A Prognostic Nutritional Index-Based Nomogram to Predict Breast Cancer Metastasis: A Retrospective Cohort Validation

Introduction

Breast cancer (BC) is the most commonly diagnosed malignancy and the leading cause of cancer-related deaths, accounting for approximately 24.5% of all cancer diagnoses globally among women.¹ Despite advances in early diagnosis and treatment, BC metastasis continues to account for the majority of cancer-related deaths and poor outcomes.² It is reported that metastases are culpable for approximately 90% of cancer-associated deaths.³ The primary sites of BC metastasis include bone, lung, liver, and brain, exhibiting a tendency to spread to different organs, a phenomenon known as metastatic heterogeneity, which may be one of the reasons for the failure of BC treatments.⁴ Therefore, Understanding the factors that drive BC metastasis and developing effective predictive models to assess metastasis risk are crucial for improving patient outcomes.

In previous studies, traditional predictors such as lymph node status, hormone receptor expression, and microRNA have been widely studied.^5,6 However, these factors are limited by practical challenges, such as the complexity of detection, in predicting BC metastasis risk. Recently, the relationship between clinically accessible indicators and cancer prognosis has attracted extensive attention. For example, the prognostic nutritional index (PNI), derived from albumin levels and lymphocyte counts (albumin[g/L] + 5×lymphocytes[×10⁹/L]), serves as a validated biomarker integrating nutritional and immunological profiles.⁷ Studies have shown that PNI is closely associated with the prognosis of breast cancer.^8–10 However, previous studies have primarily concluded that a high pre-treatment prognostic nutritional index (PNI) is associated with longer disease-free survival, overall survival, or an increased rate of pathological complete response in breast cancer.^11–14 Currently, there is few research that considers distant metastasis of breast cancer as a primary clinical outcome to analyze the relationship between PNI and metastasis.

In addition to the PNI, nutritional and inflammatory markers such as the geriatric nutritional risk index (GNRI), the neutrophil percentage-to-albumin ratio (NPAR), and the hemoglobin, albumin, lymphocyte, and platelet (HALP) ratio have also attracted widespread attention. These indices have also demonstrated prognostic potential in various malignancies.^15–17 They reflect the multifaceted interactions among cancer progression, systemic inflammation, and condition of nutrition. Although these markers have gradually been incorporated into predictive models for cancer prognosis, such as recurrence or survival,^18–21 there remains a lack of comprehensive models that integrate these markers with other clinical variables to accurately and practically predict BC metastasis.

This study investigates the clinical relevance of four readily accessible biomarkers - PNI, GNRI, NPAR, HALP - in predicting breast cancer metastasis. Leveraging their unique combination of nutritional and inflammatory profiles, we are the first to develop the comprehensive nomogram model that synergistically integrates these biomarkers with key clinical parameters. The resulting tool provides clinicians with an intuitive, evidence-based platform for individualized metastasis risk stratification, enabling timely therapeutic interventions.

Methods

Study Population

This retrospective study included 311 patients with primary breast cancer admitted to the First People’s Hospital of Yancheng between March 2015 and December 2023. The final cohort comprised 81 patients with BC metastasis and 230 without metastasis. Inclusion criteria: (1) age > 18 years, (2) female, (3) histopathologically confirmed breast cancer, (4) confirmed metastatic sites (bone, lung, liver, or brain), and diagnosed through imaging or pathological histology, (5) complete clinical and histopathological records. Exclusion criteria: (1) patients with other malignancies, (2) patients with heart, liver, kidney and other organ failure or severe infectious diseases, (3) incomplete clinical data records.

Data Collection

We collected relevant patient information from the electronic medical record system, including age, height, weight, marital status, histological grade, estrogen receptor (ER) status, progesterone receptor (PR) status, human epidermal growth factor receptor 2 (HER2) status, Ki - 67 index, T stage, N stage, molecular subtype, chronic disease history (hypertension, diabetes, and hyperlipemia), treatment regimens, Karnofsky Performance Status (KPS) score, and blood test data. Among them, marital status refers to married and divorced status, with divorced status including divorce, separation, and widowhood. Baseline hematological parameters were obtained within 24 hours of admission through standardized complete blood count (CBC) analysis and biochemical profiling, including white blood cell count (WBC), red blood cell count (RBC), lymphocytes, hemoglobin, albumin, etc.

