Development of a prediction model integrating cardiac ultrasound parameters for cardiac complications after distal cholangiocarcinoma surgery: a retrospective cohort study

Abstract

Background

Patients undergoing pancreaticoduodenectomy for distal cholangiocarcinoma (dCCA) face a substantial risk of major postoperative cardiac complications (MPCC), which significantly impact mortality and recovery. Existing risk assessment tools lack objective cardiac functional parameters. This study aimed to develop and validate a novel prediction model integrating preoperative cardiac ultrasound parameters to individually predict the risk of MPCC within 30 days after dCCA surgery.

Methods

A retrospective cohort study was conducted on 154 dCCA patients who underwent radical pancreaticoduodenectomy. Univariate and multivariate binary logistic regression analyses were performed to identify independent predictors of MPCC from clinical variables and preoperative transthoracic echocardiography parameters. A nomogram model was constructed based on the identified independent predictors. The model’s discrimination, calibration, and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA), with internal validation via bootstrapping.

Results

The incidence of MPCC was 34.4% (53/154). Multivariate analysis identified preoperative B-type natriuretic peptide (BNP), left ventricular ejection fraction (LVEF), left ventricular mass index (LVMI), and left atrial volume index (LAVI) as independent predictors. A nomogram incorporating these four factors was developed. The model demonstrated excellent discrimination, with an AUC of 0.894 (95% CI: 0.838–0.950). Calibration curves showed good agreement between predicted and observed probabilities. DCA confirmed the model’s clinical net benefit across a wide range of threshold probabilities.

Conclusion

This study presents a robust nomogram that effectively integrates cardiac ultrasound parameters (LVEF, LVMI, LAVI) and BNP to preoperatively predict the risk of cardiac complications following dCCA surgery. The model offers superior individualized risk stratification compared to traditional tools, potentially facilitating optimized perioperative management for high-risk patients.

Background

Distal cholangiocarcinoma (dCCA), a malignant tumor of the middle to lower bile duct, is relatively rare but poses significant diagnostic and therapeutic challenges. Radical pancreaticoduodenectomy (Whipple procedure) is the standard curative approach [1, 2]. However, this surgery’s complexity—involving partial gastrectomy and multiple anastomoses—results in extensive trauma, prolonged operative time, and a high postoperative complication rate [3, 4]. These complications impede recovery, increase costs, and can delay adjuvant therapy, adversely affecting long-term survival.

Among the various complications, cardiac complications—such as perioperative myocardial injury (MINS), decompensated heart failure, significant arrhythmias, and cardiac death—are key risk factors for increased perioperative mortality and poor prognosis. Their pathogenesis involves both pre-existing cardiovascular diseases and surgical stress-induced responses, including systemic inflammation, fluid shifts, hemodynamic fluctuations, and hypercoagulable states [5–7]. In dCCA patients, obstructive jaundice can induce myocardial suppression, reduce vascular resistance, and exacerbate coagulopathy and renal injury, further increasing cardiac risk. Therefore, accurate identification of high-risk patients and targeted prevention are essential in perioperative management. Current preoperative assessment relies on general tools like the American Society of Anesthesiologists (ASA) classification [8] and Revised Cardiac Risk Index (RCRI) [9]. While ASA reflects overall health status, it lacks cardiac functional specificity. Although RCRI focuses on cardiac risk, it relies heavily on medical history and surgical type, representing a macro-level clinical assessment. These tools lack objective, quantifiable functional indicators and fail to reflect the heart’s compensatory capacity under stress. For dCCA patients, obstructive jaundice often causes systemic pathophysiological alterations (e.g., malnutrition, immune suppression), which may further reduce the accuracy of general population-derived models like RCRI, highlighting their notable limitations.

In this context, transthoracic echocardiography (TTE), as a non-invasive, reproducible gold-standard for assessing cardiac structure and function, is increasingly used. It provides not only the left ventricular ejection fraction (LVEF)—a classic measure of global systolic function—but also enables evaluation of diastolic function (e.g., E/e′ ratio) via Doppler techniques, crucial for detecting elevated filling pressures and early cardiac dysfunction. Recent ultrasound advancements offer more sensitive methods; for instance, global longitudinal strain (GLS) quantifies myocardial deformation via speckle tracking and has been validated to detect subclinical dysfunction earlier than LVEF in conditions like hypertension and ischemia, proving valuable for identifying occult cardiac impairment. Additionally, right ventricular function parameters and the left atrial volume index are increasingly recognized as important prognostic predictors [10–12]. While the prognostic value of cardiac ultrasound parameters is well established in cardiovascular disease, predictive models specifically for 30-day postoperative cardiac complication risk in distal cholangiocarcinoma patients are currently lacking.

Therefore, this study aims to identify risk factors for postoperative cardiac complications in distal cholangiocarcinoma patients via a retrospective cohort analysis, focusing on the predictive value of preoperative cardiac ultrasound parameters. We will develop and validate a nomogram model integrating clinical and echocardiographic parameters to provide a practical tool for individualized risk stratification and perioperative management optimization.

