Keywords

1 Introduction

Epidemiology is the study of disease and other health-related events in populations and the factors that determine their occurrence (Thrusfield et al. 2017). Epidemiologic studies have long provided key evidence for understanding and developing management actions related to noninfectious and infectious diseases, including zoonoses, for both human and domestic animal populations. There is a clear need to understand interactions between human, wildlife, and domestic animal populations within a One Health context due to the increasing frequency of disease emergence over the past two decades (Cunningham et al. 2017; Stephen 2021). However, there are several challenges, associated with applying epidemiological techniques to wild animal populations, including issues associated with sample collection and diagnostic tests, and lack of knowledge about wildlife populations. Many of the assumptions and population characteristics needed by epidemiological methods are much more challenging to objectively measure and apply in wild populations compared to domestic animals or people. Although such studies are challenging and no single study can generate all the required information about a wildlife population, it is important to remember that it is the cumulative effort that counts, with progress being made in incremental successive studies over time (Stallknecht 2007). Wildlife epidemiological studies and methods must embrace multiple lines of evidence to triangulate toward a better understanding and be interdisciplinary in all its forms, including indigenous ways of knowing. In this chapter our objectives are to: (1) describe some of the major issues and problems experienced by researchers and health professionals when attempting to apply epidemiological principles and methods to free-ranging wildlife populations, and (2) help researchers and practitioners to understand how these methods and tools can be applied effectively in free-ranging populations.

2 Goals and Objectives for Epidemiological Studies in Wildlife Populations

Epidemiological approaches can be used to describe, understand, and ultimately inform the health management of free-ranging wildlife populations. Descriptive approaches are often used initially to determine the “what” and the “who”: which etiologic agents are involved in causing morbidity and mortality and which demographic groups are primarily affected? Excellent references exist for understanding patterns of disease occurrence and how to collect such data effectively (Delahay et al. 2009; Thrusfield et al. 2017; Wobeser 2006). Once patterns of disease and health in a population are understood, one is often interested in then generating or testing hypotheses about the ‘why’ and ‘how’: why are certain parameters more, or less, involved in causing disease or health outcomes, and how can these factors be managed? This often involves more analytical epidemiological approaches that use observational, retrospective, or prospective study designs (cross-sectional, cohort, and case-control designs).

To manage and understand health implications and move from knowledge to action (Stephen 2021), researchers often want to implement control measures in an adaptive management framework. These approaches often involve before-after-control-impact (BACI) designs (Conner et al. 2007; Shaffer and Buhl 2016; Rytwinski et al. 2015), or experimental designs where different treatments are implemented in similar populations (Cassidy 2015; Delahay et al. 2009). Regardless of the approach taken, it is important to understand the value of evidence triangulation and the importance of putting knowledge gained into action (Stephen 2021). Evidence from multiple different pathways of study is often extremely valuable for understanding complex ecological phenomena. For example, using two-eyed seeing approaches (Kutz and Tomaselli 2019), quantitative and qualitative approaches can provide complementary and corroborating information that allow us to more fully understand both human values and wildlife health determinants in a holistic framework. A two-eyed seeing approach is a collaborative, iterative, adaptive process that bridges multiple knowledge systems including western science and indigenous ways of knowing to co-generate knowledge within a rigorous, transparent, and appropriate process of knowledge acquisition and verification (Kutz and Tomaselli 2019).

3 Issues with Epidemiologic Studies in Wildlife Populations

When attempting to understand determinants of health in free-ranging fish and wildlife populations, there are many barriers that create challenges with the interpretation of surveillance outputs and epidemiologic data. These include lack of validated diagnostic tests, issues associated with sample collection, and a host of population-level issues related to determining causation. Despite these challenges, it is still feasible and quite reasonable to apply epidemiologic methods to the study of wildlife health. One just needs to understand the limitations beforehand and know how these limitations can be addressed using proper study design. Some of the major challenges and limitations when studying free-ranging wildlife populations are described below (Table 1), followed by epidemiological methods and approaches that work best for studying free-ranging wild populations.

Table 1 Common issues arising from sampling and collecting from free-ranging wildlife populations
Full size table

Collecting biological samples for wildlife health studies is challenging from a cost and welfare perspective. Live-capture studies depend on being able to capture and collect samples that do not compromise the health of the captured animal (Stallknecht 2007). These studies can be very expensive and time-consuming, so many studies rely on sampling free-ranging populations using convenience sampling frames; those samples that are easily acquired through existing means such as hunting, fishing, or citizen science efforts. This means these samples are often biased and not representative of the populations being sampled (Wobeser 2006). Sample sizes for studies involving wild populations are often small, due to the high cost of capturing and sampling remote populations, or when working with small, endangered populations with few individuals. Even with large populations of relatively common species, detecting pathogens at low prevalence requires enormous effort over long periods of time during which, many of these population characteristics change rapidly.

