A real-world clinicopathological model for predicting pathological complete response to neoadjuvant chemotherapy in breast cancer

Purpose This study aimed to develop and validate a clinicopathological model to predict pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in breast cancer patients and identify key prognostic factors. Methods This retrospective study analyzed data from 279 breast cancer patients who received NAC at Zhejiang Provincial People’s Hospital from 2011 to 2021. Additionally, an external validation dataset, comprising 50 patients from Lanxi People’s Hospital and Second Affiliated Hospital, Zhejiang University School of Medicine from 2022 to 2023 was utilized for model verification. A multivariate logistic regression model was established incorporating clinical, ultrasound features, circulating tumor cells (CTCs), and pathology variables at baseline and post-NAC. Model performance for predicting pCR was evaluated. Prognostic factors were identified using survival analysis. Results In the 279 patients enrolled, a pathologic complete response (pCR) rate of 27.96% (78 out of 279) was achieved. The predictive model incorporated independent predictors such as stromal tumor-infiltrating lymphocyte (sTIL) levels, Ki-67 expression, molecular subtype, and ultrasound echo features. The model demonstrated strong predictive accuracy for pCR (C-statistics/AUC 0.874), especially in human epidermal growth factor receptor 2 (HER2)-enriched (C-statistics/AUC 0.878) and triple-negative (C-statistics/AUC 0.870) subtypes, and the model performed well in external validation data set (C-statistics/AUC 0.836). Incorporating circulating tumor cell (CTC) changes post-NAC and tumor size changes further improved predictive performance (C-statistics/AUC 0.945) in the CTC detection subgroup. Key prognostic factors included tumor size >5cm, lymph node metastasis, sTIL levels, estrogen receptor (ER) status and pCR. Despite varied pCR rates, overall prognosis after standard systemic therapy was consistent across molecular subtypes. Conclusion The developed predictive model showcases robust performance in forecasting pCR in NAC-treated breast cancer patients, marking a step toward more personalized therapeutic strategies in breast cancer.


Introduction
Breast cancer is a primary cause of cancer-related mortality among women globally (1,2).Despite strides in therapeutic approaches, the death rate associated with advanced stages of the disease remains distressingly high (3)(4)(5).In this context, neoadjuvant chemotherapy (NAC) has risen to prominence as a crucial measure for gauging therapeutic effectiveness.It has shown notable efficacy in enhancing event-free survival (EFS) and overall survival (OS) rates, particularly among patients who achieve pathological complete response (pCR) during treatment (6)(7)(8).Additionally, the customization of post-neoadjuvant treatments has been promising in augmenting long-term outcomes for patients who do not attain pCR (9,10).This underscores the importance of accurately predicting pCR in breast cancer patients, a critical factor in optimizing treatment strategies and improving survival prospects.
Extensive research has focused on identifying the determinants of pathologic complete response (pCR) and prognosis following NAC in breast cancer.Tumor-infiltrating lymphocytes (TILs), recognized as key regulators within the tumor microenvironment (11)(12)(13)(14).The TILs could be classified into two types: intratumoral (iTILs) and stromal (sTILs), depending on whether they are located within the tumor nest or embedded in the tumor stroma (15,16).Studies have indicated that sTILs are a better and more reproducible biomarker than iTILs (15,17).TIL levels, particularly sTILs, have been highlighted as predictors of response to neoadjuvant chemotherapy in breast cancer (18,19).
Circulating tumor cells (CTCs), which are a subset of tumor cells shed into the peripheral blood as a result of tumor tissue instability, have been identified as independent predictors of pCR and prognosis (20,21).Additionally, traditional diagnostics, such as ultrasound paired with molecular markers like CA15-3 and CEA, also correlate with pCR and prognosis (22)(23)(24).Recent advances highlight the significance of combining imaging, pathology, and hematology in assessing treatment efficacy, as shown in Wang et al.'s model which integrates clinical and radiomics features such as background parenchymal enhancement (BPE), human epidermal growth factor receptor-2 (HER-2) status, and the Ki-67 index (25).
Despite the considerable accuracy of existing models, their limitations in specificity underscore the ongoing critical need for research focused on determining pCR status following NAC in breast cancer.Our study is directed towards developing an integrated predictive model that combines clinical (ultrasound), pathological (Ki-67 expression, molecular subtype), and novel biomarkers (sTIL levels, CTCs).We expect this comprehensive model to outperform individual indicators in predicting pCR with greater accuracy, thus providing a foundation for personalized breast cancer treatment strategies.

