PDF
Citation
Print
INTRODUCTION
End-stage liver disease (ESLD) represents a significant health concern. According to data from the Centers for Disease Control and Prevention, chronic liver disease and cirrhosis now rank as the 10th leading cause of mortality in the United States [1]. Despite advances in surgical techniques since the first successful liver transplantation (LT) in 1967, substantial blood loss and the consequent need for massive transfusion (MT) remain critical issues [2].
The complications associated with MT, such as dilutional coagulopathy and thrombocytopenia, add further challenges, necessitating careful volume resuscitation to avoid life-threatening conditions like disseminated intravascular coagulation (DIC) and thrombosis [3]. These issues are further compounded by metabolic derangements, including acid-base disturbances, citrate-induced hypocalcemia, hyperkalemia, and hypothermia [4]. Collectively, these complications may lead to increased postoperative interventions, prolonged mechanical ventilation, extended intensive care unit stays, and higher rates of 30-day mortality and long-term morbidity [5]. This scenario underscores the urgency of optimizing perioperative care for LT patients, particularly in managing blood loss and mitigating the risks associated with MT.
In this context, preoperative predictive models, such as the Model for End-Stage Liver Disease (MELD) and Child-Turcotte-Pugh (CTP) scores, are extensively used to assess short-term survival and inform organ allocation decisions [6]. However, these models have limitations in predicting individual transfusion risk [7]. Furthermore, geographic variations in the use of ESLD scoring systems, which weight constituent parameters differently, complicate direct comparisons of their ability to predict morbidity and mortality. This study aimed to investigate the potential associations between commonly used ESLD scoring systems, including their individual clinical parameters, and both blood loss and the need for intraoperative blood and blood product transfusions during LT.
METHODS
This study was approved by the Research Ethics Committee at the King Faisal Specialist Hospital and Research Center, Riyadh, Saudi Arabia (REC #2211161). Waiver of informed consent was also granted for retrospective examination of the routinely collected clinical data during the patient care episodes.
We conducted a retrospective study using the electronic medical records of 975 adult patients (aged ≥18 years) who underwent LT at a high-volume single center between 2011 and 2021. To ensure data accuracy, records with incomplete data (n=239) and instances of retransplantation (n=17) were excluded. Variables recorded included age, sex, body mass index (BMI), liver function tests, coagulation profiles, surgical details, and intraoperative blood transfusion requirements.
Six established scoring systems, commonly used in clinical practice to predict morbidity and mortality, were applied to assess their ability to predict blood loss in LT cases. These scoring systems included CTP, modified CTP (mCTP), MELD, MELD-Sodium (MELD-Na), integrated MELD (iMELD), and UK MELD (UKELD). Since different weightings are applied to the individual clinical parameters in these formulas, each constituent parameter was also examined separately. These parameters included serum bilirubin, albumin, prothrombin time, international normalized ratio (INR), sodium, creatinine, encephalopathy, ascites, and age.
Statistical Analysis
Data were summarized using means, standard deviations, interquartile ranges, and frequencies. Statistical significance was determined using a two-tailed P<0.05. Data were recorded in Microsoft Excel and analyzed using IBM SPSS ver. 27 (IBM Corp.), with further analyses performed in STATA ver.18 (StataCorp LLC). Q-Q plots indicated that the data were not normally distributed; therefore, nonparametric tests were applied. Data are presented to three significant figures.
The ESLD scoring systems and their constituent parameters were compared to the estimated blood loss volume (eBLV) and transfusion volume (TxV). TxV refers to the combined volume of cell-salvaged blood and the volume corresponding to the number of packed red blood cells (PRBC) units administered, with each unit approximated at 300 mL.
Spearman correlation coefficient was used to assess the associations between the scoring systems and eBLV or TxV. Multiple regression analyses were conducted to evaluate sex, age, and BMI as predictors of eBLV, as well as to analyze preoperative clinical parameters such as laboratory test values. Categorical variables, including ascites and encephalopathy, were analyzed using the chi-square test of association. Additionally, ANOVA and logistic regression analyses were employed to identify scoring system predictors using individual factors of eBLV and PRBC units. eBLV and TxV were categorized into four quartiles for logistic regression analysis to minimize variation and improve overall predictive power.