The formulas for calculating PNI, GNRI, NPAR, and HALP were as follows: PNI = albumin (g/L) + 5 * lymphocyte count(10⁹/L);²² GNRI =[1.489 * albumin (g/L)] +[41.7 *(present weight/ideal body weight)]; ideal body weight was calculated²³ for men: ideal body weight (men) = height (cm) − 100 − ((height (cm) − 150)/4), and for women: ideal body weight (women) = height (cm) − 100 − ((height (cm) − 150)/2.5). NPAR= (neutrophil percentage * 1000/ albumin (g/L));²⁴ HALP = hemoglobin (g/L) × albumin (g/L) × lymphocyte count (10⁹/L)/ platelets (10⁹/L).²⁵

Statistical Analysis

Normal continuous variables were expressed as mean ± standard deviation (SD), and the independent sample t-test was used for comparison between groups. Non-normal continuous variables were expressed as median and interquartile range (IQR), and the non-parametric test was used for comparison between groups. Categorical variables were expressed as frequency and percentage (%), and the chi-square test or Fisher’s exact test was used for comparison.

The receiver operating characteristic curve (ROC) was used to determine the optimal cut-off values of PNI, GNRI, NPAR, and HALP. The restricted cubic spline (RCS) curve was used to analyze the linear relationship between PNI, GNRI, NPAR, HALP and breast cancer. Multimodel logistic regression analysis was used to evaluate the association between indicators such as PNI and BC metastasis. Statistically significant factors were incorporated into the nomogram to predict the risk of BC metastasis. Internal validation was performed by randomly splitting the dataset into a training set and a validation set at a ratio of 7:3. The receiver operating characteristic curve (ROC), area under the curve (AUC), Hosmer-Lemeshow test, and calibration curve were used to evaluate the accuracy of the model. The decision curve analysis (DCA) was used to evaluate the clinical utility of the nomogram. All statistical analyses were performed using R software (version 4.4.0), and P < 0.05 was considered statistically significant.

Results

Baseline Characteristics

As shown in the flowchart in Figure 1, a total of 311 breast cancer patients were enrolled in this study. 81 patients with BC metastasis and 230 patients without BC metastasis were included. The average age of these patients was 53.69 ± 9.90 years old.

Figure 1 The flow diagram of sample selection in the study.

As presented in Table 1, there were statistically significant differences (P<0.05) in marital status, histological grade, PR status, T stage, N stage, surgery, endocrinotherapy, immunotherapy, hemoglobin, albumin, prealbumin, RBC, PNI, GNRI, and NPAR between the non-metastatic patients and the BC patients with metastasis.

Table 1 Baseline Characteristics of the Study Population

Influencing Factors of Breast Cancer Metastasis

As shown in Figure 2, we used restricted cubic spline (RCS) curves to analyze the nonlinear relationships between PNI, GNRI, NPAR, HALP and BC metastasis. After adjusting for age, BMI, marital status, histological grade, hypertension, diabetes, hyperlipidemia, and KPS score, PNI and HALP were nonlinearly correlated with BC metastasis (P for overall association < 0.05, P for nonlinear < 0.05) (Figure 2A–D). However, there was no linear correlation between GNRI, NPAR and BC metastasis (P for overall association > 0.05, P for nonlinear > 0.05) (Figure 2B and C).

Figure 2 Restricted cubic spline curves for analyzing the nonlinear relationships between BC metastasis and PNI (A), GNRI (B), NPAR (C), HALP (D).

Based on the receiver operating characteristic (ROC) curve analysis and the optimal Youden’s index, the optimal cut-off values of PNI, GNRI, NPAR and HALP were 43.83, 103.52, 16.67, and 30.38, respectively (Table 2). According to the cut - off value, we divided PNI into two groups (PNI< 43.83 and PNI ≥ 43.83). The relationships between PNI and clinical parameters in BC patients were detailed showed in Table 3. Compared to the low PNI group (PNI< 43.83), there were significant differences in the distribution of age, T stage, N stage, hyperlipidemia, liver metastasis, metastasis, hemoglobin, total protein, albumin, and other blood parameters in the high PNI group (PNI ≥ 43.83).

