Multicenter external validation of the FW-TRIC score for predicting red blood cell transfusion in on-pump cardiac surgery Authors: Tianlong Wang, Shifu Wang, Renbin Tian, Jing Wang, Qiaoni Zhang, Shan Guo, Bo Wang, Bing Liang, Yisheng Song, Ximeng Yao, Guanglei Li, Lian Duan, Junbo Chuai, Jianchao Li, Liping Shi, Fuqing Jiang, Yi Song, Jian Wang, Gang Liu, Shujie Yan, Yu Liu, Jianxi Ye, Bingyang Ji Published: 2026 ABSTRACT AbstractBackgroundPerioperative red blood cell (RBC) transfusion remains common in cardiac surgery and is associated with increased morbidity and mortality. Accurate preoperative risk prediction is essential for patient blood management. The Fuwai-transfusion risk of on-pump cardiac surgery (FW-TRIC) score was previously developed and internally validated as a tool to estimate transfusion risk. This study aimed to externally validate the FW-TRIC score in a multicenter cohort.MethodsA retrospective multicenter study was conducted across nine cardiovascular centers in China from January to December 2024. Adult patients undergoing on-pump cardiac surgery were included, while those undergoing transplantation, mechanical circulatory support, or preoperative transfusion were excluded. The discriminative performance of the FW-TRIC score was evaluated using receiver operating characteristic (ROC) curve analysis, and calibration was assessed through linear regression of observed versus predicted transfusion rates.ResultsA total of 3,287 patients met inclusion criteria, with an overall perioperative RBC transfusion rate of 40.1%. The FW-TRIC score demonstrated good discrimination for predicting transfusion, with an AUC of 0.713 (95% CI, 0.696–0.731) and excellent calibration. All individual predictors of FW-TRIC score, including age, sex, NYHA classification, body surface area, previous cardiac surgery, emergency surgery, and surgery type, underperformed the prediction ability of FW-TRIC score (all P < 0.05 for AUC comparison). Performance remained good across all nine centers (AUCs > 0.65). Across all transfusion risk categories in the external validation cohort, the observed transfusion rates compared well with those predicted by the FW-TRIC model.ConclusionsThe FW-TRIC score was externally validated as a robust and generalizable preoperative tool for predicting perioperative RBC transfusion in on-pump cardiac surgery. Its use enables effective transfusion risk stratification and supports personalized patient blood management strategies across diverse clinical settings.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13741-026-00707-z.Keywords: Red blood cell, Transfusion, Cardiac surgery, Cardiopulmonary bypass, External validation, Prediction model Introduction Cardiac surgery utilizing cardiopulmonary bypass (CPB) is frequently associated with a substantial requirement for allogeneic red blood cell (RBC) transfusion (Casselman et al. 2025; Salenger et al. 2025; Wahba et al. 2025; Yan et al. 2025; Vlaar et al. 2011). Despite its potential life-saving benefits, RBC transfusion is independently linked to increased risks of adverse outcomes, including infection, renal impairment, prolonged hospital stays, and higher mortality (Carson et al. 2017; Murphy et al. 2007). Consequently, perioperative transfusion practice has emerged as a critical quality indicator in cardiac surgical care. Effective patient blood management hinges on the accurate preoperative identification of patients at high risk for transfusion. This allows for the timely implementation of tailored conservation strategies, such as the use of minimized volume of extracorporeal circuits, intraoperative cell salvage, and precise transfusion triggers, thereby optimizing patient outcomes and conserving valuable blood resources. In 2023, our group developed and internally validated a transfusion risk stratification score to predict perioperative RBC transfusion in patients undergoing on-pump cardiac surgery (Zhang et al. 2023). This tool is later to be referred as Fuwai-transfusion risk of on-pump cardiac surgery score (FW-TRIC score). The score incorporates eight preoperative variables: age, sex, anemia status, New York Heart Association (NYHA) classification, body surface area (BSA), cardiac surgery history, emergency status, and type of surgery. The initial development of the FW-TRIC score was based on a large, single-center retrospective dataset, where it demonstrated good discriminative ability (c-statistic 0.77). However, as with any prediction model derived from a specific population, its generalizability