Materials and methods

Patient selection

A retrospective analysis was conducted on data from patients with distal cholangiocarcinoma (dCCA) who underwent surgical treatment in the Department of Hepatobiliary Surgery at Beijing Chaoyang Hospital between January 2018 and August 2025. Based on predefined inclusion and exclusion criteria, a total of 154 eligible patients with dCCA were screened for analysis.

Inclusion Criteria: (1) Age between 29 and 80 years; (2) Postoperative pathological confirmation of biliary epithelial carcinoma; (3) Comprehensive preoperative assessment confirming no distant metastasis; (4) Complete perioperative clinical data; (5) No contraindications to surgery based on preoperative evaluation; (6) Informed consent obtained from patients and their families regarding the surgical approach; (7) Preoperative assessment confirming no significant history of cardiovascular disease. （Note: “Significant cardiovascular disease” was defined as active or uncontrolled conditions, including unstable angina, recent myocardial infarction < 3 months, decompensated heart failure, severe valvular disease, or uncontrolled hypertension. Patients with stable, well-controlled conditions (e.g., prior coronary revascularization with no symptoms, or medically managed coronary heart disease) were not excluded, as these were not deemed to significantly increase perioperative risk. This distinction was based on preoperative evaluation involving clinical history, physical exam, ECG, and cardiac biomarkers as needed）

Exclusion Criteria: (1) Intraoperative discovery of distant metastasis; (2) Postoperative pathological confirmation of non-cholangiocarcinoma; (3) Inability to complete the surgery successfully.

The study protocol adhered to the Declaration of Helsinki and received formal approval from the Ethics Committee of Beijing Chaoyang Hospital(Approval No0.2024-D-511). Written informed consent was secured from all participants and their legal representatives prior to data inclusion.

Data collection

This study systematically collected clinical variables and preoperative cardiac ultrasound parameters from all enrolled patients. The collected clinical variables included age, sex, body mass index (BMI), American Society of Anesthesiologists (ASA) classification, preoperative comorbidities (such as hypertension, diabetes mellitus, coronary heart disease with currently stable condition, chronic kidney disease, chronic obstructive pulmonary disease), smoking history, preoperative medication history (including beta-blockers, angiotensin-converting enzyme inhibitors/angiotensin II receptor blockers [ACEI/ARBs], and statins), as well as preoperative troponin (cTn), B-type natriuretic peptide (BNP), N-terminal pro-B-type natriuretic peptide (NT-proBNP), serum creatinine levels, and electrocardiogram (ECG) findings.

Preoperative cardiac ultrasound parameters (transthoracic echocardiography, TTE) were independently reviewed and extracted from reports by two experienced sonographers blinded to the clinical outcomes. The assessment primarily covered the following aspects: (1)Systolic Function: Evaluated by left ventricular ejection fraction (LVEF, %) and ejection fraction (EF, %, calculated as EF (%) = [(EDV - ESV)/EDV] × 100%). (2) Diastolic Function: Core indicators included mitral inflow Doppler velocities - E-wave velocity (cm/s), A-wave velocity (cm/s), and the E/A ratio; tissue Doppler-derived septal e’ velocity (cm/s) and lateral e’ velocity (cm/s), from which the average E/e’ ratio was calculated; and the left atrial volume index (LAVI, ml/m², calculated as LAVI = LAV/BSA). The left atrial volume (LAV) was calculated using the formula: LAV = (π/6) × anterior-posterior diameter × medial-lateral diameter × superior-inferior diameter. Body surface area (BSA, m²) was calculated as: BSA = 0.007184 × height(cm)^0.725 × weight(kg)^0.425. (3) Right Ventricular Function: Assessment indicators included tricuspid annular plane systolic excursion (TAPSE, mm) and pulmonary artery systolic pressure (PASP, mmHg, if estimable). (4) Cardiac Morphology: The left ventricular mass (LVM) was calculated using the Devereux formula: LVM (g) = 0.8 × [1.04 (LVIDd + IVSd + PWTd)3-(LVIDd)3] + 0.6 and was normalized BSA.

Representative echocardiographic images illustrating the key measurements extracted for this study are presented in Fig. 1.

Fig. 1.

