Exposure to Inorganic Arsenic in Drinking Water and CHD
Exposure to Inorganic Arsenic in Drinking Water and CHD
We investigated the relationship between lifetime inorganic arsenic exposures and the risk of incident CHD using a case-cohort design within the San Luis Valley Diabetes Study (SLVDS). SLVDS is a population-based prospective study conducted from 1984 through 1998 in Alamosa and Conejos counties of south central Colorado. The study investigated the risk factors for diabetes mellitus (DM) and other related chronic diseases in Hispanic and non-Hispanic whites 20–74 years old. SLVDS data collection methods and participant recruitment have been described elsewhere (Hamman et al. 1989). In brief, researchers collected clinical, behavioral, and demographic data and diagnostic assessments including diagnoses of CHD from 1984 through 1988 (Hamman et al. 1989). Participants were then invited to attend follow-up visits every 4 years through 1998 to update their behavioral, demographic, and clinical assessments, along with an additional set of assessments on participants with impaired glucose tolerance at the initial visit. All participants were followed between clinic visits with telephone interviews and searches of vital statistics records to track vital status and identify underlying cause of death, where applicable (Hokanson et al. 2003). This cohort is stable, with a 98% follow-up of study participants through 1998 (Hokanson et al. 2003).
There were 1,297 SLVDS participants with no known CHD events or a diagnosis of DM before the baseline visit. Participants with a documented refusal for re-contact in SLVDS (n = 361) were excluded, leaving 936 participants eligible for this study. Cases of CHD included all eligible participants with a documented CHD event between their baseline visit and 1998. A CHD event was defined as any of the following: myocardial infarction, angioplasty, and death due to acute, subacute, or chronic ischemic heart disease [ICD-9 (International Classification of Diseases, 9th Revision) codes 410–414]. Potential CHD events were identified through self-report on yearly follow-up phone calls, obituary monitoring, and death certificate searches (Swenson et al. 2001). The medical records of identified CHD events were reviewed by a three-member committee of medical physicians for case confirmation.
The subcohort for estimating person-years of risk was randomly selected from the eligible participants without a previous diagnosis of CHD at the time of initial enrollment. The sample size of the subcohort was determined in Pass® software (www.ncss.com/software/pass/) based on recent research to estimate effect size (a relative risk of 1.4 for an increased risk > 10 μg/L) (Zierold et al. 2004) with alpha = 0.05 and power of 80%. The subcohort included 533 randomly selected participants, of whom 74 were incident CHD cases. The remaining 22 incident CHD cases not selected were added to the subcohort for a total of 555 participants in the CHD case-cohort study. Within the 555 study subjects, we had 64% (n = 357) participation rate, 33% (n = 189; 30% noncases and 46% cases) unable to locate, and 3% (n = 19) who refused participation.
Residential history (determined from structured interviews or secondary data sources) was linked to a geospatial model of predicted inorganic arsenic in groundwater to reconstruct annual estimated exposure to inorganic arsenic in drinking water over each participant's lifetime. The use of inorganic arsenic levels in residential drinking water to assess exposure was supported by findings from our research that correlated annual predicted arsenic exposure estimates with temporally concurrent speciated inorganic arsenic concentrations in historically collected samples. Between 2006 and 2008, study subjects or next of kin of deceased subjects as designated in SLVDS (14.2%) were contacted by mail with information about the study, followed by a call to set up an appointment for an interview and water sample collection. During the interviews (n = 357; 64%) we collected data about past residences, past workplace/school locations, and history of drinking-water consumption at each location. For each location, data included addresses, residence dates, water source (well or public), water treatment device (if yes, type, model number), number of glasses of water consumed per day (nonbottled water), number of glasses of beverages made with water (e.g., coffee, tea, juice), whether they typically cooked with water from the tap in the home, and whether they had a vegetable garden (if yes, which vegetables). For subjects or next of kin who were not able to be located for an interview (n = 189, 33%), we used triangulation methods incorporating records from the county assessor's office and SLVDS tracking database to reconstruct residential history. In brief, 189 participants were not interviewed (n = 2,023 person-years); therefore, the SLVDS contact tracking database was used to determine residence history back to 1975 and earlier for participants who reported living at the residence listed in 1975, before 1975. This left 914 person-years across 98 subjects still missing residential history who therefore were further investigated in county clerk records. In the end, there were 18 subjects with partial residential history (before 1975) (n = 126 person-years) with missing residential locations, and these were assigned the mean value for the last known city of residence.
