Machine Learning-Derived Personalized Fluid Intake Strategy in Sepsis-Associated AKI

Background: Sepsis associated acute kidney injury (SA-AKI) is common and associated with poor outcomes. While intravenous fluids are mainstay of therapy for these patients, recent literature suggests that restrictive fluid administration may be beneficial in certain patients with SA-AKI. This study aims to identify SA-AKI patients who would benefit from restrictive fluid administration. Methods: This retrospective study used eICU database for development and MIMIC-IV for external validation. We identified 2,086 SA-AKI patients in eICU and 6,532 patients in MIMIC-IV using KDIGO criteria for AKI. We defined restrictive fluid as administration of <500mL of fluid in 24 hours after SA-AKI. Primary outcome was early AKI reversal within 48 hours of AKI onset. Secondary outcomes included sustained AKI reversal for at least additional 48 hours after early AKI reversal and Major Adverse Kidney Events (MAKE) by discharge (defined as death, need for new dialysis, or discharge creatinine>=200% of baseline). We used causal tree and policy tree algorithms to estimate individual treatment effects and identify patients benefiting most from fluid restriction, and multivariable logistic regression to assess its impact on outcomes. Results: 38% patients in the eICU and 43% patients in MIMIC-IV were recommended restrictive fluids. Among those recommended, only 26% patients in eICU and 8% patients in MIMIC-IV received restrictive fluids. Among patients that were recommended restrictive fluids, those that received had higher rates of early AKI reversal (eICU: 73%vs 27%, p<.01; MIMIC-IV: 53%vs 44%, p<.01) and sustained AKI reversal (eICU: 52%vs 18%, p<.01; MIMIC-IV: 36%vs 29%, p<.01), and lower rates of MAKE at discharge (eICU: 26%vs 38%, p<.01 ; MIMIC-IV: 20%vs 32%, p<.01) in both eICU and MIMIC-IV (validation set). These effects were similar after adjustment of confounders (Fig 1). Conclusion: In this study using a novel, machine learning approach we have developed and validated a strategy to identify SA-AKI patients who would benefit from a restrictive fluid administration. Funding: NIDDK Support