Predicting post-liver transplantation mortality: a retrospective cohort study on risk factor identification and prognostic nomogram construction

Abstract

Background

To identify risk factors for post-transplant mortality and develop a machine learning-integrated prognostic tool to optimise clinical decision-making in liver transplantation (LT) recipients.

Methods

This retrospective cohort study analysed 173 allogeneic LT recipients at the Affiliated Hospital of Zunyi Medical University between August 2019 and December 2023. Clinical and biochemical variables were systematically collected, including recipient profiles [age, gender, prior abdominal surgery Performance Status (PS) scores], biochemical markers (serum creatinine, sodium, albumin, total bilirubin, neutrophil/lymphocyte counts), and prognostic scores [Model for End-Stage Liver Disease (MELD), MELD–sodium (MELD–Na), Child–Turcotte–Pugh (CTP), neutrophil-to-lymphocyte ratio (NLR), and albumin–bilirubin (ALBI)]. Intraoperative metrics, such as blood loss volume and anhepatic phase duration, were also recorded. Univariate and multivariate Cox regression identified mortality predictors. LASSO-regularised Cox regression facilitated variable selection and nomogram construction. Internal validation used decision curve analysis (quantifying clinical net benefit) and time-dependent receiver operating characteristic (ROC) curve analysis [12/18/24-month area under the curve (AUC)]. Kaplan–Meier survival analysis stratified patients into tertiles.

Results

Univariate analysis identified MELD score > 25, blood loss > 5 L, PS score, neutrophil count, total bilirubin level, and MELD–Na score as significant predictors (p < 0.05). Multivariate Cox regression confirmed massive haemorrhage (> 5 L) as an independent mortality predictor (p < 0.001). LASSO-selected predictors (prior abdominal surgery, blood loss > 5 L, and ALBI score) formed a prognostic nomogram demonstrating strong discrimination (1-year AUC: 0.824; 2-year AUC: 0.788). Tertile-based stratification revealed significant intergroup differences in survival (p < 0.001).

Conclusions

Massive intraoperative haemorrhage independently predicted post-LT mortality. The validated nomogram integrating surgical history, haemorrhage severity, and ALBI score enables clinically actionable risk stratification, potentially informing perioperative resource allocation and personalised management protocols.

Background

Liver transplantation remains the definitive treatment for end-stage liver disease. However, critical challenges persist, including donor organ scarcity, substantial surgical trauma, high procedural costs, and the need for lifelong immunosuppressive therapy. These challenges highlight the need for precise prognostic frameworks to optimise donor allocation and improve post-transplant outcomes through risk-stratified clinical decision-making. Despite their proliferation, current prognostic models demonstrate inconsistent predictive accuracy for mortality across heterogeneous cohorts, and existing scoring systems fail to reconcile conflicting evidence regarding key risk determinants. This study aimed to explore the predictive value of relevant indicators before and during surgery, as well as certain scoring systems, for liver transplantation mortality and to conduct a preliminary analysis of survival prognosis after liver transplantation.

Materials and methods

Study population

A retrospective analysis was conducted on the clinical data of 190 patients who underwent allogeneic liver transplantation at the Affiliated Hospital of Zunyi Medical University between August 2019 and December 2023, with 173 patients ultimately meeting the inclusion criteria. The inclusion criteria were as follows: (1) end-stage liver disease (ESLD) secondary to various hepatic pathologies, (2) deceased donor liver transplantation, (3) orthotopic liver transplantation, (4) primary liver transplantation, (5) a uniform surgical team for all procedures, (6) transplantation utilised a non-marginal donor liver, and (7) had comprehensive clinical data sets with no missing data. The exclusion criteria were as follows: (1) paediatric recipients (< 18 years), (2) incomplete clinical records, (3) retransplantation cases, (4) combined multi-organ transplantation, (5) ABO-incompatible transplantation, (6) split-liver transplantation, and (7) recipients of marginal donor grafts. The study protocol adhered to the ethical guidelines and was approved by the Institutional Review Board of Zunyi Medical University. Written informed consent was obtained from all participants. Immunosuppressive protocol: all recipients received a triple immunosuppressive regimen comprising corticosteroids, calcineurin inhibitors (tacrolimus), and mycophenolate mofetil during the early postoperative phase. Corticosteroids were discontinued within 3-month post-transplantation. Maintenance therapy beyond 3 months consisted of tacrolimus monotherapy or combined tacrolimus/mycophenolate mofetil with dosage adjustments guided by therapeutic drug monitoring: tacrolimus trough levels of 8–12 ng/mL (0–3-month post-transplantation), 6–8 ng/mL (3–12-month post-transplantation), and 3–6 ng/mL (> 12-month post-transplantation). This study was approved by the Ethics Committee of the Affiliated Hospital of Zunyi Medical University (approval number: KLLY-2024-130).

