Articles - Rogers et al.
to clean polyethylene bags for counting. All of the gamma-ray spectroscopy was per- formed using four high purity germanium detectors coupled to an 8192-channel Genie system operating on a VAX 3100 worksta- tion. Elemental concentrations were deter- mined using custom-made interactive peak fitting and analysis software (all hardware and software were from Canberra Industries, Inc, Meriden, CT). A fly ash standard reference material (SRM 1633), purchased from the National Institute of Standards and Technology (Gaithersburg, MD), was used to calculate elemental con- centrations. Two other reference materials, SRM 1571 (orchard leaves) and SRM 1 577a (bovine liver), were used to check the system stability. All samples, standards, and control samples were counted for 12 hr at a constant geometry. Additional details of the analytical procedure have been published elsewhere (26).
Exposure estimation. An accurate method for relating As and Cr concentra- tions measured in hair samples to exposure to these elements in water from wells G and H would have been to determine the amount of each element imbibed during the period when each donor's hair sample was grown. However, because As and Cr were not measured in water froms wells G and H when the wells were in use and because the volume of water from wells G and H consumed by each hair donor were also unknown, measurements of actual exposure to As and Cr could not be made. Instead, we investigated the relationship between As and Cr concentrations in hair samples and the donors' relative access to water from wells G and H. Access is defined here as the ratio of the amount of water from wells G and H to the total amount of municipal water delivered to a hair donor's house during the period when the hair sam- ple was grown. Access was calculated by combining estimates of the period of time when each hair sample was grown and the fraction of water supplied to a residence by wells G and H. The methods used in deter- mining these estimates are described below.
The hair growth period was estimated by dividing the length ofthe hair by an assumed growth rate of 1 cm/month (22). The dates corresponding to the oldest and youngest parts of the hair strands (i.e., the dates between which As and Cr cxposure would have been recorded in the hair sample) were calculated using the date that the hair was cut and estimates of the length of hair on a donor's head remaining after cutting. If the length of hair remaining after cutting was unknown (i.e., it was not supplied by the donor or the donor's parents), lengths were assumed based on the sex and age of the
donor. For boys, the length of hair remain- ing after cutting was assumed to be 4-6 cm; for girls younger and older than 2 years of age, values of 4-6 cm and 8-12 cm, respec- tively, were used.
For the 14 samples for which only the year of collection was known, the growth period was deliberately overestimated to ensure that the true growth period for the hair was included within the estimated growth period. The beginning of the hair growth period was calculated using the ear- liest possible month of hair cutting (anuary) and the longest estimate of hair length; the end of the hair growth period was calculated using the latest possible month (December) and the shortest esti- mate of hair length. The disadvantage of this technique is that the true growth peri- od for a sample was actually shorter than the estimated growth period; therefore, episodes of low or high access to wells G and H water cannot be resolved. Despite this distortion, this method was chosen over the alternative, which was to arbitrarily choose a shorter growth period that may or may not represent the true growth period.
Estimates of well water access were based on a model of the city's water distrib- ution system developed by Murphy (14). In the original model, well pumping records and data on the water distribution system in Woburn were used to estimate the rela- tive amounts of water arriving at a particu- lar location from wells G and H and the Horn Pond well fields (Fig. 1). Recently, hydraulic mixing calculations were incorpo- rated into the model, and source apportion- ment indices for various distribution nodes located throughout the city were calculated for each month that wells G and H were in use (28). In estimating the amount ofwater from wells G and H to which hair donors had access, we divided the sum of these published monthly exposure values by the number of months that the hair was esti- mated to have grown. Average monthly access estimates ranged from 0 (none of the water delivered to a residence during the period of hair growth was from wells G and H) to 1 (all of the water delivered to a resi- dence during the period of hair growth was supplied by wells G and H).
Arsenic. Overall, our results indicate that donors who had access to water from wells G and H did not have higher concentrations ofAs in their hair than donors who did not have access. A plot of As concentrations in hair versus a donor's relative access to wells G and H water is shown in Figure 2A. The figure shows that there is no apparent corre- lation between access and As concentrations
measured in hair. A least squares regression line through the data has a slope that is not significantly different from zero. Interesting- ly, a plot of the As concentrations in all hair samples versus the year that the hair was cut (Figure 3A) indicates that As levels in the hair of Woburn residents have decreased over the last 50 years. To determine whether there was a relationship between As concen- trations in hair and a donor's relative access once this temporal variation in As concen- trations was accounted for, multivariable regression was performed. Since As concen- trations are linearly related to year (Figure 3A), the form ofthe regression was
[As]hai, = Bo + B, x year + B2 x access + error
The results in Table 1 show that As concentrations in hair were statistically dependent (at the 95% confidence level) upon year, but not upon access. The R2 value for the model was 0.19.
The arithmetic and geometric mean concentrations of As in hair samples are shown in Table 2. The samples are grouped according to the year that the hair was cut and the donor's access to wells G and H water. As is evident from Figure 4A, the As concentrations in the hair of donors with access (i.e., relative access estimate >0) and donors without access to the well water are log-normally distributed; thus, the geo- metric mean concentrations are the more appropriate measure to compare the differ- ent groups. The arithmetic mean concen- trations were calculated so that we could compare our results with other studies that have reported their findings in terms of arithmetic mean concentrations. The geo- metric mean concentration (GSD) ofAs in the hair of residents who had access to water from wells G and H (1964-1979, with access) was 0.14 (2.6) j'g/g (n = 27), while for the concurrent control group (1964-1979, no access) the mean concen- tration was 0.18 (1.6) pg/g (n = 9) (Table 2). The mean concentration of As in the hair of all of the controls (1938-1994, no access) was 0.13 (3.0) j'g/g (n = 55).
Chromium. The plots of Cr concentra- tions in hair samples versus relative access to water from wells G and H and versus the year that the hair was cut are shown in Figures 2B and 3B, respectively. Although there appears to be an increase in the con- centrations of Cr in hair just after well G was turned on, well water access estimates and concentrations of Cr in hair are not significantly correlated. A least squares regression line through the data has a slope that is not significantly different from zero. Thus, as was the case for As, there is no indication that increased access to water
Volume 105, Number 10, October 1997 * Environmental Health Perspectives