Table 2.

Association of p, p’-DDT with breast cancer stratified by age of “first exposure” in 1945, the year DDT became widely available in the United States for two prospective case-control samples nested within the Child Health and Development Studies cohort

Model	Current case-control sample (n = 153 cases/432 controls) for early postmenopausal breast cancer diagnoses from ages 50 to 54 y				Prior case-control sample (n = 129 cases/129controls) for premenopausal breast cancer diagnoses before age 50 y Cohn et al., 2007 (5)
	All ages	Younger than age 3 y	Age 3 y and older	P interaction with age§	All ages	Younger than age 14 y¶	Age 14 y and older¶
	OR (95% CI)	OR (95% CI)	OR (95% CI)		OR (95% CI)	OR (95% CI)	OR (95% CI)
Model 1: includes all DDTs*
log2(p, p’-DDT)	1.95 (1.34 to 2.83)	0.69 (0.26 to 1.79)	2.71 (1.72 to 4.27)	.01	—	—	—
Model 2: excludes DDE†
log2(p, p’-DDT)	1.99 (1.48 to 2.67)	0.56 (0.26 to 1.19)	2.83 (1.96 to 4.10)	.01	—	—	—
Model 3: DDT tertiles‡
Tertile 1	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)	—	—	—	—
Tertile 2	0.97 (0.59 to 1.60)	0.11 (0.01 to 0.91)	1.30 (0.75 to 2.25)	.07	—	—	—
Tertile 3	1.52 (0.83 to 2.77)	0.10 (0.01 to 0.96)	2.17 (1.13 to 4.19)	.02	—	—	—
Model 4: DDT tertiles‖
Tertile 1	—	—	—	—	1.00 (Reference)	1.00 (Reference)	1.00 (Reference)
Tertile 2	—	—	—	—	1.92 (0.93 to 3.97)	2.77 (1.13 to 6.84)	0.68 (0.14 to 3.30)
Tertile 3	—	—	—	—	2.79 (1.15 to 6.72)	5.42 (1.71 to 17.19)	0.62 (0.12 to 3.18)

^*Model #1 includes: p, p’-DDT(log₂-transformed as a continuous variable), p, p’-DDE (log₂-transformed as a continuous variable), and o, p’-DDT (log₂-transformed as a continuous variable), year of blood draw (continuous), and parity (continuous). The DDT odds ratio represents a one-unit change in log₂(p, p’-DDT) corresponding to an estimated effect for a twofold increase in p, p’-DDT, a range encompassed within the interquartile range of the study sample (see Table 1). CI, confidence interval; OR, odds ratio estimated by conditional logistic regression.

^†Model #2 deletes p, p’-DDE as an independent variable because it was neither a confounder nor a predictor and associations for other variables are otherwise the same as in Model #1. As for Model 1, the OR represents a twofold increase in p, p’-DDT.

^‡Model #3 includes two indicator variables for tertiles 2 and 3 of p, p’-DDT where tertile 1 was the reference category (tertile 1, <8.09 μg/L; tertile 2, 8.09–13.09 μg/L; tertile 3, >13.90 μg/L as coded in our previous publication [Cohn et al., 2007 (5)], o’, p’-DDT represented as a 3-cateogry ordinal variable coded at tertile medians as described in Table 4 of our previous publication [Cohn et al., 2007 (5)], year of blood draw (continuous), and parity (continuous).

^§ P values for product terms between p, p’-DDT and age in 1945 (proxy for age at first exposure), dichotomized as younger than 3 years versus ≥3 years. For Models 1 and 2, an interaction term was added to the variables listed above to represent the product of continuous log₂(p, p’-DDT) with age at first exposure. For Model 3, interactions were estimated using two product terms for each p, p’-DDT tertile with age at first exposure, dichotomized as younger than 3 years versus ≥3 years.

^‖Model #4 included indicator variables for tertiles 2 and 3 of p, p’-DDT, where tertile 1 was the reference category (as described for Model #3), o’, p’-DDT represented as a three-cateogry ordinal variable coded at tertile medians (as described for Model #3), and year of blood draw (continuous).

^¶ P  = .02 for interaction between age at first exposure and p, p’-DDT estimated by a product term between a dichotomous variable for age in 1945 (<14 vs. ≥14 years) and p, p’-DDT (continuous) in a conditional logistic regression model that included o’, p’-DDT and year of blood draw [Cohn et al., 2007 (5)].