After the first cycle of NACT, five patients showed a significant (more than 100% compared to baseline) increase in the number of tumor vessels. We assumed this was associated with severe peritumoral inflammation around necrotic areas, and therefore these patients were not included in the final analysis. The characteristics of all analyzed subjects (103 patients included in further analysis) are shown in Table 1. When studying the pathologic tumor response (PTR), the first degree was detected in 13 patients (13%), the second degree in 21 patients (20%), the third in 24 patients (23%), and the fourth degree in 20 patients (20%). The fifth degree (pCR) was detected in 25 patients (24%). In accordance with the proposed criterion for tumor response (> 50% cell loss), 57 (55%) patients responded to treatment, 46 (45%) patients did not respond.
Table 1
Patients’ characteristics.
Description
|
Number of patients (n) (Р±σр%)
|
Age
|
< 40
|
29 (27 ± 4,3%)
|
41–50
|
31 (30 ± 4,3%)
|
51–60
|
24 (24 ± 4,1%)
|
> 61
|
19 (19 ± 3,7%)
|
BC stage
|
IIa
|
26 (25 ± 4,1%)
|
IIb
|
44 (42 ± 4,7%)
|
IIIa
|
16 (16 ± 3,5%)
|
IIIb
|
12 (12 ± 3,1%)
|
IIIc
|
5 (5 ± 2,1%)
|
State of regional lymph nodes
|
No metastases
|
33 (32 ± 4,5%)
|
With metastases
|
70 (68 ± 4,5%)
|
Immunophenotype
|
Luminal A
|
12 (12 ± 3,5%)
|
Luminal В HER2neu-negative
|
26 (25 ± 4,1%)
|
Luminal В HER2neu-positive
|
12 (12 ± 9,5%)
|
HER2neu-positive
|
18 (18 ± 3,6%)
|
Triple negative
|
35 (33 ± 4,5%)
|
Scheme of NACT
|
Anthracyclines
|
39 (41 ± 4,7%)
|
Anthracyclines and taxane
|
38 (35 ± 4,5%)
|
Anthracyclines and trastuzumab
|
25 (23 ± 4,0%)
|
Trastuzumab / pertuzumab
|
1 (1 ± 0,94%)
|
Degree of pathological tumor response (Miller & Payne scale)
|
1
|
13 (12 ± 3,1%)
|
2
|
21 (21 ± 3,9%)
|
3
|
24 (24 ± 4,1%)
|
4
|
20 (19 ± 3,7%)
|
5
|
25 (24 ± 4,1%)
|
Pathological tumor response (actual morphometric scale in our study)
|
Responders
|
57 (55 ± 4,5%)
|
Non-responders
|
46 (45 ± 4,5%)
|
Total
|
103 (100%)
|
Doppler US
In 46 patients who did not respond to treatment, a decrease in the number of tumor vessels was observed in 33 (72%), and an increase in 13 (28%) (Fig. 2a). Among 57 responders, 56 showed a decrease in tumor vascularization. Statistical analysis showed a moderate negative correlation between ΔrelVascularization index and PTR (ρ= -0.512, р≤0.01). The sensitivity, specificity, and accuracy of the indicator for predicting PTR were 75%, 59%, and 68%, respectively.
Tumor volume
In the vast majority (94 patients, 91%), after the first cycle of NACT a decrease in tumor volume has been observed regardless of PTR (Fig. 2b). Among 46 non-responders, a tumor volume increased in 6 cases and has not changed in 2 cases. Among 57 responders, 56 showed a decrease in the tumor volume. A moderate negative correlation was revealed between ΔrelTumor volume and PTR (ρ=-0.416, р≤0.01). The sensitivity, specificity, and accuracy of the index for predicting PTR to NACT were 79%, 61%, and 71%, respectively.
Diffuse Optical Spectroscopy Imaging
Breast tumors showed multidirectional dynamics of oxygenation depending on the PTR. Among 57 responders, 53 (93%) showed an increase in StO2 after the 1th cycle of NACT. In three patients, the oxygenation decreased, in one it did not change (Fig. 2c). Twenty-nine of the 46 non-responders experienced a decrease in tumor oxygenation after the 1th cycle of NACT. In 12 cases, the level of oxygenation did not change, and in 5 cases it increased. A positive strong correlation was found between the change in the oxygenation (ΔStO2) and the PTR (ρ = 0.802, р≤0.01). Early changes in the tumor oxygenation for prediction of the PTR demonstrated high sensitivity, specificity and accuracy − 95%, 89% and 95%, respectively.
Binary logistic model
Separate models were run on each of three indicators: ΔStO2, Δ relVascularization index, and ΔrelTumor volume. The fact of tumor response (decrease in the number of tumor cells after NACT by more than 50%) was taken as a binary event. The probability of an event occurring for a particular case was calculated by the formula:
$$р=\frac{1}{1+{e}^{-z}}$$
where\(z=b\times X+a\)
X — independent variable value, a, b – regression coefficients.