Study parameters

The collected data included recipient demographics (age and sex), preoperative clinical indices, and intraoperative variables, including performance status (PS) score, preoperative serum creatinine level, history of abdominal surgery, intraoperative blood loss, anhepatic phase duration, serum sodium concentration, serum albumin levels, absolute neutrophil count, absolute lymphocyte count, total bilirubin, ascites severity, hepatic encephalopathy grading, and prothrombin time. The following prognostic scoring systems were used: model for end-stage liver disease (MELD) score [1], Child–Turcotte–Pugh (CTP) score [2], model for end-stage liver disease sodium (MELD–Na) score [3], albumin–bilirubin index (ALBI), and neutrophil-to-lymphocyte ratio (NLR). The CTP score was categorised into three classes: A (5–6 points), B (7–9 points), and C (≥ 10 points);MELD Stratification defined by thresholds of 20, 25, and 30 points [4]; the model for end-stage liver disease sodium (MELD–Na) scores were dichotomised using a validated cutoff of 24.3, categorising patients into two prognostic groups: Group 1: MELD–Na ≤ 24.3,Group 2: MELD–Na > 24.3 [5]; ALBI = (log₁₀ bilirubin × 0.66) + (albumin ×  − 0.085), where bilirubin is in μmol/L and albumin in g/L [6], and stratified into three prognostic groups using −2.6–1.39 as established thresholds; Intraoperative blood loss was categorised into four groups through combined statistical and clinical stratification. A threshold of 5000 mL was used to define massive haemorrhage cases. The NLR is the ratio of the absolute value of neutral cells to the absolute value of lymphocytes and the neutrophil–lymphocyte ratio.

Study methodology

A cohort of 173 patients was stratified into post-transplant mortality and survival groups for comparative analyses. All statistical procedures were conducted using R with the following analytical workflow: (1) univariate and multivariate Cox proportional hazards regression was used to identify mortality predictors. (2) LASSO-regularised Cox regression optimised variable selection. (3) Prognostic nomogram construction with internal validation via decision curve analysis (DCA), quantifying clinical utility, and time-dependent receiver operating characteristic (ROC) analysis. (4) Kaplan–Meier survival stratification by risk analysis. Normally distributed continuous variables were expressed as mean ± standard deviation (SD), whereas non-normally distributed data were reported as median (interquartile range). Statistical significance was defined as p < 0.05.

Results

Among the 173 patients who underwent transplantation, 159 (91.9%) survived for 1 year postoperatively. Postoperative mortality occurred in 22 patients (12.7%). Early mortality (survival < 3 months [7]) accounted for 14 deaths (8.1% of total transplants; 63.6% of total deaths), with a mean survival time of 20.57 ± 19.44 days. The primary causes of early mortality were severe infections (6 cases), hepatic artery and/or portal vein thrombosis (5 cases), cardiovascular events (2 cases), and acute rejection (1 case). Late mortality (beyond 3 months) accounted for 8 deaths (36.4% of total deaths), with a mean survival of 629.88 ± 250.75 days, predominantly due to tumour recurrence with cachexia (five cases), COVID-19 (one case), chronic rejection (one case), and hepatitis B recurrence with liver failure (one case).The clinical data are summarised in Table 1. Univariate analysis results (Fig. 1) identified several factors associated with early mortality; those reaching statistical significance (p < 0.05) were included in multivariate analysis. Multivariate analysis (Fig. 2) identified intraoperative blood loss > 5000 mL (p = 0.021) as the sole independent risk factor for early post-transplant mortality. LASSO regression analysis using the Cox model further indicated that prior abdominal surgery, blood loss > 5000 mL, and albumin–bilirubin (ALBI) score were significantly associated with survival prognosis (Fig. 3). A nomogram integrating these three predictors was developed to estimate post-transplant survival probability (Fig. 4).Validation via decision curve analysis (DCA, Fig. 5) and ROC curves (Fig. 6) demonstrated clinical utility. ROC analysis showed good predictive performance for 12-, 18-, and 24-month survival, with AUC values of 0.824, 0.818, and 0.788, respectively (Fig. 6). Survival curves stratified by risk groups (Fig. 7) revealed significant differences in survival time between cohorts (p = 0.012).

