Use of machine learning models for the prognostication of liver transplantation: A systematic review

Table 1 Summary table of included studies

Ref.	Context	Aim	Methods	Results	Conclusion
Briceño et al[53], 2014	A Spanish study using a two-fold ANN model which included, the positive survival and the negative loss models were implored to predict 3 mo graft survival post LT	To test the accuracy of ANN inpredicting post-transplant outcomes and compare with other conventional models	Sixty-four donor and recipient variables from a set of 1003 LT from a multicenter study including 11 Spanish centers were included. For each D-R pair, common statistics (simple and multiple regression models) and ANN formulae for two non-complementary probability-models of 3-months graft-survival and -loss were calculated: a positive-survival (NN-CCR) and a negative-loss (NN-MS) model. The NN models were obtained by using the Neural Net Evolutionary Programming (NNEP) algorithm. Additionally, receiver-operating curves (ROC) were performed to validate ANN against other scores	Optimal results for NN-CCR and NN-MS models were obtained, with the best performance in predicting the probability of graft-survival (90.79%) and -loss (71.42%) for each D-R pair, significantly improving results from multiple regressions. ROC curves for 3- months graft-survival and –loss predictions were significantly more accurate for ANN than for other scores in both NN-CCR (AUROC-ANN = 0.80 vs –MELD = 0.50; -D-MELD = 0.54; -P- 5 SOFT = 0.54; -SOFT = 0.55; –BAR = 0.67 and -DRI = 0.42) and NN-MS (AUROC-ANN = 0.82 vs – MELD = 0.41; -D-MELD = 0.47; -P-SOFT = 0.43; -SOFT = 0.57, -BAR = 0.61 and -DRI = 0.48)	ANN maybe considered a powerful decision-making technology for this dataset, optimizing the principles of justice, efficiency and equity. This may be a useful tool for predicting 3-months outcome and a potential research area for future D-R matching models
Ershoff et al[54], 2020	An American study in which DNN was trained on pre transplant data and compared with the BAR and SOFT scores in predicting 90-d mortality post LT	The primary aim of the study was to classify recipients with 90-d post-liver transplant mortality using DNNs	In this study, we trained a DNN to predict 90-d post -transplant mortality using preoperative variables and compared the performance to that of the Survival Outcomes Following Liver Transplantation (SOFT) and Balance of Risk (BAR) scores, using United Network of Organ Sharing data on adult patients who received a deceased donor liver transplant between 2005 and 2015 (n = 57544). The DNN was trained using 202 features, and the best DNN’s architecture consisted of 5 hidden layers with 110 neurons each	The area under the receiver operating characteristics curve (AUC) of the best DNN model was 0.703 (95%CI: 0.682-0.726) as compared to 0.655 (95%CI: 0.633-0.678) and 0.688 (95%CI: 0.667-0.711) for the BAR score and SOFT score, respectively	Despite the complexity of DNN, it did not achieve a significantly higher discriminative performance than the SOFT score. Future risk models will likely benefit from the inclusion of other data sources, including high-resolution clinical features for which DNNs are particularly apt to outperform conventional statistical methods
Lau et al[55], 2015	An Australian study proposing an algorithm made from 15 donor, recipient and transplant factors selected by ML predicting mortality within 30 days after LT	To evaluate the utility of machine-learning algorithms, such as random forests and artificial neural networks, to predict outcome based on donor and recipient variables which are known before organ allocation	Liver transplant data from the Austin Hospital, Melbourne, Australia, from 2010 to 2013 has been included in the study. The top 15 donor, recipient, and transplant factors influencing the outcome of graft failure within 30 days were selected using a machine learning methodology. An algorithm predicting the outcome of interest was developed using those factors	Donor risk index predicts the outcome with an area under the receiver operating characteristic curve (AUC-ROC) value of 0.680 (95%CI: 0.669-0.690). The combination of the factors used in donor risk index with the model for end-stage liver disease score yields an AUC-ROC of 0.764 (95%CI: 0.756-0.771), whereas survival outcomes after liver transplantation (LT) score obtains an AUC-ROC of 0.638 (95%CI: 0.632-0.645). The top 15 donor and recipient characteristics within random forests results in an AUC-ROC of 0.818 (95%CI: 0.812-0.824)	This study confirms that machine-learning algorithms based on donor and recipient variables which are known before organ allocation can be utilized to predict transplant outcomes