Preoperative Echocardiographic Assessment of Cardiac Structure and Function. A & B Septal and lateral early diastolic mitral annular velocities were obtained via tissue Doppler imaging from an apical four-chamber view. C. Pulsed-wave Doppler recording of mitral inflow showing the E-wave and A-wave velocities for the assessment of left ventricular diastolic function. D. Peak tricuspid regurgitation velocity was measured using the Continuous-wave Doppler. E. M-mode echocardiogram at the lateral tricuspid annulus measured Tricuspid Annular Plane Systolic Excursion (TAPSE) for evaluating right ventricular systolic function. F. Two-dimensional image showing left atrial enlargement

Observation period and endpoint definitions

All patients were observed starting from the end of the surgery. The observation period continued until postoperative day 30, allowing for a follow-up window of ± 3 days. The primary endpoint of this study was the occurrence of surgery-related cardiac events within 30 days postoperatively. These event types included myocardial ischemia/myocardial infarction, arrhythmia, acute heart failure and cardiac death. The diagnosis of all endpoint events required comprehensive judgment based on electrocardiogram (ECG), cardiac biomarkers (troponin cTn, BNP/NT-proBNP), and imaging evidence. Events clearly secondary to non-cardiac causes (such as infection, hemorrhage, pulmonary embolism, etc.) were strictly excluded. Specific definitions for endpoint events were as follows: (1) Myocardial Ischemia/Myocardial Infarction: Diagnosis required at least one measurement of cardiac troponin (cTn) exceeding the 99th percentile upper reference limit, accompanied by a typical pattern of rise and/or fall over subsequent hours to days, PLUS at least one of the following: typical symptoms of acute myocardial ischemia (e.g., sustained chest pain); new ischemic ECG changes (ST-segment elevation ≥0.2 mV in men or ≥0.15 mV in women in leads V2–V3, or ≥0.1 mV in other adjacent leads; or new ST-segment depression ≥0.05 mV); imaging evidence of new loss of viable myocardium or new regional wall motion abnormality; or identification of a coronary thrombus by angiography. (2) Arrhythmia: Referred to postoperative occurrences of atrial fibrillation/atrial flutter (with ventricular rate > 120 beats per minute), sustained ventricular tachycardia (duration > 30 seconds), ventricular fibrillation, or high-grade atrioventricular block. This included asymptomatic or transient arrhythmias. (3) Acute Heart Failure: Required the presence of relevant symptoms or signs (e.g., acute dyspnea, orthopnea, pulmonary edema confirmed by chest X-ray, jugular venous distension, or S3 gallop). Laboratory evidence required BNP > 400 pg/mL or NT-proBNP > 1800 pg/mL. Furthermore, the condition necessitated urgent intravenous treatment with diuretics, vasodilators, or inotropic agents. (4) Cardiac Death: Defined as death occurring within 30 days postoperatively, directly caused by a cardiac reason, and strictly excluding deaths secondary to or caused by non-cardiac factors such as infection, hemorrhage, or pulmonary embolism.

Postoperatively, all patients were routinely admitted to the ICU for monitoring. According to protocol, all patients underwent systematic serial high-sensitivity cardiac troponin (cTn) measurements at 48 and 72 hours, ensuring uniform surveillance. Additional testing was triggered by any suspicious symptoms. Preoperative cTn testing, in contrast, was performed based on clinical assessment of individual risk factors. All cases of heart failure and myocardial infarction required verification via emergency transthoracic echocardiography (TTE) or chest X-ray. A blinding principle was applied. Endpoint adjudication was performed by two independent cardiologists unaware of the ultrasound predictor variables, according to predefined criteria. In case of disagreement, a third senior physician arbitrated. Strict distinction was made between primary cardiac events and events secondary to non-cardiac causes (e.g., cardiac dysfunction due to septic shock was categorized as an “infectious complication,” while hypovolemic shock due to major hemorrhage was categorized as a “hemorrhagic complication”). All original documents supporting the diagnoses (e.g., ECGs, troponin reports, ultrasound images, consultation notes) were properly archived. For patients lost to follow-up, their data were handled as “censored,” meaning they were included in the analysis until the last confirmed time of survival, and the reasons for loss to follow-up (e.g., refusal of follow-up, inability to contact) were recorded in detail.

Statistical analysis

In this study, all statistical analyses followed the procedures outlined below: Measurement data conforming to a normal distribution are presented as mean±standard deviation, while non-normally distributed data are presented as median (interquartile range). Categorical data are presented as number (percentage). For intergroup comparisons, the independent samples t-test was used for normally distributed measurement data, and the Mann-Whitney U test was used for non-normally distributed data. Comparisons of categorical data were performed using the chi-square test, or Fisher’s exact test if any expected cell count was less than 1. First, univariate binary logistic regression was employed to screen clinical variables and cardiac ultrasound parameters associated with the outcome (occurrence of Major Postoperative Cardiac Complications, MPCC), with MPCC occurrence as the dependent variable (no complications vs. complications). Variables with a P-value < 0.05 from the univariate analysis were included in a multivariate binary logistic regression model. A backward or stepwise selection method (using p < 0.05 as the criterion for retention) was used to further identify independent predictors. Variance inflation factor (VIF < 5) was examined during modeling to check for multicollinearity among variables. Based on the final set of independent predictors identified, a binary logistic regression model predicting the risk of MPCC occurrence was constructed. A nomogram was further developed to visualize the model and enable individualized risk calculation. Model performance was comprehensively evaluated using the following methods: Discriminative ability was assessed using the area under the receiver operating characteristic curve (AUC or C-index). To benchmark the performance, the Revised Cardiac Risk Index (RCRI) score was calculated for each patient and its predictive performance (AUC) for Major Postoperative Cardiac Complications (MPCC) was evaluated within the same cohort. Calibration was evaluated using calibration curves and the Hosmer-Lemeshow test to assess the agreement between predicted probabilities and observed frequencies. Internal validation was performed using the Bootstrap resampling method (200 repetitions) to correct for optimism in the model performance metrics. Given the number of candidate predictors relative to the number of events, there was a risk of model overfitting. The bootstrap procedure was employed specifically to obtain optimism-corrected estimates of the model’s performance (e.g., AUC). Clinical net benefit across different risk thresholds was assessed using decision curve analysis (DCA). Part of the analysis in this study was completed with SPSS 24.0 software. The parts involving model development and validation were implemented using R software (version 4.3.2) and relevant packages such as rms, pROC, and rmda. All hypothesis tests were two-tailed, and a P-value < 0.05 was considered statistically significant.