Table 1.

Clinical data of patients

Index	Variable	Survival group (n = 151)	Death group (n = 22)
Age (year)		48.28 ± 10.08	48.45 ± 9.52
Gender	Male	117	15
	Female	34	7
MELD grading (n)	A < 20	101	9
	B 20–24.9	25	3
	C 25–29.9	10	2
	D > 29.9	15	8
Child–Turcotte–Pugh grading (n)	A	42	5
	B	55	5
	C	54	12
Blood loss (n)	A < 1200 mL	89	9
	B 1200–3500 mL	43	4
	C 3500–5000 mL	14	3
	D 5000 mL	5	6
Preoperative creatinine (umol/L)		77.96 ± 39.85	91.14 ± 78.67
PS score (n)	0	75	4
	1	55	11
	2	13	6
	3	1	1
	4	7	0
Liver free period (min)		64.50 ± 28.12	75.86 ± 59.36
History of abdominal surgery	Yes	0	6
	No	151	16
Serum sodium concentration (mmol/L)		137.62 ± 3.99	136.90 ± 8.80
Neutrophil count (109/L)		2.85 ± 2.29	5.50 ± 5.99
Lymphocyte count (109/L)		0.75 ± 0.45	0.78 ± 0.51
Serum albumin (g/L)		35.15 ± 6.54	33.97 ± 5.74
TBIL (umol/L)		85.61 ± 118.22	167.59 ± 211.02
ALBI grading (n)	1: ≤ − 2.6	26	1
	2: − 2.6 < ALBA ≤ − 1.39	86	10
	3: > − 1.39	39	11
NLR		4.89 ± 6.22	8.00 ± 8.23
MELD–Na grading (n)	A < 24.3	119	11
	B ≥ 24.3	32	11

Fig. 1.

Hazard ratio forest plot of univariate analysis

Fig. 2.

Hazard ratio forest plot of multifactor analysis

Fig. 3.

Variable importance based on LASSO cox model

Fig. 4.

Survival probability predicted by nomogram

Fig. 5.

Decision curve analysis diagram

Fig. 6.

Receiver operating characteristic curve

Fig. 7.