Study participants
Between April 2011 and June 2021, our study, adhering to rigorous clinical, ultrasonic, and pathological criteria, enrolled 279 female patients from the breast cancer database of Zhejiang Provincial People's Hospital (ZJPPH-BCDB).These patients, all pathologically confirmed with breast cancer, had undergone neoadjuvant chemotherapy (NAC) prior to surgery.This cohort was refined from an initial pool of 347 patients, as depicted in Figure 1.To validate our model, we further gathered an external dataset comprising 50 patients from Lanxi People's Hospital (Jinhua, China) and the Second Affiliated Hospital, Zhejiang University School of Medicine (Hangzhou, China) between March 2022 and June 2023.We ensured that all participants had no prior malignant disease or contraindications to chemotherapy.A subset of 71 participants was assessed for CTCs both pre-and post-NAC.This study received approval from the ethics committees of all the institutions involved, with informed consent obtained from participants with CTC detection, and adherence to all relevant guidelines and regulations was strictly maintained throughout the research process.

Data collection
Clinical, hematological, and ultrasonic data were systematically collected.Clinical parameters included age, body mass index (BMI), menopausal status, red cell distribution width (RDW), platelet distribution width (PDW), mean platelet volume (MPV), carcinoembryonic antigen (CEA), cancer-associated antigen (CA) 125, and CA153.Metrics derived from ultrasound (US) encompass tumor dimensions, geometric form, internal echogenicity, and the ratio of width to height.Pathological data, sourced from pre-NAC core-needle biopsy specimens, encompassed Ki-67, estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) expression.These were assessed using the American Society of Clinical Oncology and the American College of Pathologist guidelines, employing immunohistochemical methods (26,27).sTIL levels, classified as low, intermediate, or high (0-10%, 11-39%, and ≥40%, respectively), were determined based on recommendations from previous research (16,17).For the CTC detection subgroup, CTC data was collected pre-and post-NAC in 71 patients.Metrics from magnetic resonance imaging (MRI) capture tumor dimensions, structural configuration, signal intensity variations, and patterns of contrast enhancement in the CTC detection subgroup.Patients with ≥1 CTC per 7.5 mL of blood were categorized as CTC-positive (28).Pathological response was assessed using the Miller-Payne system, with a grade of 5 indicating pCR (29).Follow-up was conducted through phone checks and outpatient visits, with overall survival (OS) and disease-free survival (DFS) being defined according to standard clinical endpoints (30).

Statistical analysis
Analyses were conducted using SPSS (IBM SPSS 26.0, SPSS Inc.).Categorical data, represented as percentages, were assessed using appropriate chi-square tests.Continuous data were presented either as means ± SD or as medians with interquartile ranges (P50 [P25, P75]), depending on their distribution, and analyzed using the t-test or Mann-Whitney U test, respectively.Significant variables from univariate analyses (p < 0.05) were advanced to multivariate logistic regression.Stepwise logistic regression was employed, with the Hosmer-Lemeshow test evaluating model fit.The ROC and AUC were used to appraise model calibration and discrimination.Survival analyses were conducted using Kaplan-Meier curves, with significance ascertained via the log rank (Mantel-Cox) test.A threshold of p < 0.05 was set for statistical significance.

