Regional Differences in Breast Cancer Survival Are Correlated With Differences in Differentiation and Rate of Proliferation JAN P. A. BAAK, PHD, FRCPATH, ELS C. M. WISSE-BREKELMANS, BSc, PIET H. J. KURVER, MD, LEO H. M. VAN GORP, MD, FEJA J. VOORHORST, MD, PHD, AND OLLI S. MIETTINEN, MD Patients with invasive ductal breast cancer and with 5 to 12 years of follow-up,
identified
from two pathology
laboratories
serving
hospitals in two distinct but fairly close regions, were studied for differences
in length of survival. In the years when the cases were
diagnosed,
population
screening was not performed,
temic therapy was not administered, irradiation
protocols
two laboratories.
and surgical
were similar in the hospitals
There
was a significant
survival between the two regional
groups
served by the
difference (Nl
adjuvant systreatment and in length of
= 160, N2 = 111;
P = .006). Survival rate at 10 years in the two regions was 48% and 69%. Distribution
of age, tumor size, and lymph node status (as
negative or positive as well as number of positive nodes) were similar, but quantitative and qualitative microscopic
features differed.
Patients from the region in which the prognosis
was less favorable
had the higher median values for the mitotic activity index (14 u 4;
P < .OOOl) and for nuclear area (59.2 u 38.2; P < .OOOl). Nuclear and histologic grade distributions regions.
Logistic
regression
survival differences
were also different between the
analysis confirmed
are correlated
that the regional
with the microscopic
features,
even after adjustment for age, tumor size, and lymph node status. Comparison
of breast cancers from the periods
1988 to 1989 in one of the two regions and microscopic proliferation results, apart from indicating survival rate correlated
tive microscopic significant
that the clinical
features were similar over time. These regional
to differences
teristics of the disease between port to previous
1970 to 1974 and
revealed
variation in breast cancer in the microscopic
regions,
provide
charac-
additional
sup-
reports that qualitative and, especially, quantitadifferentiation
bearing
and proliferation
on the prognosis
HUM PATHOL 23:989-992.
Copyright
features
of breast cancer 0
1992 by W.B.
have
patients. Saunders
Company
In the past decade it has become clear that multiple adjuvant chemotherapy in breast cancer improves longterm prognosis, especially for premenopausal patients.’ From the Institute of Pathology, Free University Hospital, and the Department of Theory of Medicine, Epidemiology, and Biostatistics, Medical Faculty, Free University, Amsterdam, The Netherlands; BMA, de Meern, The Netherlands; the Department of Pathology, St Anthonius Hospital, Nieuwegein, The Netherlands; and the Department of Epidemiology and Biostatistics, McGill University, Montreal, Canada. Accepted for publication October 15, 199 1. Key words: breast cancer, survival, proliferation. geography. Supported by grant 28-1398 of the Praeventiefonds Stichting Bevordering Diagnostische Morphometrie and by EMGO Institute, Free University, Amsterdam, The Netherlands. Address correspondence and reprint requests to J. P. A. Baak, MD, Free University Hospital, De Boelelaan 1117, 1081 HV Amsterdam, The Netherlands. Copyright 0 1992 by W.B. Saunders Company 0046-8177/92/2309-0004$5.00/O
989
As the side effects of therapy are serious, attempts to develop ever more accurate prognostic indicators are intensive, with a view to developing ever more discriminating indications for therapy. Lymph node status, although widely used, is not very discriminating as a predictor of clinical course.’ Judgmental microscopic features, such as nuclear and histologic grade, are wellknown prognostic indicators, although their reproducibility may be wanting. Quantitative microscopic features, especially mitotic activity index and nuclear area,’ have great prognostic value. Because of the objectivity, reproducibility, and prognostic value of quantitative cell and tissue analyses, a multicenter program (Multicenter Morphometric Mammary Carcinoma Project) designed to consolidate knowledge of and experience with morphometric and DNA flow cytometric analyses of breast cancers was started in the Netherlands in 1987.4 All patients diagnosed between 1988 and 1989 in the 32 participating centers as having an invasive breast cancer have been enrolled in that program. Quantitative pathologic assessment of the tumors was performed within 1 month of the diagnosis. Accuracy of the samples and quantification was carefully checked by means of blind duplications of the readings in a central laboratory. From this new material evidence revealed that the values of two morphometric features with an important bearing on prognosis (mitotic activity index and nuclear area) showed geographic variations, This led us to reanalyze the material of a previous retrospective study, 5 in which morphometric analyses were performed on consecutive invasive breast cancers with long follow-up, diagnosed in the early 1970s in two pathology centers serving hospitals in two distinct but fairly close (50 miles apart) geographic regions. In 1985, we did not consider the possible influence of the two regions and the breast cancers were thus grouped together. No population screening for breast cancer was performed in The Netherlands during the years when the diagnoses had been made (1970 to 1974). We therefore undertook to analyze qualitative and quantitative microscopic features of those early patients from the point of view of finding regional differences in these characteristics, which could provide an explanation for the differences in length of survival.
