Development and internal validation of a nomogram for predicting neoadjuvant therapy efficacy in breast cancer based on the HALP score Authors: Jianping Long, Xiaoke Chai, Tao Zhang, Chongyi Wei, Tao Yang, Haicun Zhou, Xiufen Ma Published: 2026 ABSTRACT AbstractBackgroundHemoglobin-albumin-lymphocyte-platelet (HALP) score is a composite index reflecting patients’ immune, nutritional, and inflammatory status, which has been associated with breast cancer prognosis. However, its value in predicting neoadjuvant therapy (NAT) efficacy in the overall breast cancer population remains controversial and understudied. This study aims to explore the relationship between preoperative HALP score and NAT efficacy in breast cancer patients and develop a predictive nomogram.MethodsClinicopathological data of 314 breast cancer patients who received NAT and mastectomy between January 2013 and November 2024 were retrospectively collected. Propensity score matching (PSM) was used to balance baseline characteristics between high-HALP and low-HALP groups. A nomogram was constructed based on multivariate Logistic regression analysis. Receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA) were used to evaluate the model’s performance. Kaplan-Meier curves were applied to compare disease-free survival (DFS) and overall survival (OS) between groups before and after PSM, as well as between different point stratification groups based on the nomogram.ResultsAfter PSM, 226 patients were included, with 113 in each group. Multivariate Logistic regression confirmed that a low HALP score was an independent negative predictor of pathological complete response (pCR, defined as ypT0/is ypN0) [odds ratio (OR) =0.42, 95% confidence interval (CI): 0.19–0.92, P=0.03], while positive human epidermal growth factor receptor-2 (HER2) status was an independent positive predictor (OR =5.00, 95% CI: 1.93–12.92, P<0.001). A higher preoperative HALP score was correlated with a greater likelihood of achieving pCR, whereas a low HALP score was associated with a lower pCR rate and poorer long-term prognosis in breast cancer patients. The nomogram integrating HALP score, HER2, estrogen receptor (ER), progesterone receptor (PR), N stage, and grade showed area under the curve (AUC) values of 0.80 (95% CI: 0.73–0.87) in the training cohort and 0.78 (95% CI: 0.66–0.91) in the validation cohort. Notably, only internal validation via random data splitting was conducted in this study, and no external validation was performed. Calibration curves demonstrated good consistency between predicted and actual pCR probabilities, and DCA confirmed favorable clinical utility. Kaplan-Meier analysis showed that the high-HALP group had significantly better DFS and OS both before and after PSM (all P<0.05), and the low-point group based on nomogram scores had longer DFS and OS (all P<0.05).ConclusionsPreoperative low HALP score is associated with lower pCR rate and poorer prognosis in breast cancer patients. The nomogram based on HALP score combined with PSM provides a reliable tool for predicting NAT efficacy and individualized treatment decision-making in breast cancer.Keywords: Breast cancer, hemoglobin-albumin-lymphocyte-platelet score (HALP score), neoadjuvant therapy (NAT), pathological complete response (pCR), nomogram Highlight box Key findings • This study confirms that preoperative hemoglobin-albumin-lymphocyte-platelet (HALP) score is an independent predictor of pathological complete response (pCR, ypT0/is ypN0) in breast cancer patients receiving neoadjuvant therapy (NAT), with low HALP score associated with lower pCR rate and poorer disease-free survival/overall survival. • A nomogram integrating HALP score and clinicopathological factors (human epidermal growth factor receptor-2, estrogen receptor, progesterone receptor, N stage, grade) was developed, showing good predictive performance (area under the curve: 0.80 in training cohort, 0.78 in validation cohort) and clinical utility. • Propensity score matching was used to balance baseline confounders, enhancing the robustness of the association between HALP score and NAT efficacy. What is known and what is new? • HALP score is associated with breast cancer prognosis, but its predictive value for NAT efficacy is controversial and mostly limited to specific subtypes. • This study focuses on the overall breast cancer population, confirming HALP score as an independent predictor of NAT-related pCR; the constructed nomogram integrates HALP score with conventional clinicopathological factors to achieve personalized efficacy and prognosis prediction. What is the implication, and what should change now? • Preoperative HALP score can be used as a low-cost, accessible indicator for NAT efficacy and prognosis prediction in breast cancer. • Incorporate HALP score into clinical assessment of breast cancer patients before NAT; validate the nomogram in