Results

General patient characteristics

A total of 154 patients with distal cholangiocarcinoma (dCCA) who underwent surgical treatment were included in this study. The cohort consisted of 86 males and 68 females, with a male-to-female ratio of 1.26:1. The age range was 32–80 years, with a mean age of 63.5 ± 10.6 years. The primary presenting symptoms included: abdominal pain in 9 patients (5.8%), jaundice in 140 patients (90.9%), gastrointestinal discomfort in 3 patients (1.9%), and incidental discovery during physical examination in 2 patients (1.3%). Preoperative percutaneous transhepatic biliary drainage (PTBD) was performed in 29 patients (18.8%), and preoperative endoscopic retrograde cholangiopancreatography (ERCP) for biliary decompression was performed in 21 patients (13.6%). Within this cohort, 38 patients (24.7%) had diabetes mellitus, and 8 patients (5.2%) had a history of previous upper abdominal surgery.

All patients successfully underwent radical pancreaticoduodenectomy, and postoperative pathology confirmed biliary epithelial carcinoma in all cases. Among these, R0 resection was achieved in 139 cases (90.3%), while 15 cases (9.7%) were R1 resections (7 with positive retroperitoneal margins, 5 with negative pancreatic circumferential margins, and 3 with negative proximal bile duct margins). Tumor location was distributed as follows: pancreatic segment of the common bile duct in 81 cases (52.6%) and intradiodeal segment in 73 cases (47.4%). Tumor differentiation was categorized as: well-differentiated in 33 cases (21.4%), moderately-differentiated in 35 cases (22.7%), moderately-to-poorly differentiated in 31 cases (20.1%), moderately-to-well differentiated in 29 cases (18.8%), and poorly differentiated in 26 cases (16.9%). Perineural invasion was present in 37 patients (24.0%), and lymph node metastasis was positive in 32 patients (20.8%). The median intraoperative blood loss for all patients was 500 ml (IQR: 300, 800 ml). Intraoperative blood transfusion was administered to 82 patients (53.2%). The mean operative duration was 9.2 ± 2.7 hours. The mean postoperative hospital stay for this cohort was 19.2 ± 8.7 days. All patients were observed for 30 days postoperatively (allowing a ± 3-day follow-up window). No patients were lost to follow-up.

A total of 38 patients experienced non-cardiac complications, yielding an incidence rate of 24.7%. There were no perioperative deaths. Specific complications included: pancreatic fistula in 14 cases (9.1%; comprising 8 biochemical leaks, 4 Grade B fistulas, and 2 Grade C fistulas), delayed gastric emptying in 17 cases (11.0%), intra-abdominal infection in 8 cases (5.2%), pulmonary infection in 3 cases (1.9%), pleural effusion in 5 cases (3.2%), and diarrhea in 21 cases (13.6%). According to the Clavien-Dindo classification, there were 17 Grade I complications, 14 Grade II complications, and 7 Grade IIIa complications.

Within 30 days postoperatively, 53 patients experienced the primary endpoint event (Major Postoperative Cardiac Complications, MPCC), resulting in an overall incidence of 34.4% (53/154). Arrhythmias were the most common type (49 cases, 31.8%), followed by acute heart failure (3 cases, 1.9%), and cardiac death (1 case, 0.6%).

Analysis of factors influencing postoperative cardiac complications

The results of the univariate analysis are shown in Table 1 below. Age, history of coronary heart disease, CA19-9, BNP, EDV, LVEF, average E/e’ ratio, LAVI, LVH, and LVMI were identified as potential risk factors influencing postoperative cardiac complications in dCCA patients. These identified indicators were included in a multivariate binary logistic regression model for multivariate analysis. BNP (Wald = 7.284, 95% CI: 1.006–1.041), LVEF (Wald = 16.975, 95% CI: 0.691–0.877), LVMI (Wald = 21.828, 95% CI: 1.087–1.225), and LAVI (Wald = 4.985, 95% CI: 1.021–1.379) were identified as independent risk factors for postoperative cardiac complications. Patients with BNP < 100 pg/mL, LVEF > 55%, LVMI < 115 g/m², and LAVI < 34 mL/m² were less likely to experience postoperative cardiac complications.

