The HALP–CONUT integrated score predicts survival and postoperative complications in locally advanced esophageal squamous cell carcinoma following neoadjuvant therapy: evidence from a multicenter retrospective cohort study

Abstract

Background

The prognostic role of nutrition–inflammation indices in esophageal squamous cell carcinoma (ESCC) requires further investigation. We developed the HALP–CONUT Integrated Score (HCIS) and evaluated its ability to predict survival and postoperative complications in locally advanced ESCC patients undergoing neoadjuvant therapy.

Methods

This multicenter retrospective study analyzed 410 patients. HCIS was calculated by integrating the HALP index and CONUT score. Patients were stratified into risk groups. Survival was analyzed using Kaplan–Meier and Cox regression. Logistic regression assessed associations with complications and pathologic complete response (pCR). Predictive performance was evaluated using ROC curves, C-index, and decision curve analysis.

Results

Patients with high HCIS scores had significantly worse overall survival (OS) compared with those in lower-risk groups (p < 0.001). Multivariate Cox analysis confirmed HCIS as an independent prognostic factor for OS (HR 4.59, 95% CI 2.64–7.97, p < 0.001). Higher HCIS scores were also independently associated with increased risk of major postoperative complications (OR 7.144, 95% CI 3.698–13.813, p < 0.001) and a lower likelihood of achieving pCR (OR 6.470, 95% CI 3.577–12.40, p = 0.005). Predictive models incorporating HCIS demonstrated superior discrimination compared with models based on HALP or CONUT alone (AUC for OS: 0.789 vs 0.711 and 0.703, respectively).

Conclusion

HCIS is a robust biomarker that integrates hematologic, nutritional, and inflammatory parameters. It independently predicts survival, postoperative complications, and pCR, providing a valuable tool for risk stratification and individualized treatment planning in locally advanced ESCC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12957-026-04413-4.

Introduction

Esophageal cancer remains one of the leading causes of cancer-related mortality worldwide, with esophageal squamous cell carcinoma (ESCC) being the predominant histological type in East Asia [1]. Despite advances in multimodal therapy, including neoadjuvant chemoradiotherapy and surgery, the prognosis of locally advanced ESCC remains unsatisfactory, with high rates of recurrence and postoperative complications [2, 3]. Reliable biomarkers for risk stratification and outcome prediction are urgently needed to guide individualized treatment.

Malnutrition and systemic inflammation have been recognized as critical determinants of treatment tolerance and long-term outcomes in cancer patients [4]. Several composite indices derived from hematological and nutritional parameters, such as the Controlling Nutritional Status (CONUT) score and the Hemoglobin, Albumin, Lymphocyte, and Platelet (HALP) index, have shown prognostic value in various malignancies [5, 6]. Although both HALP and CONUT have been independently associated with prognosis in esophageal cancer, each index provides only a partial assessment of the host’s nutritional–inflammatory status. HALP incorporates hemoglobin, albumin, lymphocytes, and platelets, thereby reflecting anemia, nutritional reserve, immunocompetence, and thrombocytosis-driven inflammation. In contrast, CONUT includes albumin, total cholesterol, and lymphocytes, placing additional emphasis on lipid metabolism and protein-energy malnutrition. Notably, HALP uniquely captures hemoglobin and platelets – both independent prognostic factors in esophageal cancer – while CONUT uniquely captures total cholesterol, a marker of hepatic synthetic function. Thus, these two indices are complementary rather than redundant. A patient with preserved HALP but abnormal CONUT may have isolated nutritional depletion without significant systemic inflammation; conversely, abnormal HALP with normal CONUT might indicate inflammation-driven thrombocytopenia or anemia without overt malnutrition. We therefore hypothesized that integrating HALP and CONUT into a single composite score (HCIS) could provide a more holistic assessment of the malnutrition–inflammation–immunity axis, potentially improving risk stratification beyond either index alone. However, each index has limitations when applied independently, as they fail to capture the multidimensional interplay between nutrition, systemic inflammation, and immunity [7, 8].

To overcome these limitations, we developed the HALP–CONUT Integrated Score (HCIS), a novel index combining HALP and CONUT to provide a more comprehensive reflection of host nutritional–inflammatory status. Preliminary evidence suggests that integrated indices may improve predictive accuracy compared with single measures [9, 10]. However, the prognostic and predictive role of HCIS in patients with locally advanced ESCC undergoing neoadjuvant therapy has not yet been investigated.

Therefore, this multicenter retrospective cohort study aimed to evaluate the predictive value of HCIS for overall survival, postoperative complications, and pathological complete response (pCR) in patients with locally advanced ESCC after neoadjuvant therapy. We further compared HCIS with existing indices to determine whether this integrated score provides superior risk stratification and clinical utility.