HUMAN PATHOLOGY
PATIENTS
Volume 23. No. 9 (September
AND METHODS
Patients The study subjects consisted of all admissible patients diagnosed as having invasive breast cancer in the Department of Pathology, Stichting Samenwerking Delftse Ziekenhuizen, Delft (n = 160) (at that time serving three hospitals with approximately 60 new breast cancer cases per year) or the Department of Pathology, the Anthonius Hospital, Utrecht (n = 111) (with approximately 50 new breast cancers annually), both in The Netherlands, between 1970 and 1974. The regions served by these two departments will be referred to henceforth respectively. Admission into the as “Delft” and “Utrecht,” study series required absence of preoperative radiotherapy and adequate axillary lymph node dissection in addition to availability of the histologic specimen and adequate follow-up (at least 5 years). Twenty patients did not fulfil these conditions. In four cases (three from Delft and one from Utrecht) axillary dissection was either not performed or was inadequate, one patient (from Delft) received preoperative radiotherapy, six patients were lost to follow-up, and nine patients (five from Delft and four from Utrecht) died from non-cancer-related causes between 12 and 30 months postdiagnosis. Although these nine patients could have been left in the study, we felt it better to disqualify them (also because of the small number involved). Therefore, only death due to breast cancer occurred in the two patient groups under study. In total, 99 of the 271 women died from breast cancer, 69 from Delft and 30 from Utrecht. Thus, 160 and 111 patients from Delft and Utrecht, respectively, were left for analysis. All patients were treated with radical or modified radical mastectomy. When central-medial tumors or positive axillary lymph nodes were found, operation was followed by adjuvant and local regional radiotherapy. Patients with subsequently diagnosed distant metastases received cytostatic and/or hormonal therapy, but none received any form of adjuvant systemic therapy. Minimum follow-up was 5.0 years after operation; maximum follow-up was 12 years. To compare stability of the different clinical and quantitative microscopic features over time, we compared the data of the 1970-1974 cancer cases from Utrecht with those of the 1988-l 989 period (unfortunately, cancer cases from the Delft region from 1988 and 1989 were not available). As the Anthonius Hospital had moved to another nearby city since 1974, we used as study material the breast cancer cases diagnosed in the pathology laboratory of the Diaconesse Ziekenhuis in Utrecht, which is very close (less than 1 km in distance) to the old Anthonius Ziekenhuis. This is acceptable, as the pathology laboratory of the old Anthonius Ziekenhuis and that of the 1988-l 989 Diaconesse Ziekenhuis serve the same area.
Microscopic
1992)
thelium (usually present at the periphery) in the invasive ductal type carcinomas. No appreciable differences in the selection of these areas were found among different observers. Counting of mitoses was performed in 10 neighboring fields defined at X40 magnification with a X40 objective (numerical aperture, 0.75; field diameter, 450 pm). Because of the prognostic significance of the mitotic activity index, strict criteria were used for the definition and counting of mitotic figures7 As it nray be difficult to distinguish between anaphase or telophase mitoses in histologic sections, two sets of chromosomes arranged in parallel were regarded as two mitotic figures. Thus, mitotic activity index is the total number of sharply defined mitoses in 10 high-power fields. The areas of nuclei were measured at X2,000 final magnification’ using a commercially available graphic tablet (MOP-Videoplan; Kontron, Munich-Eching, Germany). The mean and standard deviation values of the nuclear areas were calculated for each tumor. DNA flow cytometric analysis was not performed because tissue blocks were not available to the investigators. This seems to be of minor importance for two reasons. First, DNA flow cytometry has very limited prognostic value to the morphometric features.’ Second, in our experience, the number of inadequate histograms is considerably higher in paraffin material older than 10 years.
