In this multicenter study, we explored the combination of %TILs and change in tumor load on DCE-MRI to assess response to NAC in patients with ER+/HER2- and TN&HER2 + breast cancer. A higher CV AUC was observed for the combination of %TILs and change in tumor load on MRI compared to either one alone in the whole group ((0.75 (95% CI 0.67–0.83) vs. 0.69 for %TILS-only (95% CI 0.53–0.78) and 0.69 (95%CI 0.61–0.79) for change in MRI tumor load only). This was also observed in the ER+/HER2- group ((0.72 (95%CI 0.60–0.88) vs. 0.68 (95% CI 0.50–0.82) and vs. 0.67 (95% CI 0.51–0.82), as well as in the TN&HER2 + group ((0.70 (95% CI 0.59–0.82) vs. 0.63 (95% CI 0.49–0.74) and vs. 0.67 (95% CI 0.56–0.79).
The difference in observed discriminative ability should, however, be interpreted with caution given the wide confidence intervals.
There is a large need for improvement of response prediction, before clinical trials on postponing or omitting surgery after NAC have a good chance of succeeding (23). Our work suggests that %TILs and MRI may hold complementary information and could be a useful combined biomarker for response to NAC in different breast cancer subtypes.
TILs have been shown by others to be correlated to pCR in the HER2 + and TN subtypes, with higher TILs relating to higher pCR rates (12, 22). In the ER+/HER2- subtype, the literature is inconclusive. A large pooled analysis by Denkert et al. showed a significant positive correlation between TILs and pCR in the ER+/HER2- subtype(22). A different meta-analysis and some other smaller studies did, however, not find this correlation (12, 13, 24–26). TILs are reported to be less frequent in ER+/HER2- breast cancer compared to the other subtypes (27), which makes it less likely to find a correlation in smaller groups. We found an association between TILs and response to NAC as measured by RCB in the whole group of patients. One study found significant correlations between RCB classes and TIL CD8/FOXP3 ratio in TN breast cancer (28). A different study by Elmahs et al. did not find a correlation between TILs and RCB class, perhaps due to small sample size (29).
Immunotherapy is a topic of increasing interest in breast cancer treatment, initially investigated in the metastatic setting, followed by the neoadjuvant setting. In the TN subtype, the combination of immunotherapy and NAC was shown to improve pCR in 2020(30). In TN patients treated with this combination, higher sTILs were associated with higher pCR rates (31–33). More recently, promising results for the combination of immunotherapy and NAC have been proposed in the luminal subtype as well (34, 35). Personalizing treatment and not giving more than necessary is of special importance in the light of high costs associated with immunotherapy. In a recent study by Loi et. al., presented at SABCS 2023, higher pCR rates were seen with increasing sTILs in patients with high risk luminal breast cancer treated with NAC and nivolumab(36). Models like the one presented in this study could potentially aid clinical decision making in treatment with the combination of NAC and immunotherapy in the future. They should however be evaluated in a population treated with this regimen.
With regard to the prognostic value of TILs, we found that higher %TILs in biopsy is associated with better RFS after NAC in the whole cohort. This suggests that %TILs could also be useful for post NAC decision making, although its role in relation to other prognostic factors was not investigated here due to too few events. High TILs have been reported to be correlated to better prognosis in the TN and HER2 + subtypes (12, 22, 37). In the ER+/HER2- group, the pooled analysis by Denkert et. al. reported low TILs (0–10%) to be correlated with improved disease free survival, in contrast with our results(22). The meta-analysis by Li et. al. reported no correlation between TILs and survival(12). Our work thus contributes to the growing body of research on the prognostic role of TILs in breast cancer. We did not have enough data to evaluate the relationship with TILs in the residual tumor to RFS, but work is underway to incorporate TILs in the residual tumor after NAC in the RCB to further stratify post-neoadjuvant prognosis(38).Since MRI is more accurate in evaluating response to NAC compared to mammography, ultrasound and physical examination, it is widely used in clinical practice (39–41). Radiological assessment alone is, however, not accurate enough to guide treatment decisions (42). A (semi-) automated method for evaluating response to NAC could be of interest, since manual measurement by RECIST is associated with intra- and interobserver variability (43–45).
Tissue biopsy is always a part of the diagnostic pre-treatment work-up and assessing TILs in the biopsy is quick and easily implemented, possibly even more so when artificial intelligence algorithms are deployed (46). The combination of TILs and computer extracted MRI features may therefore be an efficient use of information that is available from the clinical workflow without additional (invasive) procedures. Our results suggest the complementary value of these different data sources in assessing response to NAC, which could ultimately help in sparing patients unnecessary treatment.
Our study has several limitations. First, there was no independent cohort to perform external validation of the developed models. Second, due to limited sample size, we were unable to account for relevant predictors such as treatment regimen and nodal status, or to evaluate the HER2 + and TN subtypes separately. Third, our two cohorts contain patients from different periods in time, which resulted in different treatment regimens that may not reflect current clinical practice. Additionally, for cohort A, not all biopsies were centrally revised for ER, HER2 and nodal status. This could result in unwanted interobserver variability in these variables, which is however a reflection of clinical practice. Lastly, the MRI processing differed between cohort A and B. In theory, this could have impacted the results, although the methods have been shown to lead to highly correlated results(20).
In conclusion, our results show that the combination of TILs and change in tumor load on MRI is informative of response after NAC overall, as well as in the ER+/HER2- and TN&HER2 + groups separately. This could be of interest for clinical trials on de-escalating surgical intervention. More work is, however, needed to reduce uncertainty and improve accuracy by modifying for other predictors as well.