The city of Fallon (population 8,000) is the only urban center in Churchill County, Nevada. The Naval Air Station–Fallon (NAS Fallon) contributes an additional 3,000 residents to the city’s population. The local community identified four major areas of concern related to environmental exposures. A pipeline runs through downtown Fallon, delivering a continuous supply of JP-8 jet fuel to the base. Both municipal and NAS water supplies have historically reported naturally occurring arsenic levels that exceed U.S. Environmental Protection Agency (U.S. EPA) standards (Welch and Lico 1998; Welch et al. 1989). Fallon is also a center for melon and alfalfa production, and agricultural pesticides are applied to fields surrounding the city. Two facilities related to tungsten refining and use are located in Churchill County: An administrative center and laboratory are in downtown Fallon, and a tungsten carbide processing plant is 11 miles north of Fallon [Agency for Toxic Substances and Disease Registry (ATSDR) 2003].

In July 2000, an astute local health care provider notified state health officials that several Churchill County children had recently been diagnosed with leukemia. In general, state cancer registries experience significant reporting lag, so state health officials conducted active case finding to verify leukemia occurrence. By February 2001, the NSHD had identified 12 children who had been diagnosed with leukemia since 1997 and who had lived in Churchill County before their diagnosis. The state convened an expert panel composed of national cancer specialists, epidemiologists, and public health officials to review these cases. The panel recommended that the state epidemiologist request technical assistance from the Centers for Disease Control and Prevention (CDC) to investigate whether ongoing exposure to environmental contaminants in Churchill County might be endangering human health (Robison et al. 2001). Two new cases were added to the state’s list shortly after the expert panel report was completed, bringing the number of cases included in the state’s investigation to 14.

The NSHD responded to recommendations of the expert panel by approaching the CDC for assistance in Churchill County. Beginning in March 2001 the CDC led a multiagency effort to conduct a comprehensive cross-sectional exposure assessment. In Table 1 we list the state and federal agencies that were involved and identify the responsibilities of each agency. All involved agencies were represented at community meetings held throughout the study period to provide updates on the status of the study. During the course of the investigation, all biologic and environmental laboratory results were reviewed as they were received and were collectively assessed for risk by multidisciplinary expert panels (CDC 2004); this joint review was conducted to fulfill the CDC’s promise to the community that any results suggesting the possibility of an ongoing health risk, regardless of known significance to the cluster analysis, would be released to the community immediately rather than waiting until the end of the study.

We defined a case family as the child diagnosed with leukemia and all other people currently living in the child’s home (i.e., all siblings, parents, guardians, and other adults), as well as biologic parents who were not living full time in the case child’s home and for whom contact information was available. At the time the CDC study was initiated, 14 case children had already been identified by NSHD. All 14 of the families of these children were approached for inclusion in the CDC study; 13 of these families agreed to participate. One of the case children from these families died before our study began. A 15th child was diagnosed and enrolled in the CDC study in December 2001. A 16th child was diagnosed after 31 December 2001, the end of the study enrollment period. In all, 15 case families were eligible to participate, and 14 were enrolled in the study. In Figure 1 we describe the date of diagnosis and leukemia cell type for all 16 children diagnosed with leukemia. Of the 16 cases, 10 (62.5%) were residents of Churchill County at the time of diagnosis.

Because leukemia etiology (e.g., genetic predisposition, prenatal exposure, postnatal exposure) may be related to age at diagnosis, leukemia type [ALL, acute myelocytic leukemia (AML)], cell type (B-cell, T-cell), or geographic location, we constructed a stricter case definition based on age (0–6 years at time of diagnosis), diagnosis (ALL pre-B), and duration of residence in Churchill County (at least 6 months before diagnosis). Nine children met the restricted case definition (Figure 1).

Comparison families from Churchill County were identified through random-digit dialing. Fifty-five (43.6%) of the 126 comparison families identified as eligible were enrolled. We attempted to match four comparison children to each case child by birth year (in 2-year increments) and sex. The comparison child was required to be cancer free at the date of the case child’s leukemia diagnosis. We defined a comparison family as the matched comparison child and that child’s parents, guardians, or other care-taking adults living full time in the comparison child’s home.

A total of 205 participants representing 14 (of the 15 eligible) case families and 55 comparison families were enrolled and asked to complete mailed questionnaires, participate in personal interviews, donate biologic samples, open their homes to environmental sampling, and provide information about previous local residences that could be sampled. We enrolled 4 comparison families matched to the case family that declined participation; these data were included in the cross-sectional analysis but not in the case-comparison analysis. Some members of one enrolled case family chose not to participate; the information from that case family is included in the case-comparison analysis. The study described here complies with all applicable requirements of the U.S. regulations including institutional review board (IRB) approval. All participants signed consent or assent forms that had been approved as part of the IRB review of the protocol for this study (CDC 2003a).