Liu et al[56], 2020	A Chinese study using ML to predict 30 d survival after LT	To use data-driven technique to develop a predictive model using ML to predict postoperative survival within 30 days for the patients who have undergone LT	We use random forest (RF) to select important features, including clinically used features and new features discovered from physiological measurement values. Moreover, we propose a new imputation method to deal with the problem of missing values and the results show that it outperforms the other alternatives. In the predictive model, we use patients’ blood test data within 1–9 d before surgery to construct the model to predict postoperative patients’ survival	The experimental results on a real data set indicate that RF outperforms the other alternatives. The experimental results on the temporal validation set show that our proposed model achieves AUC of 0.771 and specificity of 0.815	ML can detect the high risk patients in early phase after LT, and discover important factors that are essential in LT
Yang et al[57], 2022	A Chinese study in which conventional Scoring systems were compared with ML models in predicting 90 day survival in ACLF patients following LT	To compare the predictive value of conventional models and ML models for predicting 90-d post-transplant survival of ACLF patients based on preoperative variables	Preoperative data of 132 ACLF patients receiving LT at our center were investigated retrospectively. Cox regression was performed to determine the risk factors for short-term survival among ACLF patients following LT. Five conventional score systems (the MELD score, ABIC, CLIF-C OFs, CLIF-SOFAs and CLIF-C ACLFs) in forecasting short term survival were estimated through the ROC. Four machine-learning (ML) models, including support vector machine (SVM), logistic regression (LR), multi-layer perceptron (MLP) and random forest (RF), were also established for short-term survival prediction	Cox regression analysis demonstrated that creatinine (Cr) and international normalized ratio (INR) were the two independent predictors for short-term survival among ACLF patients following LT. The ROC curves showed that the AUC ML models was much larger than that of conventional models in predicting short term survival. Among conventional models the model for end stage liver disease (MELD) score had the highest AUC (0.704), while among ML models the RF model yielded the largest AUC (0.940). (AUROC) of MELDs (AUROC: 0.704) was higher than those of ABIC (AUROC: 0.607), CLIF-C OFs (AUROC: 0.606), CLIF-C ACLFs (AUROC: 0.653), and CLIF-SOFAs (AUROC: 0.633) for prediction of the 90-d outcome in ACLF patients following LT	Compared with the traditional methods, the ML models showed good performance in the prediction of short-term prognosis among ACLF patients following LT and the RF model perform the best
Andres et al[58], 2018	A United States study using ML to construct a prediction tool called PSSP using SRTR data to predict survival following LT for PSC and compared with cox regression in survival analysis	To develop ML models to predict individual survival after LT for Primary Sclerosing Cholangitis (PSC)	We applied a software tool, PSSP, to adult patients in the Scientific Registry of Transplant Recipients (n = 2769) who received a LT for PSC between 2002 and 2013; this produced a model for predicting individual survival distributions for novel patients. We also developed an appropriate evaluation measure, D-calibration, to validate this model	The learned PSSP model showed an excellent D-calibration (P = 1.0), and passed the single-time calibration test (Hosmer-Lemeshow P value of over 0.05) at 0.25, 1, 5 and 10 yr. In contrast, the model based on traditional Cox regression showed worse calibration on long-term survival and failed at 10 yr (Hosmer-Lemeshow P value = 0.027). The overall KM survival curve at 0.25, 1, 3, 5 and 10-yr showed survival probabilities of: 95.6%, 93%, 87.6%, 84.1% and 72%	Our empiricalresults show that the individual survival distributions produced by these models are well calibrated, which means they can be used for this screening task of deciding whether a candidateshould be added to the LT waiting list as they can help predict the survival of a possible recipient (or of a donor/recipient pair)