Materials and methods

Data collection

This multicenter retrospective study was approved by the ethics committee of the lead institution, Fujian Medical University Union Hospital, and received either exemption or approval from the ethics committees of the Provincial Hospital Affiliated to Fuzhou University, Fuzhou Pulmonary Hospital, and Zhangzhou Second Hospital. Written informed consent was obtained from all participants. The study included adult patients who underwent total esophagectomy between July 2012 and November 2019 at the four participating hospitals.

Participant selection

Inclusion criteria were: (1) histopathologically confirmed esophageal squamous cell carcinoma after surgery; (2) patients who received neoadjuvant therapy prior to surgery; (3) resectable tumor lesions after neoadjuvant therapy; and (4) presence of a single tumor lesion.

Exclusion criteria included: (1) CONUT score > 4; (2) history within one month before surgery of pulmonary infection, severe chronic obstructive pulmonary disease, pulmonary bullae, respiratory or cardiac failure, hyperglycemia, acute or chronic renal insufficiency, significant malnutrition, inflammatory or thrombotic diseases, or liver failure; (3) multiple primary malignancies; (4) missing preoperative or postoperative laboratory or clinical data; (5) loss to follow-up or refusal of follow-up after surgery; and (6) missing neoadjuvant-surgery interval or an interval exceeding six weeks.

Initially, 817 patients diagnosed with ESCC were identified from the databases of the four hospitals. Among them, 213 patients with advanced disease were deemed ineligible for surgical intervention. After initial screening, 604 patients remained as potential candidates. After applying stringent inclusion and exclusion criteria, 410 patients with locally advanced ESCC who underwent surgery after neoadjuvant therapy were ultimately included(Fig. 1).

Fig. 1.

Patient flow diagram showing the screening, exclusion and inclusion process that yielded the final cohort of 410 patients with locally advanced esophageal squamous cell carcinoma who received neoadjuvant therapy followed by radical esophagectomy

Operational definitions

Based on preoperative CONUT scores, patients were classified into a normal group (n = 124) and a mildly abnormal group (n = 286). The HALP score was calculated as: HALP = Hb (g/L) × Albumin (g/L) × Lymphocyte (10⁹/L)/Platelet (10⁹/L). The optimal HALP cutoff value of 39.78 was determined using the maximum Youden index (sensitivity + specificity – 1) (AUC = 0.711, 95% CI: 0.661–0.762). To evaluate the stability of this cutoff, internal validation was performed using Bootstrap resampling (1000 repetitions). Results showed a concentrated Bootstrap distribution (mean: 40.3) with a 95% confidence interval (36.37–45.08) highly consistent with the original cutoff (39.78), indicating good consistency and stability (Figure S1). Patients were divided into high HALP (n = 220) and low HALP (n = 190) groups. Before constructing HCIS, we considered alternative modeling strategies, including multivariable Cox regression with all individual components (hemoglobin, albumin, lymphocyte, platelet, and total cholesterol) as separate covariates. However, such a model was prone to multicollinearity (e.g., albumin appears in both HALP and CONUT) and overfitting, and would be less practical for routine clinical use. We also explored machine learning approaches (e.g., LASSO), but these lack transparency and require external validation. Therefore, we chose a parsimonious composite score that preserves the simplicity and clinical accessibility of the original indices while integrating their complementary information. HCIS was constructed as follows: combining CONUT and HALP, patients were further stratified into three groups:

• Low group: High HALP and normal CONUT

• Intermediate group: High HALP and mildly abnormal CONUT, or low HALP and normal CONUT

• High group: Low HALP and mildly abnormal CONUT

All minimally invasive esophagectomy (MIE) procedures were performed by senior thoracic surgeons with extensive experience in MIE. Operative protocols were standardized across surgical approaches.

The decision for postoperative discharge was based on the attending physician’s assessment and fulfillment of discharge criteria: (1) optimal general health with adapted enteral nutrition and restored bowel function; (2) smooth recovery from surgical stress with effective pain control; (3) normal body temperature, unremarkable thoracic/abdominal examinations, and laboratory results within or near normal ranges; (4) early ambulation leading to improved activities of daily living after discharge; (5) absence of wound infection symptoms and complete removal of drainage tubes; and (6) postoperative CT showing satisfactory anastomotic healing. The primary outcome was overall survival (OS), calculated from the date of discharge to the date of death or the last follow-up for surviving patients.