Statistical
Analysis
The log rank test (generalized salvage by Mantel-&x in the BMDP program, see ref 10) was used for survival analysis and the Cox proportional hazards model was used to cope with a subset of covariates. Both metastasis-free survival and survival to death from metastasis were analyzed, but as they appeared strongly correlated, only the results of survival to death will be presented. Differences in age, tumor size, lymph node involvement, and microscopic prognosticators between the two regions were tested by Student’s t-test and the chi-squared test, if appropriate. In these computations all continuous variables (except age) were used after transformation into their square roots. Differences in the microscopic prognosticators between the two regions could be confoundingly modified by variables such as age, tumor size, and lymph node involvement. The contrasts between the regions and the microscopic variables of interest were derived separately with and without adjustment for age, tumor size, and lymph node involvement. For this purpose we used a logistic regression analysis with the logit of the rate of occurrence of the outcome phenomenon (Utrecht cases) as a dependent parameter.
RESULTS The
mean length of survival in the Delft cases was and 8.9 years in the Utrecht cases. Figure 1 shows the survival functions of the patient groups in the two regions. A crude analysis of the survival patterns showed a statistically significant difference between the two regions (Mantel-Cox = 7.7; P = .006). Survival predicting the covariates of age, tumor size, and lymph node involvement concomitant with the regions were thereafter included in the Cox proportional hazards model. In this model all of these determinants but age were significantly related to length of survival (region-related adjusted relative risk: 0.48, 95% [confidence interval, 0.31 to 0.751). Inclusion of the microscopic prognosticators concomitant with region and age resulted in a model with a likelihood of the same magnitude. How-
Studies
7.7 years
Microscopic studies were based on 5 pm-thick hematoxylin-eosin-stained paraffin sections of 4% neutral formalinfixed tissue slices taken from the periphery of the tumor. Apart from axillary lymph node status (as negative or positive and number of positive nodes), tumor diameter, and nuclear and histologic grades (according to the World Health Organization criteria, including a categorized three-point evaluation of the number of mitoses as absent, moderate, or high [more than l/10 high-power fields]), the morphometric features and parameters mentioned below were investigated. For a detailed description of the measurements, reference is made to our previous publications.5,6 Briefly, all measurements were performed in the (subjectively) most cellular areas of the tumor, with at least 50% of the area occupied by epi-
990
GEOGRAPHIC VARIATIONS IN BREAST CANCER SURVIVAL (Baak et al)
TABLE 1.
Median Values and Categoric Distributions for Indicators and Probability Values, Contrasting the Two Regions
REGION 2
Delft (n ;%yO)
REGION 1 Mantel-Cox=‘7.69 P=O.O056
20
0
2
YEARS n
Covariates Age (yr) (median) Tumor size (median) Lymph node involvement (present) Microscopic variates Mitotic activity index (median) Mean nuclear area (median) Histologic grade (no. and percentage) 1 2 3 Nuclear grade (no. and percentage) 1 2 3
4
6
8
10
111
102
93
88
27
18
Region 2
160
153
125
106
56
18
Region 1
1. Survival curves of the patients from Delft (region I> and Utrecht (region 2). FIGURE
ever, the region-related adjusted relative risk in this model (0.73, 95% [confidence interval, 0.44 to 1.211) was not significant. The results of these survival analyses with covariates mean that length of survival can be predicted by the morphometric features independent of the regions. On the other hand, if tumor size and lymph nodes status are included the region-related variable should be included also to obtain an accurate prediction. Thus, the microscopic features are stronger predictors of the outcome. Table 1 shows the medians (and the numbers and percentages in the ordinal-scale categories) for various covariates and microscopic variables for each of the two regions. The normally used clinical predictors (tumor size and lymph node involvement) were not appreciably different between the regions, although age difference could not be excluded (P = .08). In contrast, the qualitative and quantitative microscopic features were quite different. The patients in Delft, the region with a worse prognosis, had higher median values for the mitotic activity index and for nuclear area. Histologic and nuclear grades (if tested over all strata) were also more ominous in that region. In both regions the women who died had higher values for the mitotic activity index and nuclear area in general. Furthermore, these features did not differ between the survivors from Delft and Utrecht or between the nonsurvivors from the two regions. Logistic regression analysis shows that the differences in microscopic prognostic indicators were still present after adjustment for age, tumor size, and lymph node involvement in all variates. Table 2 compares the most important features of the 1973-1974 and 19881989 breast cancer cases in the favorable area (Utrecht). The results were very similar. Thus, the quantitative prognosticators are stable over time in that region.
