ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: FR-PO062

Use of a Policy Tree Algorithm to Identify Maximal Treatment Effect of Crystalloid Therapy in a Cohort of Critically Ill Patients With Sepsis

Session Information

Category: Acute Kidney Injury

  • 101 AKI: Epidemiology‚ Risk Factors‚ and Prevention

Authors

  • Oh, Wonsuk, Icahn School of Medicine at Mount Sinai Department of Genetics and Genomic Sciences, New York, New York, United States
  • Kittrell, Hannah, Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Nadkarni, Girish N., Icahn School of Medicine at Mount Sinai Department of Medicine, New York, New York, United States
Background

Sepsis-associated acute kidney injury (SA-AKI) is a significant health problem in intensive care units. As SA-AKI can progress, it is important to individualize therapy early based on patient characteristics to prevent progression. We aimed to use a machine learning causal inference method to identify groups of patients that have more than the average population treatment effect, specifically in relation to intravenous crystalloids.

Methods

We identified critically ill adult patients with sepsis admitted to a tertiary academic medical center. The baseline of this study is 1 hour after sepsis onset. We excluded patients (i) who were discharged within 48 hours of onset, (ii) with a history of kidney failure, (iii) without vital signs measurement during the admission, (iv) with AKI stage 1-3 present before the baseline, and (v) without serum creatinine or urine output measurements after the baseline. The study outcome was the difference between baseline serum creatinine and peak serum creatinine during the follow-up periods. We applied a policy tree algorithm, a state-of-the-art machine learning method, to learn rule-based policy (treatment strategies) through a doubly robust estimator with a form of decision trees. We gradually increased the tree's depth (accounting for more variable interactions) and evaluated average treatment effects of crystalloids on limiting the increased peak serum creatinine levels in identified patients.

Results

We applied the policy tree algorithm on 19,179 patients. 13,204 (68.7%) patients developed AKI stage 1 or higher. As we gradually increased the policy tree, the tree identified 832, 1188, 1999, 2425, and 3384 patients who showed maximal average treatment effects (ATE) of decreasing peak serum creatinine level, and ATE showed 0.246 (95% CI: 0.008, 0.484), 0.404 (0.191, 0.617), 0.436 (0.275, 0.597), 0.526 (0.376, 0.676), and 0.553 (0.418, 0.648), respectively. Mean ATE for the entire population was -0.076 (-0.108, 0.044).

Conclusion

We used the policy tree algorithm to identify subgroups of patients with sepsis with differing benefits of crystalloid therapy on SA-AKI prevention. Our results suggest that policy learning-based patient discovery can be useful for achieving personalized therapy of sepsis to prevent SA-AKI.

Funding

  • NIDDK Support