Kong et al[59], 2020	A Chinese study in which Logistic regression and artificial neural network(ANN) analysis were used to determine the preoperative independent risk factors and protective factors for the survival or death of patients90 days after surgery	To develop a simple ML model for quick prediction of the short-term survival ofpatients after LT in the event that the donor's information is not available in advance	A total of 1495 adult patients underwent LT in the present study. Three-quarters of recipients were randomly selected into the test set (n = 1121), while the remaining 25% formed the validation set (n = 374). Univariate and multivariate analysis and machine-learning techniques were applied to evaluate possible influencing factors. To further simplify the model, a weighted-scoring system was designed considering each influencing factor and its importance in an ANN	In the test set, multivariate analysis identified creatinine, age, and total bilirubin as independent risk factors, while albumin was an independent protective factor. Logistic regression analysis showed the C-statistic to be 0.650, while ANN indicated this to be 0.698. We simplified the model to obtain the final scoring model, for which the C-statistic was 0.636, and defined four risk grades. The 90-d mortality rates corresponding to the four risk levels were 6.2%, 11.8%, 24.0%, and 34.9%, respectively. In the validation set, the C-statistic value of the original model was 0.668 and that of the simplified model was 0.647	We demonstrated that the postoperative 90-d mortality followingadult LT can be predicted using a scoring system based on recipients' preoperative characteristics
Bertsimas et al[60], 2019	An American study using Optimized prediction of mortality (OPOM) utilizing machine-learning optimal classification tree models trained to predict a candidate’s 3-months waitlist mortality or removal using the standard transplant analysis andresearch (STAR) dataset	To utilize a state-of-the-art machine-learning method-termed optimal classification trees (OCTs)-to generatea more accurate prediction of a liver candidate’s 3-months wait-list mortality or removal	An OPOM was developed (http://www.opom.online) utilizing machine-learning optimal classification tree models trained to predict a candidate’s 3-months waitlist mortality or removal utilizing the STAR dataset. The Liver Simulated Allocation Model (LSAM) was then used to compare OPOM to MELD-based allocation. Out-of-sample area under the curve (AUC) was also calculated for candidate groups of increasing disease severity	OPOM considerably outperformed both MELD variants when predicting the 3-months probability of dying or becoming unsuitable for transplant for all patients (0.859 vs 0.841 for MELD-Na, and 0.823for Match MELD) and across all exception statuses. In addition, analysis of out-of-sample AUC for OPOM, Match MELD and MELD-Na, for subpopulations of patients with increasing dis-ease severity, revealed a notable decline in predictive power for Match MELD and MELD-Na as disease severity increased, whereas OPOM’s predictive power was maintained. The largest divergence in predictive power between OPOM and MELD was at the higher disease severity brackets, with OPOM outperforming Match MELD by up to 16%	OPOM more accurately and objectively prioritizes candidates for LT based on disease severity, allowing for more equitable allocation of livers with a resultant sig- nificant number of additional lives saved every year. These data demonstrate the potential of machine learning technology to help guide clinical practice, and potentially guide national policy
He et al[61], 2021	An American study using image omics and multi-network based deep learning model that converts expertise in LT, full-slide image digitization, and deep machine learning, and integrates multimodality data of quantitative image features with relevant clinical data to identify pre-clinical and biological markers for predicting good post-transplant outcomes, regardless of size	To develop a convergent artificial intelligence (AI) model that combines transient clinical data with quantitative histologic and radiomic features for more objective risk assessment of LT for HCC patient	Patients who received a LT for HCC between 2008-2019 were eligible for inclusion in the analysis. All patients with post-LT recurrence were included, and those without recurrence were randomly selected for inclusion in the deep learning model. Pre- and post-transplant magnetic resonance imaging (MRI) scans and reports were compressed using Caps Net networks and natural language processing, respectively, as input for a multiple feature radial basis function network. We applied a histological image analysis