Table 2 Receiver Operating Characteristics Analyses of Parameters in Patients with BC Metastasis

Table 3 Relationship Between PNI and Clinical Parameters in BC Patients

Multivariate logistic regression analysis was further used to identify the influencing factors of BC metastasis (Table 4). After adjusting for age, BMI, KPS score, hypertension, diabetes, hyperlipidemia, ER, HER2, Ki67, molecular subtypes, WBC, neutrophil, neutrophil percentage, lymphocyte, lymphocyte percentage, monocyte, monocyte percentage, triglyceride (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and glucose (Model 3), marital status, T stage, N stage, RBC, total protein, and PNI were identified as independent influencing factors for BC metastasis.

Table 4 Multivariate Logistic Regression Analyses of Factors Associated with BC Metastasis

Patients with high PNI (PNI ≥ 43.83) had a significantly lower probability of BC metastasis compared to those with low PNI (PNI < 43.83) (adjusted OR 0.19, 95% CI 0.04, 0.88, P = 0.034). Patients with a higher T stage, N stage, and total protein were more likely to experience BC metastasis (adjusted OR for T3 was 11.38, 95% CI 2.78, 46.51, P < 0.001; adjusted OR for T4 was 37.08, 95% CI 2.06, 668.80, P = 0.014; adjusted OR for N3 was 6.30, 95% CI 1.71, 23.18, P = 0.006; adjusted OR for total protein was 1.16, 95% CI 1.03, 1.30, P = 0.011). Married patients had a lower probability of BC metastasis than those who were divorced (including divorced, widowed, and separated) (adjusted OR 0.16, 95% CI 0.05, 0.53, P = 0.003). The higher the RBC, the lower the risk of BC metastasis (adjusted OR 0.11, 95% CI 0.02, 0.62, P = 0.012).

Establishment and Verification of a Clinical Prediction Model for Breast Cancer Metastasis

We incorporated the independent influencing factors, namely marital status, T stage, N stage, RBC, total protein, and PNI, into the construction of a nomogram model for BC metastasis (Figure 3). Each variable was assigned a score ranging from 0 to 100. The scores corresponding to each variable were summed up to calculate the total score, and the risk of BC metastasis was located on the nomogram according to the total score level. Through this approach, the likelihood of BC metastasis can be evaluated more effectively and intuitively. For example, total scores ≥240 on the nomogram correspond to a 90% probability of BC metastasis (scale range: 0–350).

Figure 3 The nomogram for predicting the risk of BC metastasis.

To verify the robustness and clinical utility of the nomogram model, we randomly splitted the dataset into a training set and a validation set at a ratio of 7:3. The AUC of the prediction model in the training set was 0.85 (95% CI 0.79, 0.91), and that in the validation set was 0.82 (95% CI 0.71, 0.93) (Figure 4). The calibration curves of the training set and the validation set indicated that the predictive probability of the model matched the actual probability of the disease. The P values of the Hosmer-Lemeshow test were 0.704 and 0.301 respectively in the training and validation set, suggesting no differences between the predicted probability and the actual probability of the model (Figure 5A and B). The decision curves showed that the model had high clinical utility in both the training set and the validation set (Figure 5C and D).

Figure 4 The AUC of the nomogram for predicting BC metastasis in the training and validation set.

Figure 5 Calibration curves, Hosmer-Lemeshow test and decision curve analysis of the nomogram in the training set and the validation set. Calibration curves and Hosmer-Lemeshow test in the training set (A) and the validation set (B); decision curve analysis in the training set (C) and the validation set (D).

Discussion

This study identified PNI as an independent predictor of breast cancer metastasis and developed a nomogram model incorporating PNI, marital status, T stage, N stage, RBC, and total protein. The AUC values for the training and validation sets were 0.85 and 0.82, respectively, indicating strong predictive performance. Furthermore, the calibration curves and Hosmer-Lemeshow test results confirmed the model’s reliability in predicting observed outcomes, while decision curve analysis demonstrated high net benefit across a range of threshold probabilities. Collectively, these metrics highlight the model’s potential as a practical, evidence-based tool for risk stratification in clinical settings.

