Evaluation of predictive models for delayed graft function of deceased kidney transplantation

Cherry on Top or Real Need? A Review of Explainable Machine Learning in Kidney Transplantation

Research on solid organ transplantation has taken advantage of the substantial acquisition of medical data and the use of artificial intelligence (AI) and machine learning (ML) to answer diagnostic, prognostic, and therapeutic questions for many years. Nevertheless, despite the question of whether AI models add value to traditional modeling approaches, such as regression models, their "black box" nature is one of the factors that have hindered the translation from research to clinical practice. Several techniques that make such models understandable to humans were developed with the promise of increasing transparency in the support of medical decision-making. These techniques should help AI to close the gap between theory and practice by yielding trust in the model by doctors and patients, allowing model auditing, and facilitating compliance with emergent AI regulations. But is this also happening in the field of kidney transplantation? This review reports the use and explanation of "black box" models to diagnose and predict kidney allograft rejection, delayed graft function, graft failure, and other related outcomes after kidney transplantation. In particular, we emphasize the discussion on the need (or not) to explain ML models for biological discovery and clinical implementation in kidney transplantation. We also discuss promising future research paths for these computational tools.

The positive efficacy of dexmedetomidine on the clinical outcomes of patients undergoing renal transplantation: evidence from meta-analysis

INTRODUCTION

Whether dexmedetomidine (DEX), an anesthetic adjuvant, can improve renal transplant outcomes is not clear.

METHODS

We systematically identified clinical trials in which DEX was administered in renal transplantation (RT). On November 1, 2022, we searched The Cochrane Library, MEDLINE, EMBASE and https://www.

CLINICALTRIALS

gov/. The main outcomes were delayed graft function and acute rejection.

RESULTS

A total of seven studies were included in the meta-analysis. The results showed that compared with the control, DEX significantly reduced the occurrence of delayed graft function (RR 0.76; 95% CI 0.60-0.98), short-term serum creatinine [postoperative day (POD) 2: (MD -22.82; 95% CI -42.01 - -3.64)] and blood urea nitrogen [POD 2: (MD -2.90; 95% CI -5.10 - -0.70); POD 3: (MD 2.07; 95% CI -4.12 - -0.02)] levels, postoperative morphine consumption (MD -4.27; 95% CI -5.92 - -2.61) and the length of hospital stay (MD -0.85; 95% CI-1.47 - -0.23). However, DEX did not reduce the risk of postoperative acute rejection (RR 0.75; 95% CI 0.45-1.23). The results of the subgroup analysis showed that country type, donor type, and average age had a certain impact on the role of DEX.

CONCLUSIONS

DEX may improve the short-term clinical outcome of RT and shorten the length of hospital stay of patients.

A Machine Learning Prediction Model for Immediate Graft Function After Deceased Donor Kidney Transplantation

BACKGROUND

After kidney transplantation (KTx), the graft can evolve from excellent immediate graft function (IGF) to total absence of function requiring dialysis. Recipients with IGF do not seem to benefit from using machine perfusion, an expensive procedure, in the long term when compared with cold storage. This study proposes to develop a prediction model for IGF in KTx deceased donor patients using machine learning algorithms.

METHODS

Unsensitized recipients who received their first KTx deceased donor between January 1, 2010, and December 31, 2019, were classified according to the conduct of renal function after transplantation. Variables related to the donor, recipient, kidney preservation, and immunology were used. The patients were randomly divided into 2 groups: 70% were assigned to the training and 30% to the test group. Popular machine learning algorithms were used: eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine, Gradient Boosting classifier, Logistic Regression, CatBoost classifier, AdaBoost classifier, and Random Forest classifier. Comparative performance analysis on the test dataset was performed using the results of the AUC values, sensitivity, specificity, positive predictive value, negative predictive value, and F1 score.

RESULTS

Of the 859 patients, 21.7% (n = 186) had IGF. The best predictive performance resulted from the eXtreme Gradient Boosting model (AUC, 0.78; 95% CI, 0.71-0.84; sensitivity, 0.64; specificity, 0.78). Five variables with the highest predictive value were identified.

CONCLUSIONS

Our results indicated the possibility of creating a model for the prediction of IGF, enhancing the selection of patients who would benefit from an expensive treatment, as in the case of machine perfusion preservation.

Combining Clinical Parameters and Acute Tubular Injury Grading Is Superior in Predicting the Prognosis of Deceased-Donor Kidney Transplantation: A 7-Year Observational Study

Background

We developed a pragmatic dichotomous grading criterion to stratify the acute tubular injury (ATI) of deceased-donor kidneys. We intended to verify the predictive value of this criterion for the prognosis of deceased-donor kidney transplantation.

Methods

The allografts with ATI were classified into severe and mild groups. Severe ATI was defined as the presence of extreme and diffuse flattening of the tubular epithelial cells, or denudement of the tubular basement membrane. The clinical delayed graft function (DGF) risk index was calculated based on a regression model for posttransplant DGF using 17 clinical parameters related to donor–recipient characteristics.

