example is exceptional and, as a general rule, the presence of disease makes animals less likely to be captured (Wobeser 2007, p: 18). Furthermore, kill reports do not represent all mortalities, with non-reported hunter harvest, wounding loss, and depredation control hunts likely accounting for additional mortalities (Koehler & Pierce 2005).
It has been shown that a large proportion of hunters rely on open roads when travel- ling to their hunting areas, with usage increasing with the age of the hunter and varying with hunting method (Gratson & Whitman 2000). Because collecting clean samples in- volves complicated methodologies (Mason & Fleming 1999) requiring many pieces of equipment that may encumber the hunter during a stalk, the dependence of volunteer hunters with motorised access would probably increase. Consequently, hunting efforts are likely to not be spatially homogeneous (Fischer & Keith 1974, Fuller 1990, Unsworth et al. 1993, Rempel et al. 1997), showing the greatest densities along motorised access routes (Gratson & Whitman 2000).
Abernethy et al. 2003, Baker et al. 2004, The Independent Scientific Group on Cattle TB 2006). Abernethy et al. (2003) collected 543 badgers in a survey of badger traffic fatalities in Northern Ireland in an effort to assess the distribution of TB in badgers and the associa- tion between infection in carcasses and bovine herd TB history. Of the 543 badgers, 51% (n = 277) were suitable for laboratory examination. The prevalence among necropsied traffic fatalities was 18% with no evidence of spatial aggregation of TB-positive badgers. There was no correlation between herd prevalence and badger prevalence when district council regions were aggregated at the veterinary division level. However, a significantly higher proportion of herds within 3 kilometres of a TB-positive carcass was likely to have experienced a breakdown in the previous four years than herds around a TB-negative carcass. Abernethy et al. (2003) concluded that this methodology provides a relatively inexpensive means of monitoring spatial and temporal patterns in both badgers numbers and TB infection rates.
In addition to its linear characteristics (illustrated in Hubbard et al. 2000), traffic fa- tality data may be influenced by a number of other factors such as traffic volume, traf- fic speed (Joyce & Mahoney 2001, Dique et al. 2003), road characteristics (Hubbard et al. 2000, Brockie 2007), season, and weather (Case 1978, Coulson 1989). Further- more, road fatality data tend to show a skewed sex ratio and age structure, depending on the species of interest. Coulson (1997) noted a significant bias towards males in five species of macropods (i.e. kangaroos and wallabies) in southern Australia. On the other hand, males and females were equally represented in a study of road-killed armadillo
Dasypus novemcinctusin the southern United States (Loughry & McDonough 1996). In- stead, road-killed armadillos were predominantly older, with almost no juveniles. It would be also reasonable to assume that physical debility of diseased animals may increase the risk of collision with a vehicle, if road-crossing occurs.
In summary, to design effective control strategies, disease control authorities need maps that provide estimates of the spatial distribution of the disease of interest corrected for the spatial distribution of the wildlife species at risk. To achieve this objective, two pieces of information are required: a numerator, informed on the spatial distribution of disease- positive animals, and a denominator, informed on the spatial distribution of the popula- tion at risk. To have a valid numerator and denominator, it is not necessary to collect and list all individuals of the population at risk that are diseased and those that are non-
diseased. Instead, sampled individuals needs to be reliably discriminated into diseased and non-diseased, and representative of the spatial distribution of their corresponding (i.e. diseased or non-diseased) populations as a whole. Although sampling efforts aim to be homogeneous and representative of an entire study area, samples of diseased and non- diseased animals are frequently only available from portions of the study area (Muller et al. 1998, Tackmann et al. 1998). To overcome this limitation, one may: (1) compile in- formation provided by different authorities despite the potential variations in surveillance effort to detect and/or report the disease in question (see Staubach et al. 2002 for an example), and/or (2) combine data sets that have been collected using different sampling techniques (see Lamarque et al. 2000 for an example). However, compiling information from diverse sources increases the likelihood of errors that need to be taken into account when analysing the data.