Drinking-water samples were collected from the residential kitchen tap at time of interview and analyzed by the chemistry laboratory of the Colorado Department of Public Health and Environment using standard ion chromatography and inductively coupled plasma mass spectrometry with a detection limit of 1 μg/L. Samples with arsenic concentrations below the detection limit were given a value of half of the detection limit similar to other studies (Ayotte et al. 2006). Samples collected from private wells were assigned geographic coordinates using a global positioning system (GPS) unit (n = 248). Elsewhere we have provided methods for determining and validating the temporal and spatial variability of inorganic arsenic in groundwater in the SLV (James et al. 2013). In brief, findings indicate that naturally occurring inorganic arsenic concentrations in groundwater are stable over decades [consistent with other research (e.g., Steinmaus 2005)], justifying the use of geospatial models based on the mean arsenic concentration in individual private wells over decades to predict spatial variability of inorganic arsenic in groundwater. This was supported by a correlation analysis in a 10% sample of observed and predicted values (ρ = 0.715; 95% CI: 0.67, 0.75) (James et al. 2014).
An exposure matrix was developed to estimate each participant's annual exposure to arsenic in drinking water. Each record included residential, employment, and school location and an estimate of the amount of water ingested and water arsenic concentration (either observed or predicted) for each participant for each year of life from birth to death or 1998, whichever came first. Three estimated exposure values were calculated for each year of the follow-up period 1984 through 1998 or CHD diagnosis, whichever came first. The three estimated exposure values—residential arsenic concentration (arsenic concentration in drinking water at residence), residential arsenic dose (residential arsenic concentration in drinking water times the amount of water consumed in liters per day), and total arsenic dose (residential arsenic dose plus workplace and/or school dose)—were each defined based on a time-weighted average (TWA). A time-weighted average (TWA) for each exposure metric was calculated by dividing cumulative per-person arsenic exposure by the number of years in each participant's lifetime to get an annuitized exposure per year (Meliker et al. 2010).
To determine which TWA exposure estimate best approximated biologic exposure, we correlated these with speciated arsenic concentrations in historically collected urine samples (collected 1984–1991), adjusting for sex and creatinine (James et al. 2013). In brief, estimates of residential arsenic concentration (R = 0.37; ρ = 0.61) were the strongest correlates of the sum of the toxic urine arsenic species [As, As, dimethylarsinic acid (DMA), monomethylarsinic acid (MMA)], as opposed to estimates that included water consumption (residential dose) (R = 0.21; ρ = 0.46) or exposure at work or school (total dose) (R = 0.23; ρ = 0.48).
We used a Cox proportional-hazards model incorporating a robust variance estimator specific for case-cohort study designs (Barlow et al. 1999) to examine the association between TWA inorganic arsenic exposure and diagnosis of or death from CHD. We scaled the continuous arsenic exposure estimate to the interquartile range (IQR) (15 μg/L), along with other continuous covariates, similar to methods used by Lin and Huang (1995).
As described above, participants had longitudinal data from two to four study visits, including information on known risk factors for CHD. Risk factors for CHD believed to be independent of the mechanistic pathways proposed for arsenic were included in the proportional-hazards multivariate model as time-dependent covariates [lipid measurements, body mass index (BMI), physical activity, smoking, and alcohol and water consumption]. The univariate model included TWA inorganic arsenic as a continuous value scaled to the IQR (15 μg/L). Person-years and exposure were censored for CHD cases at time of diagnosis. The full model included the demographic risk factors ethnicity (white non-Hispanic/Hispanic), sex (male/female), and annual household income (high ≥ $20,000/low < $20,000); the known risk factors first-degree family history of CHD (no/yes), BMI (IQR scaled, median = 26.7, IQR = 23.8–29.3), diabetes diagnosis before CHD (no/yes); behavioral risk factors including current smoking status (no/yes), alcohol consumption (low ≤ 168 g/week/high > 168 g/week), and physical activity level (active/sedentary) (Mayer 1991); and continuous clinical risk factors including serum lipid measurements [high-density lipoprotein (HDL), triglycerides, and low-density lipoprotein (LDL) in milligrams per deciliter], hypertension (blood pressure > 140/90 or use of anti-hypertensive medicine) and folate and selenium intake (micrograms). Triglyceride and HDL levels were determined using enzymatic methods, and LDL levels were calculated using the Friedewald equation (Friedewald et al. 1972); folate and selenium intake were estimated based on 24-hr diet recall involving two- and three-dimensional visual aids for portion approximation. Nutritional analysis was based on version 14 of the Nutrition Coordinating Center's nutrient database released in 1987 (www.ncc.umn.edu/products/database.html). Vitamin supplement use was assessed through self-report using vitamin bottle labels.