Random sampling is a core assumption for many epidemiological study designs but it is often very difficult or impossible to achieve in free-ranging populations. To achieve a random sample, one must know a population’s underlying spatial and temporal variability to design sampling strategies that ensure all animals in the population have an equal (non-zero) probability of being sampled. Despite evidence that a species’ social structure can have enormous implications on disease transmission and reservoir status, the social structures of many wildlife populations are unknown, and understanding their social network is a very expensive undertaking, so it is often restricted to small, localized populations. Population characteristics such as age structure, sex ratio, recruitment, and home range and connectivity are often undetermined in wild populations, and one often has to infer these characteristics from convenience samples.

Issues with diagnostic tests also impose challenges. Validated diagnostic tests are often not available for wild species and tests developed for domestic species must often be deployed (Thomas et al. 2021). Unfortunately, this means that data on sensitivity and specificity of these tests often do not exist, so test performance (their positive or negative predictive values) is often unknown before taking it into the field. Latent class analyses and other techniques have been used to estimate test performance when these parameters are unknown when sampling from multiple populations (Richomme et al. 2019; Shury et al. 2015). Many tests developed for domestic species can cross react with similar pathogens found in wild populations, but determining which pathogen a positive test indicates requires additional cost and laboratory expertise to tease apart confounding results. One example involves Borrelia turicatae, which causes tick-borne relapsing fever and causes cross-reactions on a test meant to determine exposure to Borrelia burgdorferi, the causative agent of Lyme disease (Gettings et al. 2019).

Many countries lack the veterinary infrastructure to allow diagnostic testing in wild populations. Many of these exist in biodiversity hotspots where zoonotic spillover of pathogens to human populations may be higher. International efforts which help to establish veterinary infrastructure such as the PREDICT program (https://p2.predict.global/) have made major improvements in diagnostic capacities in recent years.

Understanding the health of populations can also be challenged by a lack of established reference ranges for parameters commonly used to assess health in domestic animals and humans, such as biochemical tests and hematology. Zoological institutions have been great sources of information in this regard in the past 30 years and there are many more established reference ranges available now. Many studies often build in a validation sample set from healthy animals to determine proper reference ranges as part of the study design. There can, however, be substantial differences between health parameters for species kept in captivity versus their free-ranging counterparts (see McAdie 2018).

4 Study Designs and Approaches for Free-Ranging Wildlife Populations

There are two broad approaches that can be applied to the study of wildlife health and disease: experimental and observational studies. Experimental studies are used relatively infrequently in free-ranging wildlife because they require the investigator to control and manipulate the exposure variable of interest (discussed below). Most wildlife health studies rely on observational studies, including descriptive and analytic studies. For a more in-depth review of current standards for reporting observational research refer to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines (http://www.strobe-statement.org) (Sargeant et al. 2016).

4.1 Descriptive Studies

Descriptive studies include case studies and case series. If one is describing a new case of a disease in a new species or location and reports this finding, this is typically described in a case report. If multiple animals having similar criteria and causes are found, this is a case series. If one goes out to look for a particular disease or presence of an etiological agent or risk factor in a population and just intends to summarize the data, these are typically also included as descriptive studies and are very common in the wildlife field (Kaur et al. 2008; Parra et al. 2006; Salb et al. 2014). This type of information is very valuable and can help to guide the efficient management of outbreaks. For example, Salb et al. (2014) summarized anthrax outbreak data from wild wood bison in northern Canada, utilizing outbreak information collected over a 46-year period, demonstrating that outbreaks had declined over time, that outbreaks peaked in early July, and primarily involved bulls. These types of studies allow researchers to identify hypotheses that can be investigated using analytic or experimental studies.

4.2 Analytic Studies

Analytic studies formally compare results between groups that differ with respect to their exposures or outcomes. Analytic or explanatory studies are used to better understand how different variables such as age, sex, or geographic area affect and interact with other variables such as disease status. They typically seek to understand causation or how an outcome variable (disease status) is related to other variables of interest.

