We quantified the relationship of acarological risk and human Lyme disease incidence for all of the known US range of I. scapularis-borne B. burgdorferi. Our study is the first to examine this association over such a large geographic range. We found that both rDON and rDIN explained a statistically significant amount of variation in human incidence in low- and high-incidence areas but that neither explained variation in incidence in rare areas. Restricting the analysis to low- and high-incidence areas, we found that the relationship of mDIN and incidence varied quantitatively by geographic location, that mDIN alone explained a statistically significant amount of variation in human incidence at the county level in only 12 of 20 states (including only states with more than eight counties), that the fit of the relationship was good in only four states (New York, Pennsylvania, Massachusetts, and Minnesota), and that fit was reversed in one of these states (Massachusetts). A subanalysis of 40 high incidence counties with NIP data identified some of the factors that contribute to geographic differences in the relationship of rDIN and incidence, most notably, genetic composition of B. burgdorferi.
One factor that has been proposed to modify the relationship between DIN and reported cases is variation in B. burgdorferi
Differences in virulence37
and transmission efficiency25,38
may lead to differences in infection risk and/or reporting rates by humans infected with different strains. Thus, areas with low incidence but high rDIN could be explained by an overrepresentation of strains with a low virulence in or transmission to humans. Although more studies are needed to confirm this hypothesis, our regression analyses support it for RST 3. We found that (1) incidence data could be partitioned similarly by region or RST 3, (2) RST 3 was more frequent in the region with lower incidence but similar rDIN, and (3) there was a negative relationship between RST 3 and incidence estimated by the Poisson regression model. Although the other two RST types each showed different regional distributions, neither of these two types alone caused deviations from the mean relationship of rDIN and reported incidence. Nevertheless, because higher levels of RST 3 (a genetically diverse group defined as non-RST 1 or 2) led to lower than expected incidence based on rDIN, by extrapolation, the sum of RST 1 and RST 2 frequencies was correlated with higher than expected incidence based on rDIN. Thus, our results suggest that the monophyletic RST 1 and 2 groups39
cause higher than expected reported cases based on rDIN, whereas a high prevalence of other strains results in fewer reported cases. More generally, our results highlight that genotypic differences are important for predicting human incidence spatially through acarological risk indices. However, more analyses of the genetic determinants of virulence (and transmission) in humans and the spatial dynamics of B. burgdorferi
genetic diversity are needed to understand the effects of B. burgdorferi
genotype on reported incidence.
In addition to B. burgdorferi
genotype, there are other factors not measured in our study that could contribute to geographical differences and weak associations in the relationship between mDIN and human incidence. One possibility is human movement. Although peridomestic exposure accounts for most cases of Lyme disease,12
some cases are a consequence of travel to other counties for outdoor work or recreation. This movement could explain the negative relationship between mDIN and cases that was found for Massachusetts. For example, if Massachusetts residents tend to get infected in counties that are densely forested but live in counties (i.e., where cases would be reported) that are mainly urban (i.e., low estimates of mDIN), then such a negative relationship could be observed. More accurate spatial risk predictions might be obtained if case reports included probable place of exposure, although accurate estimates are very challenging to obtain.40
Also, we found that spatial scale was crucial for interpreting the relationship between mDIN and incidence, because we showed that there are quantitative differences regionally (Midwest versus Northeast) as well as between states within these regions. It is possible that the county scale is appropriate in states with larger forested patches, such as New York and Pennsylvania (which showed relatively strong positive relationships between mDIN and incidence), whereas a finer scale is necessary in highly residential states with very dispersed forest cover, such as New Jersey and Maryland (which showed no significant relationship). For example, a study that measured DIN in forested areas in six Rhode Island towns found that DIN explained 97% of the variation in incidence among those towns.9
However, most of our study sites were not located in residential areas, and the incidence data corresponded to a much broader range relative to the acarological data, which increases the variance in the relationship between mDIN and incidence. Last, geographical differences in Lyme disease control efforts at the residential or personal level could also contribute to variation in the relationship between mDIN and human incidence.
There are also some experimental design caveats that should be noted when interpreting our results. First, our estimates of mDIN were based on data collected exclusively in public forested areas. Thus, our estimates of mDIN in counties that are mainly residential could be inaccurate. Second, sampling at different sites was done at different times between May and early September, and although most sites were sampled at least five times, some sites were only sampled three times. Although these sampling design caveats likely account for some of the unexplained variation in the relationship between DIN and human incidence, they were not spatially systematic (i.e., they occurred randomly with respect to state and region) and thus, should not introduce systematic bias in our results. Third, this analysis is based on original field efforts that were optimized for collections of I. scapularis
nymphs based on phenology in the northeastern United States (emergence in late May); therefore, areas in which nymphs become active earlier would have inaccurate estimates of nymphal densities. For example, a previous study found that nymphs were observed as early as April in the south.41
However, even with intensive sampling, only a few specimens were recovered at this time; thus, this difference should not introduce much bias in our results. There are also two other factors unaccounted for in our sampling design that could cause geographic variation in human incidence between northern and southern areas. First, because we only sampled I. scapularis
, the existence of an alternative vector might uncouple the relationship between mDIN and incidence. However, this possibility would only occur if the alternative vector caused a significant proportion of human cases, and there is no evidence to support this case. Second, it has been reported that nymphs in southern areas very rarely feed on humans,41
although it is still unclear whether this difference from northern areas is because of the low densities of I. scapularis
nymphs or an actual feeding preference. Future environmental surveillance should aim to quantify DIN in residential areas as well as other habitats and potential vectors to better understand how DIN translates to incident human cases across varying ecological conditions.