Unusual space-time patterning of the Fallon, Nevada leukemia cluster: Evidence of an infectious etiology

Abstract

The town of Fallon within Churchill County, Nevada exhibited an unusually high incidence of childhood leukemia during the years 1997-2003. We examined the temporal and spatial patterning of the leukemia case homes in comparison to the distribution of the general population at risk, other cancer incidence, and features of land use. Leukemia cases were predominantly diagnosed during the early to mid summer, exhibiting a seasonal bias. Leukemia cases lived outside of the "developed/urban" area of Fallon, predominantly in the "agriculture/pasture" region of Churchill County, circumscribing downtown Fallon. This pattern was different from the distribution of the underlying population (p-value<0.01) and different from the distribution of other cancers, which were evenly distributed when compared to the population (p-value=0.74). The unusual space-time patterning of childhood leukemia is consistent with the involvement of an infectious disease. A possible mode of transmission for such an infectious disease is by means of a vector, and mosquitoes are abundant in Churchill County outside of the urban area of Fallon. This region harbors a US Navy base, and a temporally concordant increase in military wide childhood leukemia rates suggests the base a possible source of the virus. Taken together, our current understanding of the etiology of childhood leukemia, the rural structure combined with temporal and geospatial patterning of these leukemia cases, and the high degree of population mixing in Fallon, suggest a possible infectious cause.

Conflict of interest statement

Conflict of interest statement

None declared.

Copyright © 2011 Elsevier Ireland Ltd. All rights reserved.

Figures

Fig. 1

Temporal pattern of leukemia cases in the Fallon, Nevada leukemia cluster. (A) Timing of leukemia over the five-year period with a notable spike in 2000. (B) Seasonal timing of leukemia cases with most cases between April and July.

Fig. 2

Age adjusted acute lymphoblastic leukemia (Age 0–19). Case counts and age, branch and year specific denominator populations are from the Tricare data base. Age adjustment was calculated using the inverse variance of the observed rates. There is minimal random variation in these estimates.

Fig. 3

Geospatial organization of cancer cases and environmental features in Churchill County. Vegetation synthesis map of Churchill County (A) Spatial organization of leukemia cases (red circles) over the cluster time period. Locations are enlarged and “jittered” to maintain confidentiality. (B) Density-equalized map of the same information as (A), leukemia cases (red circles). Non-leukemia pediatric cancers (yellow square) are also displayed. On this map, the spatial structure is altered to create an equal distance between members of the population. (C) Location of leukemia cases (red circles), non-leukemia cancer cases (green diamonds), and non-leukemia pediatric cancers (yellow square) on the DEMP projection. Non-leukemia cancer cases appear randomly distributed. The paucity of cancer cases in the lower right of the graph is due to incomplete reporting of cancer to the Nevada registry from the US Navy Air Station. (D) Tested mosquito pools (in the year 2004) are shown by crosses, and those that were negative and positive for WNV are indicated in blue and yellow, respectively. Leukemia cluster cases are indicated (red circles).

Fig. 4

Comparison of the distribution of cancer cases on the DEMP projection. The graph exhibits a step function of the occurrence of cases, working outwards towards the periphery of the DEMP projection. If cancer cases are randomly distributed, the step function will exhibit a straight line at a 45° angle. The step function can be tested for variance against the observed population, a completely random distribution would be exactly 45°. (A) Non-leukemia cancer cases, and the population at risk are graphed. By the test of variance, these distributions do not differ (p-value = 0.74). (B) Leukemia cases distribution tested against the population at risk are significantly different by the test of variance (p-value < 0.01).

Fig. 5

Comparison of the distribution of leukemia cases and tested mosquito pools. (A) Distribution of leukemia cases and mosquito pool testing locations. (B) Fig. 2 shows the same two cumulative distributions (thick lines) and 20 cumulative distribution functions (thin lines) created by a bootstrap sampling of the original data. The “band” of replicate distributions surrounding the data-estimated values provides a sense of the considerable sampling variation associated with these estimates. Therefore, a formal inference from comparing these two frequency distributions (plots) is simply that no evidence exists, that the two distributions differ.

Fig. 6

2004 West Nile Virus Human Incidence per 10,000. Churchill County had the highest human incidence of WNV in Nevada during the introductory year of the WNV epidemic.