Researchers reviewed many studies about computer models that try to predict which patients with diabetic kidney disease will develop kidney failure. They found that the models showed good accuracy in early testing. However, the studies had many serious problems that make the results hard to trust.
Most studies were small and only looked at patients from one hospital. They also often used test results that doctors do not normally have. This makes it hard to use these models in everyday care. The studies did not test the models on new groups of patients, which is very important.
The review found that all the studies had a high risk of bias. This means the results might not be correct. The models also varied a lot from one study to another. This extreme heterogeneity makes it difficult to know what the true accuracy is.
Because of these issues, the models cannot yet be used to guide patient care. More research is needed to fix these problems. Future studies should use better methods and test the models in different hospitals.