Lead Investigator: Jean-Pierre Fauvel, University Claude Bernard Lyon 1
Title of Proposal Research: Development and validation of clinical prediction tools to estimate all-cause mortality, cardiovascular and renal risk in non-dialysis chronic kidney disease (CKD) patients using machine learning
Vivli Data Request: 10116
Funding Source: French National Institute of Health and Medical Research. Dr. Tran has received a research scholarship from the Société Francophone de Néphrologie Dialyse et Transplantation. GlaxoSmithKline, AstraZeneca, Astellas.
Potential Conflicts of Interest: The research project was partially funded by commercial entities (GSK, AZ, Astellas). These research grants were allocated to the authors’ affiliated institution, the Hospices Civils de Lyon. There was no personal funding: the authors have no conflict of interest. We have declared such fundings in our previous research and will continue to declare them in our subsequent publications.
Summary of the Proposed Research:
Chronic kidney disease (CKD) is a serious condition with high levels of morbidity (other illnesses) and mortality (death), particularly of cardiovascular (CV) origin. Cardiovascular disease is the leading cause of death in patients with CKD. Although the impact of coexisting medical conditions on morbidity and mortality in patients with CKD has been studied, the individual prognosis (expected outcomes) of each patient is not yet well established.
Available tools for predicting cardiovascular risk, such as Framingham, SCORE2 or Q-risk are not adapted to CKD patients, as (i) they were developed in the general population; (ii) they predict the risk of developing CV disease within the next 10-years, whereas CV risk in CKD patients is higher, and (iii) they do not take into account variables of nephrological (kidney) interest. The most widely used tool for predicting CKD progression is the Kidney Failure Risk Equation (KFRE). This risk calculator takes into account 4 main variables: kidney function, age, sex, urinary albumin (protein in the urine). Unfortunately, it does not take into account important factors such as hypertension (high blood pressure) and diabetes (a condition causing high blood sugar levels). It therefore needs to be improved.
In a systematic review (a review of all the available literature) in 2012 it was reported that “prediction models for chronic kidney disease have often been developed using inappropriate methods and have generally been poorly reported”. In a more recent systematic review in 2020, it was stated that “future research efforts should focus on external validation and impact assessment on clinically relevant patient populations”. It is therefore necessary to carefully develop and externally validate clinical tools to predict mortality, CV risk and kidney disease progression in patients with CKD. These are the objectives of this project.
Data from the following sources will be used to construct a dataset that will be used to create and train a prediction model using machine learning techniques: the PhotoGraph 3 cohort, the ALICE-PROTECT cohort and data extracted from CKD patients consulted at the AURAL-Alsace nephrology center and the Hospices Civils de Lyon, France. The model with the best performance will be chosen to optimize the clinical prediction tool using a synthetic learning dataset and refined predictors most related to the outcome.
Optimized prediction tools will be validated externally using the CKD-REIN cohort, data from requested nephrology studies (e.g. ASCEND-ND) and the Singapore Renal Registry. CKD-REIN stands for Chronic Kidney Disease–Renal Epidemiology and Information Network. It is a cohort study that was conducted in France to investigate the causes, progression, and consequences of chronic kidney disease (CKD). The study’s goal was to identify the best strategies for preventing and managing the disease. Data from the Singapore renal registry (SRR) will be used for an external validation of our predictive tools on an independent and different population. The aim of the external validation using the Singaporean population is to ensure the generalizability of the predictive tools, since our predictive tools were developed using the French population. We have obtained permission to use the SRR data.
Requested Studies:
A Phase 3 Randomized, Open-label (Sponsor-blind), Active-controlled, Parallel-group, Multi-center, Event Driven Study in Non-dialysis Subjects With Anemia Associated With Chronic Kidney Disease to Evaluate the Safety and Efficacy of Daprodustat Compared to Darbepoetin Alfa
Data Contributor: GlaxoSmithKline
Study ID: NCT02876835
Sponsor ID: 200808