Overall predictive model of pCR
Multivariate analysis utilizing the significant factors from Table 1 identified sTIL levels, Ki-67 expression, tumor posterior echo, and molecular subtypes as independent predictors of pCR (all p < 0.05) (Table 3).The predictive logistic model based on these factors achieved a good fit (Hosmer-Lemeshow test, p = 0.962).

Predictive model of pCR with CTC detection
CTCs were detected pre-and post-NAC in the subgroup of 71 patients (Table 6).CTC-positivity decreased from 52.1% (37/71) at baseline to 23.9% (17/71) post-NAC (p < 0.05).Although no initial association between CTCs and pCR was observed (p = 0.173), a significant post-NAC difference emerged between CTC-positive and -negative patients in pCR rates (p = 0.002).Remarkably, patients converting from CTC-positive to CTC-negative after NAC had a significantly increased pCR rate (p = 0.001).
In the CTC detection subgroup, the multivariate logistic model considering CTC changes, tumor size changes under US, and sTIL levels achieved a C-statistics/AUC of 0.942 (95% CI: 0.889-0.995)for pCR prediction.Similarly, incorporating changes in tumor size as measured by MRI into the model yielded a comparable predictive performance with a C-statistics/AUC of 0.945 (95% CI: 0.894-0.997)(Figure 3, Table 7).The C-statistics/AUC values for CTC changes, tumor size changes under US, tumor size changes under MRI, and sTIL levels were 0.775, 0.7,0.798,and 0.838, respectively.The marginal increase in predictive accuracy offered by incorporating MRI-assessed tumor size changes, as evidenced by the comparable C-statistics/AUC values, suggests a limited additional value and cost-effectiveness in utilizing MRI.

Survival-related data analysis in the cohort
Over an average follow-up of 46.6 months (ranging from 4 to 143 months), a recurrence rate of 20.8% (58/279) was observed by December 2022.Kaplan-Meier curves indicated that larger tumor sizes (>5cm) and presence of lymphatic metastasis were associated with poorer DFS (p < 0.001 and p = 0.041 respectively) (Figures 4A,  B).Higher sTIL levels were linked with a reduced recurrence risk (p < 0.001) (Figure 4C).Interestingly, premenopausal patients had a slightly longer DFS (p = 0.051) (Figure 4D).A clear association between pCR and better DFS emerged (p = 0.001) ( Figures 4E, F), but molecular subtypes didn't significantly affect recurrence (Figure 5).
ER status had a significant impact on recurrence (p = 0.003).Analyzing DFS against CTC data (Figure 6) revealed that post-NAC CTC positivity correlated significantly with poorer DFS (p < 0.001).Finally, Kaplan-Meier plots for overall survival exhibited similar patterns regarding baseline tumor size, sTIL levels, and pCR (Figure 7).

Discussion
The prediction and understanding of the pCR and subsequent prognosis after NAC in breast cancer remains a significant challenge in clinical practice.The pursuit of an optimal predictive model for both pCR and prognosis is paramount, as it holds the potential to guide therapeutic decisions and enhance personalized patient care.
In this study, we integrated an array of clinical, pathological, and biomarker features to devise a comprehensive predictive model for pCR following NAC in breast cancer patients.Our integrated model, combining sTIL levels, Ki-67 expression, molecular subtypes, and ultrasound echo characteristics, demonstrated significant predictive power with an impressive C-statistics/AUC of 0.874.Its robust performance was further affirmed in the external validation set, achieving a C-statistics/AUC of 0.836, thereby underscoring the model's efficacy.Among the predictors, sTIL levels stood out as the most influential factor.While previous research has indeed underscored the correlation of sTIL levels with prognosis in tumor tissues (31)(32)(33), their relationship and prognostic value across different molecular breast cancer subtypes remain nuanced (34,35).For instance, while elevated sTIL levels have shown favorable prognosis in TNBC (36,37), their significance in other molecular subtypes, especially ER/PR+ HER2-, is less consistent (34,38).Our findings further highlight this molecular subtype-specific relationship, emphasizing the nuanced role of sTILs in predicting outcomes.The relationship between sTIL levels and prognosis across different molecular subtypes in breast cancer, and its value in prediction and evaluating treatment warrant further clarification.