algorithm to detect pathologic areas of interest from explant tissue of patients who recurred. The multilayer perceptron was designed as a feed forward, supervised neural network topology, with the final assessment of recurrence risk. We used AUC and F-1 score to assess the predictability of different network combinations	A total of 109 patients were included (87 in the training group, 22 in the testing group), of which 20 were positive for cancer recurrence. Seven models (AUC; F-1 score) were generated, including clinical features only (0.55; 0.52), MRI only (0.64; 0.61), pathological images only (0.64; 0.61), MRI plus pathology (0.68; 0.65), MRI plus clinical (0.78, 0.75), pathology plus clinical (0.77; 0.73), and a combination of clinical, MRI, and pathology features (0.87; 0.84). The final combined model showed 80% recall and 89% precision. The total accuracy of the implemented model was 82%	We validated that the deep learning model combining clinical features and multi scale histopathologic and radiomic image features can be used to discover risk factors for recurrence beyond tumor size and biomarker analysis
Pinto-Marques et al[62], 2022	A Portuguese study in which the ML model, Hepato-Predict was constructed on retrospective LT data for HCC based on the assessment of a gene expression signature plus clinical variables	To propose a new decision algorithm combining biomarkers measured in a tumor biopsy with clinical variables, to predict recurrence after LT	A literature systematic review singled out candidate biomarkers whose RNA levels were assessed by quantitative PCR in tumor tissue from 138 HCC patients submitted to LT (> 5 yr follow up, 32% beyond Milan criteria). The resulting 4 gene signature was combined with clinical variables to develop a decision algorithm using machine learning approaches. The method was named HepatoPredict	HepatoPredict identifies 99% disease-free patients (> 5 yr) including many outside clinical criteria (16%-24%). Has increased positive predictive value (88.5%-94.4%) without any loss of long-term overall survival or recurrence rates for patients deemed eligible by HepatoPredict; those deemed ineligible display marked reduction of survival and increased recurrence in the short and long term	HepatoPredict outperforms conventional clinical-pathologic selection criteria (Milan, UCSF), providing superior prognostic information
Lai et al[63], 2023	A Taiwanese study in which the ML model ResNet-18 was trained on FDG-PET-CT images to predict outcomes in HCC patients undergoing LT	To evaluate the performance of deep learning from 18F-FDG PET-CT images to predict overall survival in HCC patients before LT	We retrospectively included 304 patients with HCC who underwent 18F-FDG PET/CT before LT between January 2010 and December 2016. The hepatic areas of 273 of the patients were segmented by software, while the other 31 were delineated manually. We analyzed the predictive value of the deep learning model from both FDG PET/CT images and CT images alone	The results of the developed prognostic model were obtained by combining FDG PET-CT images and combining FDG CT images (0.807 AUC vs 0.743 AUC). The model based on FDG PET-CT images achieved somewhat better sensitivity than the model based on CT images alone (0.571 SEN vs 0.432 SEN)	Our retrospective study indicated that an automated 3D ResNet-18 convolutional neural network with FDG-PET-CT has promise for predicting clinical outcomes in patientswith HCC undergoing LDLT and that Automatic liver segmentation from 18F-FDG PET-CT images is feasible and can be utilized to train deep-learning models
Kazemi et al[64], 2019	Iranian study aimed at modelling patient survival after LT using machine-learning methods to investigate influential factors and compare the performance of these methods with a classic statistic method, cox regression	To Identify effective factors for patient survival after LT using ML techniques	Our study included 902 adults who received livers from deceased donors from March 2011 to March 2014 at the Shiraz Organ Transplant Center (Shiraz, Iran). In a 3-step feature selection method, effective features of 6-month survival were extracted by: (1) F statistics, Pearson chi-square, and likelihood ratio chi-square; (2) 5 machine earning techniques. To evaluate the performance of the machine-learning techniques, Cox regression was applied to the data set. Evaluations were based on the area under the receiver operating characteristic curve and sensitivity of models; and (3) We also constructed a model using all factors identified in the previous