multicenter prospective studies to enhance its generalizability. Introduction Breast cancer is a major global health burden that severely endangers women’s lives and health, with its incidence rising steadily worldwide (1-3). Neoadjuvant therapy (NAT) plays a significant role in reducing tumor burden, increasing the rate of breast-conserving surgery, and improving prognosis (4). The attainment of a pathological complete response (pCR) following NAT is widely recognized as a pivotal prognostic indicator associated with favorable clinical outcome in breast cancer patients (5). However, the response of different patients to NAT varies significantly. How to accurately predict the therapeutic effect before treatment is crucial for formulating individualized treatment plans. Both inflammatory and immunonutritional status are recognized as vital for assessing prognostic outcomes in breast cancer (6,7). Accumulating evidence has demonstrated that the chronic inflammatory and immunonutritional microenvironment can drive tumorigenesis and progression through diverse molecular and cellular mechanisms (8). Currently, a number of inflammatory and immunonutritional indicators associated with breast cancer have been identified such as the neutrophil-to-lymphocyte ratio (NLR), the platelet-to-lymphocyte ratio (PLR) and prognostic nutritional index (PNI) (9,10). However, these indicators only reflect one aspect of patients’ overall health, while treatment outcomes are influenced by multiple factors such as inflammation, nutritional status, and immune function. This highlights the need for a comprehensive, multifaceted approach to assessing patients’ overall health and prognosis. The hemoglobin-albumin-lymphocyte-platelet (HALP) score is a composite index that integrates hemoglobin, albumin, lymphocyte, and platelet levels, used to reflect the immune, nutritional, and inflammatory status of patients. In 2015, the HALP score was initially proposed by Chen et al. to forecast the prognosis of gastric cancer (11). Since these parameters can all be obtained through routine blood tests, HALP score has the advantages of low cost and easy accessibility, and has been extensively studied as a prognostic and diagnostic biomarker for various cancers in recent years (12-14). A plausible underlying mechanism is that the HALP score can comprehensively mirror the status of the body’s tumor immune and inflammatory microenvironment (TME) (15). Previous studies have found that preoperative HALP score can be used as a reliable independent predictor of overall survival (OS) and progression-free survival (PFS) in breast cancer patients (16). Currently, there is controversy regarding the use of HALP score to predict the efficacy of NAT in breast cancer according to the results of several prior investigations (17-20). Furthermore, previous studies on the predictive value of the HALP score for NAT efficacy in breast cancer have only focused on specific subgroups, such as triple-negative breast cancer (TNBC) and HER2-positive breast cancer, with no relevant research conducted in the overall breast cancer population (19,21). As a user-friendly visual tool, the nomogram quantifies individual clinical outcome probabilities by integrating multiple predictive factors, offering a non-invasive approach to predict pCR, identify patients most likely to benefit from NAT early in treatment, and avoid ineffective interventions (22). This study aimed to explore the relationship between preoperative HALP score and the NAT efficacy of breast cancer patients, eliminate baseline confounding factors via propensity score matching, and further develop a reliable predictive nomogram for clinical application. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0827/rc). Methods Patient selection This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Review Board at the Gansu Provincial Maternity and Child Health Hospital (Gansu Provincial Central Hospital) [No. (2025) GSFY Ethics (45)]. Written informed consent was obtained from all participants. Patient data was anonymized and de-identified prior to analysis. This study retrospectively reviewed the oncological records of 314 patients diagnosed with breast cancer between January 2013 and November 2024, who underwent pathological evaluation after NAT and had complete follow-up files in the clinic. The inclusion criteria were comprised of (I) patients diagnosed with invasive breast cancer older than 18 years old, (II) patients without metastasis, (III) patients receiving NAT in locally advanced, high-risk breast cancer. The exclusion criteria included (I) individuals with organ failure, (II) individuals whose performance status is not suitable for chemotherapy treatment, (III) individuals with autoimmune diseases, (IV) individuals with secondary malignancies, (V) individuals afflicted by inflammatory illnesses. Data collection and definition The complete