ARSENIC HEALTH EFFECTS RESEARCH PROGRAM 
PROJECTS 
Berkeley 
University of California 
Perchlorate Reanalysis 



APPENDICES: 













Appendix 1. Odds ratio calculations for perchlorate exposure and having a high TSH in measurements collected less than 18 hours of age in Kelsh et al., 2003. 

We used the following methods and the data in Tables 3, 4 and 6 of the Kelsh et al. article to estimate the odds ratio for an elevated TSH in measurements collected at age 18 hours or less comparing newborns in Redlands to those from San Bernardino and Riverside Counties as a whole. We used the counties as a whole as our comparison population since data on the number of neonates with and without elevated TSH were not provided for just those areas with no known perchlorate exposure. Including some exposed communities in our comparison group could bias results towards the null, although as we will show, this bias appeared relatively small. 

Data were not specifically provided on the numbers and proportions of perchlorate exposed and unexposed neonates who had elevated TSH in measurements collected before 18 hours of age. However, this information was provided for all neonates (regardless of the age of measurement) and for neonates who had TSH measurements collected after 18 hours of age. Subtracting the latter from the former allowed us to estimate the numbers needed for our calculations. These data are shown in Table S1. 









Table S1. The number of neonates with and without an elevated TSH in Redlands 

and in San Bernardino and Riverside Counties by age of sample collection in 

Kelsh et al., 2003. 















Redlands 

San Bernardino/Riverside 


High TSH 
Normal TSH 
Total 

High TSH 
Normal TSH 
Total 
Total 
44 
15,304 
15,348 

1,560 
694,407 
695,967 
≥ 18 hours 
6 
10,496 
10,502 

X 
461,802  X 
461,802 
< 18 hours 
38 
4,808 
4,846 

Y 
234,165  Y 
234,165 








The value of Y in Table S1 can be calculated by subtracting X from the total number of neonates with a high TSH in San Bernardino and Riverside Counties (i.e. Y = 1560  X). The value of X can be estimated using the odds ratio of 0.69 the authors reported for a high TSH level at ≥18 hours of age comparing subjects in Redlands to those in San Bernardino and Riverside communities without perchlorate. This was done using the standard odds ratio equation and solving for X. In other words, if: 

Odds ratio = ad / cb 






where the odds ratio is 0.69, a is the number of perchlorateexposed neonates with a high TSH (i.e. 6), b is the number of perchlorateexposed neonates with a normal TSH (i.e. 10,496), c = X = the number of unexposed neonates with a high TSH, and d = the number of unexposed neonates with a normal TSH (which equals 461,802 – X). Using these numbers and the odds ratio equation gives: 

0.69 = (6 * (461,802 – X))/ (X * 10,496) 




Solving for X gives a value of 385. Given this, Y = 1560 – X = 1175 and the odds ratio for a high TSH in subjects age < 18 hours comparing Redlands to San Bernardino and Riverside Counties is (38/4,808) / (1175/232,990) = 1.57 (95% CI, 1.14 – 2.16). 

It should be noted that our comparison group is all of San Bernardino and Riverside Counties while the odds ratio of 0.69 used in these calculations uses only the unexposed parts of these counties as the comparison population. Thus, our analysis is based on the assumption that the odds ratios using these two different comparison populations are similar. In fact, the authors state that the odds ratios for a high TSH using these different comparison groups were “essentially the same”, although the specific numbers are not reported. We further analyzed this assumption in several different ways. First, we calculated an unadjusted odds ratio for a high TSH for all subjects regardless of age at measurement comparing Redlands to San Bernardino and Riverside Counties as a whole. This odds ratio was (44)/(15,304) / (1,560/694,407) = 1.28 (95% CI, 0.95 – 1.73). This is very close to the odds ratio of 1.24 (95% CI, 0.89 – 1.68) reported by the authors using only those communities with no detectable perchlorate as the comparison group. This suggests that excluding those communities with detectable perchlorate levels from the comparison group causes little bias. The reason for this is unknown but it could occur if perchlorate levels in most excluded areas were much lower than in Redlands or contaminated wells in the excluded areas provided only a small fraction of the total drinking water in these areas. 

