The Peking Union Medical College Hospital (PUMCH) database has archived patient information since 1975, including age at diagnosis, functional status, comorbidities, psychological status, surgical methods, tumor histology, treatment history, recurrence/metastasis (local, regional, distant), and survival. To analyze associations between clinical information and mortality, key data were extracted from both the archive and the digital records of this study.

Older patients with breast cancer who were admitted to our center’s breast surgery department between January 2013 and December 2019 formed our study population. The inclusion criteria were: (i) patients who were ≥65 years of age, (ii) those with a preliminary clinical diagnosis of early breast cancer, (iii) those who were intended for surgical treatment, and (iv) those with no metastatic tumors in any organ. Patients who met the inclusion criteria were approached for an interview, and three geriatric details (functional status, comorbidity, and psychological state) were initially obtained. After the patients were admitted to the ward, the three components of geriatric assessment were verified again or supplemented. Moreover, the clinical information was also refined in the ward. Dedicated follow-up staff regularly collected survival information. Patients with complete information regarding the three components (GA, clinical information, and follow-up information) were included in our study.

The study was approved by the Institutional Review Board of Peking Union Medical College Hospital (ZS-2682). The principles of the Helsinki Declaration of 1964 and subsequent amendments, or comparable ethical norms, were adhered to.

The GA method used in this study was developed on the basis of previous studies and reports based on similar questionnaires (5, 11–13). The comprehensive geriatric assessment (CGA) contains a large number of domains and a cumbersome questionnaire; therefore, the use of all CGA domains to assess older patients in surgical oncology is too time-consuming to be widely practiced (14). Therefore, we used functional status, comorbidity, and psychological state, which are the three domains most closely associated with survival in older patients assessed for geriatric solid tumors. The relevance of these three domains in survival of elderly patients with breast cancer has been previously confirmed (5).

The Eastern Cooperative Oncology Group Performance Status (ECOG PS) is currently the most commonly used functional domain in clinical practice due to its simplicity. However, studies have shown that activities of daily living (ADL) are more objective and accurate than ECOG (15). Nevertheless, the ADL questionnaire content is repetitive, and the score results are not easy to interpret. Therefore, based on the Barthel Index score, we simplified the ADL questionnaire including 10 items to measure ADL performance, and interpreted the results as normal or abnormal (16). The items that were assessed included continence and independence in a range of daily activities such as bathing, feeding, dressing, going to the bathroom, getting up and moving around the house. Each patient had a score from 0 to 100, which is a measure of the patient’s level of independence. Functional status was graded as abnormal (0-60) or normal (61-100) according to each patient’s status. If insufficient, the status was marked unknown.

Comorbidities were assessed based on the Charlson Comorbidity Index (CCI) score, which includes 17 comorbid conditions (myocardial infarction, chronic pulmonary disease, rheumatologic disease, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, peptic ulcer disease, liver disease, diabetes, hemiplegia or paraplegia, renal disease, any malignancy, lymphoma, leukemia, and human immunodeficiency virus infection) (17). Metastatic cancer, paraplegia and HIV infection were excluded due to a conflict of treatment principles, so that 14 co-morbid conditions were included in this study. For each comorbidity, 4 types of weights (1, 2, 3 and 6) were assigned based on the associated mortality risk. Two categories of comorbidity were used in this study: none to mild (CCI: 0-2) and severe (CCI: ≥3).

Psychological state was evaluated according to the updated Geriatric Depression Scale (GDS-15), which helps to evaluate depression in geriatric individuals with a collection of 15 factors (18). Based on the most common clinical criteria, psychological state was categorized as normal or non-depressive, abnormal or depressive state, and unknown.

We also gathered and analyzed clinical information, such as tumor stage, lymph node stage, and intrinsic molecular subtypes. Data of all-cause mortality and breast cancer-specific mortality were obtained from our center and hospital databases, which have complete records of patient visits, medication information, and clinical referral records. Our center has a dedicated follow-up staff to capture this relevant information, and missing data were supplemented by telephone contact.

