These processes can modify the movement patterns or conspicuousness of individuals, changing their trappability, visibility, or propensity to migrate out of the study site. For example, Brazilian treefrogs Hypsiboas prasinus with more intense helminth infections exhibit reduced mating call frequencies, which could make them less detectable during surveys 23 ; Figure 1A. Similarly, house finches infected with Mycoplasma gallisepticum infection suffer impaired vision and display reduced activity levels, resulting in lower recapture rates for infected compared to uninfected individuals Figure 1B ; 7 , A Calling rate of treefrogs Hypsiboas prasinus versus total number of helminth parasites.
A further implication of differential detectability of infected vs uninfected individuals is that observed temporal patterns in disease dynamics may simply be spurious artefacts of temporal variation in the probability of detecting infected vs uninfected individuals 20 , 25 , Equally, real seasonality in disease dynamics see below and Table 2 ,F may be masked by contrasting temporal patterns of detection of hosts, vectors, or pathogens Pathogen infections in natural populations are often characterised by aggregated or over-dispersed distributions among individuals within populations 27 ; Table 2 ,A.
That is, most infected individuals harbour low parasite burdens, while relatively few harbour high parasite burdens. Although typical of macroparasite infections e. For example, Grogan et al 29 showed that the distribution of Batrachochytrium dendrobatidis Bd load between amphibian hosts is highly over-dispersed.
Despite its long-recognised importance for disease transmission rates and host-pathogen population dynamics 30 , the effect of pathogen aggregation and variation in disease intensity on the detection and estimation of disease occurrence and impacts has only recently been established. For example, Shin et al. Similarly, the probability of detecting Plasmodium infections in avian blood increases with pathogen load 12 ; Figure 2A.
The Certainty of Uncertainty
Thus, pathogen aggregation can generate bias in disease-relevant parameters via increasing the likelihood of errors in disease state assignment i. Pathogen aggregation can also generate imprecision and bias in estimates of the magnitude of disease impacts on individuals when the magnitude of disease-induced impacts varies with parasite burden or the intensity of infection 12 , 29 , 31 , For example, Grogan et al.
Just as there can be uncertainty in the assignment of disease-state for individuals in natural populations, there can also be uncertainty in assigning individuals to demographic or social classes i. For example, imperfect methods of ageing individuals will produce biased and imprecise estimates of age prevalence curves or of demographic impacts An inability to accurately sex individuals e.
For example, in badger Meles meles populations individuals that are more socially isolated from their group are at greater risk of tuberculosis Tb infection 35 , while in meerkat Suricata suricatta populations highly connected group members those that groom more and roving males are more likely to be infected with Tb 36 ; Figure 3A,B. The effect of social status on infection risk can also differ between demographic classes, as was recently shown for spotted hyenas Crocuta crocuta infected with canine distemper: Furthermore, if infection is demographically biased i.
Probability of individual meerkats testing positive for tuberculosis as a function of A the extent to which they groom others grooming outdegree and B the extent of intergroup excursions by males roving male outdegree. Summary of key sources and processes that can generate bias or lack of precision in estimates of disease-relevant parameters obtained in disease ecology studies. Erroneous or biased inferences on disease dynamics may arise when multiple, phenotypically indistinguishable but genetically distinct host, vector or parasite species are present but are cryptic, and therefore, overlooked 21 , 39 , Table 2 ,C.
Virulence can vary among parasite species Different parasite species or morphotypes can differ in the nature of their impacts on hosts 21 , or in their detectability within hosts For example, compared to uninfected individuals, blue tits Cyanistes caeruleus infected with Plasmodium circumflexum experience lower survival, while those infected with Plasmodium relictum experience lower reproductive success 21 ; Figure 4A. These contrasting impacts on blue tit fitness were obscured when the identity of the two cryptic malaria species was ignored Cryptic vector species can vary in their contribution to local infection dynamics, while the detectability of parasites in vectors can also vary among vector species For example, Gomez et al.
Crypticity among host species can also influence disease processes and may bias estimates of the distributions of hosts or disease, or provide previously unidentified explanations for observed disease distributions For example, spatial heterogeneity in Lassa fever outbreaks in humans was resolved only when the cryptic phylogeography of its reservoir host species, the rodent Mastomys natalensis , was recognised Another form of uncertainty may emerge from the incomplete characterisation of biological assemblages due to the non-detection of rare or low detectability species Table 2 ,C.