In addition, a final parsimonious model incorporating statistically significant covariates to the model based on a 10% change to the HR for TWA arsenic exposure. Known independent risk factors, including sex and family history, that were significantly associated with the outcome were maintained in all models regardless of whether or not they met statistical criteria for confounding. Covariate data were assessed at each clinic visit (up to four visits) from 1984 through 1998. The risk associated with covariates was based on the covariate value at the clinic visit before the time of the CHD event. Clinic visits assessed all behavioral and clinical values for all covariates in this analysis. Missing data occurred when participants did not attend follow-up clinic visits; however, this was a very small number (7%), so covariate values from the baseline visit were used.
We also assessed the HR for CHD across arsenic exposure groups (TWA concentration, 20–30 μg/L, 30–45 μg/L, and 45–88 μg/L relative to < 20 μg/L). The cut-points for the exposure groups are based on arsenic concentrations in past research with significant associations with CHD (Chen Y et al. 2013; Moon et al. 2013).
We assessed whether hypertension might confound the association between inorganic arsenic exposure and CHD by reanalyzing the final model with hypertension as a dichotomous covariate. Last, we completed secondary analyses to data collection methods and exposure estimates. The first was an agreement analysis on 5% of the interviewed participants (n = 28) that compared residential address and year reported in the interview by the participant with the residential county clerk records. We compared residence and year for the years 1955–1985 as reported by both sources. The next secondary analysis was completed in a limited cohort (n = 462) to confirm any association found based on residential history using mean speciated urinary arsenic concentrations (As, As, MMA, and DMA). To complete this analysis, we compared urinary arsenic concentrations between cases and noncases as a secondary analysis. We used SAS 9.2 (PROC PHREG; SAS Institute Inc, Cary, NC) for the statistical analyses. We complied with all applicable requirements of national and international regulations including approval from institutional review board, and human participants provided written informed consent before participating in the study.
Methods
We investigated the relationship between lifetime inorganic arsenic exposures and the risk of incident CHD using a case-cohort design within the San Luis Valley Diabetes Study (SLVDS). SLVDS is a population-based prospective study conducted from 1984 through 1998 in Alamosa and Conejos counties of south central Colorado. The study investigated the risk factors for diabetes mellitus (DM) and other related chronic diseases in Hispanic and non-Hispanic whites 20–74 years old. SLVDS data collection methods and participant recruitment have been described elsewhere (Hamman et al. 1989). In brief, researchers collected clinical, behavioral, and demographic data and diagnostic assessments including diagnoses of CHD from 1984 through 1988 (Hamman et al. 1989). Participants were then invited to attend follow-up visits every 4 years through 1998 to update their behavioral, demographic, and clinical assessments, along with an additional set of assessments on participants with impaired glucose tolerance at the initial visit. All participants were followed between clinic visits with telephone interviews and searches of vital statistics records to track vital status and identify underlying cause of death, where applicable (Hokanson et al. 2003). This cohort is stable, with a 98% follow-up of study participants through 1998 (Hokanson et al. 2003).
There were 1,297 SLVDS participants with no known CHD events or a diagnosis of DM before the baseline visit. Participants with a documented refusal for re-contact in SLVDS (n = 361) were excluded, leaving 936 participants eligible for this study. Cases of CHD included all eligible participants with a documented CHD event between their baseline visit and 1998. A CHD event was defined as any of the following: myocardial infarction, angioplasty, and death due to acute, subacute, or chronic ischemic heart disease [ICD-9 (International Classification of Diseases, 9th Revision) codes 410–414]. Potential CHD events were identified through self-report on yearly follow-up phone calls, obituary monitoring, and death certificate searches (Swenson et al. 2001). The medical records of identified CHD events were reviewed by a three-member committee of medical physicians for case confirmation.