Categorical variables were expressed as numbers and percentages. Pearson’s chi-square test was used to compare the results of different groups. The primary endpoints were overall survival (OS) and breast cancer specific survival (BCSS). OS was calculated as the time from the date of cancer diagnosis until the last date of vital status. BCSS was defined as the time from diagnosis of breast cancer to the breast cancer-related death. A Cox proportional hazards model was constructed for univariate and multivariate analyses of OS and BCSS. To select the variables for the nomogram, multivariate Cox regression model and backward stepwise selection were adopted based on the Akaike information criterion. A nomogram was constructed to predict the 1-, 3-, and 5-year OS and BCSS. Survival Receiver Operating Characteristic (ROC) curve and C-index were calculated to evaluate the discriminatory ability of the nomogram. Visual inspection of calibration was conducted by calibration curves. The potential clinical utility of the nomogram was evaluated by the decision curve analysis (DCA). Due to the high mortality rate among the older population, we fitted competing risk models with breast cancer-specific death and non-breast cancer-specific death as competing events. The Fine and Gray method of competing risk model was used for univariate and multivariate analyses of BCSS. A nomogram based on competing risk model was constructed to predict 3- and 5-year BCSS. To evaluate the discriminatory ability of the nomogram, C-index was produced for each model. Calibration curves were used for visual inspection of calibration. Statistical analyses were performed using the R (4.2.2) software. All statistical tests were two-tailed, and a P value less than 0.05 is considered statistical significant.

The design and procedures of this study are shown in a flowchart ( Figure 1 ). Elderly patients (≥65 years) undergoing breast surgery at our cancer unit were enrolled. Of the 641 eligible patients, 556 were enrolled. Table 1 shows the patient characteristics. Of 556 patients, 172 (31%) were diagnosed with cancer at age ≥75 years. The most common subtype of cancer was luminal B and the most common stage of the tumors was stage I. Significant differences in all-cause mortality (12.8% and 29.7% for <75 years and ≥75 years, respectively) were observed between the two different age groups during 76 months of follow-up. Breast-specific mortality (8.1% vs. 9.3%) was not significantly different between the two groups. The general distribution of patient characteristics, including geriatric assessment domains, were compared ( Table 1 ). Functional status differed significantly between the two groups but did not differ significantly with respect to the distribution of intrinsic molecular subgroups, tumor stage, nodal stage, comorbidities, or psychological status.

A Cox proportional hazards model was constructed for univariate analysis of OS using GA and clinical parameters ( Figure 2A ). Univariate analysis showed that all selected geriatric and clinical parameters were significantly associated with OS. According to the forest plot of univariate analysis, the overall death rate increased with age (HR, 1.11; 95% CI, 1.07 to 1.14), intrinsic molecular subtype (HR for Basal-like, 3.05; 95% CI, 1.35 to 6.90), tumor stage (HR for stage III, 4.88; 95% CI, 1.76 to 13.47), lymph node stage (HR for stage III, 2.55; 95% CI, 1.33 to 4.90), abnormal functional status (HR, 3.44; 95% CI, 2.20 to 5.37), severe comorbidities (HR, 2.16; 95% CI, 1.45 to 3.22), and abnormal psychological state (HR, 3.21; 95% CI, 1.78 to 5.78).

A Cox proportional hazards model was constructed for univariate analysis of BCSS using GA and clinical parameters ( Figure 2B ). Univariate analysis revealed that all clinical parameters were significantly associated with BCSS. The results showed that the risk of breast cancer-associated mortality increased with intrinsic molecular subtype (HR for HER-2 enriched subtype, 14.89; 95% CI, 1.79 to 123.73. HR for basal-like subtype, 16.35; 95% CI, 2.15 to 124.46), tumor stage (HR for stage III, 27.75; 95% CI, 3.24 to 237.89), and lymph node stage (HR for stage III, 6.56; 95% CI, 2.93 to 14.70). No significant differences were observed for BCSS in patients with different functional status (HR:1.87, 95% CI:0.87-4.01), CCI score (HR, 1.52; 95% CI, 0.83 to 2.78), and psychological state (HR, 1.40; 95% CI, 0.43 to 4.53).