Measured variables

Preoperative demographic and clinical data were collected, including blood samples, age, sex, body mass index (BMI), OS, length of hospital stay (LOS), smoking history (lifetime cumulative smoking ≥ 100 cigarettes), alcohol use (at least one drink in the past 12 months), preoperative tumor imaging, tumor location, postoperative TNM stage, presence of neural or vascular invasion, and postoperative complications. Fasting peripheral blood samples were collected 1–7 days before surgery. The normal BMI range was defined as 18.5–23.9 kg/m². Established reference intervals for blood biomarkers included: hemoglobin (female: 110–150 g/L; male: 120–165 g/L), albumin (40–55 g/L), total cholesterol (131.48–235.89 mg/dL), neutrophil count (1.5–7.0 × 10⁹/L), lymphocyte count (0.8–4.0 × 10⁹/L), monocyte count (0.12–0.8 × 10⁹/L), platelet count (100–300 × 10⁹/L), carcinoembryonic antigen (0–5 ng/mL), and CA19-9 (0–37 U/mL).

Statistical analysis

Statistical analyses were performed using R software (version 4.5.0). Continuous variables are expressed as median (interquartile range), and categorical variables as frequency (percentage). Intergroup comparisons were conducted using the Kruskal–Wallis test or χ² test, as appropriate. Receiver operating characteristic (ROC) curve analysis was employed to evaluate the predictive performance of HCIS, HALP, and CONUT for 5-year survival, and differences in area under the curve (AUC) were compared using DeLong’s test. Category-free NRI and IDI were computed with 1,000 bootstrap resamples using the survival IDINRI package (R 4.5.0). Survival analysis was performed using the Kaplan–Meier method with log-rank test. Multivariate survival analysis was carried out using Cox proportional hazards models, and logistic regression models were used to identify independent predictors of complications and pathological complete response (pCR).

To mitigate potential confounding, propensity score matching (PSM) was implemented using 1:1:1 nearest-neighbor matching with a caliper width of 0.05 standard deviations of the propensity score. The propensity score model included the following baseline covariates, which differed significantly across HCIS strata in the unmatched cohort (Table 2): age, sex, BMI, Charlson comorbidity index, smoking status, alcohol use, clinical T stage, clinical N stage, surgical approach (McKeown vs Ivor-Lewis), number of lymph nodes retrieved, and adjuvant therapy (yes/no). Post-matching balance was assessed using both p-values and standardized mean differences (SMD); an SMD < 0.1 was considered indicative of good balance. The 1:1:1 matching scheme was chosen to enable direct three-way comparisons among low-, intermediate-, and high-HCIS groups with equal sample sizes, maximizing statistical power while minimizing selection bias. A nomogram was constructed and validated through calibration curves, concordance index (C-index), time-dependent ROC analysis, and decision curve analysis (DCA).

Table 2.

Univariate and multivariate logistic regression analyses of pathological complete response (pCR) in patients with locally advanced esophageal squamous cell carcinoma following neoadjuvant therapy

	Univariate analysis		Multivariate analysis
	OR(95%CI)	P	OR(95%CI)	P
Age(years)
< 60	1.00(Reference)		1.00(Reference)
≥ 60	0.687(0.358–1.317)	0.258	0.671(0.321–1.400)	0.287
Sex
Female	1.00(Reference)		1.00(Reference)
Male	0.997(0.459–2.167)	0.994	1.161(0.342–3.943)	0.811
cT
1	1.00(Reference)		1.00(Reference)
2	1.040(0.439–2.466)	0.087	1.581(0.547–3.573)	0.398
3	1.139(0.508–2.555)	0.751	0.838(0.330–2.126)	0.710
4	2.329(0.884–6.136)	0.928	0.741(0.444–2.031)	0.560
cN
0	1.00(Reference)		1.00(Reference)
1	1.743(1.140–3.529)	0.015	1.650(0.660–4.150)	0.283
2	3.176(1.367–7.378)	0.007	1.321(0.664–2.743)	0.409
3	3.207(0.740–4.107)	0.204	1.524(0.986–2.471)	0.749
Smoke
No	1.00(Reference)		1.00(Reference)
Yes	1.417(1.014–2.547)	0.294	1.339(1.147–2.146)	0.057
Drink
No	1.00(Reference)		1.00(Reference)
Yes	0.992(0.527–1.866)	0.979	1.707(0.724–4.027)	0.222
CEA(ng/ml)	1.016(0.903–1.144)	0.786	0.966(0.846–1.104)	0.616
CA199(U/ml)	1.008(0.997–1.019)	0.170	1.010(0.997–1.022)	0.126
Platelet counts(× 109/L)	1.001(0.996–1.006)	0.805	0.998(0.992–1.004)	0.558
Neutrophil counts (× 109/L)	1.373(1.051–1.794)	0.020	1.520(1.078–2.144)	0.017
Monocyte count (× 109/L)	1.717(1.313–2.411)	0.533	1.810(1.195–3.472)	0.602
Hemoglobin(g/L)	1.001(0.980–1.022)	0.922	1.016(0.986–1.048)	0.302
Albumin(g/L)	0.999(0.920–1.086)	0.989	1.057(0.940–1.189)	0.351
Total cholesterol(mg/dl)	0.992(0.986–0.998)	0.013	0.990(0.982–0.998)	0.020
HCIS
Low	1.00(Reference)		1.00(Reference)
Intermediate	3.030(1.293–7.099)	0.011	4.930(2.733–8.146)	0.003
High	3.767(1.289–11.011)	0.015	6.470(3.577–12.40)	0.005