RESULTS
A total of 719 patients who underwent LT were analyzed. The type of donation and graft type are summarized in Table 1, and the descriptive statistics are presented in Table 2. The six commonly used ESLD scores were calculated and are detailed in Tables 3 and 4.
Spearman Correlation Coefficients for eBLV and ESLD
Table 5 shows the Spearman correlation between eBLV and ESLD. The eBLV was most strongly associated with MELD-Na (R=0.265, P<0.001) and was least associated with UKELD scores (R=0.088, P=0.018).
ESLD Scores: Regression Analysis With eBLV and TxV
Multiple regression analysis was performed to determine which ESLD scoring systems predicted eBLV, with all variables entered simultaneously. There was a statistically significant association between eBLV and all scoring systems except UKELD. The MELD-Na score was the strongest predictor, explaining 7.03% of the variance in eBLV (R2=0.070).
Additionally, multiple regression analysis was conducted to evaluate the effect of the scoring systems on TxV (Table 6). A significant association was found between TxV and the scoring systems, with the exception of UKELD. The MELD class scores emerged as the most significant predictor of variance in TxV (R2=0.032, P<0.001), while the iMELD was the weakest predictor (R2=0.008, P=0.017).
Separate multiple regression analysis was performed for the categorical variables. Both CTP and mCTP were weak but significant predictors of TxV values (R2=0.012; F2, 719=4.194; P=0.015). Although mCTP (β=0.087, P=0.171) appeared to be a stronger predictor of TxV than CTP (β=0.024, P=0.704), this difference was not statistically significant. All values were below the recommended variance inflation factor threshold of 5, and the Durbin-Watson statistic was acceptable.
Preoperative Values and Demographic Variables as Predictors of TxV
The association between TxV and preoperative values is presented in Table 7. All factors except age, albumin, and sodium significantly predicted TxV. The strongest predictor was INR (R2=0.039; β=0.238; P<0.001), accounting for 3.9% of the variance, followed by platelets (R2=0.032; β=−0.003; P<0.001). Demographic variables such as sex, BMI, and age were also analyzed as potential predictors of TxV; however, none reached statistical significance (R2=0.002; F3, 715=0.554; P=0.646). Among these, BMI was the strongest predictor (β=23.876, P=0.473), though it was not statistically significant. Collectively, these demographic variables accounted for only 0.2% of the variance.
Ascites and Encephalopathy With eBLV
The chi-square test revealed a significant association between ascites and eBLV (χ2(64)=95.9, n=719; P=0.006), whereas encephalopathy did not exhibit a significant association (χ2(64)=67.7, n=719; P=0.351).
Logistic Regression Analysis for Predictors of ESLD Scoring Systems: eBLV and Number of PRBC Units
An ordinal logistic regression analysis of the ESLD scoring systems revealed a weak association with eBLV, with the descending order of strength being MELD-Na, MELD, mCTP class, iMELD, CTP class, and UKELD (Table 8). Significant associations were observed between all ESLD scoring systems and the number of PRBC units transfused; among these, the MELD score exhibited the most substantial coefficient (β=0.067), though it explained only 3.5% of the variance.
All parameters, except age, albumin, and sodium, were significantly associated with TxV. Bilirubin exhibited the strongest association, explaining 1.73% of the variance in TxV. Encephalopathy had the highest regression coefficient (1.266), indicating that each unit increase in its severity score corresponds to a 1.266 increase in TxV.
DISCUSSION
Assessing liver disease severity in pre-transplant patients is crucial for determining eligibility for LT and guiding clinical management. Several scoring systems have been developed to predict morbidity and mortality in ESLD, thereby establishing priority on the LT waitlist; however, their adoption varies geographically. Although there is considerable overlap in the clinical and laboratory parameters used to calculate these scores, the differing weightings of these parameters result in variable predictive values for survival without LT, reflecting the complexity of liver disease pathophysiology [6,7]. Moreover, these models have limitations in predicting individual transfusion risk, emphasizing the need for improved methods to guide transfusion decisions [8]. Given the significant variability in blood loss during LT, reliably predicting recipients at risk for increased transfusion requirements is imperative [9]. Research has identified several factors that may aid in risk stratification, including advanced age, high BMI, elevated MELD and CTP scores, pre-existing coagulopathy (characterized by low platelet count, low hemoglobin, and prolonged INR), renal insufficiency, and prolonged graft ischemia time [5,10–12].