For example, when characterising the diversity of micro-organisms within a host e. This arises because more common or easily detectable species in an assemblage are catalogued early while rarer or less detectable species require greater effort 9 , For both hosts and pathogens, sampling campaigns will rarely census entire communities due to logistical constraints and as such there will be uncertainty when estimating host breadth, pathogen species richness or other diversity metrics.
For example, when characterising the viral diversity of the wild megabat Pteropus giganteus , Anthony et al. However, methods to account for imperfect detection suggested that a further 14 viruses remained undetected in this host, with the amount of testing required to detect them all estimated to be nearly seven-fold the number actually tested 9 ; Figure 5. This example illustrates how observations of host-pathogen systems can be directly biased by sampling completeness, which will always be constrained by logistical considerations.
Viral discovery curves for pathogens of the Indian Flying Fox Pteropus giganteus using PCR estimated from observed detections using three statistical models. Alongside unacknowledged taxonomic crypticity, and the presence of rare species, potential uncertainty in disease-relevant parameters can arise when multi-pathogen or multi-host dynamics are present but ignored Table 2 ,C. Coinfections can involve both antagonistic and synergistic interactions between pathogens within hosts, which can alter the outcome of infection positively or negatively and thus influence disease dynamics and host fitness 44 — For example, Budischak et al.
For example, Kilpatrick et al. Such processes broadly related to the epidemiological complexity of a disease can have follow-on effects that may also introduce other forms of bias and uncertainty; for example, mapping efforts for increasingly complex human diseases, as crudely measured by the presence and number of transmission sources e.
Although both multi-pathogen coinfections and multi-host pathogens are common, their effects and dynamics in wild host populations in natural settings remain poorly described. Uncertainty in estimates of disease-relevant parameters can arise when the temporal or spatial sampling scales do not match those at which the disease dynamics operate Table 2 ,D. For example, in many disease ecology studies the frequency of data collection occurs on a longer time scale than the disease dynamics 1 , 2 , If hosts are only monitored seasonally or annually but the progression from infection to death, or from infection to recovery, occurs over weeks or months, then individuals can acquire infection and die, or acquire and lose infections, without these events appearing in the data 16 , 17 , Inferences regarding disease dynamics may also vary as a function of the spatial extent of sampling relative to the area that determines pathogen dynamics Table 2 ,D.
For example, the relationship between biodiversity and infection risk often depends strongly on the spatial scale of sampling Similar conflicting inferences regarding the relationship between host biodiversity and the risk of infection with West Nile Virus have been demonstrated in studies conducted at small 48 and large spatial scales A mismatch of spatial sampling scale is most likely to occur for zoonotic or vector-borne pathogens or those with complex life cycles, because the production of infective-stages may be decoupled spatially from the dynamics of the infection within the target host species 48 , 54 — Disproportionate sampling effort in both space and time can be considered further examples of potential mismatches in scale that can introduce uncertainty in disease ecology studies Table 2 ,D.
Thus, the sample units are not selected according to defined rules from the pool of all possible sample units that conceptually represent the population of interest This can preclude calculating true probabilities of occurrence from a sample, resulting in biased and imprecise estimates of disease prevalence or other population parameters This approach is increasingly being applied in disease studies to map infection risk since robust records of disease absence are usually unavailable e.
In these cases, the distribution of reported occurrences are often tightly correlated with the distribution of reporting or observation effort, potentially resulting in misleading representations of disease distributions in model outputs 58 , It is well recognised that most pathogen detection methods are imperfect, resulting in errors in disease state assignment Table 2 ,E. This is true both for diagnoses made via laboratory analyses of field-collected tissue samples and those made via observational assessments of host symptoms in the field 11 , 12 , Less well recognised is how these properties interact with other potential sources of uncertainty.
For example, the accuracy of diagnostic tests can vary with the intensity of infection in hosts, meaning they can be inconsistent when pathogen distributions are aggregated [see section above and 12 , 15 , 29 ]. Quantitative PCR-based assays can fail to detect infections with low DNA copy number low parasite load , as has been demonstrated for the detection of Bd 15 and avian malaria 12 ; Figure 2A. Meanwhile, observational diagnoses may fail to detect asymptomatic individuals or those with minor symptoms 1 , 7.