step	The model predicted survival based on 26 identified effective factors. In the following order, graft failure, Aspergillus infection, acute renal failure and vascular complications after transplant, as well as graft failure diagnosis interval, previous diabetes mellitus, Model for End-Stage Liver Disease score, donor inotropic support, units of packed cell received, and previous recipient dialysis, were found to be predictive factors in patient survival. The area under the receiver operating characteristic curve and model sensitivity were 0.90 and 0.81, respectively	Data mining analyses can help identify effective features of patient survival after livertransplant and build models with equal or higher performance than Cox regression. The order ofinfluential factors identified with the machine learning model was close to clinical experiments
Nitski et al[65], 2021	An American study that examined retrospective data of transplant recipients from the SRTR and UHN to assess the role of deep learning algorithms to predict complications resulting in death after liver transplant over multiple time frames in comparison with logistic regression	To assess the ability of deep learning algorithms of longitudinal data from two prospective cohorts to predict complications resulting in death after LT over multiple timeframes, compared with logistic regression models	In this machine learning analysis, model development was done on a set of 42 146 liver transplant recipients [mean age 48.6 yr (SD 17.3); 17 196 (40.8%) women] from the Scientific Registry of Transplant Recipients (SRTR) in the United States. Transferability of the model was further evaluated by fine-tuning on a dataset from the UHN in Canada [n = 3269; mean age 52.5 yr (11.1); 1079 (33.0%) women]. The primary outcome was cause of death, as recorded in the databases, due to cardiovascular causes, infection, graft failure, or cancer, within 1 yr and 5 yr of each follow-up examination after transplantation. We compared the performance of four deep learning models against logistic regression, assessing performance using the AUROC	In both datasets, deep learning models outperformed logistic regression, with the Transformer model achieving the highest AUROCs in both datasets (P < 0.0001). The AUROC for the Transformer model across all outcomes in the SRTR dataset was 0.804 (99%CI: 0.795-0.854) for 1-yr predictions and 0.733 (0.729-0.769) for 5-yr predictions. In the UHN dataset, the AUROC for the top-performing deep learning model was 0.807 (0.795-0.842) for 1-yr predictions and 0.722 (0.705–0.764) for 5-yr predictions. AUROCs ranged from 0.695 (0.680–0.713) for prediction of death from infection within 5 yr to 0.859 (0.847-0.871) for prediction of death by graft failure within 1 yr	Deep learning algorithms can incorporate longitudinal information to continuously predict long-term outcomes after LT, outperforming logistic regression models
Ivanics et al[66], 2022	A multinational study of ML models assessing their 90-d predictive value post LT across United States, Canada and	To evaluate the feasibility of developing MLA-based models to predict 90-d post-LT mortality using 3 large nationaltransplant registries and to evaluate the external validity of the models across countries	We used data from 3 national registries and developed machine learning algorithm (MLA)–based models to predict 90-d post-LT mortality within and across countries. Predictive performance and external validity of each model were assessed. Prospectively collected data of adult patients (aged ≥ 18 yr) who underwent primary LTs between January 2008 and December 2018 from the Canadian Organ Replacement Registry (Canada), National Health Service Blood and Transplantation (United Kingdom), and United Network for Organ Sharing (United States) were used to develop MLA models to predict 90-d post-LT mortality. Models were developed using each registry individually (based on variables inherent to the individual databases) and using all 3 registries combined (variables in common between the registries [harmonized]). The model performance was evaluated using AUROC curve. The number of patients included was as follows: Canada, n = 1214; the United Kingdom, n = 5287; and the United States, n = 59558	The best performing MLA-based model was ridge regression across both individual registries and harmonized data sets. Model performance diminished from individualized to the harmonized registries, especially in Canada (individualized ridge: AUROC, 0.74; range, 0.73-0.74; harmonized: AUROC, 0.68; range, 0.50-0.73) and US (individualized ridge: AUROC, 0.71; range, 0.70-0.71; harmonized: AUROC, 0.66; range, 0.66-0.66) data sets. External model performance across countries was poor overall	External model performance across countries was poor overall. MLA-based models yield a fair discriminatory potential when used within individual databases. However, the external validity of these models is poor when applied across countries