blood counts and biochemical values were reviewed before starting NAT, along with clinicopathological information including age, menstruation, ER, PR, HER2, tumor histological grade (Grade), lymphovascular invasion (LVI), Ki67, tumor staging (T), lymph node staging (N), were extracted from the patients’ medical records. Eligible patients were staged according to the eighth edition of the American Joint Committee (AJCC) staging system (23). The expression of estrogen receptor (ER) and progesterone receptor (PR) were scored using the St. Gallen criteria. Human epidermal growth factor receptor-2 (HER2) status was assessed using immunohistochemistry (IHC) staining with a score of 0, 1+, 2+ with negative result of fluorescence in situ hybridization (FISH). HER2 positive status was defined as IHC staining with a score of 3+ or 2+ with positive result of FISH.[23] The response to NAT was evaluated using the Miller-Payne grading system through pathology reports of surgery (24). The NAT regimen and drug dosages refer to the guidelines and specifications (24). The HALP score = [hemoglobin (g/L) × albumin (g/L) × lymphocytes (g/L)]/platelets (g/L). Follow-up and endpoints All patients underwent follow-up through telephone interviews with the patient or their family members if the patient was deceased, outpatient examinations, and review of hospital follow-up records. In this study, pCR was strictly defined in accordance with internationally recognized criteria for breast cancer after neoadjuvant therapy: ypT0/is ypN0. Specifically, no invasive or in situ residual tumor cells were detected in the breast tissue (ypT0/ypTis), and no metastatic tumor cells were found in all resected regional lymph nodes (ypN0) after surgery. OS was defined as the time from initial diagnosis to death from any cause or the final follow-up time. disease-free survival (DFS) was defined as the time interval from the initiation of treatment to disease recurrence, metastasis, or patient death from any cause. Statistical analysis In this study, data processing and analysis were performed using R (version 3.4.4), along with Zstats v0.90 (www.medsta.cn/software). The optimal cut-off value for HALP score and total points were determined by receiver operating characteristic (ROC) curve analysis. Propensity score matching (PSM) was performed using 1:1 nearest neighbor matching without replacement, with a caliper width set at 0.02. This caliper value was chosen according to conventional methodological standards for oncology retrospective studies, which achieves optimal balance of baseline covariates while preserving adequate sample size. Standardized mean difference (SMD) was used to evaluate baseline balance before and after PSM; an SMD <0.1 was considered as satisfactory covariate balance. After PSM, baseline clinicopathological characteristics were well balanced between the high-HALP and low-HALP groups. Univariate and multivariate Logistic regression results were provided for the pCR on various clinical factors. Nomogram was formulated based on the results of multivariate Logistic regression analysis. The predictive performance of the Nomogram was further assessed using the ROC curve analysis, calibration curves, and decision curve analysis (DCA). Kaplan-Meier curves were performed to compare DFS and OS among different points stratification groups of the nomogram. P<0.05 was considered statistically significant. All analyses were performed based on complete-case analysis only without missing data imputation. Missing data in this cohort was identified as missing completely at random, which was not associated with patient NAT response, pCR status, survival outcomes or HALP level. Although excluding patients with incomplete data may introduce slight selection bias, the baseline profile of excluded cases was comparable to the included population, and the influence on the main research conclusions was negligible. To avoid sparse-data bias and quasi-complete separation in logistic regression, T stage and N stage subgroups with extremely small sample sizes were collapsed into adjacent categories. This category merging approach eliminated unstable OR estimates (0.00–Inf) and ensured reliable and interpretable regression results. Rationale for propensity score matching and methodological justification for HALP score. In this retrospective observational study, significant between-group differences in baseline characteristics (age and menopausal status) were observed between the high-HALP and low-HALP groups in the original cohort. We therefore adopted 1:1 PSM to balance all measurable baseline clinicopathological covariates, minimize selection bias and confounding effects, and strengthen the causal inference between HALP score and NAT efficacy. Notably, HALP score served as the grouping variable for PSM, and was simultaneously incorporated into the final predictive nomogram. This