Utrecht (n = 111) (%)
Probability Values*
53.5 2.0
58.0 2.5
.08 .ll
86 (53.8)
61 (55.0)
.85
14
4
identified
from two pathology
laboratories
serving
hospitals in two distinct but fairly close regions, were studied for differences
in length of survival. In the years when the cases were
diagnosed,
population
screening was not performed,
temic therapy was not administered, irradiation
protocols
two laboratories.
and surgical
were similar in the hospitals
There
was a significant
survival between the two regional
groups
served by the
difference (Nl
adjuvant systreatment and in length of
= 160, N2 = 111;
P = .006). Survival rate at 10 years in the two regions was 48% and 69%. Distribution
of age, tumor size, and lymph node status (as
negative or positive as well as number of positive nodes) were similar, but quantitative and qualitative microscopic
features differed.
Patients from the region in which the prognosis
was less favorable
had the higher median values for the mitotic activity index (14 u 4;
P < .OOOl) and for nuclear area (59.2 u 38.2; P < .OOOl). Nuclear and histologic grade distributions regions.
Logistic
regression
survival differences
were also different between the
analysis confirmed
are correlated
that the regional
with the microscopic
features,
even after adjustment for age, tumor size, and lymph node status. Comparison
of breast cancers from the periods
1988 to 1989 in one of the two regions and microscopic proliferation results, apart from indicating survival rate correlated
tive microscopic significant
that the clinical
features were similar over time. These regional
to differences
teristics of the disease between port to previous
1970 to 1974 and
revealed
variation in breast cancer in the microscopic
regions,
provide
charac-
additional
sup-
reports that qualitative and, especially, quantitadifferentiation
bearing
and proliferation
on the prognosis
HUM PATHOL 23:989-992.
Copyright
features
of breast cancer 0
1992 by W.B.
have
patients. Saunders
Company
In the past decade it has become clear that multiple adjuvant chemotherapy in breast cancer improves longterm prognosis, especially for premenopausal patients.’ From the Institute of Pathology, Free University Hospital, and the Department of Theory of Medicine, Epidemiology, and Biostatistics, Medical Faculty, Free University, Amsterdam, The Netherlands; BMA, de Meern, The Netherlands; the Department of Pathology, St Anthonius Hospital, Nieuwegein, The Netherlands; and the Department of Epidemiology and Biostatistics, McGill University, Montreal, Canada. Accepted for publication October 15, 199 1. Key words: breast cancer, survival, proliferation. geography. Supported by grant 28-1398 of the Praeventiefonds Stichting Bevordering Diagnostische Morphometrie and by EMGO Institute, Free University, Amsterdam, The Netherlands. Address correspondence and reprint requests to J. P. A. Baak, MD, Free University Hospital, De Boelelaan 1117, 1081 HV Amsterdam, The Netherlands. Copyright 0 1992 by W.B. Saunders Company 0046-8177/92/2309-0004$5.00/O
989
As the side effects of therapy are serious, attempts to develop ever more accurate prognostic indicators are intensive, with a view to developing ever more discriminating indications for therapy. Lymph node status, although widely used, is not very discriminating as a predictor of clinical course.’ Judgmental microscopic features, such as nuclear and histologic grade, are wellknown prognostic indicators, although their reproducibility may be wanting. Quantitative microscopic features, especially mitotic activity index and nuclear area,’ have great prognostic value. Because of the objectivity, reproducibility, and prognostic value of quantitative cell and tissue analyses, a multicenter program (Multicenter Morphometric Mammary Carcinoma Project) designed to consolidate knowledge of and experience with morphometric and DNA flow cytometric analyses of breast cancers was started in the Netherlands in 1987.4 All patients diagnosed between 1988 and 1989 in the 32 participating centers as having an invasive breast cancer have been enrolled in that program. Quantitative pathologic assessment of the tumors was performed within 1 month of the diagnosis. Accuracy of the samples and quantification was carefully checked by means of blind duplications of the readings in a central laboratory. From this new material evidence revealed that the values of two morphometric features with an important bearing on prognosis (mitotic activity index and nuclear area) showed geographic variations, This led us to reanalyze the material of a previous retrospective study, 5 in which morphometric analyses were performed on consecutive invasive breast cancers with long follow-up, diagnosed in the early 1970s in two pathology centers serving hospitals in two distinct but fairly close (50 miles apart) geographic regions. In 1985, we did not consider the possible influence of the two regions and the breast cancers were thus grouped together. No population screening for breast cancer was performed in The Netherlands during the years when the diagnoses had been made (1970 to 1974). We therefore undertook to analyze qualitative and quantitative microscopic features of those early patients from the point of view of finding regional differences in these characteristics, which could provide an explanation for the differences in length of survival.
