A Deep Learning Approach for the Estimation of Glomerular Filtration Rate

A comparison of neural networks and regression-based approaches for estimating kidney function in pediatric chronic kidney disease: Practical predictive epidemiology for clinical management of a progressive disease

PURPOSE

Clinical management of pediatric chronic kidney disease requires estimation of glomerular filtration rate (eGFR). Currently, eGFR is determined by two endogenous markers measured in blood: serum creatine (SCr) and cystatin C (CysC). Machine learning methods show promise to potentially improve eGFR, but it is unclear if they can outperform regression-based approaches under clinical constraining requiring real time measurement and only two predictors. We constructed a neural network for eGFR (NNeGFR) and compared it to the clinical standard Under 25 (U25eGFR) equations using the same data for training and validation.

METHODS

The U25eGFR data comprised 1683 training and 843 validation observations that included iohexol measured GFR (mGFR), SCr and CysC. Sex-stratified feed forward NNs included the same predictors as U25eGFR (i.e., age, height/SCr, CysC) with additional nonlinear transformations. Performance was evaluated by bias (for calibration), proportions within 10 % and 30 % of mGFR (P10 and P30, for accuracy), root mean square error (RMSE, for precision) and R2 (for discrimination).

RESULTS

NNeGFR performed comparably to the U25eGFR equations on all metrics. Biases were minimal, slightly favoring U25eGFR. NNeGFR and U25eGFR had similar P10 (>37 %), P30 (>86 %) and RMSE.

CONCLUSIONS

NNeGFR performed as well as established equations to estimate GFR. Without additional biomarkers related to kidney function, which are not currently clinically available in real time, NN methods are unlikely to substantially outperform regression derived GFR estimating equations. Implications for translation of these advanced epidemiologic methods to clinical practice are discussed.

Enhancing individual glomerular filtration rate assessment: can we trust the equation? Development and validation of machine learning models to assess the trustworthiness of estimated GFR compared to measured GFR

BACKGROUND

Creatinine-based estimated glomerular filtration rate (eGFR) equations are widely used in clinical practice but exhibit inherent limitations. On the other side, measuring GFR is time consuming and not available in routine clinical practice. We developed and validated machine learning models to assess the trustworthiness (i.e. the ability of equations to estimate measured GFR (mGFR) within 10%, 20% or 30%) of the European Kidney Function Consortium (EKFC) equation at the individual level.

METHODS

This observational study used data from European and US cohorts, comprising 22,343 participants of all ages with available mGFR results. Four machine learning and two traditional logistic regression models were trained on a cohort of 9,202 participants to predict the likelihood of the EKFC creatinine-derived eGFR falling within 30% (p30), 20% (p20) or 10% (p10) of the mGFR value. The algorithms were internally and then externally validated on cohorts of respectively 3,034 and 10,107 participants. The predictors included in the models were creatinine, age, sex, height, weight, and EKFC.

RESULTS

The random forest model was the most robust model. In the external validation cohort, the model achieved an area under the curve of 0.675 (95%CI 0.660;0.690) and an accuracy of 0.716 (95%CI 0.707;0.725) for the P30 criterion. Sensitivity was 0.756 (95%CI 0.747;0.765) and specificity was 0.485 (95%CI 0.460; 0.511) at the 80% probability level that EKFC falls within 30% of mGFR. At the population level, the PPV of this machine learning model was 89.5%, higher than the EKFC P30 of 85.2%. A free web-application was developed to allow the physician to assess the trustworthiness of EKFC at the individual level.

CONCLUSIONS

A strategy using machine learning model marginally improves the trustworthiness of GFR estimation at the population level. An additional value of this approach lies in its ability to provide assessments at the individual level.

Precision Adverse Drug Reactions Prediction with Heterogeneous Graph Neural Network

Accurate prediction of Adverse Drug Reactions (ADRs) at the patient level is essential for ensuring patient safety and optimizing healthcare outcomes. Traditional machine learning-based methods primarily focus on predicting potential ADRs for drugs, but they often fall short of capturing the complexity of individual demographics and the variations in ADRs experienced by different people. In this study, a novel framework called Precise Adverse Drug Reaction (PreciseADR) for patient-level ADR prediction is proposed. The approach effectively integrates relations between patients and ADRs, and harnesses the power of heterogeneous Graph Neural Networks (GNNs) to address the limitations of traditional methods. Specifically, a heterogeneous graph representation of patients is constructed, encompassing nodes that represent patients, diseases, drugs, and ADRs. By leveraging edges in the graph, crucial connections are captured such as a patient being affected by diseases, taking specific drugs, and experiencing ADRs. Next, a GNN-based model is utilized to learn latent representations of the patient nodes and facilitate the propagation of information throughout the graph structure. By employing patient embeddings that consider their diseases and drugs, potential ADRs can be accurately predicted. The PreciseADR is dedicated to effectively capturing both local and global dependencies within the heterogeneous graph, allowing for the identification of subtle patterns and interactions that play a significant role in ADRs. To evaluate the performance of the approach, extensive experiments are conducted on a large-scale real-world healthcare dataset with adverse reports from the FDA Adverse Event Reporting System (FAERS). Experimental results demonstrate that the PreciseADR achieves superior predictive performance in identifying patient-level ADRs, surpassing the strongest baseline by 3.2% in AUC score and by 4.9% in Hit@10.