Survival curves of different risk groups

Discussion

Liver transplantation remains the most effective treatment for end-stage liver disease, with reported 1-year and 5-year survival rates of 91% [8] and 73% [9], respectively. A significant proportion of mortalities occur within the first postoperative months [7], after which survival probabilities markedly improve. Consistent with established evidence, our cohort exhibited a 1-year survival rate of 91.2%, with no mortality observed within the initial 3 months. Predominant causes of post-transplant mortality included infection, thrombosis, cardiovascular events, and tumour recurrence [10]. Notably, infection constituted the leading cause of early mortality [7, 11]. In our cohort, infection-related deaths accounted for 42.9% of early post-transplant mortality, potentially attributable to immunosuppression, postoperative ventilation, substantial intraoperative haemorrhage, and pre-existing comorbidities. Regarding late mortality (> 3 months), tumour recurrence was the primary cause. Given that hepatocellular carcinoma (HCC) represented 27.2% of transplant indications, tumour recurrence caused 62.5% (5/8) of late deaths, occurring at 19.65 ± 9.45-month post-transplant. Despite surgical advancements, graft failure remains critical, causing irreversible donor organ loss and reducing transplantation opportunities for other candidates. Thus, meticulous recipient selection is essential for optimizing outcomes. The NLR, a systemic inflammatory marker [12, 13], showed no significant intergroup difference (p > 0.05) despite its established prognostic value in HCC recipients [14], This discrepancy may reflect etiological heterogeneity (metabolic disorders, cirrhosis, drug-induced failure, polycystic disease, and hemangiomatosis), wherein NLR may inadequately capture disease-specific immune dynamics or be confounded by therapeutic interventions. Similarly, the CTP score, though widely used, demonstrated limited prognostic utility due to subjective parameters and unweighted variables [15], showing no intergroup difference. Conversely, the PS score, a validated metric for surgical stress tolerance [16, 17], showed significant intergroup disparity (p < 0.05), aligning with evidence that poorer preoperative functional status correlates with diminished long-term survival. MELD scores, the primary allocation metric, exhibited inconsistent prognostic correlations [18–20], potentially due to treatment-altered biomarkers and interlaboratory variability. Although MELD > 25 and MELD–Na showed intergroup differences, neither emerged as independent mortality predictors in multivariate analysis. Intraoperative haemorrhage, a well-established mortality risk factor [21, 22], in addition to the operative factors, originates from multiple mechanisms: (1) inherent coagulopathies in patients with hepatic failure; (2) surgical adhesions in reoperative fields (particularly tumour cases); (3) portal hypertension-induced collateral circulation; and (4) interrupted fibrinogen production and enhanced fibrinolysis during the anhepatic phase [23]. The albumin–bilirubin (ALBI) index, which incorporates weighted albumin and bilirubin measurements, provides an objective reflection of hepatic synthetic function. Although ALBI grade 3 has been linked to short-term mortality [24], this study did not establish a linear correlation with post-transplant death. Nevertheless, LASSO–Cox modelling identified ALBI as a predictive parameter, suggesting its prognostic relevance in multivariable contexts. Although the absence of prior abdominal surgery in the survival cohort precluded formal intergroup comparison, this variable was incorporated into the LASSO-regularised Cox proportional hazards model, potentially reflecting residual confounding effects in multivariable analysis. These findings require cautious interpretation considering the sample size constraints and inherent assumptions of feature selection algorithms. The nomogram indicates that elevated cumulative risk scores correlate with reduced survival probabilities. For instance, at a total score of 80, the 24-month survival rate exceeds 80%, whereas at a score of 180, the 24-month survival rate is approximately 50%. Risk stratification revealed significantly reduced survival in high-risk versus low-risk groups (p = 0.012). Despite limited sample size, all time-dependent AUC values exceeded 0.75, with minimal decline (Δ = 0.036 between 12 and 24 months) indicating sustained discriminative ability for post-transplant monitoring. When combined with DCA results, the model demonstrated a net benefit > 0.15 at the 30% risk threshold and achieved a 22% reduction in unnecessary interventions compared to the treat-all strategy.

Conclusion

In summary, MELD scores, intraoperative haemorrhage, PS scores, neutrophil counts, total bilirubin, and MELD–Na emerged as prognostic risk factors. The developed nomogram, integrating surgical history, haemorrhage severity, and ALBI score, provides a practical tool for risk stratification, potentially guiding perioperative resource allocation and personalized management. However, our findings indicate that preoperative parameters alone are insufficient for reliably predicting post-transplant mortality. Comprehensive prognostication should incorporate systemic multi-organ function assessment and perioperative management strategies, moving beyond isolated hepatic parameters. As a retrospective single-centre study with inherent limitations, including sample size constraints and variable laboratory standards, our conclusions require validation in large-scale, multi-centre cohorts to establish broader applicability. Although internal validation demonstrated good discriminative ability (AUC 0.788–0.824), the model's performance in external populations remains untested; therefore, extrapolation requires caution, and we strongly encourage independent external validation. Despite these limitations, this study offers valuable insights and a foundation for future prospective research aimed at refining prognostic evaluation in liver transplantation.

Author contributions

Tukui and luodan wrote the main manuscript text, Statistics made by Gu Xuanyu, Clinical data collected by Zheng Zhihong and Jiang jicang, Zhao Lijin is the writing director. All authors reviewed the manuscript.

Funding

Supported by Zunyi City Science and Technology Cooperation HZ Document No. 212 (2024) and Zunyi City Science and Technology Cooperation HZ Document No. 342 (2022).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Affiliated Hospital of Zunyi Medical University (approval number: KLLY-2024-130) and conducted in accordance with the Declaration of Helsinki.

Competing interests

The authors declare no competing interests.