Accounting for uncertainty due to diagnostic test accuracy will be especially necessary in studies of pathogens with over-dispersed distributions among hosts Issues of low sensitivity and specificity are particularly problematic in studies that use serological data to infer infection status because state assignment is based on arbitrary threshold values, which can increase the likelihood of false negatives and consequently bias estimates of disease-relevant parameters In addition, cross-reactivity in serology can occur in the presence of unidentified pathogen diversity, increasing the likelihood of false positives 60 , Indeed, the presence of cryptic pathogen species as discussed above could lower the specificity of many diagnostic tests There can be additional uncertainty in inferences obtained from studies that use serology-derived measures of disease when knowledge of the serological outcomes following infection is lacking e.
Another potentially common, but rarely considered, source of uncertainty in the diagnosis of infection is the variability in diagnostic accuracy that can exist between different laboratories, technicians or observers because of differences in expertise, equipment or procedures 63 ; Table 2 ,E. Such variation in diagnostic accuracy may lead to erroneous inference when comparing prevalence and disease dynamics across studies and regions 18 , Finally, the choice of which tissues to sample within hosts can also induce uncertainty in parameter estimates because infection or the composition of pathogen assemblages may vary among tissue types 64 ; Table 2 ,E.
For example, avian plasmodium is less detectable in blood than in other tissues [via PCR; 65 ]. External factors related to the environment often have pervasive effects, both proximal and distal, on the components of host-pathogen systems, and as such, environmental covariates can be considered a crosscutting source of potential uncertainty in disease ecology studies Table 2 ,F. For example, seasonal changes in the incidence of infectious diseases is common.
Seasonal forcing of disease dynamics occurs for a variety of ecological reasons, including seasonal pulses of births and deaths, seasonal changes in host immunity or parasite vigour, or because of seasonal changes in host behaviours e. For instance, seasonal transmission of the fungus Pseudogymnoascus destructans, the cause of bat white nose syndrome, is primarily driven by changes in host physiology related to hibernation, which facilitates fungal growth in North American caves during winter 70 ; Figure 7.
Less regular or longer-term environmental influences can similarly affect disease dynamics: For example, outbreaks of the coral disease, atramentous necrosis, are associated with increased rainfall, greater particulate runoff, and higher water temperatures, all of which are predicted to increase under future climate change Disease ecology surveys that overlook the effects of seasonality and other environmental factors on disease transmission may thus produce biased and imprecise estimates of disease-relevant parameters, poorly characterised disease dynamics, or fail to identify important mechanisms driving them 20 , 26 , At worst, such studies may entirely fail to detect pathogens; for example, by only sampling annually or in a temporally ad-hoc manner see also mismatch in scale section above; Spatial variation in environmental covariates can similarly introduce uncertainty in wildlife disease studies.
For example, host species may utilise different habitat types. This could lead to differences in host detectability during disease surveys 52 or could directly affect the host-pathogen relationship. Our review demonstrates that uncertainty in disease ecology studies arises because of the sampling and diagnostic procedures used and due to factors inherent to the biology and ecology of host-pathogen systems. Accordingly, the first step of any disease ecology study should be to identify potential sources of uncertainty and their likely magnitude Figure 8. This will involve consulting prior information previous studies, similar studies, historical literature and, if resources permit, conducting a pilot study or power analysis to help determine optimal or efficient sampling strategies.
Several aspects of a well-designed sampling strategy can a priori reduce the extent to which uncertainty plagues parameter estimates 72 , For example, minimising stochastic variation among samples e. Schematic representation of the process of accounting for sources of uncertainty in disease ecology studies, including examples of ideal and alternative methods that can be used to address each step of the process. Subsequently, researchers should employ a sampling strategy that enables the application of statistical tools that can help adjust parameter estimates to account for uncertainty arising via imperfect detection including incomplete sampling and state misclassification 11 , 15 , 76 ; Figure 8.
The statistical tools most commonly used fall under two broad frameworks. Occupancy models use repeated spatio-temporal sampling to estimate detection probabilities at multiple hierarchical levels and are a flexible means of obtaining estimates of disease parameters adjusted for multiple levels of uncertainty 11 , 12 , For example, DiRenzo et al.
Hidden Markov models also termed multi-event mark-recapture models , meanwhile, model both individual detection probabilities and uncertainty in state assignment 16 , 78 , and are a powerful means of linking disease dynamics estimates of transmission and recovery rates to impacts on hosts and populations [via estimates of vital rates; 21 , 79 ]. Moreover, multi-event models can also provide a robust framework for improving diagnostic accuracy when diagnosis is imperfect, by enabling a probability-based, rather than a binary, classification of infection status.
