Step-by-Step Guidance to Differential Anemia Diagnosis with Real-World Data and Deep Reinforcement Learning
Lillian Muyama, Estelle Lu, Geoffrey Cheminet, Jacques Pouchot, Bastien Rance, Anne-Isabelle Tropeano, Antoine Neuraz, Adrien Coulet
TL;DR
This paper presents a deep reinforcement learning framework to learn step-by-step diagnostic pathways for anemia using electronic health records. By formulating anemia diagnosis as an MDP and employing Duelling DQN architectures with PER, the authors generate transparent action sequences that guide laboratory testing and final diagnoses. The evaluation uses a synthetic, expert-defined dataset and a real-world hospital dataset, comparing synthetic-only training, real-world training, and transfer learning, and shows DRL can match or exceed baseline methods while providing interpretable diagnostic pathways. The work highlights the value of synthetic data for low-resource scenarios, demonstrates gains from fine-tuning on real data, and emphasizes pathway explainability as a practical benefit for clinical decision support.
Abstract
Clinical diagnostic guidelines outline the key questions to answer to reach a diagnosis. Inspired by guidelines, we aim to develop a model that learns from electronic health records to determine the optimal sequence of actions for accurate diagnosis. Focusing on anemia and its sub-types, we employ deep reinforcement learning (DRL) algorithms and evaluate their performance on both a synthetic dataset, which is based on expert-defined diagnostic pathways, and a real-world dataset. We investigate the performance of these algorithms across various scenarios. Our experimental results demonstrate that DRL algorithms perform competitively with state-of-the-art methods while offering the significant advantage of progressively generating pathways to the suggested diagnosis, providing a transparent decision-making process that can guide and explain diagnostic reasoning.