Abstract: FR-PO0039
Equivalence of Artificial Intelligence (AI) and Physician-Guided Erythropoiesis-Stimulating Agent (ESA) Dosing in Hemodialysis: A Randomized Controlled Trial
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Chiu, Yi-Wen, Kaohsiung Medical University, Kaohsiung, Kaohsiung City, Taiwan
- Hsu, Chan, National Sun Yat-sen University, Kaohsiung, Kaohsiung City, Taiwan
- Lin, Ming-Yen, Kaohsiung Medical University Chung Ho Memorial Hospital, Kaohsiung, Kaohsiung City, Taiwan
- Kang, Yihuang, National Sun Yat-sen University, Kaohsiung, Kaohsiung City, Taiwan
Background
To optimize ESA dosing in hemodialysis (HD), we tested a top-ranked meta-learning model (TH-PO042, 2023 ASN KW) against physician-guided dosing in a clinical trial.
Methods
In this double-blind, crossover RCT (NCT05032651), patients on maintenance HD were randomized into two arms: Arm 1 began as the AI group, receiving ESA dosing recommended by the model for three months (Phase I), then switched to the Dr group after a one-month washout (Phase II); Arm 2 followed the reverse sequence. Hemoglobin (Hb) was assessed twice monthly for equivalence comparison. The primary outcome was maintaining Hb near 11 g/dL, with an equivalence margin of ±0.25 g/dL; the secondary outcome was maintaining Hb within 10–12 g/dL, with a margin of ±15%.
Results
We enrolled 175 participants who provided legal consent (Arm 1: 77; Arm 2: 78), with a mean age of 64.2 ± 12.5 years; 45.8% were female, and 44.5% had diabetes. The baseline Hb level was 10.8 ± 0.7 g/dL, with 80.4% within the 10–12 g/dL. In Arm 1, 70 and 63 participants completed Phases I and II, respectively; in Arm 2, 75 and 71 completed the corresponding phases. Withdrawal rates did not differ significantly between Arms (p=0.11) or groups (p=1). The Table summarizes the primary and secondary outcomes. Equivalence in ESA dosing between the AI and Dr groups was demonstrated by two one-sided tests. The Figure illustrates the equivalence results across 12 assessments (primary outcome shown). SAEs were similar between groups (AI vs. Dr: 22 vs. 18; p=0.63).
Conclusion
AI-guided ESA dosing was equivalent to physician guidance in maintaining Hb targets in patients undergoing HD.
Outcome
| AI group | Dr group | Two One-sided test (upper bound ) 95% CI (p value) | Two One-sided test (lower bound ) 95% CI (p value) | |
| Overall | N= 862 tests | N=853 tests | ||
| primary outcome (Hb deviated from 11) | 0.68 ± 0.57 | 0.69 ± 0.55 | ~ to 0.03 (P<0.001) | -0.05 to ~ (p<0.001) |
| secondary outcome (portion of Hb within 10 and 12) | 77.7% | 76.0% | ~ to 5% (p<0.001) | - 1% to ~(p<0.001) |
| Phase I | N=436 tests | N=465 tests | ||
| primary outcome | 0.72 ± 0.60 | 0.70 ± 0.56 | ~ to 0.08 (p<0.01) | -0.04 to ~ (p<0.001) |
| secondary outcome | 74.1% | 75.9% | ~ to 3% (<0.001) | -7% to ~ (p<0.001) |
| Phase II | N=426 tests | N=388 tests | ||
| primary outcome | 0.65 ± 0.54 | 0.69 ± 0.54 | ~ to 0.02 (p<0.001) | -0.10 to ~ (p<0.001) |
| secondary outcome | 81.5% | 76.0 % | ~ to 10% (p<0.001) | 1% to ~ (p<0.001) |
Funding
- Private Foundation Support