Reappraisal of the Dilution and Amplification Effect Framework: A Case Study in Lyme Disease

Abstract

The role of biodiversity in regulating zoonotic disease in ecological communities has been broadly referred to as the biodiversity-disease relationship in disease ecology. Whether biodiversity decreases or increases disease risk, known as a dilution or amplification effect respectively, remains unclear. The literature has focused on the strength, generality, nature, and context dependencies that could explain contradictory evidence. We suggest that a continued focus on this approach to resolving the biodiversity-disease debate detracts from a more foundational problem with testing these dilution and amplification hypotheses, in that these hypotheses are not falsifiable as proposed. When tested and interpreted as net effects in a system, these hypotheses do not possess a true null outcome; they are vulnerable to ad hoc explanations. Specifically, that an empirical null outcome can be explained by multiple processes (i.e., a true null vs. a canceling out of amplification and dilution effects) means that process cannot be inferred from pattern. To remedy this problem, we propose that biodiversity and disease risk can be modeled as latent variables in multivariate causal models to reframe how we understand them and test the relationship between them. We present a case study on Lyme disease (LD) through a systematic review, concluding that testing these net effect hypotheses falls short of providing robust evidence for its underlying mechanisms. While these hypotheses have previously been helpful in conceptualizing this idea of biodiversity as a potentially protective factor for human health, they require further specificity moving forward in order to appropriately test the relationship.

Keywords: Lyme disease; amplification effect; biodiversity–disease; dilution effect; disease ecology; hypothesis testing; latent variables; philosophy of science.

FIGURE 1

A change in biodiversity, represented specifically by host diversity, can affect the strength of component effects that can increase (amplify) or decrease (dilute) disease transmission. These component effects represent mechanisms operating at a community level and have variable influence on disease transmission. The relative strength of specific mechanisms depends on whether frequency‐dependent or density‐dependent transmission characterizes the system (Rudolf and Antonovics 2005). In frequency‐dependent transmission, host competence and vector characteristics are more important for promoting transmission, independent of their overall population abundances (i.e., population density). On the other hand, density‐dependent systems are more impacted by the population abundances of hosts and vectors to promote transmission through increased contact rates (Dobson ; Ostfeld and Keesing 2012). The outcome of all component effects on transmission occurring simultaneously in a system will produce a net increase or decrease in transmission to affect disease prevalence (i.e., disease proportion).

FIGURE 2

Path diagram depicting a hypothetical vector‐borne disease system where the effect of biodiversity (total species richness) on disease risk (density of infected vectors) is mediated by host competence (competent or non‐competent) and total vector density (gray boxes). Total vector density is also influenced by the competence of the host, such that competent hosts are also poor reproductive hosts for the vectors and vice versa. Each path represents a proposed causal path between variables with the relationship β. Signs on the direct paths (solid arrows) between variables show the directionality of the relationship between them (positive, negative, or either). The net indirect effect between total species richness and density of infected vectors (dashed arrow) is the product of the coefficients along all component paths (Wood and Lafferty 2013).

FIGURE 3

Observed variable model of a simplified diversity‐disease relationship that includes only measured variables. Observed variables (species richness, species evenness, disease prevalence, and disease abundance) are represented as rectangles. Single‐headed arrows represent proposed causal paths between variables, with β representing the standardized or unstandardized bivariate regression coefficients.

FIGURE 4

Latent variable model of the diversity‐disease relationship with two components: (1) a structural model describing the relationship between latent variables (dashed black outline) and (2) a measurement model for individual latent variables (solid black outline). Unobserved latent variables (biodiversity and disease risk) are represented as ovals and observed measured variables (species richness, species evenness, disease prevalence, and disease abundance) are represented as rectangles. Single‐headed arrows represent proposed causal paths between variables, with β representing the standardized or unstandardized bivariate regression coefficients. Ɛ represents error, including measurement error and error that is introduced via stochasticity in the system (i.e., irreducible error).

FIGURE 5

Generalized latent variable model of the proposed path diagram in Figure 2 (relationships not specified). Unobserved latent variables are represented as ovals and observed measured variables are represented as rectangles. Mediator variables in the relationship between biodiversity and disease risk are represented as white rectangles. Proxy metrics representing the latent constructs are represented as blue rectangles. Direct paths are represented by solid arrows and the net indirect effect is represented by a dashed arrow. Each path represents a proposed causal path between variables with the relationship β. The net indirect effect (dashed arrow) is the product of the coefficients along all component paths. A positive β on the net indirect effect would indicate a net amplification effect, whereas a negative β would indicate a net dilution effect. A zero β would indicate a net null effect of biodiversity on disease risk. Ɛ represents error, including measurement error and error that is introduced via stochasticity in the system (i.e., irreducible error).

FIGURE 6

A PRISMA flow chart outlining the inclusion process for the systematic review (Page et al. 2021). Reports assessed for eligibility (n = 32) from the database searches include those that underwent abstract and full‐text screening due to a large number of reports being excluded (n = 488) during title screening. From the reports assessed for eligibility from the database searches (n = 32), 15 were included. Total studies included in the systematic review (n = 19) include the addition of eligible reports from citation searching (n = 4).

FIGURE 7

Proportional bar plots showing studies that use (a) site proxies (site size or habitat type), (b) simulations in analyses of the diversity‐disease relationship for LD over time. (c) represents studies that did not use site proxies or simulations (i.e., were otherwise empirical studies that used direct biodiversity metrics). The color‐coded proportions represent the outcomes (amplification [blue], dilution [orange], none [green]) for each analysis.

FIGURE 8

Latent variable model considering Peromyscus leucopus (white‐footed mice), Odocoileus virginianus (white‐tailed deer) density, and Ixodes scapularis (black‐legged tick) density as mediators in the relationship between host biodiversity and Lyme disease (LD) risk. Unobserved latent variables are represented as ovals and observed measured variables are represented as rectangles. Mediator variables in the relationship between biodiversity and disease risk are represented as white rectangles. Proxy metrics representing the latent constructs are represented as blue rectangles. Direct paths are represented by solid arrows and the net indirect effect is represented by a dashed arrow. Each path represents a proposed causal path between variables with the relationship β. A positive β on the net indirect effect (dashed arrow) would indicate a net amplification effect, whereas a negative β would indicate a net dilution effect. A zero β would indicate a net null effect of biodiversity on disease risk. Signs on the direct paths between variables show the directionality of the relationship between them (positive, either, or unknown) as evidenced in the literature (Elias et al. ; Kilpatrick et al. ; Linske et al. ; LoGiudice et al. , ; Mason et al. ; Pepin et al. ; Werden et al. 2014). Ɛ represents error, including measurement error and error that is introduced via stochasticity in the system (i.e., irreducible error).