Results

A total of 140 recipients were enrolled: 18 severe and 122 mild ATI. Compared with the mild ATI group, the severe ATI group had more donors after cardiac death, higher median donor terminal serum creatinine level (dScr), and longer median cold ischemia time. Severe ATI had a higher DGF rate (55.6% vs 14.6%, p < 0.001), longer DGF recovery time (49.6 vs 26.3 days, p < 0.001), and a lower estimated glomerular filtration rate (eGFR) at 1 month (23.5 vs 54.0 ml/min/1.73 m2, p < 0.001), 3 months (40.4 vs 59.0, p = 0.001), and 6 months after transplant (46.8 vs 60.3, p = 0.033). However, there was no significant difference in eGFR at 1 year or beyond, graft, and patient survival. The predictive value of combined dScr with ATI severity for DGF rate and DGF recovery time was superior to that of dScr alone. The predictive value of the combined DGF risk index with ATI severity for DGF was also better than that of the DGF risk index alone; however, the association of the DGF risk index with DGF recovery time was not identified. Chronic lesions including glomerulosclerosis, interstitial fibrosis, arterial intimal fibrosis, and arteriolar hyalinosis were associated with declined posttransplant 1-year eGFR.

Conclusion

Based on our pragmatic dichotomous grading criterion for ATI in a preimplantation biopsy, donor kidneys with severe ATI increased DGF risk, prolonged DGF recovery, and decreased short-term graft function but demonstrated favorable long-term graft function. Our grading method can offer additive valuable information for assessing donor kidneys with acute kidney injury and may act as an effective supplementary index of the Banff criteria.

Risk Prediction for Delayed Allograft Function: Analysis of the Deterioration of Kidney Allograft Function (DeKAF) Study Data

BACKGROUND

Delayed graft function (DGF) of a kidney transplant results in increased cost and complexity of management. For clinical care or a DGF trial, it would be ideal to accurately predict individual DGF risk and provide preemptive treatment. A calculator developed by Irish et al has been useful for predicting population but not individual risk.

METHODS

We analyzed the Irish calculator (IC) in the DeKAF prospective cohort (incidence of DGF = 20.4%) and investigated potential improvements.

RESULTS

We found that the predictive performance of the calculator in those meeting Irish inclusion criteria was comparable with that reported by Irish et al. For cohorts excluded by Irish: (a) in pump-perfused kidneys, the IC overestimated DGF risk; (b) in simultaneous pancreas kidney transplants, the DGF risk was exceptionally low. For all 3 cohorts, there was considerable overlap in IC scores between those with and those without DGF. Using a modified definition of DGF-excluding those with single dialysis in the first 24 h posttransplant-we found that the calculator had similar performance as with the traditional DGF definition. Studying whether DGF prediction could be improved, we found that recipient cardiovascular disease was strongly associated with DGF even after accounting for IC-predicted risk.

CONCLUSIONS

The IC can be a useful population guide for predicting DGF in the population for which it was intended but has limited scope in expanded populations (SPK, pump) and for individual risk prediction. DGF risk prediction can be improved by inclusion of recipient cardiovascular disease.

Risk factors and outcomes of prolonged recovery from delayed graft function after deceased kidney transplantation

Objective

We aimed to evaluate the effect of prolonged recovery from DGF on outcomes, using a new definition of DGF recovery time, among deceased donor kidney transplant recipients with DGF, and to examine the risk factors for prolonged recovery.

Methods

From 2007 to 2016, 91 deceased donor kidney transplant recipients with DGF were retrospectively analyzed. DGF recovery time was defined as the time from transplantation to achieve a stable estimated glomerular filtration rate (eGFR). Recipients with a DGF recovery time greater than or equal to the median were assigned to the prolonged recovery group, while the others were assigned to the rapid recovery group.

Result

The median DGF recovery time was 27 days. Donor terminal eGFR was significantly lower in the prolonged recovery group (n = 46) compared with the rapid recovery group (n = 45) (median 24.9 vs. 65.4 ml/min/1.73m², p = 0.004). The eGFR at 1 year post-transplant in the prolonged recovery group was significantly lower than that in the rapid recovery group (50.6 ± 20.0 vs. 63.5 ± 21.4 ml/min/1.73m², p = 0.005). The risk of adverse outcomes (acute rejection, pneumonia, graft failure, and death) was significantly greater in the prolonged recovery group (hazard ratio 2.604, 95% confidence interval 1.102–6.150, p = 0.029) compared with the rapid recovery group.

Conclusion

Decreased donor terminal eGFR is a risk factor for prolonged recovery from DGF after deceased kidney transplantation. Prolonged DGF recovery time is associated with reduced graft function at 1-year post-transplant, and poor transplant outcome.

A prediction model of delayed graft function in deceased donor for renal transplant: a multi-center study from China

Background

Kidneys obtained from deceased donors increase the incidence of delayed graft function (DGF) after renal transplantation. Here we investigated the influence of the risk factors of donors with DGF, and developed a donor risk scoring system for DGF prediction.

Methods

This retrospective study was conducted in 1807 deceased kidney donors and 3599 recipients who received donor kidneys via transplants in 29 centers in China. We quantified DGF associations with donor clinical characteristics. A donor risk scoring system was developed and validated using an independent sample set.

