In general, greater sampling effort or frequency is perceived to result in more accurate quantification of organism dynamics over time than sampling less frequently. Here we used a longitudinal study of deer mice and SNV to examine how a reduction in the frequency of sampling can influence detection of coupled host population demographics and pathogen infection. In general, sampling less frequently (bi-monthly, quarterly, semiannually, and annually) produced underestimates (10%–20%) of deer mouse abundance (MNA) derived from monthly sampling. Overall estimates of the number of infected deer mice (MNI) and standing infection prevalence (ESP), detected by less frequent sampling, were similar to monthly MNI and ESP values (monthly, mean, and weighted mean). However, monthly values for ESP were predominantly overestimated when values of ESP from less frequent sampling were high (particularly when >0.15). As such the effect of sampling less frequently than each month had a nonlinear effect on estimates of infection (i.e., less frequent sampling produced similar estimates of ESP as monthly when values were ≤0.10, but monthly ESP was frequently overestimated above this value).
Overestimation of high levels of ESP is of direct interest and application to human HPS exposure risk. Human incidence of HPS in Montana is not detected when ESP of deer mice, from monthly sampling, is below 10% (Madhav et al. 2007
), suggesting that investment in public health awareness programs or interventions are best targeted when deer mouse ESP exceeds this value. This study demonstrates that if this knowledge were currently applied to direct health awareness programs or interventions under a less frequent sampling regime (such as would be the case with semiannual and annual sampling frequencies), values of deer mouse ESP exceeding 10% would be overestimated and would result in premature investment in public health campaigns or interventions. Conversely, intervention when it may not be necessary is far superior, in terms of public health, to inaction. Of course, less frequent sampling is also likely to delay detection of epizootic conditions.
Resources to invest in host–pathogen monitoring programs are likely limited in many instances. As such, short-term studies, with limited spatial replication, which seek general information about host abundance and pathogen dynamics over time, should be interpreted with caution. We observed substantial fluctuations in deer mouse MNA, MNI, and ESP over 14 years, a feature characteristic of highly fecund animals (Krebs 1966
, Singleton 1989
, Carver et al. 2008
). However, SNV dynamics did not always reflect deer mouse abundance on the same grid and were not always similar between grids. These findings support those of Douglass et al. (2001
), who found large spatial and temporal variation in MNA (near 0 to over 120 per 100-trap grid) and MNI (0 to over 60) among 18 grids in Montana. Despite the variable nature of SNV over time, prevalence among deer mouse populations tends to peak in spring, across disparate geographic locations and variable habitat types (Douglass et al. 2001
, Mills et al. 1999b
and references therein). While it was not the specific focus of this study, understanding processes driving population fluctuations of deer mice (e.g., survival and dispersal) would likely contribute to a better understanding of the dynamics of infection (Douglass et al. 2007
, Madhav et al. 2007
, Lonner et al. 2008
, Luis et al. 2010
). Our results indicate that to capture natural variation in deer mouse abundance and dynamics of SNV, surveys need to be spatially replicated and of long duration (Douglass et al. 2001
). We intentionally based our less frequent sampling, particularly semiannual and annual sampling, on prior knowledge of deer mouse abundance and SNV dynamics at our grids. We acknowledge that someone establishing a new sampling regime would not have the benefit of this prior knowledge, likely resulting in the detection of fewer annual minima and maxima of MNA, MNI, and ESP. Investigations, with limited resources, that aim to understand spatial variation in host populations and risk of zoonotic pathogen exposure (without a priori
knowledge of host–pathogen dynamics) should first seek to understand annual variation in infection and then target spatial surveys during seasons when infection is at a maxima. In such cases, investigators should keep in mind that estimates of high infection prevalence are likely to be inflated. As a corollary, estimates of high infection prevalence could be improved by increasing sampling frequency during periods (epizootics) where ESP exceeds 10%.
In this study, underestimation of deer mouse abundance is important in understanding SNV infection dynamics, because this inflates estimates of ESP. In general, determining the sampling interval to adequately estimate host abundance is likely to depend on the developmental rate of juveniles and the duration and magnitude of natural population fluctuations (Krebs and Myers 1974
). For example, there is a 1–2-month lag between the birth of pups and first detection of deer mice by trapping (King et al. 1963
, Kirkland and Layne 1989
). We have trapped deer mice monthly and this appears to have been a reasonable frequency to capture abundance and follow temporal fluctuations in their abundance, a conclusion also supported by Parmenter et al. (1999
). Naturally, the ability to document fluctuations in population abundance and to investigate processes that underlie fluctuations (e.g., survival and recruitment) diminishes as the frequency of sampling decreases (Krebs and Myers 1974
). Similar to our findings, that sampling less frequently than monthly leads to underestimates of MNA, attempts to model temporal fluctuations in deer mouse population abundance, using a range of predictors (e.g., temperature, precipitation, survival, and maturation), have also required a monthly sampling to acquire any level of accuracy (Yaffee et al. 2008
, Luis et al. 2010
). Further, accurate estimation of population demographic parameters [i.e., survival, recruitment (Krebs and Myers 1974
), and dispersal (Lonner et al. 2008
, Waltee et al. 2009
)], all necessary to understand disease dynamics (Keeling and Rohani 2008
), require frequent sampling. Understanding the dynamics of SNV among deer mice would greatly benefit from investigations, based on frequent sampling, which link environmental determinants of deer mouse abundance (Luis et al. 2010
) and infection.
Detecting annual extremes in host or vector abundance, particularly maxima, is an important facet in predicting infection prevalence of many diseases, including SNV, reflecting the density- or frequency-dependent nature of transmission (Mills et al. 1999b
, Madhav et al. 2007
, Keeling and Rohani 2008
, Carver et al. 2009
and references therein). Similarly, detecting annual extremes, particularly maxima, in infection is important to allocate public-health-related resources. Not surprisingly, a reduction in the frequency of sampling resulted in a consistent reduction in the detection of annual maxima of MNA, MNI, and ESP, and minima of MNA. It should also be acknowledged that annual maxima in deer mouse abundance and infection are observed in winter with reasonable frequency. For example, Calisher et al. (2005
) documented a 5-year high in the trap success of deer mice in the winter of 1999. Over the 14 years of our study, we were able to access trapping grids each month, even when environmental conditions, such as deep snow, made site access and trapping challenging. Although trapping in winter is not always possible due to inaccessibility of trapping locations (e.g., Douglass et al. 2001
, Pearce-Duvet et al. 2006
), investigators should recognize that the detection of annual highs in host abundance and infection is reduced when sampling is less frequent than monthly, and when sampling is not undertaken during winter. Nevertheless, this study suggests that annual highs in deer mouse MNA are most frequently detected between August and October and ESP between April and August, likely reflecting a delayed transmission relationship between deer mouse abundance and infection (Madhav et al. 2007
). Our results indicate that missed maxima in host abundance and infection prevalence (such as would likely be the case with semiannual or annual sampling frequencies) would result in inflated predictions and underestimation of true infection prevalence, respectively. This information is of direct relevance to the timing and allocation of management and public health resources.