Acute Kidney Injury (AKI) is a common and highly lethal health problem, affecting 10-15% of all hospitalized patients and >50% of patients in intensive care units (ICUs). It has been shown that a small increase in serum creatinine (SCr) of ?0.5 mg/dl was associated with a 6.5-fold increase in the odds of death, a 3.5-day increase in length of stay, and nearly $7,500 in excess hospital costs. Unfortunately, no specific treatment exists to cure AKI once it has developed. The ability to predict AKI in hospitalized patients would provide clinicians the opportunity to modify care pathways and implement interventions, which could in turn prevent AKI and yield better outcomes. Although electronic medical record (EMR) based monitoring systems for AKI have led to expedited interventions and may increase the percentage of patients returning to baseline kidney function, most of these systems are reactive rather than proactive, with little or no contribution to AKI prevention. Moreover, our current knowledge of AKI risk factors is far from complete, especially in the ICU and general inpatient populations, characterized by numerous deficiencies and systematic failings that may be avoidable To transform the reactive AKI care to proactive and personalized care, early identification of high risk patients and better understanding of individual modifiable risk factors for AKI is the key.
In Aim 1, to discover novel risk factors predictive of AKI, we propose to develop an ensemble multi-view feature selection framework to simultaneously consider the differences and interrelations between feature spaces and obtain robust knowledge by synthesizing findings from diverse patient populations across multiple institutions in nine US states.
In Aim 2, to discover general modifiable causes of AKI to help physicians design more effective AKI prevention policies, we propose to develop a novel multi-cause inference method to identify causal relationships between modifiable factors and AKI for susceptible patient subgroups.
In Aim 3, to explain what caused AKI in individual patients to support physicians in designing personalized AKI intervention, we propose to develop a new causal explanation method by integrating causal inference and case based reasoning to quantify patient-level causal significance of modifiable factors. The proposed study will have a significant clinical impact by not only expanding the capacity of clinicians to identify high risk patients for AKI early and advancing the general knowledge on causal and modifiable risk factors for AKI but also supporting personalized AKI intervention with suggestions on potential patient-specific actionable items. The work will not only advance AKI but also the machine learning and clinical research informatics community and the methodology developed is generalizable to other clinical domains.
The proposed research is to identify clinical risk factors of acute kidney injury (AKI) in hospitalized patients from electronic medical records (EMRs) with machine learning. AKI risk factors discovered from EMR of diverse populations from multiple institutions across nine US states will be reliable and robust and can assist clinicians in providing proactive and personalized care to high-risk patients.