Table 1.

Univariate and multivariate analysis of cardiac complications within 30 days postoperatively

	Univariate analysis		Multivariate analysis
Factors	χ2 value	P-value	OR value(95% CI)	Wald	P-value
Age	7.917	0.005	1.037(0.985–1.091)	1.911	0.167
Gender	0.336	0.562
BMI(kg/m2)	1.851	0.174
ASA grade (III/IV)	1.010	0.315
Smoking History	1.077	0.299
Diabetes History	0.644	0.422
Coronary heart disease History*	4.577	0.032	1.021(0.283–3.685)	0.001	0.974
Hypertension History	3.030	0.082
CA199 (U/mL)	5.376	0.020	1.014(0.384–2.012)	1.281	0.258
TB (umol/L)	1.866	0.172
GGT (U/L)	0.073	0.787
BNP (pg/mL)	13.233	0.000	1.024(1.006–1.041)	7.284	0.007
cTnI (ng/mL)	1.369	0.242
EDV (ml)	5.672	0.017	1.012(0.984–1.040)	0.688	0.407
ESV (ml)	2.247	0.134
EF（%）	0.011	0.916
LVEF (%)	17.599	0.000	0.778(0.691–0.877)	16.975	0.000
E/A ratio	0.161	0.688
Average E/e’ ratio	7.501	0.006	1.203(0.994–1.456)	3.614	0.057
LAVI (ml/m2)	17.143	0.000	1.187(1.021–1.379)	4.985	0.026
LVH Yes/No	5.329	0.021	0.307(0.088–1.075)	3.408	0.065
LVMI(g/m2)	18.446	0.000	1.154(1.087–1.225)	21.828	0.000
TAPSE (mm)	1.571	0.210
S` (cm/s)	0.061	0.805
Postoperative abdominal complications	0.138	0.710

Variables with a P-value < 0.05 in the univariate analysis were included in the multivariate binary logistic regression model. Empty cells in the multivariate columns indicate variables that were not included in the multivariate analysis as they did not meet the entry criterion (P-value ≥ 0.05 in univariate analysis)

*Coronary heart disease history was recorded only patients with non-significant or stable disease were included per criterion

Establishment of a prediction model for postoperative cardiac complications

Through univariate and multivariate binary logistic regression analysis, four variables were ultimately identified as independent predictive factors for postoperative cardiac complications: BNP (Wald = 7.284, 95% CI: 1.006–1.041), LVEF (Wald = 16.975, 95% CI: 0.691–0.877), LVMI (Wald = 21.828, 95% CI: 1.087–1.225), and LAVI (Wald = 4.985, 95% CI: 1.021–1.379). This nomogram integrates these four independent predictors to construct a predictive model for assessing the risk of cardiac complications within 30 days postoperatively.

As shown in Fig. 2, this scoring system integrates four key clinical parameters: BNP (0–73 points), LVEF (0–95 points), LVMI (0–100 points), and LAVI (0–63 points). The model generates a total score ranging from 0 to 180 points. The scoring procedure involves three steps: First, assign points for each variable. BNP scoring starts at 20 pg/mL, with a value less than 35 pg/mL (5.6 points) considered normal; LVMI scoring starts at 80 g/m², with a value less than 95 g/m² (30 points) considered normal; LVEF scoring starts at 62%, with a value greater than 55% (26 points) considered normal; LAVI scoring starts at 24 mL/m², with a value less than 34 mL/m² (36 points) considered normal. Second, calculate the total score. When all four parameters are at their ideal normal limits, the total score is 97.6 points. Finally, predict the probability. This total score corresponds to a 20% probability of postoperative cardiac complications. The total score corresponds to the probability of postoperative cardiac complications on the bottom axis of the nomogram. For illustration, a total score below 80 points corresponds to a predicted risk of < 10%, while a score above 160 points corresponds to a risk > 90%. Note that these thresholds are descriptive examples from the nomogram and not validated clinical decision rules.

Fig. 2.

Nomogram of the prediction model for cardiac complications within 30 days postoperatively

Specific example: The nomogram provides a user-friendly interface for clinical application. For instance, consider a patient with a preoperative BNP of 60 pg/mL, LVMI of 90 g/m², LVEF of 50%, and LAVI of 36 mL/m². Using the nomogram (Fig. 2): ①For BNP = 60, draw a line upwards to the ‘Points’ axis to assign approximately 15 points. ②For LVMI = 90, assign approximately 20 points. ③For LVEF = 50%, assign approximately 43 points. ④For LAVI = 36, assign approximately 42 points. The total points sum to 120. Drawing a line down from the ‘Total Points’ axis at 120 to the ‘Risk of Complications’ axis yields a predicted probability of approximately 50%. This visual calculation can be performed rapidly at the bedside.