Results

The incidence of DGF from donors was 19.0%. Six of the donor characteristics analyzed, i.e., age, cause of death, history of hypertension, terminal serum creatinine, persistence of hypotension, and cardiopulmonary resuscitation (CPR) time were risk factors for DGF. A 49-point scoring system of donor risk was established for DGF prediction and exhibited a superior degree of discrimination. External validation of DGF prediction revealed area under the receiver-operating characteristic (AUC) curves of 0.7552.

Conclusions

Our study determined the deceased donor risk factors related to DGF after renal transplantation pertinent to the Chinese cohort. The scoring system developed here had superior diagnostic significance and consistency and can be used by clinicians to make evidence-based decisions on the quality of kidneys from deceased donors and guide renal transplantation therapy.

Automated En Masse Machine Learning Model Generation Shows Comparable Performance as Classic Regression Models for Predicting Delayed Graft Function in Renal Allografts

BACKGROUND

Several groups have previously developed logistic regression models for predicting delayed graft function (DGF). In this study, we used an automated machine learning (ML) modeling pipeline to generate and optimize DGF prediction models en masse.

METHODS

Deceased donor renal transplants at our institution from 2010-2018 were included. Input data consisted of 21 donor features from United Network for Organ Sharing. A training set composed of ~50%/50% split in DGF-positive and DGF-negative cases was used to generate 400,869 models. Each model was based on one of seven ML algorithms (gradient boosting machine, k-nearest neighbor, logistic regression, neural network, naïve Bayes, random forest, support vector machine) with various combinations of feature sets and hyperparameter values. Each model's performance was based on a separate secondary test dataset and assessed by common statistical metrics.

RESULTS

The best performing models were based on neural network algorithms, with the highest area under the receiver operating characteristic (AUROC) curve of 0.7595. This model used 10 out of the original 21 donor features, including age, height, weight, ethnicity, serum creatinine, blood urea nitrogen, hypertension history, donation after cardiac death status, cause of death, and cold ischemia time. With the same donor data, the highest AUROC curve for logistic regression models was 0.7484, using all donor features.

CONCLUSION

Our automated en masse ML modeling approach was able to rapidly generate ML models for DGF prediction. The performance of the ML models was comparable to classic logistic regression models.

Personalized prediction of delayed graft function for recipients of deceased donor kidney transplants with machine learning

Machine learning (ML) has shown its potential to improve patient care over the last decade. In organ transplantation, delayed graft function (DGF) remains a major concern in deceased donor kidney transplantation (DDKT). To this end, we harnessed ML to build personalized prognostic models to predict DGF. Registry data were obtained on adult DDKT recipients for model development (n = 55,044) and validation (n = 6176). Incidence rates of DGF were 25.1% and 26.3% for the development and validation sets, respectively. Twenty-six predictors were identified via recursive feature elimination with random forest. Five widely-used ML algorithms-logistic regression (LR), elastic net, random forest, artificial neural network (ANN), and extreme gradient boosting (XGB) were trained and compared with a baseline LR model fitted with previously identified risk factors. The new ML models, particularly ANN with the area under the receiver operating characteristic curve (ROC-AUC) of 0.732 and XGB with ROC-AUC of 0.735, exhibited superior performance to the baseline model (ROC-AUC = 0.705). This study demonstrates the use of ML as a viable strategy to enable personalized risk quantification for medical applications. If successfully implemented, our models may aid in both risk quantification for DGF prevention clinical trials and personalized clinical decision making.

The value of cystatin C and urinary and serum neutrophil gelatinase-associated lipocalin during the perioperative period of renal transplantation

Background

The perioperative management of renal transplantation is complex. Our research aimed to study the clinical value of cystatin-C (Cys-C) and urinary and serum neutrophil gelatinase-associated lipocalin (NGAL) during the perioperative period of renal transplantation.

Methods

We collected the clinical information of 47 renal transplantation patients. Urine and serum samples were collected daily until the second week and then weekly until discharge to determine serum NGAL (s-NGAL), urine NGAL (u-NGAL), serum creatinine (s-Cr), and Cys-C levels. Receiver-operating characteristic (ROC) analysis was used, and the area under the curve (AUC) was compared to evaluate the accuracy of the diagnosis of delayed graft function (DGF). Multivariable analysis was used to find the association between the markers and renal function at discharge.

Results

In our research, the value of Cys-C, serum NGAL, and urine NGAL were higher in DGF group. In the ROC analysis, Cys-C had the highest AUC (0.939) compared with s-NGAL (0.909), u-NGAL (0.856), and s-Cr (0.747). Multivariable analysis showed that Cys-C levels in the first week after the operation and cold ischemia time were independently associated with estimated glomerular filtration rate (eGFR) at discharge (P<0.05).

Conclusions

Our results showed that Cys-C, serum NGAL, and urine NGAL could reflect renal function sensitively. Cys-C had the highest sum of sensitivity and specificity at 4.77 mg/L, with a sensitivity of 0.818 and specificity of 0.889. The Cys-C level during the first week after the operation was independently associated with eGFR at discharge and could predict the short-term prognosis of renal transplantation patients.