4.2.1 Cross-Sectional Studies

Cross-sectional studies look across populations at one point in time to record information about their subjects without manipulating the study environment. They can be classified as descriptive or analytical, depending on whether the outcome variable is only being documented and measured or assessed for potential associations with exposures or risk factors. They may be concerned with single or multiple variables of interest. Cross-sectional studies are one of the most common epidemiological study designs used in wildlife as no prior knowledge (e.g., health/disease status, age, and sex) is needed about subject animals before capture and sampling. Repeated cross-sectional studies are studies repeated over time using different individuals in the population. Longitudinal studies are cross-sectional studies that involve the same individuals repeatedly captured and sampled over time. These types of studies are relatively easier to implement relative to other types of epidemiological study designs, such as cohort studies, but researchers have less control over how animals are categorized or grouped and it is not possible to show proof of causal association with these studies, as one is measuring both the outcome and a set of explanatory variables at the same time. As a result of exposure and outcome being measured at the same time, causality cannot be reliably inferred from cross-sectional studies (see Chap. 6), unlike with cohort and case-control designs. The outcome being measured in cross-sectional studies is typically prevalence of a disease or health parameter being studied as one often does not know the actual population structure or exactly how individuals comprise a population of wild animals (i.e., the denominator), making calculation of rates (incidence, rate ratio) difficult or impossible.

4.2.2 Case-Control Study Designs

Case-control studies are a very powerful method of studying health determinants in human and domestic animal populations but are rarely used in wildlife populations as some population characteristics need to be known beforehand. Typically, with these studies, a subset of the population with a particular outcome (the cases) is compared to another subset of the population that has not experienced the outcome (controls). No intervention is attempted and no attempt is made to alter the course of the disease. Most often, case-control studies are retrospective as they look back in time to compare how frequently the exposure to a risk factor is present in each group to determine the relationship between the risk factor and the disease. Prospective case-control studies are less common and rarely possible for free-ranging wildlife as they involve following the sample group over time while monitoring their health and exposures. Cases emerge when animals develop the disease or condition under investigation as the study progresses. Challenges in identifying cases, finding retrospective exposure information, or prospectively following animals over the time limit the application of these methods.

Examples, where case-control studies have been successful, include closely monitored populations such as lowland gorillas (Haggblade et al. 2019), African buffalo (le Roex et al. 2013), and sea otters (Shockling Dent et al. 2019). This was possible in the case of lowland gorillas because there was a preexisting dataset of 132 “clinical interventions” available over a 20-year period, several, but not all of which involved snaring and subsequent treatment. This type of design was only possible in this case because of very close monitoring of this population with controls being clinical interventions other than snaring. In the paper published by Shockling Dent et al. (2019), a similar retrospective case-control design worked effectively as there was a large necropsy database from a closely monitored population where findings from nasal mite-infested sea otters (cases) could be compared with a large number of controls who did not have the nasal mites at necropsy, allowing researchers to determine that older sea otters were 9.4 times more likely to be infested with mites than younger otters.

4.3 Cohort Study Design

A cohort study selects animals based on exposure and then studies the development of disease in the exposed and unexposed groups of animals. Cohort studies can be logistically difficult because of the need to identify animals initially free of the outcome and then follow them over time to determine the development of the outcome (Caswell et al. 2018). This can be particularly challenging for wildlife studies. Nonetheless, cohort studies have been used to study wildlife disease. For example, Miller et al. (2008) used a cohort study design to compare the annual survival of prion infected and apparently uninfected adult mule deer. They found that prion infection dramatically lowered the survival of free-ranging adult mule deer (Odocoileus hemionus). Cohort studies are powerful because exposure is identified before the outcome, which confirms that the proposed cause preceded the development of the outcome.

4.4 Experimental Designs

Experimental studies, where the investigator manipulates the exposure variable of interest, are of primary importance in understanding pathogenesis, validating diagnostics, and providing a necessary perspective for interpreting data (Stallknecht 2007). Experimental studies are often conducted in laboratory settings which have limited applicability in real-world situations, but they can also be carried out in wild settings. There are numerous examples that experimentally manipulate various factors using rodents (Behnke et al. 2001; Dantzer et al. 2020; Sweeny et al. 2020), but fewer with birds, large mammals, and other taxa. For example, several recent studies have explored the relationship between gastrointestinal nematodes and bovine tuberculosis in wild African buffalo (Beechler et al. 2017; Ezenwa et al. 2010; Jolles and Ezenwa 2015; Seguel et al. 2019), but these types of studies in large mammals are relatively rare.

Despite challenges, experimental approaches have also been successfully applied in the field. An increasing number of studies have used antiparasitic treatment experiments of wildlife hosts to assess the impacts of parasites on health and fitness (reviewed by Pedersen and Fenton 2015). For example, Newey and Thirgood (2004) experimentally reduced parasite burdens in mountain hares (Lepus timidus) to test the hypothesis that parasites reduce hare fecundity. They found that treatment with ivermectin significantly reduced the abundance of Trichostrongylus retortaeformis and increased the fecundity of the hares. Experimental studies allow us to assess cause and effect in a way that is not possible using observational studies alone (Pedersen and Fenton 2015).