Cheong et al[67], 2021	A Korean study assessing the role of pre LT hyperlactatemia in early mortality post LT	To study important variables for pre-LT hyperlactatemia and examine the impact of preoperative hyperlactatemia on 30 and 90 d mortality after LT	A total of 2002 patients from LT registry between January 2008 and February 2019 were analyzed. Six organ failures (liver, kidney, brain, coagulation, circulation, and lung) were defined by criteria of EASL-CLIF ACLF Consortium. Variable importance of pre-operative hyperlactatemia was examined by machine learning using random survival forest (RSF). Kaplan-Meier Survival curve analysis was performed to assess 90-d mortality	Median lactate level was 1.9 mmol/L (interquartile range: 1.4, 2.4 mmol/L) and 107 (5.3%) patients showed > 4.0 mmol/L. RSF analysis revealed that the four most important variables for hyperlactatemia were MELD score, circulatory failure, hemoglobin, and respiratory failure. The 30-d and 90-d mortality rates were 2.7% and 5.1%, whereas patients with lactate > 4.0 mmol/L showed increased rate of 15.0% and 19.6%, respectively	Pre-LT lactate > 4.0 mmol/L was associated with increased early post-LT mortality. Our results suggest that future study of correcting modifiable risk factors may play a role in preventing hyperlactatemia and lowering early mortality after LT
Kulkarni et al[68], 2021	An American study using Random Forest approach to identify key predictors of outcomes in pediatric candidates less than 2 yr of age undergoing LT	To identify key predictors of LT outcomes in Pediatric candidates less than 2 yr of age using random forest approach	SRTR database was queried for children < 2 yr listed for initial LT during 2002-17 (n = 4973). Subjects were divided into three outcome groups; bad (death or removal for too sick to transplant), good (spontaneous improvement) and transplant. Demographic, clinical, listing history and laboratory variables at the time of listing (baseline variables), and changes in variables between listing and prior to outcome (trajectory variables) were analyzed using random forest analysis	81.5% candidates underwent LT, 12.3% had bad outcome. RF model including both baseline and trajectory variables improved prediction compared to model using baseline variables alone. RF analyses identified change in serum creatinine and listing status as the most predictive variables. 80% of subjects listed with a PELD score at time of listing and outcome underwent LT, while 70% of subjects in both bad and good outcome groups were listed with either Status 1 (A or B) prior to an outcome, regardless of initial listing status. Increase in creatinine on LT waitlist was predictive of bad outcome. Longer time spent on WL was predictive of good outcome. Subjects with biliary atresia, liver tumors and metabolic disease had LT rate > 85%; while > 20% of subjects with acute liver failure had a bad outcome	Change in creatinine, listing status, need for RRT, time spent on LT waitlist and diagnoses were the most predictive variables
Molinari et al[69], 2019	An American study using ML techniques to identify predictors of short and long term mortality post cadaveric LT	To develop a scoring system using ML that could stratify patients by their risk of death after LT based only on preoperative variables. Secondary aims were to assess whether the model could also predict 1- and 5-yr patient survival	The study population was represented by 30458 adults who underwent LT in the United States between January 2002 and June 2013. Machine learning techniques identified recipient age, Model for End-Stage Liver Disease score, body mass index, diabetes, and dialysis before LT as the strongest predictors for 90-d postoperative mortality. A weighted scoring system (minimum of 0 to a maximum of 6 points) was subsequently developed	Recipients with 0, 1, 2, 3, 4, 5, and 6 points had an observed 90-d mortality of 6.0%, 8.7%, 10.4%, 11.9%, 15.7%, 16.0%, and 19.7%, respectively (P ≤ 0.001). One-year mortality was 9.8%, 13.4%, 15.8%, 17.2%, 23.0%, 25.2%, and 35.8% (P ≤ 0.001) and five-year survival was 78%, 73%, 72%, 71%, 65%, 59%, and 48%, respectively (P = 0.001). The mean 90-d mortality for the cohort was 9%. The area under the curve of the model was 0.952 for the discrimination of patients with 90-day mortality risk ≥ 10%	Short- and long-term outcomes of patients undergoing cadaveric LT can be predicted using a scoring system based on recipients’ preoperative characteristics
