Comparison of Artificial Neural Network with Logistic Regression as Classification Models for Variable Selection for Prediction of Breast Cancer Patient Outcomes
Table 3
Main characteristics of the cohort.
Clinical parameters
Age
Mean SD
59.8 12.1
Median [range]
54.0 [23.2; 91.7]
Breast side
Left
50.1%
Right
49.9%
Personal cancer history*
2.7%
Menopausal status
68.7%
Skin invasion
4.4%
Cutaneous inflammation
1.9%
Clinical number of nodules
0
22.9%
1
74.3%
2
2.7%
Clinical T stage (size)
T0
22.2%
T1
44.9%
T2
22.0%
T3
4.9%
T4
5.9%
Clinical N stage
N+
17.3%
Histological parameters
Number of tumours (histology)
0 or 1
88.2%
2
11.8%
Tumour size (mm) (histology)
Mean SD
24.1 ± 20.1
Median (range)
20.0 [0.4; 250.0]
Histological type
Lobular
10.0%
Ductal
73.6%
Mixed
3.7%
Micro invasive
1.5%
Others
11.2%
Histological grade (SBR)
1
23.5%
2
47.2%
3
29.3%
Lymphatic embolus
36.8%
Nervous colonisation
13.7%
Necrosis (histology)
No in situ component
19.5%
In situ component without necrosis
35.0%
In situ component with necrosis
45.4%
Infiltrating tumour associated with CIS
75% in situ
5.5%
Limits of exeresis—infiltrating carcinoma
Not in sano
2.5%
Skin—infiltrating tumour
5.6%
Skin—embolus
2.0%
Nipple—infiltrating tumour
7.7%
Nipple—cancer in situ
6.8%
Nipple—embolus
4.0%
Nipple—Paget’s disease
1.3%
Pectoral muscle invaded
1.4%
Number of invaded nodes:
0
52.0%
1 or 2
28.2%
3
19.8%
Invaded nodes:
N0
8.5%
Micrometastasis
58.9%
Macrometastasis
32.7%
Capsular breaking
26.0%
Immunohistochemical parameters
Oestrogen receptor: % marked cells (count):
10%
17.1%
10–50%
7.2%
50%
75.8%
Oestrogen receptor (intensity):
+
25.8%
++ and +++
74.2%
Progesterone receptor: % marked cells (count)
10%
28.2%
10%–50%
17.4%
50%
54.4%
Progesterone receptor (intensity):
+
17.4%
++ and +++
82.6%
9 patients had cancer history. Localizations were 16 gynaecologic, 11 digestive, 11 Hodgkin/LMNH, 8 melanoma, 7 thyroid, 4 lung, 4 head and neck, 4 urinary, and 3 others.