design was pre-specified and methodologically reasonable: (I) PSM only balanced confounding clinical variables rather than removing the research exposure factor (HALP). After matching, baseline covariates were well-balanced, while the inherent difference in HALP levels between the two groups was retained, which guaranteed the validity of subsequent association analysis; (II) the primary objective of this study was to develop a comprehensive prediction model for NAT efficacy. As an independent immune-nutritional-inflammatory biomarker closely related to pCR and long-term prognosis, HALP score is an indispensable core predictor. Excluding HALP from the model would lead to loss of key predictive information and reduce the clinical value of the nomogram; (III) this analytical strategy is widely applied in similar retrospective studies focusing on inflammatory and nutritional scoring systems, which ensures the comparability of our results with published literature. Given the relatively limited sample size after PSM (n=226) and the multiple clinicopathological variables incorporated into the multivariate model, the risk of model overfitting was fully considered in this study. We adopted a conventional 7:3 random split for internal validation to preliminarily assess model performance. However, bootstrap validation, a robust method for detecting overfitting and optimizing predictive models, was not performed in the current analysis due to limited computing resources and research conditions. Random split validation can reduce overfitting to a certain extent but has limitations in fully evaluating model stability. Results Baseline characteristic After screening according to the inclusion and exclusion criteria, a total of 314 patients with non-metastatic breast cancer were finally enrolled, of which 118 patients achieved pCR, but 196 patients did not achieve pCR (Figure 1). The cut-off value was determined to be 53.29 for the HALP score by ROC analysis. According to the cut-off value, patients were divided into the high-HALP group (≥53.29, n=115) and the low-HALP group (<53.29, n=199). The characteristics of the two groups were similar, except for age (P=0.001), menstruation (P<0.001) (Table 1). PSM with a 1:1 ratio was conducted to compensate for selection bias and avoid potential confounding effects, ultimately including 113 breast cancer patients in the low-HALP group and 113 breast cancer patients in the high-HALP group (Table 1). The results after PSM were shown that the clinical characteristics among the two groups were well-balanced (all P>0.05). After matching, the patients were randomly divided into a training cohort (n=158) and a validation cohort (n=68) at a ratio of 7:3, with well-balanced (all P>0.05) baseline characteristics between the two groups (Table 2). This splitting strategy was used for internal validation. Given the limitations of single-center data resources, external validation with independent patient cohorts from other medical centers was not carried out in the present study. Flowchart of patient selection and grouping. pCR, pathological complete response. Table 1Baseline clinicopathological characteristics before and after PSMVariableBefore PSMAfter PSMTotal (n=314)High (n=115)Low (n=199)Statistic χ2PSMDTotal (n=226)High (n=113)Low (n=113)Statistic χ2PSMDMenstruation14.319<0.0010.1770.67 Postmenopause169 (53.82)78 (67.83)91 (45.73)−0.444149 (65.93)76 (67.26)73 (64.60)−0.056 Premenopause145 (46.18)37 (32.17)108 (54.27)0.44477 (34.07)37 (32.74)40 (35.40)0.056Age, years10.2650.0010.3000.58 ≤50152 (48.41)42 (36.52)110 (55.28)0.37786 (38.05)41 (36.28)45 (39.82)0.072 >50162 (51.59)73 (63.48)89 (44.72)−0.377140 (61.95)72 (63.72)68 (60.18)−0.072ER0.1230.730.8700.35 Negative157 (50)59 (51.30)98 (49.25)−0.041107 (47.35)57 (50.44)50 (44.25)−0.125 Positive157 (50)56 (48.70)101 (50.75)0.041119 (52.65)56 (49.56)63 (55.75)0.125PR0.9040.341.0400.31 Negative222 (70.7)85 (73.91)137 (68.84)−0.109159 (70.35)83 (73.45)76 (67.26)−0.132 Positive92 (29.3)30 (26.09)62 (31.16)0.10967 (29.65)30 (26.55)37 (32.74)0.132HER23.5120.170.8950.64 0102 (32.59)30 (26.09)72 (36.36)0.21465 (28.76)30 (26.55)35 (30.97)0.096 Low143 (45.69)58 (50.43)85 (42.93)−0.152112 (49.56)56 (49.56)56 (49.56)0.000 Positive68 (21.73)27 (23.48)41 (20.71)−0.06849 (21.68)27 (23.89)22 (19.47)−0.112T0.7990.370.5300.47 T1-T2217 (69.11)83 (72.17)134 (67.34)−0.103159 (70.35)82 (72.57)77 (68.14)−0.095 T3-T497 (30.89)32 (27.83)65 (32.66)0.10367 (29.65)31 (27.43)36 (31.86)0.095N5.2300.070.9680.62 N096 (30.57)43 (37.39)53 (26.63)−0.24378 (34.51)41 (36.28)37 (32.74)−0.075 N1100 (31.85)37 (32.17)63 (31.66)−0.01171 (31.42)37 (32.74)34 (30.09)−0.058 N2-N3118 (37.58)35 (30.43)83 (41.71)0.22977 (34.07)35 (30.97)42 (37.17)0.128Grade0.1860.911.0570.59 I27 (8.6)10 (8.70)17 (8.54)−0.00523 (10.18)10 (8.85)13 (11.50)0.083 II228 (72.61)82 (71.30)146 (73.37)0.047164 (72.57)81 (71.68)83 (73.45)0.040 III59 (18.79)23 (20.00)36 (18.09)−0.05039 (17.26)22 (19.47)17 (15.04)−0.124LVI0.6810.410.3630.55 Negative229 (72.93)87 (75.65)142 (71.36)−0.095166 (73.45)85 (75.22)81 (71.68)−0.079 Positive85 (27.07)28 (24.35)57 (28.64)0.09560 (26.55)28 (24.78)32 (28.32)0.079Ki670.1460.700.1030.75 Low69 (22.04)24 (20.87)45 (22.73)0.04450 (22.12)24 (21.24)26 (23.01)0.042 High244 (77.96)91 (79.13)153 (77.27)−0.044176 (77.88)89 (78.76)87 (76.99)−0.042Data are presented as n (%). ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; PR, progesterone receptor; PSM, propensity score matching; SMD, standardized mean difference; T, tumor. Baseline clinicopathological characteristics before and after PSM Data are presented as n (%). ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; PR, progesterone receptor; PSM, propensity score matching; SMD, standardized mean difference; T, tumor. Table 2Comparison of baseline characteristics of patients in training and validation cohortVariablesTotal (n=226)Validation (n=68)Training (n=158)Statistic χ2PMenstruation2.500.11 Postmenopause149 (65.93)50 (73.53)99 (62.66) Premenopause77 (34.07)18 (26.47)59 (37.34)Age, years0.740.39 ≤5086 (38.05)23 (33.82)63 (39.87) >50140 (61.95)45 (66.18)95 (60.13)ER0.270.60 Negative107 (47.35)34 (50.00)73 (46.20) Positive119 (52.65)34 (50.00)85 (53.80)PR1.010.32 Negative159 (70.35)51 (75.00)108 (68.35) Positive67 (29.65)17 (25.00)50 (31.65)HER21.840.40 065 (28.76)20 (29.41)45 (28.48) Low112 (49.56)37 (54.41)75 (47.47) Positive49 (21.68)11 (16.18)38 (24.05)T1.740.19 T1-T2159 (70.35)52 (76.47)107 (67.72) T3-T467 (29.65)16 (23.53)51 (32.28)N5.790.055 N078 (34.51)20 (29.41)58 (36.71) N171 (31.42)17 (25.00)54 (34.18) N2-N377 (34.07)31 (45.59)46 (29.11)Grade0.310.86 I23 (10.18)8 (11.76)15 (9.49) II164 (72.57)49 (72.06)115 (72.78) III39 (17.26)11 (16.18)28 (17.72)LVI0.450.50 Negative166 (73.45)52 (76.47)114 (72.15) Positive60 (26.55)16 (23.53)44 (27.85)Ki670.110.74 Low50 (22.12)16 (23.53)34 (21.52) High176 (77.88)52 (76.47)124 (78.48)Data are presented as n (%). χ2, Chi-squared test. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; PR, progesterone receptor; T, tumor. Comparison of baseline characteristics of patients in training and validation cohort Data are presented as n (%). χ2, Chi-squared test. ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; PR, progesterone receptor; T, tumor. Univariate and multivariate Logistic regression analyses Univariate Logistic regression showed that ER (OR: 0.27, 95% CI: 0.14–0.54, P<0.001), PR (OR: 0.16, 95% CI: 0.06–0.42, P<0.001), HER2 (OR: 4.76, 95% CI: 2.10–10.79, P<0.001), N (OR: 0.42, 95% CI: 0.19–0.92, P=0.03), grade (OR: 7.29, 95% CI: 1.39–38.15, P=0.02), HALP (OR: 0.49, 95% CI: 0.25–0.95, P=0.03) were associated with a higher pCR. Age, menstruation, T, LVI, and Ki67 showed no significant associations with pCR (all P>0.05). Multivariate Logistic regression confirmed HER2 (OR: 5.00, 95% CI: 1.93–12.92, P<0.001) and HALP as independent predictor of pCR (OR: 0.42, 95% CI: 0.19–0.92, P=0.03), while N, Grade, ER, and PR were no longer statistically significant (all P>0.05) (Table 3). Table 3Univariate and multivariate Logistic regression analyses of factors associated with pCRVariablesUnivariateMultivariateβSEZPOR (95% CI)βSEZPOR (95% CI)Menstruation Postmenopause1.00 (reference) Premenopause0.260.330.780.431.30 (0.68–2.49)Age, years ≤501.00 (reference) >50−0.290.33−0.880.380.75 (0.39–1.42)ER Negative1.00 (reference)1.00 (reference) Positive−1.840.36−5.14<0.0010.16 (0.08–0.32)−1.260.53−2.370.020.28 (0.10–0.81)PR Negative1.00 (reference)1.00 (reference) Positive−2.220.48−4.65<0.0010.11 (0.04–0.28)−1.360.63−2.150.030.26 (0.07–0.88)HER2 01.00 (reference)1.00 (reference) Low−0.860.40−2.140.030.42 (0.19–0.93)0.060.490.130.901.07 (0.41–2.76) Positive1.120.472.400.023.07 (1.23–7.66)1.930.593.30<0.0016.90 (2.19–21.77)T T1-T21.00 (reference) T3-T4−0.280.35−0.790.430.76 (0.38–1.50)N N01.00 (reference)1.00 (reference) N1−0.370.38−0.970.330.69 (0.33–1.46)−0.180.48−0.380.700.83 (0.33–2.12) N2-N3−1.000.42−2.380.020.37 (0.16–0.84)−0.900.50−1.800.070.41 (0.15–1.09)Grade I1.00 (reference)1.00 (reference) II1.540.781.970.0494.66 (1.01–21.59)−0.520.95−0.550.580.59 (0.09–3.80) III2.160.852.540.018.67 (1.64–45.87)−0.421.07−0.390.700.66 (0.08–5.33)LVI Negative1.00 (reference) Positive−0.450.37−1.210.230.64 (0.31–1.32)Ki67 Low1.00 (reference)1.00 (reference) High1.250.462.720.0073.50 (1.42–8.64)1.100.621.780.083.00 (0.90–10.07)CI, confidence interval; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; OR, odds ratio; pCR, pathological complete response; PR, progesterone receptor; SE, standard error; T, tumor. Univariate and multivariate Logistic regression analyses of factors associated with pCR CI, confidence interval; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; LVI, lymphovascular invasion; N, node; OR, odds ratio; pCR, pathological complete