Results with borderline statistical significance (i.e., P-values approaching 0.05) should be interpreted with caution. A P-value < 0.05 was considered statistically significant.

Note on multiple comparisons

The multivariate Cox and logistic regression models in this study were designed to validate the independent predictive value of prespecified core variables (such as HCIS score and pathological stage) based on prior research and clinical rationale, rather than to perform exploratory testing of a large number of hypotheses. Given the limited number of variables included in the final models (Cox model: 4; complication logistic model: 2; pCR logistic model: 3), the risk of Type I error inflation due to multiple comparisons was considered low. Therefore, to maintain consistency with comparable studies and to directly report adjusted effect estimates, no additional multiplicity adjustment was applied to the P-values in the multivariate models. Sensitivity analyses were subsequently conducted to assess the robustness of the findings (see Sects. 3.4, 3.5, and 3.6).

Results

Baseline characteristics and treatment analysis of esophageal cancer patients

We analyzed 410 patients with locally advanced ESCC treated with neoadjuvant therapy followed by esophagectomy; cohort demographics, tumor location, comorbidity burden, and laboratory indices are summarized in Table S1. The cohort was predominantly male with a median age of 60 years, and tumors most commonly arose in the middle thoracic esophagus.

By preoperative HCIS stratification, 46.10% were low-risk, 31.22% intermediate-risk, and 22.68% high-risk (Table S1). Across HCIS strata, baseline and treatment features differed for BMI (p = 0.012), sex (p = 0.004), Charlson index (p = 0.013), clinical T and N stage (both p ≤ 0.006), surgical approach (p = 0.026), number of retrieved nodes (p = 0.019), and adjuvant regimen (p < 0.001), with higher HCIS aligning with lower BMI and more advanced cT/cN (Table S2).

All patients underwent esophagectomy (McKeown and Ivor-Lewis were the most common approaches) with a median of 34 lymph nodes examined (Table S1). A pathologic complete response was achieved in 10.49% overall. Group-wise matched comparisons are presented in Table S3.

Comparison of predictive performance of HCIS, HALP, and CONUT scores for 5-year survival

Using ROC analysis with DeLong’s test for between-model comparison, HCIS demonstrated the highest discriminative ability for 5-year overall survival, with an AUC of 0.789, compared with 0.649 for PNI,0.711 for HALP and 0.703 for CONUT (Figure S2). To quantify reclassification and discrimination gains, we calculated NRI and IDI. Versus HALP, HCIS improved reclassification by 0.124 (95% CI 0.021–0.235, P < 0.05) and discrimination by an IDI of 0.143 (95% CI 0.070–0.213, P < 0.05). Versus CONUT, NRI was 0.485 (95% CI 0.404–0.564, P < 0.001) and IDI was 0.097 (95% CI 0.038–0.160, P < 0.05). Versus PNI, NRI was 0.351 (95% CI 0.275–0.412, P < 0.001) and IDI was 0.157 (95% CI 0.042–0.183, P < 0.05). Collectively, these metrics indicate that HCIS outperforms its component indices not only in overall accuracy but also in patient-level risk reclassification, supporting its use for prognostic stratification after neoadjuvant therapy.

Survival differences by HCIS stratification and impact of adjuvant therapy

Patients were stratified by preoperative HCIS into low- (n = 189, 46.1%), intermediate- (n = 128, 31.2%), and high-score (n = 93, 22.7%) groups. Baseline characteristics differed significantly across the three strata for BMI, sex, Charlson comorbidity index, clinical stage, surgical approach, number of lymph nodes retrieved, and adjuvant regimen (all p < 0.05; Table S2). To reduce confounding, propensity score matching (PSM) was performed, yielding 64 well-balanced patients per group (192 total; Table S3). Thus, 218 patients (53.2% of the original cohort) were excluded because they could not be matched within the prespecified caliper. After matching, all covariates were adequately balanced, confirming good model calibration.