Validation and clinical utility assessment of the model

Based on the comprehensive analysis presented in Fig. 3, the prediction model for cardiac complications within 30 days postoperatively demonstrates excellent performance characteristics across multiple validation metrics. The calibration curve (Fig. 3A) reveals nearly perfect alignment between predicted and observed probabilities, with the Bias-corrected line (solid line) and the Ideal line (dashed line) showing remarkable overlap throughout the entire prediction spectrum. This high degree of consistency underscores the model’s reliability for clinical application. The Bootstrap validation with 200 resamples yielded a Mean Absolute Error of 0.012, well below the 0.05 threshold, and the Hosmer-Lemeshow goodness-of-fit test showed no significant deviation (χ² = 17.93, p = 0.397), collectively confirming excellent model calibration.

Fig. 3.

Calibration curve and ROC curve of the prediction model for cardiac complications within 30 days postoperatively. (A) calibration curve. (B) receiver operating characteristic (ROC) curve. (C) density distribution of independent predictors. (D) risk stratification analysis

The ROC analysis (Fig. 3B) demonstrates outstanding discriminative ability, with an Area Under the Curve of 0.894 (95% CI: 0.838–0.950). The corresponding Somers’ Dxy statistic of 0.788 further validates the model’s strong predictive capability for distinguishing between patients with and without cardiac complications. The density distributions of independent predictors (Fig. 3C) illustrate the relative contributions of BNP, LAVI, LVEF, and LVMI, showing distinct distribution patterns that support their individual predictive value within the model framework.

Risk stratification analysis (Fig. 3D) confirms the model’s clinical utility, effectively categorizing patients into four distinct risk groups with progressively increasing event rates, enabling precise risk assessment and tailored clinical management strategies.

To evaluate the incremental value of the proposed model, its performance was compared against the Revised Cardiac Risk Index (RCRI). The RCRI yielded an AUC of 0.642 (95% CI: 0.556–0.728) for predicting MPCC in our cohort. The nomogram model demonstrated significantly superior discriminative ability compared to the RCRI (AUC: 0.894 vs. 0.642, p < 0.001).

These results collectively affirm that the prediction model incorporating BNP, LAVI, LVMI, and LVEF exhibits robust calibration, excellent discrimination, and meaningful clinical stratification capacity for identifying patients at risk of postoperative cardiac complications following distal cholangiocarcinoma surgery.

The decision curve analysis (Fig. 4A) demonstrates that the “Treat All” strategy’s net benefit decreases linearly with rising threshold probabilities, while “Treat None” remains at a zero baseline. Single-predictor models (BNP Only, LVMI Only, etc.) show limited utility—providing minimal net benefit only at low thresholds (0–20%) before declining rapidly. In contrast, the full model exhibits superior performance across a broad threshold range (10–100%), consistently surpassing both baseline strategies and all single-predictor models. The peak net benefit occurs at 10–60% thresholds, indicating this interval as the optimal clinical decision point. The clinical impact curve (Fig. 4B) further validates the model’s practical value. At lower thresholds (10–20%), it identifies more high-risk patients but with increased false positives. In the optimal 30–50% threshold range, the model maintains an ideal balance—detecting 70–80% of true positives while keeping false positives at clinically manageable levels. This equilibrium is critical for resource optimization and personalized care. Beyond 50% thresholds, the sharp decline in false positives confirms the model’s high specificity, ensuring reliable risk stratification without unnecessary interventions.

Fig. 4.

Clinical decision curve analysis (DCA) of the prediction model for cardiac complications within 30 days postoperatively. (A) decision curve analysis. (B) clinical impact curve

Discussion

This study, through a retrospective cohort analysis, explored the application of a nomogram model integrating preoperative cardiac ultrasound parameters for predicting the risk of major postoperative cardiac complications (MPCC) within 30 days in patients with distal cholangiocarcinoma (dCCA) undergoing radical pancreaticoduodenectomy. The study included 154 dCCA patients and focused on evaluating the predictive value of clinical variables and cardiac ultrasound parameters (such as BNP, LVEF, LVMI, and LAVI). The results showed a postoperative cardiac complication rate of 34.4%, with arrhythmias being the most common. Multivariate analysis identified BNP, LVEF, LVMI, and LAVI as independent predictors, and a nomogram model was constructed based on these factors. Model validation revealed an AUC of 0.894. Both the calibration curve and decision curve analysis (DCA) indicated that the model possesses good discriminative ability, calibration, and clinical utility. This study provides, for the first time, an individualized cardiac risk prediction tool for the specific dCCA population, filling the gap left by existing risk assessment models (such as ASA classification or RCRI) and emphasizing the core role of cardiac ultrasound parameters in perioperative management.

