Chronic kidney disease (CKD) is a prevalent and severe health condition associated with a high burden of morbidity and mortality. However, existing classification systems for CKD, typically based on estimated glomerular filtration rate (eGFR), proteinuria, or urine albumin-creatinine ratio, still need to fully capture the complexity of this condition or its associated risk factors and outcomes.
This limited classification and risk prediction approach can lead to adequate or appropriate management of patients with CKD and suboptimal outcomes. Therefore, there is a pressing need to improve the subtype definition of CKD to enhance the prediction of outcomes and inform effective interventions. This paper explores the importance of improving the CKD subtype definition and the potential implications for patient care and outcomes.
According to eBioMedicine, chronic kidney disease (CKD) affects 9.1% of the global population, approximately 700 million individuals, and is a significant cause of morbidity and mortality, with 1.2 million annual deaths worldwide. Despite calls for “precision nephrology,” existing classification systems and risk prediction models largely ignore the complexity of CKD and its risk factors, highlighting the need for improved subtyping.
Electronic health records (EHR) and machine learning (ML) methods provide opportunities for better subtype definition and risk prediction across diseases. Still, observational studies in CKD have not fully used available longitudinal EHR data.
A new study in the UK used seven ML methods to generate clinically relevant subtypes of CKD across 2670 variables in a population-based cohort of 350,067 individuals with the incident and prevalent CKD. The study aimed to demonstrate validity, investigate the distribution of prescription medication classes at baseline and over time, and improve risk prediction for CKD-related outcomes such as hospital admissions, mortality, and incident diseases over five years.
The results showed that ML methods could identify distinct subgroups of CKD associated with future risks of CKD and cardiovascular events, independent of established CKD risk factors. The study also highlighted the importance of considering all prescription medication classes in the context of CKD, as medications may need to be adjusted or stopped due to nephrotoxicity.
Overall, this study provides promising evidence for the potential of ML methods to improve subtyping and risk prediction for CKD, which could ultimately lead to better-targeted management and early intervention.
A study using machine learning to identify subtypes of chronic kidney disease (CKD) has identified five distinct subtypes with clinically significant differences in baseline characteristics, including early-onset, late-onset, metabolic, cancer, and cardiometabolic subtypes. Individuals with both incident and prevalent CKD showed high 5-year rates of all-cause hospital admissions, mortality, and incident chronic diseases, with significant differences across subtypes.
The study also highlighted a high medication burden, with differences across CKD subtypes, incidents, and prevalent CKD. The authors suggest that the findings highlight integrated CKD primary prevention, including Type 2 diabetes, hypertension, cardiovascular disease, cancer, and age. Further research is required to validate the subtypes externally and assess their effectiveness and cost-effectiveness.
The methodology and framework used are applicable, generalizable, and scalable to other diseases where electronic health record data is available. The cardiometabolic subtype was found to have the worst prognosis, with the highest incidence of cancer, cardiovascular disease, and mortality.
The study suggests that the findings could have direct clinical applications and inform guidelines where inappropriate prescribing before and after diagnosis is common. The authors highlight the need for more evidence to inform management guidelines and facilitate knowledge and action to prevent medication-induced kidney disease.