With a p-value of less than 0.5, it can be assumed that the event will not occur (non-responders); otherwise, an event is expected (responders).
Model 1 "ΔrelTumor volume"
The overall number of patients correctly assigned to the groups of "responders" and "non-responders" using the variable "ΔrelTumor volume" was 70.9% (Table 2).
Table 2
Classification table for Model 1 (“ΔrelTumor volume”)
Observed
|
Predicted
|
Percentage of correct
|
Non-responders
|
Responders
|
Non-responders (46)
|
28
|
18
|
60.9
|
Responders (57)
|
12
|
45
|
78.9
|
Total
|
70.9
|
Table 3
Variables in Model 1 Equation (“ΔrelTumor volume”)
Variable
|
Value
|
Standard deviation
|
Significance level p
|
b
|
-0.052
|
0.016
|
0.001
|
a
|
-0.831
|
0.376
|
0.027
|
Despite the presence of a fairly large number of false positive cases, all variables in the regression equation appeared to be statistically significant (Table 3): а = -0.831 (р=0.027) и b = -0.052 (р=0.001).
z = -0.052×Х(ΔrelTumor volume) – 0.831 (Model 1 "ΔrelTumor volume")
To test a Model 1, a patient was randomly selected with “ΔrelTumor volume” = -3.
z = -0.052×(-3) – 0.831 = -0.675
$$р=\frac{1}{1+{e}^{-(-0.675)}}=0.337$$
According to the model the patient will respond to treatment with a probability of 33.7%. This was confirmed by a pathomorphological study of the residual tumor (number of viable tumor cells was 90% which corresponds to the 2nd degree of PTR).
Model 2 "ΔrelVascularization index"
When using the "ΔrelVascularization index" model, the total percentage of correct predictions decreases up to 68% (Table. 4).
Table 4
Model 2 Classification Table (“ΔrelVascularization index”)
Observed
|
Predicted
|
Percentage of correct
|
Non-responders
|
Responders
|
Non-responders (46)
|
27
|
19
|
58.7
|
Responders (57)
|
14
|
43
|
75.4
|
Total
|
68.0
|
Table 5
Variables in Model 2 Equation (“ΔrelVascularization index”)
Variable
|
Value
|
Standard deviation
|
Significance level p
|
b
|
-0.043
|
0.011
|
0.001
|
a
|
-0.989
|
0.366
|
0.070
|
In model 2 (Table 5), only the variable b = -0.043 (p = 0.001) turned out to be statistically significant. For the constant a = -0.989, the critical level of significance was not overcomed (p = 0.07), so it is not advisable to test model 2.
z = -0.043×Х(ΔrelVascularization index) − 0.989 (Model 2 (“ΔrelVascularization index”))
Model 3 "ΔStO2"
The last step of the regression analysis was the introduction of the variable “oxygenation level” (ΔStO2) into the model.
Table 6
Model 2 Classification Table (“ΔStO2”).
Observed
|
Predicted
|
Percentage of correct
|
Non-responders
|
Responders
|
Non-responders (46)
|
41
|
5
|
89.1
|
Responders (57)
|
6
|
51
|
89.5
|
Total
|
89.3
|
The “ΔStO2” variable significantly increases the percentage of correct prediction of non-response to treatment. The overall percentage of correct forecasts was 89.3% (Table 6).
Table 7
Variables in the Model 3 Equation (“ΔStO2”)
Variable
|
Value
|
Standard deviation
|
Significance level p
|
b
|
0.318
|
0.064
|
0.001
|
a
|
-0.773
|
0.367
|
0.035
|
Both variables of the regression equation turned out to be statistically significant: a = -0.773 (p = 0.035) and b = 0.318 (p = 0.001) (Table 7).
z = 0.318×Х(ΔStO2) – 0.773 (Model 3 “ΔStO2”)
To test the oxygenation level model, we have randomly used data from patients from the two groups.
Patient 1 (ΔStO2 = -15)
z = 0.318×(-15) – 0.773 = -4.91
$$р=\frac{1}{1+{e}^{-(-4.91)}}=0.0073$$
This patient would respond to treatment with a probability of 99.27%, which was confirmed by a pathological response (number of viable malignant cells was 94% which corresponded to the 2nd degree of PTR).
Patient 2 (ΔStO2 = 18)
z = 0.318×(18) – 0.773 = 4.95
$$р=\frac{1}{1+{e}^{-4.95}}=0.993$$
The second patient with a probability of 99.3% could be attributed to the responders group, which was confirmed by a 5th degree of PTR.
Thus, Model 1 "ΔrelTumor volume" and Model 2 "ΔrelVascularization index" have the same limitation: they do not accurately predict a lack of response to treatment, while predicting a positive response is relatively good. The combined inclusion of all of the above variables (ΔStO2, ΔrelVascularization index, and ΔrelTumor volume) into the model did not show an improvement in predictive values.