Table 3.

Association between absolute change in mammographic percent density and breast cancer-specific death among ER-positive patients treated with tamoxifen, overall and stratified by baseline percent density

Change in percent density	Case patients (N) / Control patients (N)	OR (95% CI)*	P†	OR (95% CI)*‡	P†‡	OR (95% CI)*§	P†§
Overall
T1||: >-0.5	36 / 86	Ref.	.005	Ref.	.01	Ref.	.04
T2: -8.7 to -0.5	45 / 82	1.36 (0.79 to 2.34)		1.38 (0.80 to 2.40)		1.27 (0.71 to 2.25)
T3: <-8.7	16 / 84	0.44 (0.22 to 0.88)		0.49 (0.23 to 1.02)		0.47 (0.21 to 1.03)
Stratified by baseline percent density
Baseline density <20%
T1: >-0.5	16 / 46	Ref.	—¶	Ref.	—¶	Ref.	—¶
T2: -8.7 to -0.5	22 / 29	2.23 (0.99 to 5.03)		2.82 (1.17 to 6.76)		2.22 (0.88 to 5.62)
T3: <-8.7	0 / 11	—¶		—¶		—¶
Baseline density 20% to <37%
T1: >-0.5	15 / 23	Ref.	.16	Ref.	.16	Ref.	.23
T2: -8.7 to -0.5	15 / 32	0.69 (0.28 to 1.71)		0.70 (0.28 to 1.72)		0.78 (0.31 to 1.96)
T3: <-8.7	7 / 29	0.35 (0.12 to 1.02)		0.35 (0.12 to 1.02)		0.38 (0.13 to 1.15)
Baseline density ≥37%
T1: >-0.5	5 / 17	Ref.	.38	Ref.	.38	Ref.	.43
T2: -8.7 to -0.5	8 / 21	1.32 (0.35 to 4.94)		1.34 (0.36 to 5.02)		1.31 (0.30 to 5.68)
T3: <-8.7	9 / 44	0.60 (0.17 to 2.12)		0.59 (0.17 to 2.11)		0.57 (0.14 to 2.38)

* Odds ratio and 95% confidence interval. Overall associations estimated using conditional logistic regression; stratified associations were estimated using unconditional logistic regression adjusted for age at diagnosis, stage at diagnosis, year of diagnosis, and case follow-up time. CI = confidence interval; ER = estrogen receptor; OR = odds ratio.

† Two-sided P _{heterogeneity}, based on the Wald test.

‡ Multivariable model, adjusted for baseline percent density.

§ Multivariable model, adjusted for baseline percent density and duration of tamoxifen use.

|| Tertiles, based on the distribution of change in percent density among control patients. Positive values indicate an increase in percent density and negative values indicate a reduction in percent density.

¶ Not estimated.