The subcohort for estimating person-years of risk was randomly selected from the eligible participants without a previous diagnosis of CHD at the time of initial enrollment. The sample size of the subcohort was determined in Pass® software (www.ncss.com/software/pass/) based on recent research to estimate effect size (a relative risk of 1.4 for an increased risk > 10 μg/L) (Zierold et al. 2004) with alpha = 0.05 and power of 80%. The subcohort included 533 randomly selected participants, of whom 74 were incident CHD cases. The remaining 22 incident CHD cases not selected were added to the subcohort for a total of 555 participants in the CHD case-cohort study. Within the 555 study subjects, we had 64% (n = 357) participation rate, 33% (n = 189; 30% noncases and 46% cases) unable to locate, and 3% (n = 19) who refused participation.
Estimating Arsenic Exposures
Residential history (determined from structured interviews or secondary data sources) was linked to a geospatial model of predicted inorganic arsenic in groundwater to reconstruct annual estimated exposure to inorganic arsenic in drinking water over each participant's lifetime. The use of inorganic arsenic levels in residential drinking water to assess exposure was supported by findings from our research that correlated annual predicted arsenic exposure estimates with temporally concurrent speciated inorganic arsenic concentrations in historically collected samples. Between 2006 and 2008, study subjects or next of kin of deceased subjects as designated in SLVDS (14.2%) were contacted by mail with information about the study, followed by a call to set up an appointment for an interview and water sample collection. During the interviews (n = 357; 64%) we collected data about past residences, past workplace/school locations, and history of drinking-water consumption at each location. For each location, data included addresses, residence dates, water source (well or public), water treatment device (if yes, type, model number), number of glasses of water consumed per day (nonbottled water), number of glasses of beverages made with water (e.g., coffee, tea, juice), whether they typically cooked with water from the tap in the home, and whether they had a vegetable garden (if yes, which vegetables). For subjects or next of kin who were not able to be located for an interview (n = 189, 33%), we used triangulation methods incorporating records from the county assessor's office and SLVDS tracking database to reconstruct residential history. In brief, 189 participants were not interviewed (n = 2,023 person-years); therefore, the SLVDS contact tracking database was used to determine residence history back to 1975 and earlier for participants who reported living at the residence listed in 1975, before 1975. This left 914 person-years across 98 subjects still missing residential history who therefore were further investigated in county clerk records. In the end, there were 18 subjects with partial residential history (before 1975) (n = 126 person-years) with missing residential locations, and these were assigned the mean value for the last known city of residence.
Drinking-water samples were collected from the residential kitchen tap at time of interview and analyzed by the chemistry laboratory of the Colorado Department of Public Health and Environment using standard ion chromatography and inductively coupled plasma mass spectrometry with a detection limit of 1 μg/L. Samples with arsenic concentrations below the detection limit were given a value of half of the detection limit similar to other studies (Ayotte et al. 2006). Samples collected from private wells were assigned geographic coordinates using a global positioning system (GPS) unit (n = 248). Elsewhere we have provided methods for determining and validating the temporal and spatial variability of inorganic arsenic in groundwater in the SLV (James et al. 2013). In brief, findings indicate that naturally occurring inorganic arsenic concentrations in groundwater are stable over decades [consistent with other research (e.g., Steinmaus 2005)], justifying the use of geospatial models based on the mean arsenic concentration in individual private wells over decades to predict spatial variability of inorganic arsenic in groundwater. This was supported by a correlation analysis in a 10% sample of observed and predicted values (ρ = 0.715; 95% CI: 0.67, 0.75) (James et al. 2014).
An exposure matrix was developed to estimate each participant's annual exposure to arsenic in drinking water. Each record included residential, employment, and school location and an estimate of the amount of water ingested and water arsenic concentration (either observed or predicted) for each participant for each year of life from birth to death or 1998, whichever came first. Three estimated exposure values were calculated for each year of the follow-up period 1984 through 1998 or CHD diagnosis, whichever came first. The three estimated exposure values—residential arsenic concentration (arsenic concentration in drinking water at residence), residential arsenic dose (residential arsenic concentration in drinking water times the amount of water consumed in liters per day), and total arsenic dose (residential arsenic dose plus workplace and/or school dose)—were each defined based on a time-weighted average (TWA). A time-weighted average (TWA) for each exposure metric was calculated by dividing cumulative per-person arsenic exposure by the number of years in each participant's lifetime to get an annuitized exposure per year (Meliker et al. 2010).