We also evaluated the possibility that using an odds ratio of 0.69 introduced error by performing a sensitivity analysis to see what would happen if values other than 0.69 were used. If we replaced the odds ratio of 0.69 with a lower odds ratio, the high TSH odds ratio for perchlorate exposure for subjects with measurements before 18 hours of age would be even higher than 1.57. If we entered an odds ratio of 1.0 instead of 0.69, our odds ratio for a high TSH measurement before 18 hours of age would be lower but still statistically significant (OR = 1.42; 95% CI, 1.03 – 1.96). 









Appendix 2. Odds ratio calculations for perchlorate exposure and having a low T4 in Kelsh et al., 2003. 

The estimated odds ratio for having a low T4 (typically defined as a T4 value below 9.0 ug/dl or part of the lowest 5% of the remaining daily tray samples) in Kelsh et al. (2003) was calculated using the data in their Table 3. As discussed above, T4 was measured in all newborns, but TSH was only measured if T4 levels were low. Thus, the number of people who had “TSH levels screened” represents the number of neonates with low T4 levels. These data are shown in Table S2.


Table S2. Data on estimated odds ratio calculations for low T4 in Redlands and in San Bernardino and Riverside Counties in Kelsh et al., 2003. 





Community 
Low T4 
Normal T4 
Total 




Redlands 
2,081 
13,267 
15,348 




San Bernardino/Riverside 
81,401 
614,566 
695,967 




Total 
83,482 
627,833 
711,315 












Based on the data in Table S2, the odds ratio for having a low T4 in Redlands compared to San Bernardino and Riverside Counties is (2,081/13,267) / (81,401 / 614,833) = 1.18 (95% CI, 1.13 – 1.24). 









Appendix 3. Odds ratio calculations for perchlorate exposure and having a high TSH in all subjects and in subjects with TSH measurements collected before 24 hours of age in Buffler et al., 2006. 

The data used to estimate the odds ratio for having a high TSH regardless of age at measurement comparing perchlorate exposed to unexposed communities are provided in Table 1 of Buffler et al. and shown in Table S3 below. Based on these data, the odds ratio for having a high TSH in all subjects comparing perchlorate exposed to unexposed communities is (147/50,179) / (537/291,394) = 1.59 (95% CI, 1.331.91). 

Table S3. Data for estimated odds ratio calculations for all subjects regardless of age at measurement in Buffler et al., 2006. 






TSH levels 






Perchlorate 
Elevated 
Normal 
Totals 




> 5 µg/L 
147 
50,179 
50,326 




Ł 5 µg/L 
537 
291,394 
291,931 




Totals 
684 
341,573 
342,257 












The corresponding odds ratio for only those subjects with TSH measurements before 24 hours of age can be estimated by subtracting the number of subjects with measurements collected at 24 hours or later (given in their Table 4) from the total number of subjects (given in their Table 1 and our Table S3). These data are shown in Table S4. (These are estimates since their Table 1 appears to include all subjects whereas their Table 4 appears to only include subjects who have data on all the covariates used in their adjusted analyses). Based on these data the odds ratio for having a high TSH at < 24 hours of age comparing perchlorate exposed and unexposed communities is (133/21,079)/(418/105,985) = 1.60 (95% CI, 1.321.94)


Table S4. Data for estimated odds ratio calculations for subjects with TSH measurements collected < 24 hours after birth in Buffler et al., 2006. 






TSH levels 






Perchlorate 
Elevated 
Normal 
Totals 




> 5 µg/L 
133 
21,079 
21,212 




< 5 µg/L 
418 
105,985 
106,403 




Totals 
551 
127,064 
127,615 




2470 Telegraph Avenue, Suite 301, Berkeley, CA 94704 Tel: (510) 9908354 Email: asrg@berkeley.edu 