Beginning in August 2001, all participants were invited to visit a CDC clinic in Fallon, where blood, urine, and cheek swab samples were collected in a highly controlled setting (CDC 2003b) and personal interviews were administered. The biologic samples were aliquoted on site in a biological safety cabinet to maintain sample integrity and to keep the samples free from chemical contamination and infectious agents. These samples were sent daily to CDC laboratories in Atlanta, Georgia, where they were analyzed for essential and toxic metals (CDC 2003b), nonpersistent pesticides (Baker et al. 2004; Bravo et al. 2002; Olsson et al. 2004), persistent pesticides and polychlorinated biphenyls (PCBs) (Barr et al. 2003), VOCs (Cardinali et al. 2000), and markers of past exposure to infectious diseases. Concentrations less than the limit of detection (LOD) were assigned a value equal to the LOD divided by the square root of 2.

We considered both creatinine-corrected and non-creatinine-corrected results for analyses of urine samples and lipid-adjusted and non-lipid-adjusted results for analyses of blood samples. In this article we present non-creatinine-corrected results for metal and nonpersistent pesticide levels in urine samples because creatinine correction may not be reliable in children (Barr et al. 2005; O’Rourke et al. 2000) and because non-creatinine-corrected comparison values for most analytes were available in the second National Report on Human Exposure to Environmental Chemicals (second NER) (CDC 2003d). We present lipid-adjusted results for persistent pesticide and PCB levels in serum samples, and results not adjusted for lipids for VOCs analyzed in blood samples.

Using standardized protocols (CDC/Nevada Department of Environmental Protection 2005), the Nevada Department of Environmental Protection collected samples of indoor air, play yard soil, and household dust from case families’ current and previous residences in Churchill County, current residences of all comparison families, and previous residences for one randomly selected comparison child for each case child. Tap water samples from all of these residences were collected by the U.S. Geological Survey (USGS). Environmental samples were successfully collected from all 11 homes of the case families currently living in Churchill County, all 8 previous homes in Churchill County in which case families had lived before the date of birth of the case child (i.e., during pregnancy), current homes of all 55 comparison families, and 6 previous residences of 15 comparison families. Of 36 previous residences eligible for sampling, 8 case homes and 14 comparison homes were not accessible (e.g., house was destroyed or vacant, or current owner was nonresponsive). Current and previous residences outside of Churchill County were not sampled. Household samples were tested for heavy metals, persistent and nonpersistent pesticides, PCBs, VOCs, radon, and radionuclides. Environmental samples were analyzed by the USGS, U.S. EPA Region 9, the Nevada Department of Agriculture, and several contract laboratories. In this article we present results of environmental sample analysis only when they are relevant to biologic sample results.

In our primary analysis, we used univariate statistics to describe each of the exposures analyzed. For continuous variables, we used geometric means and selected percentiles to summarize the range and distribution of the data for the various subpopulations of interest. We calculated geometric means [with 95% confidence intervals (CIs)] assuming that the data approximated a log-normal distribution, and only when the proportion of results above the LOD was at least 60%. The estimates of the mean and 95% CI are based on a statistical model that controlled for the possible correlation of observations within a family (i.e., a variance components model), when appropriate. Categorical variables, frequency counts, and percentages are presented as summary statistics for the subpopulations of interest. In the secondary analysis, we compared exposure among the case and comparison populations using conditional logistic regression; we initially analyzed all cases and their corresponding controls, and then only those cases and their corresponding controls meeting the restricted case definition.

For categorical exposure variables, odds ratios (ORs) are used to assess the association between disease and exposure between two specific levels of the categorical variable. For continuous exposure variables, the ORs are based on data that were standardized before analysis. The exposure measures were standardized by dividing each individual response by the standard deviation observed among the entire study population. Because of the limited sample size, ORs were not adjusted for potential confounders. The LogXact software from Cytel Corporation (version 4.0; Cytel Software Corp., Cambridge, MA) was used to fit the conditional logistic regression models.

For this investigation, many of the logistic regression models compared current levels of exposure among the case and comparison populations. Because both populations were sampled after diagnosis of leukemia among the case population, treatment, past diagnostic procedures, changes in behavior, and changes in chemical exposures over time may all have significant, although immeasurable, impact on the relationships that were explored in our secondary analysis.