The combination of both lab and field experimental studies can lead to the discovery of additional relationships that are relevant to epidemiology (Stallknecht 2007). A great example of this can be found in Ezenwa et al. (2010) where experimental data was combined with longitudinal field studies to make inferences about co-infections and population health for African Buffalo.

4.5 Other Approaches

Although it is beyond the scope of this chapter to delve into the world of mathematical modeling, we wanted to highlight the role those mathematical models can play in identifying knowledge gaps, assessing possible management strategies, and understanding the spatial and temporal factors of disease emergence (Alexander et al. 2012; Ryser-Degiorgis 2013). Despite limitations, including limited data for wildlife populations, a broad range of modeling approaches has been applied to support decision-making in wildlife management problems (McCallum 2016). For example, modeling different vaccination strategies for preventing brucellosis in bison in the Greater Yellowstone Ecosystem (GYE) suggested that vaccinating all female bison captured during boundary operations for bison leaving the park combined with remote darting of female bison in the park would be the most effective alternative (Treanor et al. 2010). Moreover, although it was clear from the modeling that brucellosis could not be eradicated in GYE bison populations using vaccination, it could be useful to reduce prevalence over a 30+ year time period.

Molecular epidemiology has been a rapidly expanding field in the last two decades due to improvements in computer analysis power and rapidly decreasing costs of whole-genome sequencing methodologies. As a result, there have been many advances in understanding the global distribution and epidemiology of infectious pathogens in recent years involving wildlife populations. A thorough discussion of this field is beyond this chapter, but interested readers are referred to the following references as examples of this for more information (Carlson et al. 2019; Thompson and Ash 2016; Wong et al. 2019).

4.6 Approaches to Spatial Data

Spatial health analysis focuses on mapping diseases, risk factors, and other health outcomes and analyzing them in comparison to two or more variables. Due to the importance of how wildlife assemble and move in shared places, the interactions of people, domestic animals, and wildlife in spaces and the effects of landscape features on health outcomes, spatial analysis is growing in importance. It is beyond the scope of this chapter to delve into this topic in detail, but more is available in Chap. 14, as well as the following references (White et al. 2018; Pfeiffer and Hugh-Jones 2002; Cunningham et al. 2021; Moustakas 2017; Baratchi et al. 2013).

5 Validity, Bias, and Confounding

To make inferences about a study population, one generally must take a sample of the population of interest, measure a set of variables, and then analyze those variables to infer what the health of the population may be or understand how the health parameters influence an outcome of interest (typically disease prevalence or rate). The ways in which we undertake these steps influence how far our studies systematically (as opposed to randomly) deviate our measurements or observations from the truth. In free-ranging wildlife populations, many biases result from our data collection methods because logistic and financial considerations drive us to compromise our studies away from epidemiological study ideals and assumptions (Table 2). Reducing the amount of, and understanding the nature and direction of biases, during and after data collection, allow us to measure these inferences and understand the validity of our data more precisely.

Table 2 Examples of types of biases common to studies of health and disease in free-ranging wildlife populations
Full size table

Bias affects the validity of a study. Validity generally refers to how well our sample population (generally referred to as the target population) reflects the true nature of the overall (source) population about which we are making inferences. Internal validity refers to how unbiased our inferences about the association between an exposure (e.g., geographic location or home range) and a health outcome (e.g., disease status or serological exposure) truly are for the study population. External validity refers to how generalizable our findings about these associations are to other populations, situations, or species. Poorly designed studies often lead to erroneous associations, resulting in indefensible conclusions and poorly targeted management interventions. Having valid measures are very important to ensure management approaches are targeted appropriately and can be used to ensure the effectiveness of management interventions over time.

Selection bias is likely the most common and important source of bias in most wildlife studies and results from the target population not accurately reflecting the attributes of the source population. Many datasets acquired from wildlife are convenience samples (acquired when animals are reported by the public or staff) or are acquired through hunting, fishing, or other consumptive means that do not offer all animals in the population an equal likelihood of being sampled.