Cooper et al[70], 2022	A United States study predicting the risk of GVHD among patients undergoing OLT using ML models	To develop ML algorithms for predicting the risk of GVHD among patients undergoing OLT	To develop a predictive model, we retrospectively evaluated the clinical features of 1938 donor-recipient pairs at the time they underwent OLT at our center; 19 (1.0%) of these recipients developed GVHD. This population was divided into training (70%) and test (30%) sets. A total of 7 machine-learning classification algorithms were built based on the training data set to identify patients at high risk for GVHD	The C5.0, heterogeneous ensemble, and generalized gradient boosting machine (GGBM) algorithms predicted that 21% to 28% of the recipients in the test data set were at high risk for developing GVHD, with an AUROC of 0.83 to 0.86. The 7 algorithms were then evaluated in a validation data set of 75 more recent donor-recipient pairs who underwent OLT at our center; 2 of these recipients developed GVHD. The logistic regression, heterogeneous ensemble, and GGBM algorithms predicted that 9% to 11% of the validation recipients were at high risk for developing GVHD, with an AUROC of 0.93 to 0.96 that included the 2 recipients who developed GVHD	we show that a machine-learning approach can predict which recipients are at high risk for developing GVHD after OLT based on factors known or measurable at the time of transplantation
He et al[71], 2021	A Chinese study comparing the predicting power of ML models and logistic regression for AKI among patients undergoing DCDLT	To compare the performance of ML algorithms to that of a logistic regression model for predicting AKI after LT using preoperative and intraoperative data	A total of 493 patients with donation after cardiac death LT (DCDLT) were enrolled. AKI was defined according to the clinical practice guidelines of kidney disease: improving global outcomes (KDIGO). The clinical data of patients with AKI (AKI group) and without AKI (non-AKI group) were compared. With logistic regression analysis as a conventional model, four predictive machine learning models were developed using the following algorithms: Random forest, support vector machine, classical decision tree, and conditional inference tree. The predictive power of these models was then evaluated using the AUC	The incidence of AKI was 35.7% (176/493) during the follow-up period. Compared with the non AKI group, the AKI group showed a remarkably lower survival rate (P < 0.001). The random forest model demonstrated the highest prediction accuracy of 0.79 with AUC of 0.850 (95%CI: 0.794-0.905), which was significantly higher than the AUCs of the other machine learning algorithms and logistic regression models (P < 0.001)	The random forest model based on machine learning algorithms for predicting AKI occurring after DCDLT demonstrated stronger predictive power than other models in our study
Chen et al[72], 2023	A Chinese study predicting the risk of sepsis within 7 days post LT	Our study aimed to develop and validate a predictive model for postoperative sepsis within 7 days in LT recipients using ML technology	Data of 786 patients who received LT from January 2015 to January 2020 was retrospectively extracted from the big data platform of Third Affiliated Hospital of Sun Yat-sen University. Seven ML models were developed to predict postoperative sepsis. The AUC, sensitivity, specificity, accuracy, and f1-score were evaluated as the model performances. The model with the best performance was validated in an independent dataset involving 118 adult LT cases from February 2020 to April 2021. The postoperative sepsis-associated outcomes were also explored in the study	After excluding 109 patients according to the exclusion criteria, 677 patients who underwent LT were finally included in the analysis. Among them, 216 (31.9%) were diagnosed with sepsis after LT, which were related to more perioperative complications, increased postoperative hospital stay and mortality after LT (all P < 0.05). Our results revealed that a larger volume of red blood cell infusion, ascitic removal, blood loss and gastric drainage, less volume of crystalloid infusion and urine, longer anesthesia time, higher level of preoperative TBIL were the top 8 important variables contributing to the prediction of post-LT sepsis. The RF model showed the best overall performance to predict sepsis after LT among the seven ML models developed in the study, with an AUC of 0.731, an accuracy of 71.6%, the sensitivity of 62.1%, and specificity of 76.1% in the internal validation set, and a comparable AUC of 0.755 in the external validation set	The random forest classifier model showed the best overall performance to predict sepsis after LT