Surgery for distal cholangiocarcinoma is associated with a high rate of postoperative complications due to its anatomical complexity and significant physiological trauma, among which cardiac complications are key factors leading to perioperative mortality and poor prognosis. The MPCC incidence of 34.4% reported in this study is consistent with complication rates for similar major surgeries (30%-50%) reported in the literature by Devereaux et al. [13], but higher than that for general abdominal surgeries. This may be related to myocardial suppression and hemodynamic instability often caused by obstructive jaundice in dCCA patients. And the research reviewed by Devereaux et al., also emphasizes that cardiac events are the main cause of perioperative death in non-cardiac surgeries, particularly in hepatobiliary-pancreatic surgeries, where the cardiac risk in jaundiced patients is underestimated. Existing risk assessment tools, such as the American Society of Anesthesiologists (ASA) classification and the Revised Cardiac Risk Index (RCRI), although widely used in clinical practice, rely primarily on physicians’ subjective medical history collection and macro-level clinical indicators, lacking objective and quantitative assessment of organ function. This limitation is particularly prominent in the distal cholangiocarcinoma (dCCA) patient population. Because traditional tools like the RCRI fail to incorporate specific pathophysiological changes related to obstructive jaundice (such as myocardial suppression and coagulation dysfunction), they exhibit significant insufficiency in specificity when predicting perioperative cardiac risk in dCCA patients. This study successfully overcame this limitation by introducing cardiac ultrasound parameters. This innovative approach highly aligns with the conclusions of a multicenter study by Wang et al. [14] published in JAMA Cardiology. That study, through artificial intelligence-based analysis of ultrasound parameters, confirmed that cardiac ultrasound indicators can predict cardiac decompensation 48 hours earlier than traditional clinical assessment (HR = 2.31, 95% CI: 1.89–2.82), significantly improving the timeliness and accuracy of risk prediction. The value of ultrasound parameters lies in their ability to directly and quantitatively assess the heart’s compensatory potential under stress, rather than relying on indirect clinical symptoms or historical indicators.

This study identified LVEF, LVMI, and LAVI as independent predictors of MPCC, reinforcing the value of cardiac ultrasound in risk assessment. Our methodology is supported by recent advances, such as the multimodular AI algorithm developed by Gül et al. [15] for automated left ventricular assessment, which demonstrates the potential of machine learning to enhance predictive accuracy. While our study relied on manual interpretation by expert sonographers—the current clinical gold standard—the emerging role of artificial intelligence(AI) warrants consideration. AI-based algorithms promise to standardize measurements, reduce inter-observer variability, and enable high-throughput feature extraction. However, in current practice, fully automated systems often still require expert verification, particularly for complex cases or suboptimal image quality. Therefore, future iterations of our model could explore integrating AI-derived parameters to optimize predictive precision and workflow integration, while building upon the expert-driven foundation established here. Our study further revealed that LVEF > 55% serves as a protective factor, suggesting that dCCA patients may require a higher threshold to mitigate risks. This stems from the potential impact of jaundice on cardiac function, which aligns with Yun et al. [16] in Annals of Surgery. While their study primarily focused on resection margin status, it comprehensively explored multiple prognostic factors in distal cholangiocarcinoma patients. Importantly, the inclusion of diastolic function parameters such as LAVI and LVMI underscores the significance of subclinical cardiac dysfunction. Kim et al. [17] in Scientific Reportsconfirmed that left atrial reservoir strain predicts left ventricular filling pressure, while Arslan et al. [18] in the American Journal of Hypertensionutilized speckle-tracking echocardiography to investigate subclinical left ventricular dysfunction in hypertensive patients. Their work further validates the value of diastolic parameters, which is highly consistent with our findings. Collectively, our model achieves a comprehensive evaluation of cardiac function through multiparameter integration, aligning with current innovations in ultrasound technology.

BNP emerged as an independent predictor in this study, reflecting ventricular wall stress and early cardiac dysfunction. We established BNP < 100 pg/mL as the low-risk threshold, consistent with recent research. Poredoš et al. [19] in Cellsestablished a link between systemic inflammation and cardiac biomarkers in their study on inflammation and perioperative cardiovascular events. The inclusion of BNP compensates for limitations in ultrasound parameters by dynamically monitoring neurohumoral activation, particularly in dCCA patients whose jaundice may mask heart failure symptoms. Furthermore, while troponin and other biomarkers were collected in this study, they did not show significance in multivariate analysis, possibly due to sample size or timing of detection. Future studies could incorporate high-frequency monitoring to enhance sensitivity. The synergy between BNP and ultrasound parameters exemplifies the advantage of multimodal assessment, akin to the prognostic stratification strategy based on lymph node metastasis proposed by Hirose et al. [20] in Surgery. Our model’s combination of BNP and ultrasound parameters enables more precise risk stratification, reducing the limitations of single indicators.