and performance in broader, diverse clinical settings require external validation. The primary endpoint of this multicenter external validation study is to assess the performance of the FW-TRIC score as a predictor of perioperative RBC transfusion in an external, multi-institutional cohort of patients undergoing on-pump cardiac surgery. Methods Study design and ethical approval This multicenter study was approved by the local ethics committee (Approval No. 2024–2379, July 2024). Patient population Data were collected from 9 cardiovascular centers across China between January 1, 2024, and December 31, 2024. Adult patients (age > 18 years) who underwent on-pump cardiac surgery during the study period were eligible for inclusion. Patients were excluded if they had undergone heart transplantation, mechanical circulatory support procedures, or had received any preoperative blood transfusion. Cases with missing data that could not be retrieved from the original medical records were also excluded. Study outcome The primary endpoint of this study was to validate the FW-TRIC score as a predictor of perioperative RBC transfusion in patients undergoing on-pump cardiac surgery using an external dataset. Data collection Data collection included patient demographics, comorbidities, intraoperative variables, and perioperative RBC transfusion data. The details of the FW-TRIC score and transfusion risk stratification criteria are provided in Supplementary Table 1. Perioperative RBC transfusion was defined as any intraoperative or postoperative transfusion of RBCs administered in the operating room, intensive care unit, or general ward after surgery. Sample size The sample size was calculated based on the primary endpoint of perioperative RBC transfusion. According to historical institutional data, the expected perioperative RBC transfusion rate for on-pump cardiac surgery was 23% (Zhang et al. 2023). With an alpha level of 0.001, a power of 80%, a target c-statistic of 0.75, and a null hypothesis c-statistic of 0.50, the required number of events was estimated at 240, corresponding to a total study population of approximately 1043 patients. Statistical analysis The discriminative ability of the FW-TRIC score was assessed using Receiver Operating Characteristic (ROC) curve analysis. The area under the curve (AUC) value, which is identical to the c-statistic for the binary outcome, was used to represent model performance. Calibration was evaluated using linear regression analysis of observed versus predicted perioperative RBC transfusion events. Continuous variables were expressed as mean ± standard deviation or median (Q1–Q3), depending on the normality of distribution as determined by the Shapiro–Wilk test. Between-group differences were analyzed using the Student’s t-test for normally distributed data or the Mann–Whitney U test for non-normally distributed data. Categorical variables were expressed as counts and percentages, and comparisons between groups were made using the Pearson’s chi-squared test. All statistical analyses were performed using R software (version 4.3.3; R Foundation for Statistical Computing, Vienna, Austria). A two-sided P value < 0.05 was considered statistically significant for all analyses. Results A total of 3,287 patients from 9 medical centers in China met the inclusion and exclusion criteria and were included in the study (Supplemental Fig. 1). The final sample size satisfied the initial sample size estimation requirements. The perioperative RBC transfusion rate in this cohort was 40.10%. The general characteristics of the study population are summarized in Table 1. Several preoperative and intraoperative risk factors differed significantly between transfused and non-transfused patients, including age (P < 0.001), sex (P < 0.001), body surface area (BSA) (P < 0.001), NYHA class III/IV (P = 0.02), smoking status (P = 0.04), hypertension (P = 0.001), cerebrovascular disease (P = 0.002), preoperative hemoglobin (Hb) level (P < 0.001), previous cardiac surgery (P < 0.001), emergency surgery (P < 0.001), type of surgery (P < 0.001), cardiopulmonary bypass (CPB) time (P < 0.001), aortic cross-clamp (ACC) time (P < 0.001), and total operative time (P < 0.001). Table 1The general characteristics of the study populationaVariablesTransfusion (1318)No Transfusion (1969)Sex (female, %)676 (51.3)603 (30.6)< 0.001Age (year, median [IQR])60 (16)58 (16)< 0.001Age (year, %)< 0.001 < 60623 (47.3)1141 (57.9) 60–74619 (47.0)774(39.3) ≥ 7576 (5.8)54 (2.7)BSA (m2, %)< 0.001 ≥ 253 (4.0)237 (12.0) 