When preliminary study results suggested unusually high levels of tungsten in urine and water samples collected from Churchill County, we decided to expand our study to include tungsten measurements in three other Nevada communities to determine if the findings in Churchill County were unique or represented levels characteristic of the region. The towns of Lovelock, Yerington, and Pahrump were chosen for a cross-sectional tungsten exposure based on hydrogeologic criteria and history of tungsten mining (CDC 2003c). A geographically random sample pool was selected in each town, including 30 households each from Yerington and Pahrump and 11 households from Lovelock. For sample selection, coordinate grids were superimposed over maps of a given city. A random list of X and Y grid coordinates was generated; the first 30 coordinates falling within the city limits were selected, and the household closest to each coordinate was approached for recruitment. If the first house refused participation or was ineligible, the next closest house was approached. The eligibility criteria for households are residence in the city for at least 1 month before the interview and the presence of one adult and one child younger than 18 years who consent to participate in the study. Eligibility criteria included residence for at least 1 month before the interview. Environmental sampling included tap water, floor dust, and yard soil from each household. Urine samples were collected from one adult and one child younger than 18 years in each family. A total of 141 participants were recruited, ranging in age from 2 to 65 years (children younger than 18 averaged 9.4 years of age and adults averaged 38.9 years of age). Sample collection procedures and their analytical parameters were identical to those used to measure tungsten in the Churchill County study.

The median tungsten level found in urine samples from the entire study population was 0.97 μg/L (geometric mean, 1.19 μg/L), compared with 0.07 μg/L (geometric mean, 0.08 μg/L) in the 1999 National Health and Nutrition Examination Survey (NHANES), the most current population reference available in August 2002 or in the second NER (CDC 2003d). In our secondary analysis, median levels were similar for case and comparison children (1.93 and 2.35 μg/L) and case and comparison families (0.61 and 0.62 μg/L). Almost 80% of the Churchill County participants had tungsten levels above the 95th percentile (0.48 μg/L) reported in the second NER. The level of exposure to tungsten that may cause health effects is not known. The CDC successfully petitioned the National Toxicology Program of the National Institute of Environmental Health Sciences, National Institutes of Health, to prioritize research regarding the health effects of tungsten exposure. The CDC contracted with USGS to collect and measure tungsten in tap water samples from participants’ homes. Results ranged from < LOD to 290 μg/L. There is no regulatory limit for tungsten in drinking water.

In the three additional Nevada communities that were sampled for tungsten, 68% of the participants had geometric mean levels of urinary tungsten at or above the 95th percentile second NER levels (Table 2). Yerington, the town most similar to Churchill County with respect to hydrology, geology, and land use, had levels of tungsten in urine and water samples that were statistically similar to Churchill County’s samples. Few household soil and dust samples from the three comparison communities yielded detectable tungsten levels; however, the method used had a high LOD (50 μg/g).

In our cross-sectional analysis, arsenic levels in urine ranged from < LOD to 1,180 μg/L, with a geometric mean of 34.6 μg/L and median of 37.4 μg/L. A national reference value for urinary arsenic levels is currently unavailable. However, a study of arsenic exposure in Washington State showed urinary arsenic levels ranging from 19.6 μg/L (associated with high exposure) to 9.4 μg/L (associated with low exposure) (Kalman et al. 1990). Furthermore, the 95th percentile for nonrandom samples of the U.S. population is 22 μg/L (CDC, unpublished data). Noncarcinogenic clinical effects are observed at urinary concentrations > 200 μg/L (Hall 1998); however, clinical effects from concentrations < 200 μg/L may be possible. Although not a statistically significant finding, levels of arsenic were nominally higher (p = 0.25) in children (geometric mean, 38.9 μg/L) than in their parents (geometric mean, 32.3 μg/L). In our secondary analysis, arsenic levels did not differ significantly (p = 0.29) between case children (geometric mean, 29.8 μg/L) and comparison children (geometric mean, 41.3 μg/L) or between case families (geometric mean, 23.9 μg/L) and comparison families (geometric mean, 36.0 μg/L) (p = 0.18). Arsenic was also measured in tap water samples collected from 70 current and previous residences. Overall results ranged from < LOD to 874 μg/L (median, 50.9 μg/L). At the time of this study, the U.S. EPA regulatory limit for arsenic in municipal drinking water systems was 50 μg/L; as of 1 January 2006, the regulatory limit is 10 μg/L.

Biologically plausible environmental risk factors of concern in Churchill County included exposure to benzene and other VOCs from JP-8 fuel and also exposure to persistent and nonpersistent pesticides. We analyzed other lifestyle and demographic risk factors that have been implicated in the development of leukemia, including exposure to ionizing radiation (Doll 1995; Miller 1967; Preston et al. 1994), parental age at child’s birth (Dockerty et al. 2001), birth weight (Hjalgrim et al. 2004; Robison et al. 1987), breast-feeding (Kwan et al. 2004), history of allergies (Schuz et al. 1999; Wen et al. 2000a), and parental military service (CDC 1988; Wen et al. 2000b).