Kaplan–Meier analysis demonstrated significant differences in overall survival (OS) among HCIS groups after matching (p < 0.001; Figure S3A). Patients with higher HCIS had progressively poorer OS, validating HCIS as a strong prognostic stratification tool. Length-of-stay (LOS) analyses further showed longer postoperative hospitalization in the high HCIS group, whereas low HCIS patients experienced the shortest LOS (Figure S3B).

Within-group evaluation of adjuvant therapy effects revealed no survival advantage in the low- or intermediate-HCIS groups (p = 0.72 and 0.88, respectively). In contrast, adjuvant therapy significantly improved OS in high-HCIS patients (p = 0.0041), suggesting that those with the most unfavorable inflammatory-nutritional profiles derive the greatest benefit. Kaplan–Meier curves depicting these differences are presented in Figures S4A-C.

Predictors of overall survival

On univariate analysis, multiple clinicopathologic and hematologic variables were associated with overall survival (OS), including sex, BMI, smoking status, ypT stage, ypN stage, perineural and lymphovascular invasion, HCIS category, neutrophil and monocyte counts, hemoglobin, albumin, and total cholesterol (all P < 0.05).

In the multivariable Cox model, four factors remained independently prognostic (Figure S5).

1. HCIS category: Relative to the low-risk group, the intermediate-risk group had a 90% higher mortality risk (HR 1.898, 95% CI 1.256–2.868, P = 0.002), and the high-risk group had an approximately 4.6-fold higher risk (HR 4.588, 95% CI 2.461–8.533, P < 0.001).

2. ypT stage: Mortality risk increased stepwise with higher pathologic T stage versus ypT0—ypT1 (HR 2.323, 95% CI 1.147–4.704, P = 0.019); ypT2 (HR 2.956, 95% CI 1.584–5.517, P < 0.001); ypT3 (HR 3.331, 95% CI 1.849–6.002, P < 0.001); ypT4 (HR 3.729, 95% CI 2.158–7.467, P = 0.015).

3. ypN stage: A graded association was also observed versus ypN0—ypN1 (HR 1.420, 95% CI 1.121–2.184, P = 0.005); ypN2 (HR 1.721, 95% CI 1.080–2.743, P = 0.022); ypN3 (HR 2.492, 95% CI 1.266–4.905, P = 0.008).

4. BMI: BMI > 23.9 kg/m² correlated with shorter OS (HR 1.470, 95% CI 1.042–2.072, P = 0.028). The detail results are summarized in Table S4.

To account for multiple comparisons, we applied false discovery rate (FDR) correction to the four significant signals from the multivariable model. All associations remained significant after adjustment, supporting robustness of the primary findings: HCIS (intermediate vs low: P < sub > FDR </sub > = 0.015; high vs low: P < sub > FDR </sub > = 0.009), ypT (ypT1 vs ypT0: P < sub > FDR </sub > = 0.027; ypT2: P < sub > FDR </sub > = 0.043; ypT3: P < sub > FDR </sub > = 0.036; ypT4: P < sub > FDR </sub > = 0.028), ypN (ypN1 vs ypN0: P < sub > FDR </sub > = 0.033; ypN2: P < sub > FDR </sub > = 0.028; ypN3: P < sub > FDR </sub > = 0.041), and BMI (P < sub > FDR </sub > = 0.034).

Subgroup analyses stratified by sex, BMI, surgical approach, adjuvant-therapy type, tumor location, and clinical T/N stage showed a consistent gradient of risk across HCIS strata: compared with the low-HCIS group, both intermediate- and high-HCIS groups exhibited higher OS mortality across most subgroups, with the high-HCIS group uniformly at greatest risk (Figure S6).

For clinical translation, we integrated HCIS, ypT, ypN, and BMI into a nomogram derived from the multivariable model (C-index 0.747; 95% CI 0.713–0.781) (Fig. 2A). Internal validation demonstrated close agreement between predicted and observed 3- and 5-year OS (Fig. 2B-C). Using the scoring system, total score < 100 identified patients with > 90% 3-year OS, while total score < 40 corresponded to > 90% 5-year OS.

Fig. 2.

