A Machine Learning-Driven Virtual Biopsy System For Kidney Transplant Patients

Abstract

In kidney transplantation, day-zero biopsies are used to assess organ quality and discriminate between donor-inherited lesions and those acquired post-transplantation. However, many centers do not perform such biopsies since they are invasive, costly and may delay the transplant procedure. We aim to generate a non-invasive virtual biopsy system using routinely collected donor parameters. Using 14,032 day-zero kidney biopsies from 17 international centers, we develop a virtual biopsy system. 11 basic donor parameters are used to predict four Banff kidney lesions: arteriosclerosis, arteriolar hyalinosis, interstitial fibrosis and tubular atrophy, and the percentage of renal sclerotic glomeruli. Six machine learning models are aggregated into an ensemble model. The virtual biopsy system shows good performance in the internal and external validation sets. We confirm the generalizability of the system in various scenarios. This system could assist physicians in assessing organ quality, optimizing allograft allocation together with discriminating between donor derived and acquired lesions post-transplantation.

Fig. 1. Clinical and biological parameters’ importance.

We performed random forest, gradient boosting machine, extreme gradient boosting tree, linear discriminant analysis, model averaged neural network, and multinomial logistic regression to measure the parameter importance for predicting the day-zero biopsy histological lesion scores during the derivation process. The importance was then averaged for the ensemble model. a Donor parameter importance for arteriosclerosis (cv Banff score). b Donor parameter importance for arteriolar hyalinosis (ah Banff score). c Donor parameter importance for interstitial fibrosis and tubular atrophy (IFTA Banff score). d Donor parameter importance for the percentage of sclerotic glomeruli (glomerulosclerosis score). Banff scores: cv arteriosclerosis, ah arteriolar hyalinosis, IFTA interstitial fibrosis and tubular atrophy. BMI body mass index, DCD donation after circulatory death, HCV hepatitis C virus. Source data are provided as a Source Data file.

Fig. 2. Performance metrics of ensemble models across internal and external validation cohorts.

Ensemble models were internally and externally validated on the 3-times repeated 10-folds cross-validation and the external validation cohorts comprising Columbia university from the USA and Sun Yat-sen university from China. For multi-AUC, the full lesion scores were used. For other metrics, such as AUROC and sensitivity, categorical Banff scores (arteriosclerosis [cv Banff score], arteriolar hyalinosis [ah Banff score], and interstitial fibrosis and tubular atrophy [IFTA Banff score]) were dichotomized. Cut-offs were calibrated based on internal validation (cross-validation): 0.582, 0.596, 0.637 for cv, ah, IFTA lesions, respectively. For internal validations, performance was assessed in 30 resamples during cross-validation. For external validations, performance was assessed using 1,000 times bootstrapping. All box plots comprise the median line, the box indicated the interquartile range (IQR), whiskers denote the rest of the data distribution and outliers are denoted by points greater than ±1.5 × IQR. * For sensitivity, specificity, balanced accuracy, accuracy, and AUROC, the Banff lesion scores, cv, ah, and IFTA were dichotomized (scores 0–1 as negative and 2-3 as positive). Banff scores: cv arteriosclerosis, ah arteriolar hyalinosis, IFTA interstitial fibrosis and tubular atrophy. multi-AUC multi-area under the receiver operating characteristic curve, AUROC area under the receiver operating characteristic curve, MAE mean absolute error, RMSE root mean square error. Source data are provided as a Source Data file.

Fig. 3. Ready-to-use online application for physicians.

The online application aims to help physicians freely use the virtual day-zero biopsy findings for post-transplant patient management. a A virtual biopsy finding from 63-year-old female donor from circulatory cause of death with moderate BMI but poor kidney function (creatinine). b A virtual biopsy finding from 51-year-old male donor from cerebrovascular cause of death with high BMI and hypertension but moderate kidney function. Banff scores: cv arteriosclerosis, ah arteriolar hyalinosis, IFTA interstitial fibrosis and tubular atrophy, ci interstitial fibrosis, CT tubular atrophy, BMI body mass index, DCD donation after circulatory death, HCV hepatitis C virus Link to the app: https://transplant-prediction-system.shinyapps.io/Virtual_Biopsy_System.

Fig. 4. Flow chart of virtual biopsy system machine learning pipeline.

The study comprises three main processes to develop and validate the virtual biopsy system for kidney transplant patients. Each step also comprises three sub-processes. multi-AUC multi-area under the receiver operating characteristic curve.