We analyzed blood samples for 12 VOCs, including benzene, which is a minor component of JP-8 fuel and gasoline. Most study participants had blood benzene levels below method LODs (0.06 ng/mL). Median VOC levels in Churchill County were similar to those reported in NHANES III (Churchill et al. 2001) and other peer-reviewed reference levels (Table 3). Levels of 2,5-dimethylfuran among the study population were consistent with levels reported in the literature (Ashley et al. 1996) for smokers and nonsmokers. In the Churchill County population, smokers had a median level of 0.08 μg/L and nonsmokers had a median level below the LOD (= 0.024). In our secondary analysis, 7 VOCs were detectable in a high enough percentage of samples to calculate ORs. Exposure to ethylbenzene suggested increased risk for leukemia among case children using the broader case definition (OR 2.67; 95% CI, 1.04–6.84) as well as the restricted case definition (OR 6.13; 95% CI, 1.29–29.00). When comparing case and control families using the broad definition of a case, we found a slightly positive, although not statistically significant, association between leukemia status and exposure to ethylbenzene (OR 1.14; 95% CI, 0.73–1.78). This OR was somewhat lower and still not statistically significant when we analyzed levels of ethylbenzene among case and control families using the restricted case definition (OR 1.08; 95% CI, 0.60–1.94).

Tetrachloroethylene was the only VOC that showed a slightly significant protective OR (0.35; 95% CI, 0.14–0.86) between case and comparison families (including case and comparison children and their family members but excluding siblings). When comparing case and comparison children, however, we detected no statistical differences in VOC levels using either the broad case definition (OR 0.32; 95% CI, 0.05–2.14) or the restricted case definition (OR 0.39; 95% CI, 0.04–3.91).

Geometric mean levels of 5 of the 31 nonpersistent pesticides that we measured in urine samples were above the reference geometric mean (Table 4); 5 of the pesticides were well below the national reference values, and the rest were very similar to reference levels. In our secondary analysis, none of the nonpersistent pesticides were associated with leukemia risk.

Of the 11 persistent pesticides measured in serum samples, only dichlorodiphenyldichloroethylene [DDE; a breakdown product of dichlorodiphenyltrichloroethane (DDT)] was elevated in our study population (median, 445.3 ng/g lipid; geometric mean, 447.1; range, < LOD to 8169.95) when compared with the second NER (median, 226.0 ng/g lipid; geometric mean, 260). However, when we analyzed DDE levels stratified by birthplace, we found that the elevation persisted only among parents born outside the United States and among children breast-fed by mothers born outside the United States (Figure 2). In the secondary analysis, we found no relationship between leukemia status and exposure to any of the persistent pesticides.

AML and ALL are widely recognized as biologically different diseases and are therefore likely to have different etiologies (Kersey 1997). Moreover, even subgroups of ALL, or widely discrepant age groups within a cell-type subgroup, may manifest different treatment responses and disease outcomes (Greaves 2002). Our study may be limited because we included a variety of leukemia types in some parts of our analysis. Similarly, some parts of our analysis are possibly skewed by inclusion of case children who had only recently moved to Churchill County, had been tangentially exposed to the Churchill County environment, or had moved away from Churchill County long before their diagnosis. To overcome these limitations, we conducted separate statistical analyses on a restricted number of cases that appeared to be most alike. Although these subanalyses did not yield meaningful results, we encourage researchers to combine data from this and future leukemia studies to obtain a larger sample size of cases meeting the strict case definition. Other limitations, particularly of the secondary analyses in which we compared case with comparison subjects, reflect the issue of temporality and our limited ability to control confounding factors.

Cancer clusters can occur by chance. However, the statistical process of fully evaluating the role of chance in a suspected cluster occurrence is daunting and lacks standardization, leaving political and emotional pressures to drive cluster investigations. Ideally, a cancer cluster should be evaluated not just on whether it exceeds previous cancer incidence in a specific defined geographical area during a specific time frame but whether it exceeds expected cancer incidence after accounting for the many ways to define “area,” the large number of areas, as well as the many ways that one could choose a time period for evaluation (Wartenberg 2001).

Cancer cluster investigations are highly charged with emotional and political overtones that inevitably challenge the validity of results (Aldrich and Sinks 2002; Neutra 1990; Rothman 1990; Thun and Sinks 2004; Wartenberg 1995). Affected families and communities are frustrated by a lack of conclusive findings, whereas public health investigators are equally frustrated by their inability to identify a causative agent. In an attempt to mitigate these frustrations, we convened several external peer-review and community-based panels to review and deliberate on our results. This ongoing peer review enhanced communication and increased our confidence in the methods and conclusions of this study.