Table 4. A subgroup analysis was completed based on ER/PR+ and triple-negative tumours.
| Subgroup | Best feature | Model | %Sn | %Sp | AUC |
|---|---|---|---|---|---|
| ER/PR+ | Hb-con | Logistic regression | 76.2 | 66.7 | 0.746 |
| HbO2-hom | Naive Bayes | 93.3 | 90.1 | 0.883 | |
| HbO2-con | k-NN | 85.8 | 82.5 | 0.851 | |
| Triple negative | Hb-hom | Logistic regression | 100.0 | 33.3 | 0.917 |
| Hb-ene | Naive Bayes | 100.0 | 66.7 | 0.667 | |
| Hb-hom | k-NN | 75.0 | 66.7 | 0.917 | |
| FEC-D | TOI-hom | Logistic regression | 100.0 | 92.3 | 0.949 |
| Hb-con | Naive Bayes | 60.0 | 81.7 | 0.722 | |
| Hb-hom | k-NN | 80.0 | 80.0 | 0.806 | |
| AC-T | HbO2-cor | Logistic regression | 100.0 | 71.4 | 0.837 |
| HbO2-hom | Naive Bayes | 96.4 | 90.7 | 0.882 | |
| HbO2-hom | k-NN | 83.6 | 85.0 | 0.896 |
Abbreviations: AC-T=adriamycin, cyclophosphamide, taxol; AUC=area under curve; ER=oestrogen receptor; FEC-D=fluorouracil, epirubicin, cyclophosphamide, docetaxel; Hb=deoxy-haemoglobin; HbO2=oxy-haemoglobin; k-NN=k-nearest neighbour; PR=progesterone receptor; Sn=sensitivity; Sp=specificity; TOI=tissue optical index.
Patients were also grouped according to chemotherapy type for analysis. Three classification models were used (logistic regression, naive Bayes, and k-NN) and the best predictive features are presented.