Our motivating interest is to predict kidney disease in its early stages, before serum creatinine levels are elevated. Our ultimate goal is to create a tool that applied to a patient’s routine labs would predict the likelihood of kidney disease. Such a tool could lead to early diagnosis.

The MIMIC IV dataset contains a full range of routine labs and many kidney patients in the ICU. We seek to answer the question: Could kidney disease patients be distinguished from non-kidney disease patients in the ICU on the basis of a set of routine labs that does not include serum creatinine, the lab test whose confirmed elevated level defines kidney disease?

If we can answer this question in the affirmative in the more compromised ICU setting, then it would look very promising that our question, with the appropriate data, could be answered with our methods in the affirmative in the outpatient setting.