Comparison between the EKFC-equation and machine learning models to predict Glomerular Filtration Rate

In clinical practice, the glomerular filtration rate (GFR), a measurement of kidney functioning, is normally calculated using equations, such as the European Kidney Function Consortium (EKFC) equation. Despite being the most general equation, EKFC, just like previously proposed approaches, can still struggle to achieve satisfactory performance, limiting its clinical applicability. As a possible solution, recently machine learning (ML) has been investigated to improve GFR prediction, nonetheless the literature still lacks a general and multi-center study. Using a dataset with 19,629 patients from 13 cohorts, we investigate if ML can improve GFR prediction in comparison to EKFC. More specifically, we compare diverse ML methods, which were allowed to use age, sex, serum creatinine, cystatin C, height, weight and BMI as features, in internal and external cohorts against EKFC. The results show that the most performing ML method, random forest (RF), and EKFC are very competitive where RF and EKFC achieved respectively P10 and P30 values of 0.45 (95% CI 0.44;0.46) and 0.89 (95% CI 0.88;0.90), whereas EKFC yielded 0.44 (95% CI 0.43; 0.44) and 0.89 (95% CI 0.88; 0.90), considering the entire cohort. Small differences were, however, observed in patients younger than 12 years where RF slightly outperformed EKFC.

Diagnostic standard: assessing glomerular filtration rate

Creatinine-based estimated glomerular filtration rate (eGFR) is imprecise at individual level, due to non-GFR-related serum creatinine determinants, including atypical muscle mass. Cystatin C has the advantage of being independent of muscle mass, a feature that led to the development of race- and sex-free equations. Yet, cystatin C-based equations do not perform better than creatinine-based equations for estimating GFR unless both variables are included together. The new race-free Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation had slight opposite biases between Black and non-Black subjects in the USA, but has poorer performance than that the previous version in European populations. The European Kidney Function Consortium (EKFC) equation developed in 2021 can be used in both children and adults, is more accurate in young and old adults, and is applicable to non-white European populations, by rescaling the Q factor, i.e. population median creatinine, in a potentially universal way. A sex- and race-free cystatin C-based EKFC, with the same mathematical design, has also be defined. New developments in the field of GFR estimation would be standardization of cystatin C assays, development of creatinine-based eGFR equations that incorporate muscle mass data, implementation of new endogenous biomarkers and the use of artificial intelligence. Standardization of different GFR measurement methods would also be a future challenge, as well as new technologies for measuring GFR. Future research is also needed into discrepancies between cystatin C and creatinine, which is associated with high risk of adverse events: we need to standardize the definition of discrepancy and understand its determinants.

UroAngel: a single-kidney function prediction system based on computed tomography urography using deep learning

BACKGROUND

Accurate estimation of the glomerular filtration rate (GFR) is clinically crucial for determining the status of obstruction, developing treatment strategies, and predicting prognosis in obstructive nephropathy (ON). We aimed to develop a deep learning-based system, named UroAngel, for non-invasive and convenient prediction of single-kidney function level.

METHODS

We retrospectively collected computed tomography urography (CTU) images and emission computed tomography diagnostic reports of 520 ON patients. A 3D U-Net model was used to segment the renal parenchyma, and a logistic regression multi-classification model was used to predict renal function level. We compared the predictive performance of UroAngel with the Modification of Diet in Renal Disease (MDRD), Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations, and two expert radiologists in an additional 40 ON patients to validate clinical effectiveness.

RESULTS

UroAngel based on 3D U-Net convolutional neural network could segment the renal cortex accurately, with a Dice similarity coefficient of 0.861. Using the segmented renal cortex to predict renal function stage had high performance with an accuracy of 0.918, outperforming MDRD and CKD-EPI and two radiologists.

CONCLUSIONS

We proposed an automated 3D U-Net-based analysis system for direct prediction of single-kidney function stage from CTU images. UroAngel could accurately predict single-kidney function in ON patients, providing a novel, reliable, convenient, and non-invasive method.

Evaluating and Improving the Performance and Racial Fairness of Algorithms for GFR Estimation

Data-driven clinical prediction algorithms are used widely by clinicians. Understanding what factors can impact the performance and fairness of data-driven algorithms is an important step towards achieving equitable healthcare. To investigate the impact of modeling choices on the algorithmic performance and fairness, we make use of a case study to build a prediction algorithm for estimating glomerular filtration rate (GFR) based on the patient's electronic health record (EHR). We compare three distinct approaches for estimating GFR: CKD-EPI equations, epidemiological models, and EHR-based models. For epidemiological models and EHR-based models, four machine learning models of varying computational complexity (i.e., linear regression, support vector machine, random forest regression, and neural network) were compared. Performance metrics included root mean squared error (RMSE), median difference, and the proportion of GFR estimates within 30% of the measured GFR value (P30). Differential performance between non-African American and African American group was used to assess algorithmic fairness with respect to race. Our study showed that the variable race had a negligible effect on error, accuracy, and differential performance. Furthermore, including more relevant clinical features (e.g., common comorbidities of chronic kidney disease) and using more complex machine learning models, namely random forest regression, significantly lowered the estimation error of GFR. However, the difference in performance between African American and non-African American patients did not decrease, where the estimation error for African American patients remained consistently higher than non-African American patients, indicating that more objective patient characteristics should be discovered and included to improve algorithm performance.