Table 2.

Odds of a square becoming the location of a subsequent breast tumour in relation to its percent mammographic density 5 years prior to diagnosis

Grid length (cm)	Within-breast quartiles of square-specific percent MD	No. of squares in which tumour arises a	OR (95% CI) b
4	1 (lowest)	26	1
	2	48	1.5 (0.9, 2.5)
	3	66	2.5 (1.6, 4.1)
	4 (highest)	74	2.6 (1.6, 4.2)
			P for linear trend < 0.001
3	1 (lowest)	19	1
	2	44	2.3 (1.3, 4.0)
	3	69	3.9 (2.3, 6.4)
	4 (highest)	91	4.6 (2.8, 7.6)
			P for linear trend < 0.001
2	1 (lowest)	15	1
	2	32	2.1 (1.1, 3.8)
	3	71	4.9 (2.8, 8.6)
	4 (highest)	111	6.4 (3.7, 11.1)
			P for linear trend < 0.001
1	1 (lowest)	3	1
	2	32	6.1 (1.9, 20.1)
	3	63	16.6 (5.2, 53.2)
	4 (highest)	130	25.5 (8.1, 80.3)
			P for linear trend < 0.001

MD=mammographic density

^aThe total number is less than 231 because tumour-squares for which no control squares of similar size (i.e. within 10%) could be identified were excluded (see Methods section).

^bOdds ratio (OR) and 95% confidence interval (CI) estimated using a conditional logistic regression model where the matching set is a woman’s pre-diagnostic breast consisting of a square where the tumour will subsequently originate (tumour-square) and several tumour-free squares (control-squares) (see Methods section).