A Nomogram to predict OS in patients with locally advanced ESCC undergoing neoadjuvant therapy. Points assignment: For each patient, locate their value on the axis for each prognostic factor (HCIS, ypT, ypN, BMI). Draw a line upward to the ‘Points’ axis to determine the corresponding score for each variable. Sum the scores for all variables to obtain the ‘Total Points’. Finally, locate the total points on the ‘Total Points’ axis and draw a line straight down to the ‘3-Year OS Probability’ and ‘5-Year OS Probability’ axes to estimate the survival probability. B-C Calibration curves for 3-year OS (B) and 5-year OS (C) prediction based on nomogram. D Compare the ability to predict OS with the pathological TNM stage by t-AUC. E DCA curve of nomogram-HCIS vs ypTNM.The horizontal black line represents the net benefit when all patients are classified as negative (no intervention), whereas the slanted green line represents the net benefit when all patients are classified as positive (all receive intervention)

Time-dependent ROC analyses confirmed the nomogram’s superior discrimination versus traditional ypTNM across time points (AUC 0.777–0.875 vs 0.571–0.690) (Fig. 2D). Decision curve analysis further showed a consistently higher net clinical benefit for the nomogram across threshold probabilities of 0.20–0.80 (Fig. 2E). Collectively, these findings indicate that the nomogram offers a more accurate and clinically practical approach to survival prediction after neoadjuvant therapy than conventional staging.

Predictive factors for postoperative complications

To develop a risk tool for severe postoperative complications (Clavien–Dindo grade ≥ III, the distribution of complication types across HCIS strata is summarized in Supplementary Table S5), we performed univariate and multivariable logistic regression (Table 1). On univariate analysis, lymphocyte count, hemoglobin, albumin, and HCIS category were associated with complications. After multivariable adjustment, only lymphocyte count (OR 1.729, 95% CI 1.114–2.682, P = 0.015) and HCIS category (intermediate vs low: OR 2.393, 95% CI 1.128–5.076; high vs low: OR 7.771, 95% CI 3.521–17.150; both P < 0.05) remained independently predictive.

Table 1.

Univariate and multivariate logistic regression analyses assessing the association between clinical factors and postoperative Clavien-Dindo grade ≥ III complications, presented as OR (95% CI) and P-value

	Univariate analysis		Multivariate analysis
	OR(95%CI)	P	OR(95%CI)	P
Age(years)
< 60	1.00(Reference)		1.00(Reference)
≥ 60	1.622(0.973–2.701)	0.063	1.612(0.915–2.838)	0.098
BMI(kg/m2)	1.010(0.929–1.098)	0.824	1.065(0.971–1.167)	0.18
Sex
Female	1.00(Reference)		1.00(Reference)
Male	1.143(0.614–2.126)	0.674	1.989(0.826–4.786)	0.125
Smoke
No	1.00(Reference)		1.00(Reference)
Yes	1.013(0.612–1.677)	0.959	1.344(0.629–2.874)	0.445
Drink
No	1.00(Reference)		1.00(Reference)
Yes	0.894(0.546–1.465)	0.658	0.712(0.363–1.400)	0.325
Surgery
Ivor-Lewis	1.00(Reference)		1.00(Reference)
McKeown	1.204(0.733–1.978)	0.464	0.951(0.491–1.840)	0.881
Number of retrieved nodes(n)	1.014(0.986–1.043)	0.318	1.006(0.970–1.044)	0.742
Platelet counts(× 109/L)	1.000(0.996–1.003)	0.854	0.998(0.993–1.002)	0.356
Neutrophil counts (× 109/L)	1.110(0.962–1.281)	0.154	1.041(0.869–1.247)	0.661
Lymphocyte count (× 109/L)	1.583(1.090–2.298)	0.016	1.729(1.114–2.682)	0.015
Monocyte count (× 109/L)	1.594(0.540–4.704)	0.398	1.034(0.231–4.627)	0.965
Hemoglobin(g/L)	0.973(0.957–0.989)	0.001	0.984(0.963–1.005)	0.137
Albumin(g/L)	0.933(0.876–0.993)	0.030	1.018(0.943–1.099)	0.646
Total cholesterol(mg/dl)	0.995(0.990–1.001)	0.083	1.002(0.996–1.009)	0.424
HCIS
Low	1.00(Reference)		1.00(Reference)
Intermediate	2.624(1.339–5.145)	0.005	2.393(1.128–5.076)	0.023
High	7.144(3.695–13.813)	< 0.001	7.771(3.521–17.150)	< 0.001

To address multiplicity, FDR correction was applied to the two significant signals; both retained statistical significance (lymphocyte count: P < sub > FDR </sub > = 0.030; HCIS—intermediate vs low: P < sub > FDR </sub > = 0.046; high vs low: P < sub > FDR </sub > < 0.001), supporting their robustness.

A two-factor nomogram incorporating lymphocyte count and HCIS demonstrated good performance (C-index/AUC 0.733, 95% CI 0.675–0.796) with excellent calibration (Fig. 3A-C). Decision-curve analysis showed higher net clinical benefit than “treat-all” or “treat-none” strategies for threshold probabilities 0.20–0.60 (Fig. 3D). This model provides a concise, practical tool for preoperative risk stratification and individualized perioperative management in locally advanced ESCC after neoadjuvant therapy.

