UC Berkeley

Millions of people are exposed to arsenic in drinking water. An ancient poison, arsenic occurs naturally in groundwater and geothermal springs. Removing arsenic from drinking water costs about $200 million every year in the United States alone. The brunt of this is borne by California and other western states, where groundwater is needed more for drinking water.

Arsenic in drinking water causes cardiovascular death, cognitive deficits in children, reproductive problems, and cancer. Surprisingly, many studies have shown that the human lung is especially susceptible to ingested arsenic. After being consumed in drinking water, arsenic accumulates in the lungs. Lung cancer is now believed to be the most common cause of death from this widespread contaminant.

Most lung carcinogens, including tobacco smoke, asbestos, and silica, also cause non-malignant respiratory effects. Evidence suggests that arsenic in drinking water follows this pattern, but nearly all data involve adults with recent exposures. The impacts of early-life arsenic exposures on nonmalignant lung disease are largely unknown. In northern Chile, the city of Antofagasta (population 390,000 in 2014) had high concentrations of arsenic in drinking water (>800 µg/L) from 1958 until 1970, when a new treatment plant was installed. This scenario, with its large population, distinct period of high exposure, and accurate data on past exposure, is virtually unprecedented in environmental epidemiology. Chapter 2 of this dissertation describes a pilot study on early-life arsenic exposure and long-term lung function. We recruited a convenience sample consisting primarily of nursing school employees in Antofagasta and Arica (population 160,000) a city with low drinking water arsenic. Lung function and respiratory symptoms in 32 adults exposed to >800 µg/L arsenic before age 10 were compared to 65 adults without high early-life exposure. Early-life arsenic exposure was associated with 11.5% lower forced expiratory volume in one second (FEV1) (p = 0.04), 12.2% lower forced vital capacity (FVC) (p = 0.04), and increased breathlessness (prevalence odds ratio = 5.94, 95% confidence interval 1.36–26.02). Exposure-response relationships between early-life arsenic concentration and adult FEV1 and FVC were also identified (p trend = 0.03). These results suggest that early-life exposure to arsenic in drinking water may have irreversible respiratory effects of a magnitude similar to smoking throughout adulthood. Given the small study size and non-random recruitment methods, further research is needed to confirm these findings.

The arsenic concentrations >800 µg/L in Chile can reveal previously unknown health outcomes, but they do not shed much light on what the drinking water standard should be. Arsenic is known to cause lung cancer at concentrations above about 200 µg/L. The effects of lower exposures are unknown. This uncertainty has created controversy over the 10 µg/L World Health Organization guideline and U.S. regulatory limit for public water supplies because arsenic is widespread in groundwater naturally and expensive to remove. In Chapter 3, I present the first lung cancer study in the largest U.S. populations with exposures between 50 and 100 µg/L. This was also the first U.S. lung cancer study with individual data on past drinking water arsenic concentrations. We enrolled 196 lung cancer cases and 359 controls, matched on age and sex, from western Nevada and Kings County, California in 2002–2005. After adjusting for age, sex, education, smoking and occupational exposures, odds ratios for arsenic concentrations ≥85 μg/L (median = 110 μg/L, mean = 173 μg/L, maximum = 1,460 μg/L) more than 40 years before enrollment were 1.39 (95% CI = 0.55–3.53) in all subjects and 1.61 (95% CI = 0.59–4.38) in smokers. Although odds ratios were greater than 1.0, these increases may have been due to chance given the small number of subjects exposed more than 40 years before enrollment. The findings suggest that concentrations near 100 μg/L are not associated with markedly high relative risks.

The California Nevada lung cancer case-control study, designed before research in Chile suggested arsenic-related cancer latencies of 40 years or more, illustrates the difficulties of identifying arsenic-related health effects in low-exposure countries with mobile populations like the U.S. Making matters worse, many wells used by participants 20–60 years ago could not be measured because they could not be located or had fallen into disrepair. A major problem for health studies of arsenic and other drinking water contaminants is a lack of measurements for unregulated water sources such as private wells, which are much more likely to have high arsenic than public supplies. For Chapter 4, I developed models to estimate arsenic in unmeasured wells in the western Nevada study area. In cross-validations with five partitions of 3138 residential wells, inverse distance weighted (IDW) averaging was the best categorizer of arsenic, averaging 71% correct assignment to the high, medium and low categories used in the lung cancer study. The next best categorizers were the random forest model (70% correct) and the median of measurements within 408 m (69%). These models categorized wells better than kriging (67%), consistent with findings in other areas. Nearest neighbors averaged 66% correct assignment. For exact arsenic concentrations, the best models were the random forest model (Arsenic = 0.95*RFM, R2 = 0.34) and kriging (Arsenic = 1.00*OK, R2 = 0.30). The RFM might be considered the best model overall, since it categorized measurements almost as well as IDW, had a higher R2 (0.34 versus 0.28), and a linear slope closer to 1 (0.95 versus 0.80). In terms of detecting high arsenic wells, RFM was also just behind IDW. These models can be substantially improved, but IDW and RFM already nearly doubled the number of unmeasured wells correctly categorized around Fallon, compared to the alternative of treating them as zeroes in effect estimates based on participants’ highest known concentrations. Given that the direction of the bias is the same whether unmeasured wells are estimated or excluded, the smaller bias using estimates is preferable.

Chapter 5 presents meta-analyses of low exposure lung cancer studies, and assesses differences in relative risks between men and women. Pooled results of six independent studies suggest that increased lung cancer risks reported for concentrations between 10 and 100 µg/L are not likely due to chance (pooled relative risk = 1.09, 95% CI = 1.02-1.20), but there was substantial heterogeneity between studies (p<0.001). The higher relative risk of 4.1 (95% CI = 1.8-9.6) in the case-control study by Ferreccio et al. (2000) might be explained by effects in ecological studies being diluted by in-migration or uncontrolled confounding. Relative risks were higher in men in Chile, in women in Taiwan, and similar for all countries combined. The gender differences in Chile and Taiwan were not likely due to chance, and may be explained by differences in study participation, case ascertainment, smoking, occupational, or household smoke exposures.

Chapter 6 concludes with a summary of recent research and ideas for future studies. New findings of health effects below 100 µg/L, including lung cancer in people with early life exposures (Steinmaus et al. 2014), need to be confirmed. However, increasing evidence supports the 2001 decision to tighten the U.S. regulatory from 50 to 10 µg/L. Detecting the increases in health risk expected below 50 µg/L (e.g., odds ratios below 1.5) will likely require very large sample sizes, but may be possible in studies with accurate data on early life exposures, inter-individual differences in arsenic metabolism, and related susceptibility factors (e.g., age, genetics, gender, diet, health status, smoking, and occupational exposures). On the environmental side, high arsenic concentrations have been predicted in many regions that lack available measurements, including much of the Amazon Basin (Amini et al. 2008), contrary to the expectations of some geologists (Ravenscroft et al. 2009). Measurements in these areas can test and improve models of arsenic in groundwater, improve exposure assessment, and help ensure the safety of millions of people.