Within this context, we identified a small subset of risk factors that were statistically associated with increased transfusion requirements above the population mean for each variable. In general, ESLD scoring systems exhibited a weak correlation with the eBLV. Specifically, the UKELD score demonstrated the weakest correlation, whereas the MELD-Na scoring system showed the strongest correlation with eBLV (P<0.001), accounting for only 7.03% of the variance (R2=0.070). Notably, both MELD-Na and UKELD employ the same parameters (INR, bilirubin, creatinine, and sodium) but assign them different weightings (Supplementary Material 1).
Although some ESLD scoring systems demonstrated statistical significance in predicting TxV, the overall variance explained by these models was low. For instance, MELD class scores were the best predictors for TxV, explaining only 3.24% of the variance (R2=0.032). In contrast, both the UKELD and iMELD scores did not show a significant predictive relationship with TxV, further highlighting the multifactorial nature of intraoperative hemorrhage. A significant association was observed between all ESLD scoring systems and the number of PRBC units transfused, with the MELD score exhibiting the most substantial coefficient (β=0.067), though it accounted for only 3.5% of the variance.
Regression analysis of preoperative laboratory parameters identified bilirubin and INR as the most influential factors for TxV, accounting for 1.73% and 3.9% of the variance, respectively (P<0.001). The platelet count exhibited an inverse relationship (P<0.001), suggesting that lower platelet counts contribute to higher transfusion needs, consistent with the role of coagulation status in managing intraoperative bleeding risks. Compared to encephalopathy, the presence of ascites was significantly associated with eBLV (P<0.006), with ascites showing the highest regression coefficient (1.266) for TxV, thereby underscoring its relevance as a marker for advanced liver dysfunction. Interestingly, age, sex, BMI, albumin, and sodium levels did not significantly predict TxV.
While existing scoring systems offer a foundational understanding of patient risk, our results suggest that they may require calibration or the inclusion of additional factors to improve their predictive validity regarding intraoperative outcomes. The modest correlations observed—especially with the UKELD scoring system—indicate that relying solely on any single scoring system may be insufficient for capturing the dynamic variables influencing intraoperative bleeding and transfusion requirements. Incorporating tools such as thromboelastography and rotational thromboelastometry could improve the identification of patients at higher risk for bleeding, providing insights beyond those offered by traditional scoring systems [13].
Current clinical practices underscore the importance of personalized care in high-risk surgical populations. This approach integrates a comprehensive array of patient-specific metrics, preoperative laboratory hemostatic profiles, and intraoperative factors to develop more accurate predictive models that align with patient blood management initiatives [14,15]. The use of machine learning models to predict eBLV and TxV during LT shows promise for future improvements in this area [16,17].
The strengths of this study include a comprehensive statistical analysis of a large sample size from a single center, which minimizes variations in surgical and anesthetic practices (e.g., transfusion thresholds targeting a hemoglobin of 70–80 g/L, rotational thromboelastometry-guided transfusion protocols, a predominance of donations from living donors [80%] compared to donations after brain death [20%], and the absence of perfusion machine use). However, the single-center design may limit the generalizability of the findings to other patient populations. Additional limitations include the exclusion of several cases due to missing laboratory data and the retrospective design of the study.
A modest correlation was identified between blood loss and several scoring systems (including MELD-Na, MELD, mCTP, etc.) as well as individual parameters (such as ascites, INR, and platelets). However, the overall predictive accuracy for blood loss was only 2.4% (R2=0.024). This finding underscores the complexity of forecasting blood loss during surgery and highlights the need for ongoing research to improve predictive models. Such advancements are essential for optimizing blood transfusion protocols in the context of LT.
ARTICLE INFORMATION
Conflict of Interest
No potential conflict of interest relevant to this article was reported.
Author Contributions
Conceptualization: AM. Data curation: AM, MS, MI, MSN, BT. Formal analysis: AM, MA, LA, AAGA. Investigation: AM. Methodology: AM. Validation: AM, BT. Writing–original draft: NEA. Writing–review & editing: all authors. All authors read and approved the final manuscript.
XML Download