Fig. 3.

A Nomogram to predict Clavien-Dindo grade ≥ III complications after surgery. Points assignment: For each patient, determine the points for ‘Lymphocyte count’ and ‘HCIS’ by projecting their values to the ‘Points’ axis. Sum these points to get the ‘Total Points’. The predicted probability of complications can then be read by projecting the ‘Total Points’ down to the ‘Predicted Probability’ axis. B Calibration curves for Clavien-Dindo grade ≥ III complications based on nomogram. C ROC curve of the nomogram for predicting postoperative Clavien–Dindo grade ≥ III complications. Area under the curve (AUC) = 0.733 (95% CI 0.675–0.796). D DCA curve of nomogram for Clavien-Dindo grade ≥ III complications.The horizontal red line represents the net benefit when all patients are classified as negative (no intervention), whereas the slanted black line represents the net benefit when all patients are classified as positive (all receive intervention)

Predictive Factors for Pathological Complete Response (pCR)

Univariate logistic regression identified neutrophil count, total cholesterol, and HCIS category as significant correlates of pathological complete response (pCR). In multivariable analysis, all three remained independent predictors: neutrophil count (OR 1.520, P = 0.017), total cholesterol (OR 0.990, P = 0.020), and HCIS (intermediate vs low: OR 4.930, P = 0.003; high vs low: OR 6.470, P = 0.005) (Table 2).

After FDR correction, significance persisted for neutrophil count (P < sub > FDR </sub > = 0.043) and HCIS (intermediate vs low and high vs low, each P < sub > FDR </sub > = 0.015), while total cholesterol approached significance (P < sub > FDR </sub > = 0.060). These findings indicate that elevated neutrophil levels and higher inflammatory–nutritional burden, reflected by HCIS, are robust predictors of reduced pCR probability.

A three-factor nomogram incorporating neutrophil count, total cholesterol, and HCIS achieved a C-index of 0.717 (95% CI 0.650–0.798) (Fig. 4A). The calibration curve showed close concordance between predicted and observed pCR probabilities (Fig. 4B). ROC analysis confirmed good discrimination with an AUC of 0.717 (95% CI 0.650–0.798) (Fig. 4C). Finally, decision-curve analysis (DCA) demonstrated a clear net clinical benefit for the nomogram across a decision-threshold range of 0.10–0.50, outperforming “treat-all” and “treat-none” strategies (Fig. 4D). Taken together, this model offers a concise, interpretable, and clinically applicable tool for preoperative identification of patients most likely to achieve pCR following neoadjuvant therapy in locally advanced ESCC.

Fig. 4.

A Nomogram to predict pCR after surgery. Points assignment: The points for ‘HCIS’, ‘Total Cholesterol’, and ‘Neutrophil count’ are assigned based on patient values. The sum of these points yields the ‘Total Points’, which corresponds to the predicted probability of achieving pCR on the bottom axis. B Calibration curves for pCR based on nomogram. C ROC curve of the nomogram for predicting pCR. Area under the curve (AUC) = 0.717 (95% CI 0.650–0.798). D DCA curve of nomogram for pCR.The horizontal red line represents the net benefit when all patients are classified as negative (no intervention), whereas the slanted black line represents the net benefit when all patients are classified as positive (all receive intervention)

Discussion

In this multicenter retrospective cohort of locally advanced ESCC treated with neoadjuvant therapy followed by esophagectomy, we show that the HALP–CONUT Integrated Score (HCIS) offers reproducible and clinically meaningful risk stratification. HCIS outperformed HALP and CONUT for 5-year survival discrimination and remained independently associated with overall survival after adjustment for pathologic factors. Importantly, HCIS also predicted major postoperative complications and pathologic complete response (pCR), indicating utility across perioperative safety and oncologic efficacy domains. Together, these results support HCIS as a pragmatic biomarker for individualized risk assessment in ESCC.

The biological rationale for integrating HALP and CONUT lies in their complementary domains. HALP captures anemia (hemoglobin), nutritional reserve (albumin), immunocompetence (lymphocyte), and platelet-driven inflammation, whereas CONUT additionally incorporates total cholesterol – a marker of hepatic lipid metabolism and long-term nutritional status. By combining them, HCIS identifies patients with multidimensional host deterioration (malnutrition + inflammation + immune dysfunction + metabolic dysregulation), a state that likely promotes tumor progression, reduces treatment tolerance, and increases perioperative complications. This integrated interpretation is supported by the stepwise increase in mortality and complication risk across HCIS strata observed in our study.