HUMAN PATHOLOGY
PATIENTS
Volume 23. No. 9 (September
AND METHODS
Patients The study subjects consisted of all admissible patients diagnosed as having invasive breast cancer in the Department of Pathology, Stichting Samenwerking Delftse Ziekenhuizen, Delft (n = 160) (at that time serving three hospitals with approximately 60 new breast cancer cases per year) or the Department of Pathology, the Anthonius Hospital, Utrecht (n = 111) (with approximately 50 new breast cancers annually), both in The Netherlands, between 1970 and 1974. The regions served by these two departments will be referred to henceforth respectively. Admission into the as “Delft” and “Utrecht,” study series required absence of preoperative radiotherapy and adequate axillary lymph node dissection in addition to availability of the histologic specimen and adequate follow-up (at least 5 years). Twenty patients did not fulfil these conditions. In four cases (three from Delft and one from Utrecht) axillary dissection was either not performed or was inadequate, one patient (from Delft) received preoperative radiotherapy, six patients were lost to follow-up, and nine patients (five from Delft and four from Utrecht) died from non-cancer-related causes between 12 and 30 months postdiagnosis. Although these nine patients could have been left in the study, we felt it better to disqualify them (also because of the small number involved). Therefore, only death due to breast cancer occurred in the two patient groups under study. In total, 99 of the 271 women died from breast cancer, 69 from Delft and 30 from Utrecht. Thus, 160 and 111 patients from Delft and Utrecht, respectively, were left for analysis. All patients were treated with radical or modified radical mastectomy. When central-medial tumors or positive axillary lymph nodes were found, operation was followed by adjuvant and local regional radiotherapy. Patients with subsequently diagnosed distant metastases received cytostatic and/or hormonal therapy, but none received any form of adjuvant systemic therapy. Minimum follow-up was 5.0 years after operation; maximum follow-up was 12 years. To compare stability of the different clinical and quantitative microscopic features over time, we compared the data of the 1970-1974 cancer cases from Utrecht with those of the 1988-l 989 period (unfortunately, cancer cases from the Delft region from 1988 and 1989 were not available). As the Anthonius Hospital had moved to another nearby city since 1974, we used as study material the breast cancer cases diagnosed in the pathology laboratory of the Diaconesse Ziekenhuis in Utrecht, which is very close (less than 1 km in distance) to the old Anthonius Ziekenhuis. This is acceptable, as the pathology laboratory of the old Anthonius Ziekenhuis and that of the 1988-l 989 Diaconesse Ziekenhuis serve the same area.
Microscopic
1992)
thelium (usually present at the periphery) in the invasive ductal type carcinomas. No appreciable differences in the selection of these areas were found among different observers. Counting of mitoses was performed in 10 neighboring fields defined at X40 magnification with a X40 objective (numerical aperture, 0.75; field diameter, 450 pm). Because of the prognostic significance of the mitotic activity index, strict criteria were used for the definition and counting of mitotic figures7 As it nray be difficult to distinguish between anaphase or telophase mitoses in histologic sections, two sets of chromosomes arranged in parallel were regarded as two mitotic figures. Thus, mitotic activity index is the total number of sharply defined mitoses in 10 high-power fields. The areas of nuclei were measured at X2,000 final magnification’ using a commercially available graphic tablet (MOP-Videoplan; Kontron, Munich-Eching, Germany). The mean and standard deviation values of the nuclear areas were calculated for each tumor. DNA flow cytometric analysis was not performed because tissue blocks were not available to the investigators. This seems to be of minor importance for two reasons. First, DNA flow cytometry has very limited prognostic value to the morphometric features.’ Second, in our experience, the number of inadequate histograms is considerably higher in paraffin material older than 10 years.