Lee et al[73], 2018	A Korean study comparing the predicting power for AKI post LT of ML models and logistic regression	To compare the performance of machine learning approaches with that of logistic regression analysis to predict AKI after LT	We reviewed 1211 patients and preoperative and intraoperative anesthesia and surgery-related variables were obtained. The primary outcome was postoperative AKI defined by acute kidney injury network criteria. The following machine learning techniques were used: decision tree, random forest, gradient boosting machine, support vector machine, naïve Bayes, multilayer perceptron, and deep belief networks. These techniques were compared with logistic regression analysis regarding the AUROC	AKI developed in 365 patients (30.1%). The performance in terms of AUROC was best in gradient boosting machine among all analyses to predict AKI of all stages (0.90, 95%CI: 0.86-0.93) or stage 2 or 3 AKI. The AUROC of logistic regression analysis was 0.61 (95%CI: 0.56-0.66). Decision tree and random forest techniques showed moderate performance (AUROC 0.86 and 0.85, respectively)	In our comparison of seven machine learning approaches with logistic regression analysis, the gradient boosting machine showed the best performance with the highest AUROC
Bredt et al[74], 2022	A Brazilian study investigating risk factors of AKI post DDLT using ML and Logistic regression	To identify the risk factors of AKI after deceased-donor LT (DDLT) and compare the prediction performance of ANN with that of LR for this complication	Adult patients with no evidence of end-stage kidney dysfunction (KD) who underwent the first DDLT according to model for end-stage liver disease (MELD) score allocation system were evaluated. AKI was defined according to the International Club of Ascites criteria, and potential predictors of postoperative AKI were identified by LR. The prediction performance of both ANN and LR was tested	The incidence of AKI was 60.6% (n = 88/145) and the following predictors were identified by LR: MELD score > 25 (OR = 1.999), preoperative kidney dysfunction (OR = 1.279), extended criteria donors (OR = 1.191), intraoperative arterial hypotension (OR = 1.935), intraoperative massive blood transfusion (MBT) (OR = 1.830), and postoperative serum lactate (SL) (OR = 2.001). The area under the receiver-operating characteristic curve was best for ANN (0.81, 95%CI: 0.75-0.83) than for LR (0.71, 95%CI: 0.67-0.76). The root-mean-square error and mean absolute error in the ANN model were 0.47 and 0.38, respectively	The severity of liver disease, pre-existing kidney dysfunction, marginal grafts, hemodynamic instability, MBT, and SL are predictors of postoperative AKI, and ANN has better prediction performance than LR in this scenario

ANN: Artificial neural network; DRI: Donor risk index; D-R: Donor-recipient; LT: Liver transplantation; MELD: Model of end-stage liver disease; NN-CCR: Neural network for correct classification rate; NN-MS: Neural network for minimum sensitivity; SOFT: Survival outcomes following liver transplantation score; ROC: Receiver-operating curves; BAR: Balance of risk score; DNN: Deep neural network; ML: Machine Learning; ACLF: Acute-on-chronic liver failure; CLIF-C OFs: Chronic liver failure consortium organ failure scores: CLIF-SOFAs: CLIF sequential organ failure assessment scores; CLIF-C ACLFs: CLIF consortium ACLF scores; RF: Random forest; SRTR: Scientific registry of transplant recipients; PSSP: Patient-specific survival prediction; PSC: Primary sclerosing cholangitis; KM: Kaplan meier; OPOM: Optimized prediction of mortality; STAR: transplant analysis and research; HCC: Hepatocellular carcinoma; Milan-UCSF: Milan-University of California San Francisco criteria; 18F-FDG: 18F-fluorodeoxyglucose; PET-CT: Positron emission tomography and computed tomography; LDLT: Live donor liver transplantation; DDLT: Deceased donor liver transplant; EASL-CLIF ACLF Consortium: European Association for the Study of the Liver-CLIF ACLF; PELD: Pediatric end stage liver disease; WL: Wait list; GVHD: Graft-versus-host disease; OLT: Orthotropic liver transplant.