Download the Current Annual Report (PDF, 8.5 MB)

It’s Not Just for Sci-Fi: Using Artificial Intelligence to Identify Kidney Disease

It’s Not Just for Sci-Fi: Using Artificial Intelligence to Identify Kidney Disease

Successful development and implementation of an artificial intelligence-driven clinical decision support system for detection and treatment of acute kidney injury in the emergency department may improve the quality of kidney care and generate best practice methods to advance the application of artificial intelligence as well as develop a scalable CDS product.

Kidney disease is common but difficult to detect

Providers delivering care in the emergency department (ED) are faced with making high-stakes clinical decisions in a high-volume, time-sensitive, and hectic environment. Acute kidney injury (AKI) is a very common condition for patients presenting in the ED and is associated with adverse clinical outcomes. Unlike conditions such as heart attack, where a provider witnesses a patient wincing in pain, or stroke, where a provider can rapidly measure a patient’s strength and assess their speech, it is difficult to identify kidney injury quickly. Detection and assessment of the severity of AKI in the ED setting is limited due to laboratory-based diagnostic criteria and a general lack of syndrome recognition. Diagnosis of AKI currently relies on detection of changes in serum creatinine (sCr) concentration and urine output, both of which may take several days to manifest and identify.

Oftentimes, the signal that is used to detect kidney injury is delayed. So, no matter what we would do in response to that signal, we don't have a way to know as emergency physicians that this person's going to develop acute kidney injury. There is a protein product that builds up in the blood and it can be measured, but that doesn't happen for days after the initial injury, so we can totally miss it.”
– Dr. Hinson

Complex pattern matching

To address this diagnosing challenge, Drs. Jeremiah Hinson and Scott Levin and a Johns Hopkins University-based research team are betting that artificial intelligence (AI) techniques will be able to extract and analyze relevant electronic health record (EHR) data for AKI detection and treatment in emergency point-of-care clinical decision support (CDS). They are developing an EHR-based algorithm to estimate AKI risk and flag patients at high risk for AKI. This algorithm will then be translated into an AKI-CDS system and pilot-tested among emergency department providers. While it sounds very complicated, Dr. Hinson describes it as pattern matching that is very complex.

We have a database with over 300,000 patient encounters from people who have come into the emergency department in the past, some of whom developed AKI. We can use computers and algorithms to really mine all of these data, and look at patterns in patients who developed AKI. We can leverage the information that we've gained over time, to identify patients who are at really high risk, based on other patients in the past, and develop an estimate of how sure we are about that, or what their probability is of developing AKI. All the data come together and they tell a story, but it can be too complex for us to do that pattern matching using a human brain.”
– Dr. Hinson

Innovative research that can be scaled and disseminated

Dr. Hinson and team believe this research is highly innovative and will have an impact, as it will generate important knowledge and tools to advance the study and application of AI in the ED; this will result in a CDS product that is scalable via distribution platforms such as AHRQ’s CDS Connect. This research has the capacity to improve the quality of kidney care delivered to more than 1 million patients affected by AKI in the United States every year.