Based on the results of univariate analysis, cox proportional hazards models were constructed for multivariate analysis of OS ( Figure 3A ). Similarly, multivariate analysis of BCSS was also performed ( Figure 3B ). Although comorbidities were not significantly associated with BCSS in univariate analysis, they were included in multivariate analysis because coexisting conditions may affect breast cancer management.

Multivariate analysis showed that age, intrinsic molecular subtype, tumor stage, lymph node stage, functional status, comorbidities, and psychological state significantly affected OS in an independent manner. The results showed that patients with Basal-like subtype (HR, 2.41; 95% CI, 1.23 to 4.71), tumor stage III (HR, 6.92; 95% CI, 2.28 to 21.03), lymph node stage III (HR, 2.66; 95% CI, 1.31 to 5.38), abnormal physical status (HR, 2.06; 95% CI, 1.49 to 4.01), abnormal psychological state (HR, 2.12; 95% CI, 1.09 to 4.09), and severe comorbidities (HR, 1.60; 95% CI, 1.01 to 2.53) tended to have shorter OS. It was also noted that the risk of all-cause mortality increased with age (HR, 1.11; 95% CI, 1.07 to 1.15).

Multivariate analysis showed that the intrinsic molecular subtype, tumor stage, and lymph node stage were independent prognostic parameters. The result showed that patients with HER-2 enriched subtype (HR, 5.83; 95% CI, 1.55 to 21.98) and Basal-like subtype (HR, 4.81; 95% CI, 1.49 to 15.39) tended to have shorter BCSS. Compared to other tumor stages, stage III was significantly associated with higher breast cancer specific mortality rate (HR, 17.44, 95% CI:1.85‐164.118). Likewise, patients with lymph node stage III showed a higher risk of breast cancer specific mortality (HR:4.40, 95% CI:1.88‐10.29). However, severe comorbidities did not show a significant correlation with BCSS.

A prognostic model for OS was developed based on the multivariate Cox regression analysis. A nomogram including age, intrinsic molecular subtype, tumor stage, lymph node stage, functional status, comorbidities, and psychological state was constructed to predict the 1-, 3-, and 5-year OS by summing each factor score to calculate the total score ( Figure 4 ). The predicted C-index of the OS prognostic model was 0.748 (95% CI:0.695-0.800). The 1-year OS prediction (AUC=0.769), 3-year OS prediction (AUC = 0.764), and 5-year OS prediction (AUC = 0.745) are shown in the ROC plot ( Figure 5A ). The calibration plot revealed good predictive accuracy between the actual and predicted probabilities for 1-year, 3-year and 5- years OS ( Figures 5B–D , respectively).

A DCA of the nomogram was conducted, which indicated that the nomogram had potential for clinical utility. The DCA indicated that when the threshold probability for 3-year and 5- years OS ( Figures 6A, B , respectively) were within the range (2.5%-17% and 8%–50%, respectively), the nomogram added more net benefit than “all or none” strategy. Enough number of events did not occur to support the DCA of the nomogram for 1-year.

A prognostic model for BCSS was developed based on the multivariate Cox regression analysis. A nomogram, including age, intrinsic molecular subtype, tumor stage, lymph node stage, and comorbidities, was built to predict the 1-, 3- and 5-year BCSS by summing each factor score to calculate the total score ( Figure 7 ).

The predicted C-index of the BCSS prognostic model was 0.760 (95% CI:0.753-0.834). The 1-year (AUC=0.772), 3-year (AUC = 0.773), and 5-year BCSS risk predictions (AUC = 0.761) are shown in the ROC plot ( Figure 8A ). The calibration plot revealed good predictive accuracy between the actual and predicted probabilities for 1year, 3-year and 5- year BCSS. ( Figures 8B–D , respectively).

DCA of the BCSS nomogram was also performed, indicating potential clinical utility of the nomogram. The DCA indicated that the nomogram added more net benefit than the “all or none” strategy when the threshold probabilities for 3-year and 5-year BCSS ( Figures 8E, F ) were in the range (1.5%-4.5% and 5%-45%, respectively).Enough number of events did not occur to support the DCA of the nomogram for 1-year.