Purpose of review
Recent years have witnessed the development of kidney risk prediction models which diverge from traditional model designs to incorporate novel approaches along with a focus on earlier outcomes. This review summarizes these recent advances, evaluates their pros and cons, and discusses their potential implications.
Several kidney risk prediction models have recently been developed utilizing machine learning rather than traditional Cox regression. These models have demonstrated accurate prediction of kidney disease progression, often beyond that of traditional models, in both internal and external validation. On the opposite end of the spectrum, a simplified kidney risk prediction model was recently developed which minimized the need for laboratory data and instead relies primarily on self-reported data. While internal testing showed good overall predictive performance, the generalizability of this model remains uncertain. Finally, there is a growing trend toward prediction of earlier kidney outcomes (e.g., incident chronic kidney disease [CKD]) and away from a sole focus on kidney failure.
Newer approaches and outcomes now being incorporated into kidney risk prediction modeling may enhance prediction and benefit a broader patient population. However, future work should address how best to implement these models into practice and assess their long-term clinical effectiveness.