response; PR, progesterone receptor; SE, standard error; T, tumor. Establishment and evaluation of predictive models Based on the above mentioned multivariate Logistic regression analyses, nomogram for the prediction of the pCR was constructed based on independent predictive factors, including HER2 and HALP, and meaningful factors including ER, PR, N, and grade (Figure 2). The nomogram showed excellent discrimination with AUC of 0.80 (95% CI: 0.73–0.87) for training cohort and 0.78 (95% CI: 0.66–0.91) for validation cohort (Figure 3). The calibration plots of pCR demonstrated a strong consistency between predicted survival probability and actual survival proportion in both the training (Hosmer-Lemeshow test, P=0.10) and validation (Hosmer-Lemeshow test, P=0.20) cohorts (Figure 4). The DCA demonstrated that the predictive model yielded a favorable net benefit in the training cohorts when the predicted risk ranged from approximately 10% to 60%, and in the validation cohorts when the predicted risk was roughly 20% to 90%, indicating that the model has certain clinical applicability. The total points of each patient in the training cohorts and validation cohorts was calculated separately using the prediction model, and the optimal cut-off values for DFS (169.5044) and OS (117.5106) were determined respectively based on the ROC curves. Patients were then divided into high-point and low-point groups according to the optimal cut-off values (Figure 5). All model performance assessments were based on internal validation. Further external validation is required to confirm the general applicability of this nomogram. Nomograms for predicting pCR of breast cancer. The nomogram integrates six predictive factors: N stage, Grade, ER, PR, HER2, and HALP score. Clinical interpretation: (I) locate the value of each factor for an individual patient on the corresponding axis; (II) draw a vertical line upward to the Points axis to obtain the score for each factor; (III) sum the scores of all factors to get the Total Points; (IV) draw a vertical line downward from the Total Points axis to the pCR axis to obtain the predicted pCR probability. ER, estrogen receptor; HALP, hemoglobin-albumin-lymphocyte-platelet; HER2, human epidermal growth factor receptor 2; N, node; pCR, pathological complete response; PR, progesterone receptor. ROC curves of the nomogram for predicting pCR in the training and validation cohorts. The ROC curve shows the discriminative ability of the nomogram. The AUC of the training cohort is 0.80 (95% CI: 0.73–0.87), and the AUC of the validation cohort is 0.78 (95% CI: 0.66–0.91). AUC, area under the curve; CI, confidence interval; pCR, pathological complete response; ROC, receiver operating characteristic. Calibration plots of the nomogram for predicting pCR in the training (A) and validation (B) cohorts. pCR, pathological complete response. Decision curve analysis of the nomogram for predicting pCR in the training (A) and validation (B) cohorts. The nomogram shows a higher net benefit within the threshold range of 10–60% (training cohort) and 20–90% (validation cohort), indicating favorable clinical utility. pCR, pathological complete response. Prognostic value of HALP for DFS and OS In the primary cohort, the high-HALP group exhibited a significant advantage over the low-HALP group in both DFS (HR: 3.016, 95% CI: 1.329–6.843, P=0.005) and OS (HR: 3.565, 95% CI: 1.226–10.369, P=0.01), respectively. After PSM adjustment, the similar result was observed in both DFS (HR: 0.242, 95% CI: 0.104–0.566, P<0.001) and OS (HR: 0.188, 95% CI: 0.063–0.560, P<0.001) compared the high-HALP group with the low-HALP group (Figure 6). DFS (HR: 0.341, 95% CI: 0.118–0.981, P=0.04) and OS (HR: 0.000, 95% CI: 0.000–0.Inf, P=0.04) of patients in the low-point group were significantly longer than those in the high-point group, with statistically significant differences (Figure 7). Kaplan-Meier curves for DFS and OS between the low-HALP and high-HALP groups before PSM and after PSM: DFS before PSM (A), OS before PSM (B), DFS after PSM (C), OS after PSM (D). The high-HALP group shows significantly better DFS and OS than the low-HALP group both before and after PSM (all P<0.05). DFS, disease-free survival; HALP, hemoglobin-albumin-lymphocyte-platelet; OS, overall survival; PSM, propensity score matching. Kaplan-Meier curves for DFS (A) and OS (B) between the low-point and high-point groups. The low-point group (based on nomogram total points) has significantly longer DFS and OS than the high-point group (all P<0.05). DFS, disease-free survival; OS, overall survival. Discussion pCR is a commonly used core efficacy endpoint for breast cancer patients after receiving NAT, and it is closely related to improved survival outcomes (5,25). In clinical practice, pCR is not only an important indicator for evaluating therapeutic efficacy but also a reference for prognosis judgment and