The nomogram model in this study visually integrates four independent predictors, allowing clinicians to rapidly calculate individual risk. This approach aligns with the risk assessment methodology used by Gao et al. [21] in Surgical Endoscopywhen comparing laparoscopic and open pancreaticoduodenectomy. Model validation demonstrated an AUC of 0.894, surpassing traditional tools and indicating superior discriminative ability. We have integrated this comparison into the discussion: “Furthermore, our model demonstrated significantly superior discriminative ability (AUC: 0.894) compared to the established RCRI (AUC: 0.642) in our cohort. This direct comparison underscores the added value of integrating objective cardiac ultrasound parameters over traditional, history-based risk indices for the specific dCCA population, potentially leading to more accurate risk stratification. Calibration curves showed close alignment between predicted probabilities and actual observations, supporting model reliability. This rigorous validation framework mirrors that employed by Umino et al. [22] in Annals of Surgical Oncologywhen reappraising the clinical utility of Hi-Cut pancreaticoduodenectomy. Decision curve analysis (DCA) further confirmed the model’s net clinical benefit within the 10–60% threshold range, outperforming single indicators or baseline strategies. This implies that clinicians can adjust preventive measures—such as enhanced monitoring or pharmacological interventions for high-risk patients—based on model scores. Such individualized approaches align with precision medicine principles and reflect trends in recent studies on surgical management of distal cholangiocarcinoma. Overall, our model not only enhances predictive accuracy but also facilitates clinical translation through its user-friendly interface.

This study has several limitations. Its retrospective, monocentric design with a relatively small sample size (n = 154) may introduce selection bias and limit generalizability. Preoperative assessment, including troponin testing, was based on clinical judgment, potentially creating heterogeneity in baseline risk. Furthermore, the exclusion of patients with significant cardiovascular disease might underestimate the true risk, and the model did not incorporate emerging parameters like GLS. Future multicenter prospective studies, standardized protocols, and model updates including more parameters are warranted for validation and improvement. Furthermore, while bootstrap validation corrected for optimism, the variable selection process itself remains vulnerable to overfitting—a limitation inherent in our sample size. External validation in a larger cohort is therefore essential to confirm the generalizability and stability of the predictors.

In summary, this study successfully developed and validated a nomogram model integrating preoperative cardiac ultrasound parameters for predicting the risk of cardiac complications after dCCA surgery. The model emphasizes the synergistic effect of BNP, LVEF, LVMI, and LAVI, providing an individualized tool that compensates for the deficiencies of traditional risk assessment. The results support the promotion of cardiac ultrasound as a routine preoperative assessment in clinical practice, enabling early intervention, especially for high-risk patients. Future work should focus on external validation and clinical integration of the model to ultimately reduce perioperative morbidity and mortality. By incorporating evidence from the literature, this study not only advances perioperative management for dCCA but also provides a reference framework for risk prediction in other high-risk surgeries.

Abbreviations

MPCC

Major Postoperative Cardiac Complications

dCCA

Distal Cholangiocarcinoma

TTE

Transthoracic Echocardiography

LVEF

Left Ventricular Ejection Fraction

LVMI

Left Ventricular Mass Index

LAVI

Left Atrial Volume Index

BNP

B-type Natriuretic Peptide

NT-proBNP

N-terminal pro-B-type Natriuretic Peptide

cTn

Cardiac Troponin

ECG

Electrocardiogram

ASA

American Society of Anesthesiologists

RCRI

Revised Cardiac Risk Index

PTBD

Percutaneous Transhepatic Biliary Drainage

ERCP

Endoscopic Retrograde Cholangiopancreatography

BMI

Body Mass Index

BSA

Body Surface Area

EDV

End-Diastolic Volume

ESV

End-Systolic Volume

GLS

Global Longitudinal Strain

MINS

Myocardial Injury After Non-cardiac Surgery

DCA

Decision Curve Analysis

CI

Confidence Interval

SD

Standard Deviation

Author contributions

Writing-original draft: Wang FF, Jin S; Investigation, Data curation: Wang FF, Jin S, Zhao X, Lyu SC; Project administration: Wang FF, Jin S, Lyu XZ, He Q; Supervision, Writing-review and editing: Lyu XZ and He Q; All authors wrote the manuscript.

Funding

No.

Data availability

The datasets generated and analyzed during the current study are not publicly available due to patient privacy and confidentiality protections under the ethical approval granted by the Ethics Committee of Beijing Chaoyang Hospital (Approval Number: 2024-D-511). All patient identifiers were removed prior to analysis, and data were stored on password-protected, encrypted servers accessible only to the research team. Researchers accessing the data completed training on patient confidentiality. Anonymized data can be made available from the corresponding author (Prof. Qiang He) upon reasonable request, subject to compliance with institutional data-sharing policies, ethical guidelines, and a formal data use agreement. All requests will undergo review to ensure adherence to confidentiality agreements and regulatory requirements.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Beijing Chaoyang Hospital (Approval Number: 2024-D-511). All procedures adhered to the ethical standards of the 1964 Helsinki Declaration.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.