The prognostic nutritional index (PNI), calculated from serum albumin levels and lymphocyte counts, reflects a patient’s immune and nutritional status, which are critical determinants of cancer progression and metastasis.^26,27 Research by Xiang et al has found that low serum albumin level is associated with poor overall survival (OS) in patients with metastatic breast cancer and served as a prognostic factor.²⁸ Previous studies have shown that lymphocyte count can prevent tumor progression by activating the host immune response.²⁹ Low PNI reflects the presence of malnutrition or impaired immune function in patients. This condition may weaken the body’s immune surveillance, increasing the potential for tumor immune evasion and metastasis.¹⁰ PNI has been reported as independent predictors of prognosis in various cancers, including gastric, colorectal, and lung cancers.^30–32 In the context of breast cancer, existing studies have demonstrated the prognostic value of PNI in predicting disease-free survival and overall survival, as well as its utility in guiding treatment strategies.^33,34 For example, Qu et al have confirmed that PNI is positively correlated with pathological complete response rate in breast cancer patients.¹⁴ Furthermore, systematic reviews have also highlighted the predictive ability of the PNI for the prognosis of breast cancer patients.⁸ This study further expands the application of PNI in assessing the risk of metastasis in breast cancer. It clarifies the non-linear relationship between PNI and breast cancer metastasis. By incorporating PNI into a predictive model, the actual clinical risk stratification has been improved. Additionally, dynamic monitoring of PNI may also assist in guiding nutritional support and immunomodulatory therapy, thereby improving patient prognosis.

In addition to PNI, other nutritional and inflammatory indices, including HALP, also showed nonlinear association with BC metastasis in our analysis. Our results align with previous studies, which have identified HALP as a reliable prognostic indicator for various malignancies.^17,35 However, in the final logistic regression model, HALP, GNRI, and NPAR were not associated with BC metastasis, with PNI emerging as the stronger independent predictor. It is well known that cancer-related inflammation promotes tumor growth, invasion, and metastasis, while malnutrition may exacerbate these processes by weakening immune surveillance and promoting inflammation.^36,37 The clinical advantage of PNI may lie in its biological foundation, which is closely related to the dual functions of immune surveillance and nutritional support within the tumor microenvironment. Other indicators may be limited by their definitions or the sample size, failing to account for multiple mechanisms.

The nomogram model developed in this study integrates PNI along with other independent predictive factors, such as marital status, T stage, N stage, RBC, and total protein, providing a personalized risk prediction tool for BC metastasis. The inclusion of marital status in the model may appear unconventional. However, previous studies have demonstrated that marital status is an independent prognostic indicator for survival in patients with breast cancer.^38,39 Marital status may influence cancer prognosis through mechanisms including infection, immune response regulation, social support and treatment adherence.^40–42 Although the biological basis of this association remains unclear, its inclusion underscores the multifactorial nature of breast cancer metastasis.

Despite these advantages, this study has several limitations. Firstly, while retrospective design allows for quick and effective analysis of existing clinical data, it may introduce selection bias, thereby limiting the generalizability of the findings and the exploration of causal relationships. Prospective studies are needed in the future to confirm the predictive value of PNI. Secondly, the relatively small sample size may affect the extrapolation of the results. Larger-scale, multicenter studies and external validation are required to further conform these findings. Lastly, although this study focused on objective and easily accessible laboratory parameters, other potentially relevant biomarkers, such as C-reactive protein or interleukin-6, were not included. Incorporating these biomarkers could further enhance the model’s accuracy and provide additional insights into the biological mechanisms driving BC metastasis.

Conclusions

In conclusion, this study identifies PNI as an independent predictor of BC metastasis and introduces a novel PNI-based nomogram model with significant predictive accuracy and clinical applicability. By integrating nutritional, inflammatory, and clinical factors into a user-friendly tool, this model enables personalized risk assessment and enhances clinical decision-making. Given the limitations of our retrospective design and sample size, future studies should validate these findings in large-scale prospective multicenter cohorts and explore interventions targeting nutritional and inflammatory pathways to mitigate metastasis risk (such as protocolized albumin support and lymphocyte-boosting diets for PNI < 43.83). Compared to traditional prediction models like TNM stage, this PNI-based nomogram model may be more advantageous in improving the prognosis of breast cancer patients.