treatment decision-making (26). However, accurately predicting whether patients can achieve pCR before treatment is a key challenge in achieving individualized treatment stratification and optimizing prognosis (27). Currently, many reported predictive biomarkers have limitations in terms of economic accessibility and technical complexity, and the pCR status is significantly influenced by the molecular subtype of breast cancer and specific NAT regimens (28). A single indicator may be insufficient to fully capture its comprehensive impact on pCR, especially in patients with comorbidities, where the complexity of the systemic state further increases the difficulty of prediction. The HALP score comprehensively reflects the host’s anemia severity, nutritional status, immune function, and coagulation status through four parameters: hemoglobin, albumin, lymphocytes, and platelets. Among these, hemoglobin levels are closely associated with the tumor hypoxic microenvironment, and low hemoglobin may induce tumor hypoxia and treatment resistance; as a core indicator of nutritional status, reduced albumin suggests malnutrition and systemic inflammation activation, which may impair the body’s therapeutic response; lymphocytes, as the central component of anti-tumor immunity, their decreased count reduces the efficiency of tumor cell killing; platelets participate in tumor progression by secreting pro-angiogenic and pro-metastatic factors. The interaction of these indicators forms a complex host-tumor crosstalk network. Multiple recent studies have consistently linked high HALP score to favorable survival outcomes in breast cancer (28). In the NAT setting, clinical studies have identified HALP score as a promising candidate for predicting treatment response: when integrated with other biomarkers to construct nomogram models, HALP can improve the accuracy of predicting pCR and graded pathological responses to NAT (29). Despite these findings, critical knowledge gaps persist regarding the precise prognostic value of the HALP score and its potential direct association with pCR in breast cancer patients receiving NAT. PSM can reduce selection bias in retrospective studies and enhance the robustness of the model. Combining PSM and machine learning is expected to enhance the generalization ability and clinical practicality of prediction model (30). In terms of methodology, we fully explained the application of PSM in this study. Given the inherent limitations of retrospective research, unbalanced baseline characteristics may interfere with the true association between HALP score and NAT response. PSM was used to balance demographic and clinicopathological variables to control confounding bias. Differing from conventional grouping indicators, HALP is the core research variable of this study. Retaining HALP in the final prediction model after PSM is a necessary design to construct a multi-factor predictive tool. Our results verified that HALP was an independent predictor of pCR after adjusting for multiple confounders, which further confirmed the stable predictive effect of this biomarker. This combined strategy of PSM and nomogram construction provides a rigorous methodological framework for exploring similar composite scoring systems in oncology. Previous studies have constructed prognostic nomograms incorporating NAT response, which exhibit moderate predictive value for pCR in breast cancer (31-34). Therefore, we conducted a retrospective study involving 314 breast cancer patients to assess the value of the HALP score in predicting pCR in breast cancer. After balancing baseline differences between the high and low HALP groups using the PSM method, the independence of HALP score in predicting pCR can be validated. Combined with Logistic regression analysis, a nomogram can be constructed to achieve individualized survival prediction. A meta-analysis incorporating four retrospective cohort studies with pCR as the primary outcome indicator revealed that AUC of the HALP score for pCR prediction was only 0.57, indicating its limited ability to distinguish between patients with and without pCR (28). However, we found that an elevated HALP score predicted improved pCR in breast cancer patients (P=0.009). This may be attributed to the heterogeneous molecular subtypes of breast cancer, definition of pCR, and variations in treatment regimens. Consistent with the regression results (OR <1), low HALP score was associated with reduced odds of pCR, demonstrating that a higher preoperative HALP score independently predicts a better NAT response and higher pCR probability in breast cancer patients. In this study, the HALP OR value was less than 1, suggesting that elevated HALP level confers a greater chance of achieving pCR, while decreased HALP level correlates with impaired treatment response to NAT. The nomogram, combined the HALP score with conventional clinicopathological