854 J. Manjer et al.
RESULTS
Comedo carcinoma and treatment by complete
mastectomy was most common among never smokers
with stage 0-I tumours, while comedo carcinoma and
treatment with adjuvant chemotherapy was most
common among ex-smokers in stage II tumours,
(Table III).
Breast cancer mortality in women with stage 0-I
tumours was 443/105 person-years, in stage II: 2225/
105, in stage III: 11374/105 and in stage IV: 32927/105.
In all these groups, some women died of breast
cancer within three years of diagnosis but others
survived until the end of follow-up, more than 10
years (among women with stage IV tumours the
longest survival was 7.6 years).
The odds ratio for a stage II compared with a stage
0-I tumour was 0.79 (0.52 to 1.21) in smokers
compared with 1.00 in never smokers and adjusted
for age at diagnosis, menopausal status, BMI, and
mode of detection. Corresponding comparison for ex-
smokers showed an odds ratio of 0.90 (0.56 to 1.44).
Smokers who died of breast cancer were compared
with those who did not, to assess potential modifiers,
(Tables IV and V). The influence of these factors on
survival was evaluated in a Cox’s proportional hazards
model with adjustment for age and stage. In the
analysis of the effect of adjuvant hormonal and
chemotherapy, menopausal status was added as a third
covariate. The only factor that significantly affected
survival in smokers was use of HRT, RR for ever use
compared with never use: 0.34 (0.12 to 0.97).
Breast cancer mortality was 1347/105 in those who
had never smoked, 1941/105 in smokers, and 1493/105
person-years in ex-smokers. Crude relative risk,
compared with those who had never smoked, were
1.44 (1.01 to 2.06) and 1.13 (0.66 to 1.94) for smokers
and ex-smokers, respectively, (Table I).
All-cause mortality was 3508/105 in never smokers,
4246/105 in smokers, and 2892/105 person-years in ex-
smokers. The age-adjusted relative risk was 1.46 (1.15
to 1.86) for smokers and 0.98 (0.67 to 1.44) for ex-
smokers.
The increased breast cancer mortality associated
with smoking remained significant after adjustment for
age and stage at diagnosis, RR 2.14 (1.47 to 3.10),
(Fig. 1), and other potential confounders, (Tables I and
II). The extended Cox’s analysis was limited to 555
cases because of missing information. Exclusion of 34
cases who had been diagnosed with breast cancer
before the screening trial did not change the associa-
tion. The analyses were repeated using only the 35
deaths from breast cancer that had been classified
according to ICD-9 and had been confirmed by
necropsy. In these 35 women the RR of death from
breast cancer, as compared to never smokers was 3.31
(1.57 to 6.95) for current smokers and 1.62 (0.51 to
5.07) for ex-smokers.
Breast cancer mortality was higher among the 242
cases for whom there was no information on smoking
5 person-years compared with 1517/105
habits, 2521/10
for the study group. The prevalence of stage III and IV
tumours was similarly higher in this group, (Table VI).
If all 242 women with missing information on smoking
had been never smokers, the RR of breast cancer death,
adjusted for stage and menopausal status, in current
compared with never smokers would have been 1.61
(1.15 to 2.26).
DISCUSSION
We conclude that differences in exposure to smoking
contribute to the heterogeneity in long term survival of
women with breast cancer. The appropriateness of that
conclusion should be assessed in relation to certain
methodological issues.
Differences between groups with regard to com-
pleteness of follow up and confirmation of end-points
could have confounded the results. As vital status was
updated on each patient and it has been confirmed that
cause of death in women with breast cancer is accurate
(3, 16), we consider it unlikely that the results were
confounded by biased retrieval or low validity of end-
points. In addition, when we restricted the analysis to
deaths that had been coded according to ICD-9 and
which had been confirmed by necropsy, the main
findings remained.
Misclassification with regard to exposure to smoking
is another relevant issue. Some of the women who did
not smoke were described as non-smokers in the
hospital records. These women were counted as never
Fig. 1. Survival from breast cancer according to smoking
status at the time of diagnosis.
Eur J Surg 166