1.8–1.99259 (19.7)579 (29.4) 1.6–1.79466 (35.4)725 (36.8)< 1.6540 (41.0)428 (21.7)NYHA III/IV (%)787 (59.7)958 (48.7)0.02COPD (%)22 (1.7)34 (1.7)0.901Smoke (%)325 (26.7)591 (30.0)0.04Hypertension (%)581 (44.1)759 (38.5)0.001CKD (%)42 (3.2)57 (2.9)0.631Cerebrovascular diseases (%)169 (12.8)185 (9.4)0.002Diabetes (%)210 (15.9)297 (15.1)0.304Preoperative Hb (g/L, median [IQR])125 [23.0]137 [21.8]< 0.001Anemia (%)b< 0.001 None955 (72.5)1676 (85.1) Mild330 (25.0)284 (14.4) Moderate and severe33 (2.5)9 (0.5)Previous cardiac surgery (%)84 (6.4)72 (3.7)< 0.001Emergency surgery (%)116 (8.8)72 (3.7)< 0.001Surgery type (%)< 0.001 Value582 (44.2)920 (46.7) CABG301 (22.8)465 (23.6) CABG + Valve83 (6.3)70 (3.6) CABG + Others21 (1.6)26 (1.3) Aortic surgery216 (16.4)124 (6.3) Others115 (8.7)364 (18.5)CPB time133 [89]114 [79]< 0.001ACC time91 [64]80 [59]< 0.001Operative time315 [150]280 [154]< 0.001FW-TRIC-ScoreScore12 (5)9 (6)< 0.001Class< 0.001 Ⅰ876 (66.5)1724 (87.6) Ⅱ396 (30.0)236 (12.0) Ⅲ46 (3.5)9 (0.5)FW-TRIC score Fuwai-transfusion risk of on-pump cardiac surgery score, BSA Body surface area, NYHA New York Heart Association, COPD Chronic obstructive pulmonary disease, CKD Chronic kidney disease, CABG Coronary artery bypass grafting, CPB Cardiopulmonary bypass, ACC Aortic cross clampingaContinuous variables reported as mean ± standard deviation or median (interquartile range), categorical variables reported as n (%)bAnemia was defined as preoperative hemoglobin value < 110 g/L for women and < 120 g/L for men. Mild anemia was defined as pre-operative hemoglobin value 90–109 g/L for women and 90–119 g/L for men. Moderate and severe anemia was defined as preoperative hemoglobin value < 90 g/L The general characteristics of the study populationa FW-TRIC score Fuwai-transfusion risk of on-pump cardiac surgery score, BSA Body surface area, NYHA New York Heart Association, COPD Chronic obstructive pulmonary disease, CKD Chronic kidney disease, CABG Coronary artery bypass grafting, CPB Cardiopulmonary bypass, ACC Aortic cross clamping aContinuous variables reported as mean ± standard deviation or median (interquartile range), categorical variables reported as n (%) bAnemia was defined as preoperative hemoglobin value < 110 g/L for women and < 120 g/L for men. Mild anemia was defined as pre-operative hemoglobin value 90–109 g/L for women and 90–119 g/L for men. Moderate and severe anemia was defined as preoperative hemoglobin value < 90 g/L The results of univariate logistic regression analyses for the variables related to FW-TRIC score are presented in Supplemental Table 1, showing discrimination consistent with that observed in the modeling dataset (Zhang et al. 2023). The discriminative performance of the FW-TRIC score and its individual components was assessed using receiver operating characteristic analyses for perioperative RBC transfusion (Table 2). Among all variables, the FW-TRIC score demonstrated the highest discriminative ability. In predicting perioperative RBC transfusion, the FW-TRIC score was significantly superior to other predictors (P < 0.001 vs. age, sex, NYHA III/IV, BSA, previous cardiac surgery, emergency surgery, and surgery type; P = 0.023 vs. preoperative Hb). Table 2The discriminative performance of the FW-TRIC score and its individual componentsItemAUC value (95% CI)P-value*Age (year)0.583 (0.563, 0.603) < 0.001Female0.603 (0.586, 0.620) < 0.001Preoperative Hb0.688 (0.669, 0.706)0.023III/IV NYHA level0.539 (0.522, 0.556) < 0.001BSA (m2)0.645 (0.626, 0.664) < 0.001Previous cardiac surgery0.514 (0.506, 0.521) < 0.001Emergency surgery0.530 (0.521, 0.539) < 0.001Surgery type0.599 (0.581, 0.618) < 0.001FW-TRIC score0.713 (0.696, 0.731)RefFW-TRIC score Fuwai-transfusion risk of on-pump cardiac surgery score, BSA Body surface area, NYHA New York Heart Association, AUC Area under the curve*The AUC values of each variable were compared with those of the FW-TRIC score using the Delong nonparametric method The discriminative performance of the FW-TRIC score and its individual components FW-TRIC score Fuwai-transfusion risk of on-pump cardiac surgery score, BSA Body surface area, NYHA New York Heart Association, AUC Area under the curve *The AUC values of each variable were compared with those of the FW-TRIC score using the Delong nonparametric method The ROC analysis for the FW-TRIC score showed an AUC of 0.713 (95% CI 0.696—0.731) (Fig. 1A). The calibration plot indicated that the FW-TRIC prediction model was well-calibrated, with the calibration curve closely following the diagonal reference