For example, Buzdugan et al. For studies that evaluate hosts or pathogen communities, existing tools stemming from biodiversity studies, such as non-parametric richness estimators, can be readily applied to estimate or account for the non-detection of species within species assemblages Several previous studies provide comprehensive guidance to the design and application of these models in disease ecology and other studies 11 , 15 , 52 , 75 , 80 — 82 , and we refer the interested reader to the more detailed discussions therein to develop further applications in disease systems.
Despite being widely advocated and employed in disease ecology studies, the application of analytical tools to obtain unbiased and more precise parameter estimates may not always be possible. Multi-event models require long-term, and often large, datasets on marked individuals 21 , 51 , 83 , while occupancy models demand repeated sampling at every level of inference: Numerous financial, logistic and even biological constraints can prevent such rigorous hierarchical sampling from being undertaken.
Paradoxically, these methods also require reasonable detection probabilities of hosts and pathogens to estimate parameters of interest 11 , Logistic or financial constraints that limit the scale or frequency of sampling or the type or quality of data collected, or the presence of rare species, will therefore preclude the use of statistical adjustment to account for uncertainty in many cases.
When such constraints preclude the use of such models to adjust parameter estimates for heterogeneous detectability or state misclassification, it may be possible to use prior information on detectability or state uncertainty from other studies or similar systems to adjust parameter estimates via Bayesian methods, or to use simulations and sensitivity analyses to assess the influence of a range of detectabilities or misclassification bias on parameter estimates and inferences 74 , 75 ; Figure 8.
Ultimately, statistical tools can only adjust for uncertainty due to measured and identified sources of observation error i. A post-hoc statistical adjustment of infection rates will remain biased when infection is caused by multiple unidentified pathogen species, if the time span between sampling periods is greater than the average infection time for hosts, or if important species interactions within a multi-host disease system are overlooked. Thus, to truly account for multiple sources of uncertainty in disease ecology studies researchers must i have an intimate knowledge of the host-pathogen dynamics, the aetiology of the disease and the ecology of the system, ii employ a rigorous, biologically sound, replicated survey design, iii statistically adjust parameter estimates for known sources of uncertainty or assess its influence on parameters otherwise , and iv acknowledge remaining sources of uncertainty Figure 8.
Where possible, the influence of remaining potential sources of uncertainty should be evaluated via simulation or sensitivity analyses e. Somewhat ironically, the four steps listed above might be exactly what a study is trying to ascertain in the first place, paving the way for studies and methods that scale with, or incrementally improve, knowledge of a system utilising an adaptive or iterative approach.
As such, depending on the state of knowledge of a particular system, each of the steps in this hierarchy of actions Figure 8 represent potential future directions or avenues of enquiry for the system at hand. Reliable, unbiased and precise estimates of disease-relevant parameters are critical for disease monitoring and risk analysis; for predicting disease spread and dynamics; for understanding the ecological and evolutionary implications of pathogens in host populations; and for ensuring the success of conservation interventions and management actions 11 , 15 , 20 , 85 ; Table 1.
Over the last decade, disease ecologists have begun to acknowledge the importance of accounting for uncertainty when making inferences on natural disease systems 9 , 11 , 12 , 14 , 15 , 20 , 26 , 52 , 53 , To date, however, uncertainty in disease ecology studies has been considered primarily in terms of imperfect detection of hosts or pathogens or disease-state misclassification.
In this review, we show that uncertainty in disease ecology studies extends beyond these components of observation error and can arise from multiple varied processes that pertain to aspects of the disease system, the study design, the methods used to study the system, and the state of knowledge of the system. Some of these processes, such as unidentified crypticity among vectors, hosts or pathogens, or a mismatch of sampling scales, may not be immediately apparent, and may not be adequately accounted for via statistical adjustments 11 , In this review, we have discussed the processes by which these varied sources of uncertainty can reduce the precision of, and introduce bias in, estimates of disease-relevant parameters.
Importantly, we show that uncertainties in parameter estimates generated via one process may propagate through to others because of interactions between the numerous biological, methodological and environmental factors at play. Understanding how these interactions among sources of uncertainty affect the degree and direction of bias in disease-relevant parameters is a key challenge for this field, and we present a hierarchy of needs that could be tailored to individual study contexts in order to reveal next steps and future directions towards improving estimates of disease-relevant parameters.
Given the diverse set of factors that can contribute to uncertainty in disease ecology studies Table 2 , the extent of ecological variation in host-pathogen systems e. Nevertheless, some general guidelines are possible.
