Table 3. Multivariatea Cox regression analysis.

Variable	HR (95% CI)b	P
Base model
Age (years)		0.146
⩾50 vs <50	0.61 (0.31–1.19)
Tumour size		0.140
pT2 vs pT1	1.95 (0.98–3.86)
pT3+4 vs pT1	1.96 (0.90–4.28)
Histological grade		0.019
III vs I/II	2.20 (1.21–4.03)
No. of involved lymph nodes		<0.001
1–3 vs node negative	6.27 (2.89–13.6)
⩾4 vs node negative	13.1 (6.12–27.9)
ER statusc		0.852
Positive vs negative	1.08 (0.49–2.39)
PgR statusc		0.858
Positive vs negative	0.96 (0.58–1.58)
Additions to model
CA9		0.018
Log-transformed	0.77 (0.62–0.96)
Adjuvant radiotherapy		0.128
Yes vs no	0.60 (0.31–1.16)
Adjuvant chemotherapy		0.828
Yes vs no	0.76 (0.06–9.25)
Adjuvant endocrine therapy		0.889
Yes vs no	1.15 (0.17–7.96)
Interaction, radiotherapy × CA9		0.624
Yes and log-transformed vs no and log-transformed	0.89 (0.54–1.44)
Interaction, chemotherapy × CA9		0.009
Yes and log-transformed vs no and log-transformed	1.31 (1.07–1.61)
Interaction, endocrine therapy × CA9		<0.001
Yes and log-transformed vs no and log-transformed	1.41 (1.20–1.66)

The analysis was conducted in two blocks. Firstly, a model including all established clinicopathological factors was fitted. In a second block, the contributions ofCA9 (as a continuous, log-transformed variable), treatment and interactions were investigated.

^aThe final multivariate model included 220 patients.

^bHazard ratio (95% CI) of multivariate analysis.

^cCut-off point used for ER and PgR, 10 fmol mg⁻¹ protein.