Reason2Decide: Rationale-Driven Multi-Task Learning

TL;DR

Reason2Decide introduces a two-stage training framework that first learns rationale generation and then jointly optimizes prediction and explanation with task-level scheduled sampling to mitigate exposure bias. By conditioning explanations on predicted labels during training, the approach aligns rationales with predictions and improves fidelity, achieving superior results on clinical triage and biomedical QA tasks across multiple model sizes. The method demonstrates robustness to rationale source, showing that LLM-generated rationales can effectively pretrain smaller models, enabling high-quality explainable decision support with substantially smaller architectures. These findings advance practical, explainable clinical decision support, particularly in resource-constrained settings, and point to future work in broader domains and human-centered evaluation.

Abstract

Despite the wide adoption of Large Language Models (LLM)s, clinical decision support systems face a critical challenge: achieving high predictive accuracy while generating explanations aligned with the predictions. Current approaches suffer from exposure bias leading to misaligned explanations. We propose Reason2Decide, a two-stage training framework that addresses key challenges in self-rationalization, including exposure bias and task separation. In Stage-1, our model is trained on rationale generation, while in Stage-2, we jointly train on label prediction and rationale generation, applying scheduled sampling to gradually transition from conditioning on gold labels to model predictions. We evaluate Reason2Decide on three medical datasets, including a proprietary triage dataset and public biomedical QA datasets. Across model sizes, Reason2Decide outperforms other fine-tuning baselines and some zero-shot LLMs in prediction (F1) and rationale fidelity (BERTScore, BLEU, LLM-as-a-Judge). In triage, Reason2Decide is rationale source-robust across LLM-generated, nurse-authored, and nurse-post-processed rationales. In our experiments, while using only LLM-generated rationales in Stage-1, Reason2Decide outperforms other fine-tuning variants. This indicates that LLM-generated rationales are suitable for pretraining models, reducing reliance on human annotations. Remarkably, Reason2Decide achieves these gains with models 40x smaller than contemporary foundation models, making clinical reasoning more accessible for resource-constrained deployments while still providing explainable decision support.

Figures (1)

• Figure 1: Overview of Reason2Decide. Stage-1 trains rationale generation. Stage-2 jointly predicts labels and generates label-conditioned explanations with task-level scheduled sampling. A single T5 model is used throughout; inference conditions explanations on the model’s own prediction.