line (Fig. 1B). The same analysis performed separately in 9 participating medical centers demonstrated consistent results, with all AUC values exceeding 0.65 and calibration curves closely approximating the reference line (Supplemental Figs. 2 and 3). External validation results. A Receiver Operating Characteristics analysis for the Fuwai-transfusion risk of on-pump cardiac surgery score (FW-TRIC score). B Calibration curve in the external validation set The relationship between the FW-TRIC score and the probability of perioperative RBC transfusion is presented in Fig. 2 (logistic regression with 95% confidence interval). When the FW-TRIC score exceeded 13, the probability of transfusion surpassed 50%, and when the score exceeded 19, the probability was greater than 80%. Across all risk groups in the external validation cohort, the observed transfusion rates compared well with those predicted by the FW-TRIC model (Fig. 3). Perioperative RBC transfusion risk predicted rate based on the Fuwai-transfusion risk of on-pump cardiac surgery score (FW-TRIC score). Logistic regression with 95% confidence interval. Dots indicate cases with or without perioperative RBC transfusion Observed versus predicted RBC transfusion rate among three groups in the external validation set Discussion Our multicenter external validation study confirms the robust performance and generalizability of the FW-TRIC score as a predictive tool for perioperative RBC transfusion in patients undergoing on-pump cardiac surgery. The score demonstrated good discrimination (AUC > 0.7) across the entire validation cohort and maintained acceptable performance (AUC > 0.65) in all nine participating centers, underscoring its reliability across diverse practice settings. Importantly, the calibration of the model was excellent, with predicted risks closely aligning with observed transfusion rates across all risk strata. The discriminative power achieved in this external cohort, while slightly lower than in the developmental series, is consistent with the expected performance attenuation when a model is applied to new populations. This successful validation is noteworthy. It can be largely attributed to the multi-center design and the objective data capture from a large sample size, which mitigates the biases inherent in single-center retrospective analyses. Many preoperative factors have been already highlighted as risk factors for perioperative RBC transfusion of on-pump cardiac surgery in previous studies and are confirmed in our study, such as elderly patients, female population, low BSA, preoperative anemia and complex surgery (Salenger et al. 2025; Cavalli et al. 2023; Munoz et al. 2018; 2021; 2016). However, taken alone, they have a poor predictivity for perioperative RBC transfusion. The FW-TRIC score's performance was significantly superior to that of its individual components, highlighting the additive value of combining multiple risk factors into a unified, multifactorial model. This underscores the complex, multi-etiological nature of transfusion requirements in cardiac surgery. The clinical utility of the FW-TRIC score is enhanced by its effective risk stratification. The inflection points identified—where a score > 13 corresponds to a > 50% transfusion probability and a score > 19 corresponds to a > 80% probability—provide clear, actionable thresholds for clinicians. This allows perfusionists and surgical teams to proactively identify high-risk patients before the operation begins. Previous research has confirmed that unnecessary RBC transfusion raised the risk of in-hospital major adverse events across all transfusion-risk patients undergoing on-pump cardiac surgery (Wang et al. 2025). Additionally, the high medical expenses associated with RBC transfusion seriously increased the medical burden (Shamshirian et al. 2020). For those in the medium- and high-risk transfusion categories, targeted blood conservation measures, such as the prepared use of a modified minimal volume extracorporeal circulation system or aggressive cell salvage protocols (Anastasiadis et al. 2021; Gao et al. 2024; Lloyd et al. 2023; Roman et al. 2021), can be readied to avoid unnecessary RBC transfusion risk, moving practice from a reactive to a proactive paradigm. Unlike many existing risk scores which are often limited to isolated coronary artery bypass grafting or incorporate intraoperative variables (Karkouti et al. 2001; Likosky et al. 2017), the FW-TRIC score utilizes only preoperative parameters and is applicable to a wide range of