To determine which TWA exposure estimate best approximated biologic exposure, we correlated these with speciated arsenic concentrations in historically collected urine samples (collected 1984–1991), adjusting for sex and creatinine (James et al. 2013). In brief, estimates of residential arsenic concentration (R = 0.37; ρ = 0.61) were the strongest correlates of the sum of the toxic urine arsenic species [As, As, dimethylarsinic acid (DMA), monomethylarsinic acid (MMA)], as opposed to estimates that included water consumption (residential dose) (R = 0.21; ρ = 0.46) or exposure at work or school (total dose) (R = 0.23; ρ = 0.48).
Statistical Analyses
We used a Cox proportional-hazards model incorporating a robust variance estimator specific for case-cohort study designs (Barlow et al. 1999) to examine the association between TWA inorganic arsenic exposure and diagnosis of or death from CHD. We scaled the continuous arsenic exposure estimate to the interquartile range (IQR) (15 μg/L), along with other continuous covariates, similar to methods used by Lin and Huang (1995).
As described above, participants had longitudinal data from two to four study visits, including information on known risk factors for CHD. Risk factors for CHD believed to be independent of the mechanistic pathways proposed for arsenic were included in the proportional-hazards multivariate model as time-dependent covariates [lipid measurements, body mass index (BMI), physical activity, smoking, and alcohol and water consumption]. The univariate model included TWA inorganic arsenic as a continuous value scaled to the IQR (15 μg/L). Person-years and exposure were censored for CHD cases at time of diagnosis. The full model included the demographic risk factors ethnicity (white non-Hispanic/Hispanic), sex (male/female), and annual household income (high ≥ $20,000/low < $20,000); the known risk factors first-degree family history of CHD (no/yes), BMI (IQR scaled, median = 26.7, IQR = 23.8–29.3), diabetes diagnosis before CHD (no/yes); behavioral risk factors including current smoking status (no/yes), alcohol consumption (low ≤ 168 g/week/high > 168 g/week), and physical activity level (active/sedentary) (Mayer 1991); and continuous clinical risk factors including serum lipid measurements [high-density lipoprotein (HDL), triglycerides, and low-density lipoprotein (LDL) in milligrams per deciliter], hypertension (blood pressure > 140/90 or use of anti-hypertensive medicine) and folate and selenium intake (micrograms). Triglyceride and HDL levels were determined using enzymatic methods, and LDL levels were calculated using the Friedewald equation (Friedewald et al. 1972); folate and selenium intake were estimated based on 24-hr diet recall involving two- and three-dimensional visual aids for portion approximation. Nutritional analysis was based on version 14 of the Nutrition Coordinating Center's nutrient database released in 1987 (www.ncc.umn.edu/products/database.html). Vitamin supplement use was assessed through self-report using vitamin bottle labels.
In addition, a final parsimonious model incorporating statistically significant covariates to the model based on a 10% change to the HR for TWA arsenic exposure. Known independent risk factors, including sex and family history, that were significantly associated with the outcome were maintained in all models regardless of whether or not they met statistical criteria for confounding. Covariate data were assessed at each clinic visit (up to four visits) from 1984 through 1998. The risk associated with covariates was based on the covariate value at the clinic visit before the time of the CHD event. Clinic visits assessed all behavioral and clinical values for all covariates in this analysis. Missing data occurred when participants did not attend follow-up clinic visits; however, this was a very small number (7%), so covariate values from the baseline visit were used.
We also assessed the HR for CHD across arsenic exposure groups (TWA concentration, 20–30 μg/L, 30–45 μg/L, and 45–88 μg/L relative to < 20 μg/L). The cut-points for the exposure groups are based on arsenic concentrations in past research with significant associations with CHD (Chen Y et al. 2013; Moon et al. 2013).
We assessed whether hypertension might confound the association between inorganic arsenic exposure and CHD by reanalyzing the final model with hypertension as a dichotomous covariate. Last, we completed secondary analyses to data collection methods and exposure estimates. The first was an agreement analysis on 5% of the interviewed participants (n = 28) that compared residential address and year reported in the interview by the participant with the residential county clerk records. We compared residence and year for the years 1955–1985 as reported by both sources. The next secondary analysis was completed in a limited cohort (n = 462) to confirm any association found based on residential history using mean speciated urinary arsenic concentrations (As, As, MMA, and DMA). To complete this analysis, we compared urinary arsenic concentrations between cases and noncases as a secondary analysis. We used SAS 9.2 (PROC PHREG; SAS Institute Inc, Cary, NC) for the statistical analyses. We complied with all applicable requirements of national and international regulations including approval from institutional review board, and human participants provided written informed consent before participating in the study.