Confounding is a source of bias in wildlife studies which must be recognized and controlled for either during sampling or post hoc during data analysis. It results from situations where one or more variables are associated with both the exposure of interest and the outcome variable. When these confounding variables are not measured, we can get a biased relationship between other variables and the outcome. For example, suppose you wanted to understand the relationship between tuberculosis status and survival. You logically would want to know about the outcome variable of survival, the animals’ tuberculosis status, and a determined number of other health-related variables that could influence survival. Increasing age is often associated with increasing exposure to infectious diseases, so it is likely associated with TB status. But age is also likely associated with the probability of survival as the older an animal gets the less likely it is to survive the coming year. Age would, therefore, act as a confounder and muddle the relationships between survival and TB status in this example. Fortunately, there are many sampling methodologies that can account for confounding during sampling, as well as many analytical methods to control these relationships.

6 Overcoming Biases and Other Challenges

Despite the problems identified above, there is still value in using epidemiological approaches to understand health in wildlife populations. There are many strategies that can be employed, but sources of variability or error must first be identified so they can be dealt with. Previously published information on the species or population of interest from other geographic areas can often give a sense of these sources. Pilot studies using potential diagnostic tests on smaller subsets of individuals can help to refine parameter estimates a priori. Matching on known confounders such as age, sex, and species can also be used to control for confounding and should be considered during study design and analysis. Matching can be done through either frequency or individual matching (thorough explanations can be found in Dohoo et al. (2009)).

Once potential sources of bias or error have been identified, appropriate sampling strategies can be devised to reduce sources of error and bias. Identification of important covariates is an important first step to ensure parameters that can be measured are measured during capture or field sampling. The preparation of causal diagrams is an excellent way to visualize and identify potential covariates (Greenland et al. 1999; Dohoo et al. 2009). Sampling strategies to reduce bias involve hierarchical or probability sampling. For cohort studies, ensuring that exposed and unexposed groups are essentially as similar as possible in all important parameters of interest is important to be able to reduce bias to the extent possible.

Stallknecht (2007) also provides several excellent suggestions for both conducting and analyzing wildlife data that include: (1) developing integrated plans (use a variety of approaches involving both field and lab studies), (2) archive samples (important for future studies and molecular epidemiology approaches), (3) maintain quality control (integrate different diagnostic tools, understand data limitations), (4) interpret data carefully (need to question results because wildlife systems are complex), (5) don’t restrict yourself to traditional approaches (unique challenges require unique approaches; innovate), and (6) don’t be intimidated (individual studies contribute to our overall understanding). Similarly, Lachish and Murray (2018) suggest several considerations to reduce uncertainty and bias in disease ecology studies including: (1) rigorous identification of sources of uncertainty (use of pilot studies, a priori information), (2) employing rigorous sampling strategies (probability sampling, hierarchical levels, control for measured covariates), (3) statistical adjustment of parameter estimates for observation error where possible (mark recapture, occupancy models, simulations, sensitivity analyses), and (4) acknowledging remaining uncertainty (temper inferences and conduct sensitivity analyses).

6.1 Mixed Methods and Participatory Epidemiological Approaches in Wild Populations

Participatory epidemiology involves the participation of communities or human populations in studying the wildlife health parameters of interest. This is discussed in more detail in Chap. 5. Combining different epidemiological approaches, both quantitative and qualitative, to better understand complex wildlife population systems can be very powerful and lead to strong outcomes when communities are directly involved in understanding wildlife health. One example is the management and control of bovine tuberculosis in the area around Riding Mountain National Park in Manitoba, Canada. Understanding farmer and rancher attitudes toward elk and deer management (Brook 2015; Brook et al. 2013; Brook 2008, 2010; Brook and McLachlan 2009; Brook and McLachlan 2006) on their lands along with field epidemiological investigations (Nishi et al. 2006; Shury 2015; Shury et al. 2014; Shury and Bergeson 2011) allowed a holistic view of the bovine tuberculosis problem in the region and narrowed down potential solutions that would be ultimately successful in eradicating the disease amongst sympatric cattle and wildlife populations over time. Similar research in northern Canada involving local indigenous communities to understand caribou and muskoxen diseases have been very successful in building bridges between scientists and community members and providing key solutions to understand emerging pathogens and parasites and how these are associated with climate change (Forde et al. 2016; Hoberg et al. 2008; Keatts et al. 2021; Tomaselli et al. 2019). The common denominator amongst these examples is the importance of transdisciplinary research leading to concrete impacts on the health of both humans and wildlife.

7 Conclusion

Hopefully, we have managed to demonstrate how epidemiological approaches can provide necessary tools to describe, monitor, and understand the role of health outcomes in wild animal populations. Although there are numerous challenges that need to be overcome to understand and manage health in wildlife populations, one should not be intimidated to undertake such research, as it is critically important for society as we face major biodiversity and climate crises. Epidemiological approaches have and will continue to provide important and much-needed data for managing future pandemics and to prevent and manage risks to both human and wildlife populations in coming decades.