cardiac procedures. This facilitates its integration into preoperative planning and clinical decision-support systems. The finding that the transfusion rate in our contemporary cohort (40.1%) was higher than the historical rate used for sample size calculation (23%) suggests an evolving clinical landscape, possibly reflecting changes in patient complexity or transfusion triggers, yet the model maintained its predictive accuracy. Several limitations of this study should be acknowledged. First, the validation was conducted exclusively within Chinese cardiovascular centers, and its performance in other ethnic and healthcare systems requires further investigation. Second, transfusion triggers may not have been fully standardized across the participating centers, which may have introduced heterogeneity in transfusion practice. Nevertheless, center-specific analyses showed that the FW-TRIC score retained good discriminative ability across all participating centers, suggesting that its predictive performance was relatively robust despite variations in real-world clinical practice. Third, the score predicts the binary outcome of transfusion rather than the volume of RBC units transfused; predicting transfusion quantity represents an important direction for future model refinement. Forth, certain subgroups were underrepresented in this validation dataset, leading to minor discrepancies compared with the original model derivation. For example, the odds ratio for the “CABG + Others” category was higher than for “CABG + Valve,” differing slightly from the original model definition. However, this subgroup of “CABG + Others” represented only 1.4% of the cohort, limiting the impact of this variation. Finally, the highest FW-TRIC score observed in this dataset was 26, suggesting limited validation power for extremely high-risk patients. Nevertheless, the observed transfusion rates among high-risk individuals closely aligned with model predictions, supporting the robustness of the FW-TRIC score in preoperative transfusion risk stratification for on-pump cardiac surgery patients. Conclusion In this study, the FW-TRIC score is validated as a reliable preoperative prediction tool for perioperative RBC transfusion in on-pump cardiac surgery patients, demonstrating good discrimination and excellent calibration across multiple centers. It incorporates readily available clinical variables and facilitates a practical risk stratification that can guide personalized blood management strategies. The integration of the FW-TRIC score into electronic health records for real-time risk calculation and its role in guiding the application of specific blood conservation techniques warrant further research. This tool holds significant promise for enhancing the quality management of perfusion practice and improving patient outcomes. Supplementary Information Supplementary Material 1. Acknowledgements Not applied Authors’ contributions Conceptualization, B.J. and T.W.; methodology, T.W., S.W., Q.Z., J.W., and Y.L.; software, T.W., Q.Z., J.W., G,L; validation, T.W., J.W. and S.Y.; formal analysis, T.W and J.W.; investigation, Y.S., F.J., L.S., J.L., J.C., S.W., R.T., Y.L., J.X. and J.B.; resources, S.G., B.W., B.L., Y.S., X.Y., G.L., L.D., J.C., J.L., and X.Y.; data curation, T.W and S.W.; writing—original draft preparation, T.W.; writing—review and editing, B.J.; visualization, S.W., Q.Z. and R.T.; supervision, B.J.; project administration, B.J.; funding acquisition, B.J. Funding This study was supported by the National High Level Hospital Clinical Research Funding (Grant number: 2025-GSP-GG-9) and Natural Science Foundation of Beijing Municipality (Grant number: 2025-BZJ17). Data availability The original data are available by contact to the corresponding author or first author. Declarations Ethics approval and consent to participate This study received multicenter ethics approval from the Chinese Academy of Medical Sciences&Peking Union Medical College (CAMS&PUMC) Human Research Ethics Committee (Approval No. 2024–2379, July 2024). All patients were exempt from informed consent due to the retrospective nature of this study. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Tianlong Wang, Shifu Wang and Renbin Tian contributed equally to this work. Contributor Information Yu Liu, Email: heroliu2000@sina.com. Jianxi Ye, Email: 13509338687@163.com. Bingyang Ji, Email: jibingyang@fuwai.com. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The original data are available by contact to the corresponding author or first author.