A risk stratification model to predict the benefit of chemotherapy in medullary breast carcinoma: a population-based SEER database

Characteristics of the patient

A total of 681 eligible patients with MBC from the SEER database (2010–2018) were enrolled, including 174 MBC patients who did not receive chemotherapy (median follow-up time: 51.5 months) and 507 MBC patients who received chemotherapy (median follow-up time: 51.5 months). uptime: 58 months). A screening flowchart of our study population is shown in Fig. 1. The comparison of demographic and clinicopathological characteristics between the groups is shown in Table 1, showing that compared to the non-chemotherapy group, patients in the chemotherapy group were younger, married, had a larger tumor size, had a higher rate of lymph node metastases, and were more likely to undergo radiotherapy. However, there were no differences in the distribution of race, tumor grade, hormone receptor status, human epidermal growth factor receptor 2 (HER2) status, or molecular subtypes between the two groups.

The flowchart of the population that participated in our study. ER estrogen receptor, PR progesterone receptor, HER2 human epidermal growth factor receptor.

In addition, as presented in Fig. 2A, B, patients with MBC who did not undergo chemotherapy had unfavorable overall survival (OS) (P < 0.001) and breast cancer specific survival (BCSS) (P = 0.036) compared to patients receiving chemotherapy. Considering the impact of the difference in baseline distribution between the two groups, we performed propensity score matching (PSM) analysis with a ratio of 1:1, and the difference disappeared after PSM (Table 1). At this point, we observed that patients in the chemotherapy group maintained a better OS after PSM (P = 0.005) than those in the non-chemotherapy group (Fig. 2C), while the difference in BCSS (P = 0.061) between the two groups was not statistically significant (Figure 2D).

Kaplan-Meier curves of OS and BCSS of the chemotherapy and non-chemotherapy groups in the overall population. (a,B) for PSM. (C,D) After PSM.

Independent predictive factors

We further randomly split the total population into a training set (n = 545) and a validation set (n = 136) in a ratio of 8:2. As shown in Table 2, there was no variation in the demographic and clinicopathological characteristics distributed between the two groups. Subsequently, univariate and multivariate Cox analysis were developed sequentially in the training set (Table 3), which revealed that age at diagnosis > 65 years old (< 40 jaar als referentie; > 65 years old: HR = 7.03, 95% CI (2.36–21.34), P= 0.001), T stage with T3–T4 (T1 as reference; T3–T4: HR = 5.49, 95% CI (1.96–15.4), P= 0.001), and N stage with N1-N3 (N0 as reference; N1-N3: HR = 2.39, 95% CI (1.19-4.8), P= 0.014) were independent adverse features for OS of patients with MBC, while patients with Luminal A (TNBC as reference; Luminal A: HR = 0.34, 95% CI (0.15-0.79), P= 0.012), and receiving radiotherapy (no radiation as a reference; radiation: HR = 0.34, 95% CI (0.15–0.79), P= 0.012) had a better OS compared to that of TNBC and those who did not receive radiotherapy, respectively.

Development and validation of nomograms

We incorporated five independent prognostic factors of OS by the Cox regression model into a nomogram predicting the probability of 3- and 5-year OS for the MBC population (Fig. 3). Age at diagnosis was the factor that had the greatest impact on survival rate, followed closely by T-stage, irradiation, subtype, and N-stage. Subsequently, the discriminating power of the nomogram was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves. The AUCs for the 3- and 5-year OS in the training set were 0.793 and 0.797, respectively (Fig. 4A). The AUCs for the 3- and 5-year OS were 0.781 and 0.823, respectively, in the validation set (Fig. 4B). The above findings showed that the prediction accuracy of the nomogram was high. At the same time, the calibration plots of the training and validation sets (1000 bootstraps) showed that the predicted survival rate of the nomogram was consistent with the actual prognostic results (Fig. 5). To illustrate the application of this nomogram, we included five patients and certain values ​​of the five independent prognostic factors to show readers how to predict survival rates at 3 or 5 years follow-up using the nomogram (Supplementary Table 1).

Nomogram for predicting OS at 3 and 5 years in patients with MBC. To use the nomogram, an individual patient’s value is placed on each variable axis and a line is drawn upwards to determine the number of points received for each variable value. The sum of these numbers is on the total points axis and a line is drawn down to the survival axes to determine the probability of survival at 3 or 5 years.

The ROC curves of 3- and 5-year OS from the training set (a) and validation set (B). ROCreceiver operating characteristic, AUCarea under the curve, TPtrue positive rate, FPfalse positive rate.

The calibration curve of OS at 3 and 5 years. The calibration curve for predicting patient survival at 3 years (a) and 5 years (C) in the training set and at 3 years (B) and 5 years (D) in the validation set. The probability of OS predicted by a nomogram is plotted on the x-axis; the actual operating system is plotted on the y-axis. The gray dotted lines show the ideal reference line where the predicted probabilities match the observed survival rates. The blue line shows the relationship between the actual and the predicted survival rate. Blue dots represent apparent calibration accuracy obtained by stratifying into intervals of predicted 0.5-y survival with 40 events per interval and plotting the mean predicted value within the interval against the Kaplan-Meier estimate of the stratum. The blue cross represents bootstrap bias-corrected Kaplan-Meier estimates. Dxy = 2 × (C–0.5). The C statistic is simply AUC. The “R2” index ranges from 0 to 1 and is interpreted as the fraction of the explained variance.

Risk stratification analysis Dxy

In addition, each variable was scored in accordance with the nomogram (Table 4) and the total score for each patient was obtained. Based on the total nomogram values ​​for each patient, we created a risk classification model. Thereafter, the optimal cut-off value of the total score was assessed via X-tile software (Supplementary Figure 1), and patients with MBC were then divided into a low-risk (573/681, 84.14%, score ≤ 186) group and a high-risk group (108/681, 15.86%, score ≥ 187) based on this optimal cut-off value. Kaplan-Meier curves were generated in the overall population (P< 0.001, Fig. 6A), training set (P< 0.001, Fig. 6B) and validation set (P= 0.035, FIG. 6C), demonstrating that the new risk stratification framework can accurately distinguish between the two prognostic categories for OS of patients with MBC.

Kaplan-Meier curves of OS for patients with MBC in the low- and high-risk groups. (a) Total population, (B) Training set, (C) Validation set.

The effects of chemotherapy on survival benefits in different stratifications

To further assess the survival benefit of chemotherapy, Kaplan-Meier curves were generated in the two stratified risk groups. The results showed that in both the overall population and the training set, patients with MBC in the low-risk group benefited from chemotherapy (overall population: P= 0.001, Figure 7A; training set: P= 0.001, FIG. 7C), while there is no evidence that chemotherapy can improve the OS of patients with MBC in the high-risk group, as the result is not statistically significant (overall population: P= 0.180, Figure 7B; training set: P= 0.340, Figure 7D).

Survival benefits of chemotherapy in low- and high-risk groups of patients with MBC. (a,B) Total population, (C,D) Training set.