Our findings fit within a broad evidence base linking systemic inflammation and nutritional depletion to adverse cancer outcomes and treatment intolerance [11–15]. By integrating hematologic and nutritional components, HCIS captures the inflammation–malnutrition–immune dysregulation axis that shapes host–tumor interactions [7–10, 16]. Elevated neutrophils and relative lymphopenia are associated with immunosuppressive tumor microenvironments [17, 18], while hypoalbuminemia and disordered lipid metabolism reflect impaired reserve and repair [4, 19, 20]; these mechanisms plausibly underlie the observed associations between higher HCIS and worse OS, greater complication burden, and lower pCR. In this context, HCIS provides a single, accessible index that consolidates signals previously reported across multiple inflammation–nutrition metrics [3, 5, 6, 8, 10, 21].

We also observed heterogeneity of adjuvant benefit by baseline inflammatory–nutritional risk: adjuvant therapy conferred no measurable advantage in low- and intermediate-HCIS strata but was associated with significantly improved survival among high-HCIS patients. This pattern aligns with literature suggesting that systemic inflammation and malnutrition amplify perioperative risk and blunt treatment efficacy, yet may also identify patients who derive outsized benefit from intensified postoperative strategies when appropriately selected [21–23]. Clinically, these data argue for risk-adapted pathways: (i) prehabilitation and targeted nutritional/immunonutritional optimization for patients with elevated HCIS [21, 24]; and (ii) shared decision-making around adjuvant intensification that prioritizes the high-HCIS subgroup [2, 22, 23].

To facilitate translation, we combined HCIS, ypT, ypN, and BMI into a parsimonious nomogram that showed strong discrimination, excellent calibration at 3 and 5 years, and greater net clinical benefit than ypTNM across relevant thresholds. Complementary models for severe complications and pCR achieved acceptable accuracy and decision-analytic utility, supporting use of HCIS for preoperative counseling and optimization as well as postoperative treatment selection. The robustness of the principal associations after FDR correction further mitigates concerns about multiplicity.

Our study has several strengths: (i) a relatively large, multicenter cohort representative of real-world ESCC care; (ii) comprehensive endpoints spanning survival, perioperative safety, and treatment response; (iii) rigorous statistics, including propensity score matching, internal validation, and decision-curve analysis; and (iv) a clinically deployable tool that synthesizes host and tumor information. Limitations should also be acknowledged. The retrospective design introduces potential residual confounding despite matching and multivariable adjustment. External validation was not performed, limiting generalizability beyond participating centers. Laboratory indices were assessed at baseline; whether dynamic HCIS changes during treatment further refine prediction warrants investigation. Additionally, while we explored heterogeneity of the adjuvant effect across HCIS strata, treatment allocation was not randomized.

These findings collectively position HCIS as a pragmatic biomarker to bridge preoperative assessment and postoperative decision-making in ESCC. Future work should prospectively validate HCIS and the derived nomograms across diverse populations, assess dynamic (time-updated) trajectories of HCIS during neoadjuvant and adjuvant therapy, and test HCIS-guided strategies—such as targeted nutritional optimization, immunonutrition, and risk-adapted adjuvant intensification—in randomized or adaptive trial designs. If confirmed, HCIS-based stratification could help clinicians identify patients who will benefit most from perioperative optimization and adjuvant therapy while sparing low-risk patients from unnecessary toxicity.

Conclusion

In conclusion, HCIS integrates nutritional and inflammatory information into a single, clinically accessible index that outperforms HALP and CONUT, independently predicts survival, major complications, and pCR, and identifies a subgroup most likely to benefit from adjuvant therapy. Used alone or within our nomogram, HCIS provides a practical framework for individualized, evidence-based management of locally advanced ESCC after neoadjuvant therapy.

Authors’ contributions

Conception and design: HC,XH,ZY,CX; Administrative support: CC,BZ; Provision of study materials or patients:HC, YW, YL, RH, LC; Collection and assembly of data: HC,XH,ZL;Data analysis and interpretation: All authors; Manuscript writing: All authors; Final approval of manuscript: All authors.

Funding

This research was funded by a grant from Clinical Research Center for Thoracic Tumors of Fujian Province(2022Y2007); National Natural Science Foundation of China (82203307); Fujian provincial health technology project (2022GGA021); Talent Fund Project of Fujian Medical University Union Hospital (2021XH029) and Joint Fund for the innovation of science and Technology, Fujian province (Grant number: 2023Y9204).

Data availability

The datasets used and analysed during the current study available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Approval of the research protocol by an Institutional Reviewer Board: The Ethics Committee of Union Hospital, Fujian Medical University (No:2020WSJK072).Informed Consent: All participants were informed about the new system and provided written informed consent. Registry and the Registration No. of the study/trial: N/A. Animal Studies: N/A.

Competing interests

The authors declare no competing interests.