factors, was established for the prediction of patients’ pCR based on the results of the multivariate Logistic regression analysis. Previous studies have demonstrated that an increased HALP score is associated with favorable OS (HR: 0.61, 95% CI: 0.39–0.97, P<0.05), while no statistically significant correlation was observed between the HALP score and DFS in breast cancer patients (28). In addition to directly predicting the efficacy of NAT, this study has also explored the relationship between HALP score and prognosis. This study demonstrated that the high-HALP score was significantly associated with better OS and DFS both before and after PSM. Survival prognosis is the result of the long-term interplay of multiple factors, including tumor biological characteristics and the host’s systemic status, and the nutrition-immunity-inflammation status integrated by the HALP score exerts a crucial impact on patients’ survival prognosis. Clinical implications This study explored the specific value of the HALP score in predicting pCR in breast cancer, addressing the research gap whereby existing studies focus predominantly on survival prognosis but pay insufficient attention to immediate treatment response prediction. It clarifies the applicable scope of the HALP score for evaluating breast cancer treatment efficacy, and systematically analyzes the effects of key confounding factors, thus providing clear guidance for subsequent standardized research. Additionally, this study compared the predictive performance of the HALP score for pCR and survival prognosis, yielding clinical evidence to deepen the understanding of its underlying mechanism and suggesting that the score is more suitable as a tool for long-term prognostic assessment rather than for predicting immediate treatment responses. Study limitations Although this study focuses on the predictive value of the HALP score for pCR after NAT in breast cancer and obtains clinically available data, it has several limitations. First, this is a single-center retrospective study, and we only performed internal validation by randomly splitting the enrolled participants into training and validation cohorts. No external validation using independent multicenter patient datasets was conducted. As a result, the generalizability of the established nomogram to patients from other regions, medical institutions or different ethnic groups is restricted. Second, the sample size was relatively small after PSM (n=226), while six predictive variables were included in the multivariate Logistic regression and nomogram. This mismatch between sample size and variable quantity brings about a potential risk of model overfitting. The model may perform well in the current cohort but show decreased discrimination and calibration when applied to new populations. In addition, we did not conduct bootstrap validation, a powerful internal validation approach to assess model stability and detect overfitting, which further limits the reliability of our model evaluation. Third, the enrolled population is relatively homogeneous in ethnicity, which fails to reflect the impact of regional and ethnic differences in nutritional status, immune function and disease characteristics on the predictive efficacy of the HALP score. Fourth, the retrospective design inevitably leads to selection and information biases, and the lack of systematic monitoring data on the dynamic changes of the HALP score throughout treatment makes it difficult to clarify its dynamic correlation with pCR. The molecular subtypes of the study population, treatment regimens, and the optimal cut-off value of HALP will all affect the results. This study adopted complete-case analysis and excluded patients with missing data, which may bring potential subtle selection bias. In addition, we did not perform multiple imputation for missing data; further studies could apply missing data imputation methods to validate the robustness of our findings. Fifth, although PSM was applied with a caliper of 0.02 and 1:1 nearest neighbor matching, we have supplemented SMD evaluation and detailed matching strategy justification in the revised manuscript to improve methodological transparency. Conclusions In summary, our study confirms that a higher preoperative HALP score is independently associated with a higher pCR rate and favorable DFS and OS, whereas low HALP score correlates with reduced pCR odds and inferior survival outcomes. The nomogram based on the HALP score combined with PSM provides a reliable tool for predicting NAT efficacy and facilitating individualized treatment decision-making for breast cancer. Future studies with enlarged sample size, bootstrap internal validation and multicenter external validation are required to eliminate overfitting risk and further verify the stability and generalizability of this nomogram. Supplementary The article’s supplementary files as