A B
In addition to incorporating pre-and post-NAC dynamics of CTCs, our model highlighted the significant predictive capacity of combining CTC changes with US-assessed tumor size alterations and sTIL levels, achieving a C-statistics/AUC of 0.942.The model incorporating MRI for tumor size changes showed a similar predictive performance.The C-statistics/AUC of the model with US and the model with MRI were closely comparable, being 0.942 and 0.945 respectively.Considering the complexity and higher cost associated with MRI, ultrasound emerges as the more practical and cost-effective evaluation method in this predictive model.The observed strong correlation between post-NAC CTC positivity with both non-pCR and decreased DFS highlights the utility of CTCs as a prognostic marker.Often referred to as 'liquid biopsies', CTCs are rapidly gaining recognition as vital components in cancer management, known for their roles in early detection, prognostic evaluation, and monitoring for recurrence (39)(40)(41).Our study reinforces their significance in breast cancer, underscoring their potential as key predictors for both pCR and overall prognosis.
In addition to the above features, our data accentuated traditional clinical features, such as large tumor sizes (> 5 cm) and lymph node metastasis, remain crucial in prognosis.Furthermore, low sTIL levels and negative ER expression were all linked with poor DFS, indicating their potential as prognostic markers, and providing a holistic view of the complex prognostic landscape of breast cancer.
One intriguing observation was the differential pCR rates across molecular subtypes.While the ER-PR-HER2+ subtype exhibited a higher rate of pCR, the ER/PR+ HER2-group showed a lower rate.However, when assessing DFS post-NAC, no significant differences were observed between these molecular subtypes.This raises pertinent questions about the interplay between short-term treatment responses and long-term survival outcomes, suggesting that achieving pCR does not always equate to improved longterm prognosis.
Our study contributes to the broader understanding of pCR and prognosis in breast cancer; however, it is not without its limitations.The sample size, especially within the CTC subgroup, may limit the generalizability of our findings, and the retrospective design of our analysis could potentially influence the results.These constraints underscore the necessity for larger, prospective studies to confirm and expand upon our findings.Looking ahead, we envision a comprehensive future trial that would not only validate our current results but also enhance the predictive model for broader clinical utility.

Conclusions
Our study has developed a sophisticated model adeptly predicting pCR post-NAC in breast cancer, integrating sTILs, Ki-67 expression, molecular subtypes, and ultrasound features with a C-statistics/AUC of 0.874.Particularly in the CTC-detected subgroup, the model combining CTC changes, tumor size changes, and sTILs achieved an impressive C-statistics/AUC of 0.945.Key determinants like tumor > 5 cm, presence of lymph node

FIGURE 1
FIGURE 1Study profile of 347 patients who received NAC and subsequently underwent surgery between April 2011 and June 2021.279 patients met the eligibility criteria and were enrolled in this study.Additionally, a validation set comprising 50 patients was included.Subgroup analysis was performed in 71 patients with CTC data.NAC, Neoadjuvant chemotherapy; CTC, circulating tumor cell.
metastasis, reduced sTILs, and negative ER expression were identified as pivotal to diminished DFS.Essentially, our findings offer an enriched pCR prediction model, highlight salient factors affecting prognosis, and underscore the potential for individualized breast cancer treatments.

TABLE 1
Patient characteristics of pCR and non-pCR.

TABLE 2
Molecular subtypes and pCR rate.
ROC analysis of the Logistic model and 4 consisting factors for prediction of pCR.US: ultrasound; sTILs: stromal tumor infiltrating lymphocytes.

TABLE 5
Characteristics of 279 patients and external validation data set.

TABLE 6
The relationship between CTCs and pCR rate.

TABLE 7
ROC analysis of the Logistic model and 4 factors for prediction of pCR(n=71).