The degree of uncertainty in disease ecology studies will be higher when the biology, ecology and dynamics of the system are complex or unresolved, when sampling effort is low, when sampling strategies are poorly designed e. Moreover, studies of endemic, invasive, or novel diseases may be at higher risk of uncertainty because the detectability of pathogens is often lower where disease prevalence and infection intensity are low or patchily distributed such as at invasion fronts. Uncertainty in disease ecology studies is a certainty. In this review, we have identified a myriad of ways in which uncertainty can manifest when attempting to monitor pathogens and characterise disease dynamics in natural populations and have discussed appropriate sampling and analytical methods to account for or minimise their influence on estimates of disease-relevant parameters and identify future research priorities.
We acknowledge that our list is not exhaustive, and that studies, particularly in novel systems, or that apply novel methodologies and technologies, will continue to encounter additional considerations. Nevertheless, this review should assist researchers and practitioners to navigate the pitfalls of uncertainty and strive towards more robust parameter estimates from which to make sound inferences and predictions in disease ecology.
SL and KM conceived of the design and ideas in this manuscript. SL performed the literature review. SL and KM wrote the manuscript. SL is funded by a Daphne Jackson Fellowship. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We are grateful to Andrew Tyre and sarah Benhaiem for their comments on an earlier version of this manuscript. The Supplementary Material for this article can be found online at: The impact of disease on the survival and population growth rate of the Tasmanian devil.
101 Ways to Get Inspired About Agile
J Anim Ecol 76 5: Impact and dynamics of disease in species threatened by the amphibian chytrid fungus, Batrachochytrium dendrobatidis. Health risks associated with wild animal translocation: Invasions 19 4: Using occupancy models to understand the distribution of an amphibian pathogen, Batrachochytrium dendrobatidis. Ecol Appl 20 1: Positive and negative effects of widespread badger culling on tuberculosis in cattle.
Nature Spread of white-nose syndrome on a network regulated by geography and climate.
- I capricci dello sceicco (Italian Edition);
- Embracing the certainty of uncertainty: implications for health care and research..
- Cuba, the Island I Treasure.
- Registered Dietitian Exam Practice Questions (First Set): Dietitian Practice Test & Exam Review for the Registered Dietitian Exam?
Nat Commun 3 1: Mycoplasma gallisepticum infection dynamics in a house finch population: J Anim Ecology 73 4: Wilderness in the city: A strategy to estimate unknown viral diversity in mammals. MBio 4 5: Seroprevalence of infectious diseases in saiga antelope Saiga tatarica tatarica in Kazakhstan Prev Vet Med Seeking a second opinion: Site-occupancy modelling as a novel framework for assessing test sensitivity and estimating wildlife disease prevalence from imperfect diagnostic tests. Methods in Ecology and Evolution 3 2: All That glisters is not gold: A primer on the application of Markov chains to the study of wildlife disease dynamics.
Methods in Ecology and Evolution 1 2: Estimating patterns and drivers of infection prevalence and intensity when detection is imperfect and sampling error occurs. Methods in Ecology and Evolution 3 5: Multistate capture-recapture analysis under imperfect state observation: We may chip away at it with science and other tools of rational investigation, but there is always another void of unknown in the aftermath. Meditation , contemplation, and reflection are ways to investigate our thoughts and feelings around uncertainty in non-verbal ways. I think poetry, music and art are ways to comfort us in the face of the distress evoked by this unknown by evoking shared experiences of love or feelings of connection that transcend space and time.
And religion may have evolved to help humans cope with this awareness by providing a constancy in a supreme being theistic religions or a unified whole from the complementary nature of certainty and uncertainty non-theistic religions. As we consider certainty and uncertainty in our individual lives perhaps we can see it from the perspective of our changing nature across millenia as well.
It seems clear there is a certainty of uncertainty that, in part, defines us. Mindfulness taught over mobile phones increases compassionate behavior. Back Find a Therapist. Fertilization Not Random After All? Are Psychopaths Unfairly Stigmatized? Inequality as a Lethal Disorder.
Login using
The certainty of uncertainty The nature-nurture debate on human behavior has changed. Comment Post Comment Your name. E-mail The content of this field is kept private and will not be shown publicly. Notify me when new comments are posted. Replies to my comment. Leave this field blank. Our digital devices might make us kinder Mindfulness taught over mobile phones increases compassionate behavior.
Getting Enlightened About Prostitution Let's think about what it means to buy sex. Why People Lose Interest in You. When Is It Emotional Abuse?