Statistical
Analysis
The log rank test (generalized salvage by Mantel-&x in the BMDP program, see ref 10) was used for survival analysis and the Cox proportional hazards model was used to cope with a subset of covariates. Both metastasis-free survival and survival to death from metastasis were analyzed, but as they appeared strongly correlated, only the results of survival to death will be presented. Differences in age, tumor size, lymph node involvement, and microscopic prognosticators between the two regions were tested by Student’s t-test and the chi-squared test, if appropriate. In these computations all continuous variables (except age) were used after transformation into their square roots. Differences in the microscopic prognosticators between the two regions could be confoundingly modified by variables such as age, tumor size, and lymph node involvement. The contrasts between the regions and the microscopic variables of interest were derived separately with and without adjustment for age, tumor size, and lymph node involvement. For this purpose we used a logistic regression analysis with the logit of the rate of occurrence of the outcome phenomenon (Utrecht cases) as a dependent parameter.
RESULTS The
mean length of survival in the Delft cases was and 8.9 years in the Utrecht cases. Figure 1 shows the survival functions of the patient groups in the two regions. A crude analysis of the survival patterns showed a statistically significant difference between the two regions (Mantel-Cox = 7.7; P = .006). Survival predicting the covariates of age, tumor size, and lymph node involvement concomitant with the regions were thereafter included in the Cox proportional hazards model. In this model all of these determinants but age were significantly related to length of survival (region-related adjusted relative risk: 0.48, 95% [confidence interval, 0.31 to 0.751). Inclusion of the microscopic prognosticators concomitant with region and age resulted in a model with a likelihood of the same magnitude. How-
Studies
7.7 years
Microscopic studies were based on 5 pm-thick hematoxylin-eosin-stained paraffin sections of 4% neutral formalinfixed tissue slices taken from the periphery of the tumor. Apart from axillary lymph node status (as negative or positive and number of positive nodes), tumor diameter, and nuclear and histologic grades (according to the World Health Organization criteria, including a categorized three-point evaluation of the number of mitoses as absent, moderate, or high [more than l/10 high-power fields]), the morphometric features and parameters mentioned below were investigated. For a detailed description of the measurements, reference is made to our previous publications.5,6 Briefly, all measurements were performed in the (subjectively) most cellular areas of the tumor, with at least 50% of the area occupied by epi-
990
GEOGRAPHIC VARIATIONS IN BREAST CANCER SURVIVAL (Baak et al)
TABLE 1.
Median Values and Categoric Distributions for Indicators and Probability Values, Contrasting the Two Regions
REGION 2
Delft (n ;%yO)
REGION 1 Mantel-Cox=‘7.69 P=O.O056
20
0
2
YEARS n
Covariates Age (yr) (median) Tumor size (median) Lymph node involvement (present) Microscopic variates Mitotic activity index (median) Mean nuclear area (median) Histologic grade (no. and percentage) 1 2 3 Nuclear grade (no. and percentage) 1 2 3
4
6
8
10
111
102
93
88
27
18
Region 2
160
153
125
106
56
18
Region 1
1. Survival curves of the patients from Delft (region I> and Utrecht (region 2). FIGURE
ever, the region-related adjusted relative risk in this model (0.73, 95% [confidence interval, 0.44 to 1.211) was not significant. The results of these survival analyses with covariates mean that length of survival can be predicted by the morphometric features independent of the regions. On the other hand, if tumor size and lymph nodes status are included the region-related variable should be included also to obtain an accurate prediction. Thus, the microscopic features are stronger predictors of the outcome. Table 1 shows the medians (and the numbers and percentages in the ordinal-scale categories) for various covariates and microscopic variables for each of the two regions. The normally used clinical predictors (tumor size and lymph node involvement) were not appreciably different between the regions, although age difference could not be excluded (P = .08). In contrast, the qualitative and quantitative microscopic features were quite different. The patients in Delft, the region with a worse prognosis, had higher median values for the mitotic activity index and for nuclear area. Histologic and nuclear grades (if tested over all strata) were also more ominous in that region. In both regions the women who died had higher values for the mitotic activity index and nuclear area in general. Furthermore, these features did not differ between the survivors from Delft and Utrecht or between the nonsurvivors from the two regions. Logistic regression analysis shows that the differences in microscopic prognostic indicators were still present after adjustment for age, tumor size, and lymph node involvement in all variates. Table 2 compares the most important features of the 1973-1974 and 19881989 breast cancer cases in the favorable area (Utrecht). The results were very similar. Thus, the quantitative prognosticators are stable over time in that region.
Utrecht (n = 111) (%)
Probability Values*
53.5 2.0
58.0 2.5
.08 .ll
86 (53.8)
61 (55.0)
.85
14
4
