# Safe to Stop? Risk-Constrained Stopping for Sequential Clinical Diagnosis Agents

arXiv: https://arxiv.org/abs/2609.09678 · HTML: https://mlsanity.com/paper/2609.09678/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 26 of its 31 references that are in the map.
- First posted on arXiv 9 Sep 2026.

## Builds on (most cited first, 26 of 26)

- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [SelectiveNet: A Deep Neural Network with an Integrated Reject Option](https://mlsanity.com/paper/1901.09192/) (2019)
- [Conformal Risk Control](https://mlsanity.com/paper/2208.02814/) (2022)
- [Distribution-Free, Risk-Controlling Prediction Sets](https://mlsanity.com/paper/2101.02703/) (2021)
- [Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Control](https://mlsanity.com/paper/2110.01052/) (2021)
- [AgentClinic: a multimodal agent benchmark to evaluate AI in simulated clinical environments](https://mlsanity.com/paper/2405.07960/) (2024)
- [MediQ: Question-Asking LLMs and a Benchmark for Reliable Interactive Clinical Reasoning](https://mlsanity.com/paper/2406.00922/) (2024)
- [Sequential Diagnosis with Language Models](https://mlsanity.com/paper/2506.22405/) (2025)
- [Efficiently Controlling Multiple Risks with Pareto Testing](https://mlsanity.com/paper/2210.07913/) (2022)
- [Deep Reinforcement Learning for Cost-Effective Medical Diagnosis](https://mlsanity.com/paper/2302.10261/) (2023)
- [Medchain: Bridging the Gap Between LLM Agents and Clinical Practice with Interactive Sequence](https://mlsanity.com/paper/2412.01605/) (2024)
- [Language Agents for Hypothesis-driven Clinical Decision Making with Reinforcement Learning](https://mlsanity.com/paper/2506.13474/) (2025)
- [Conformal Selective Prediction with General Risk Control](https://mlsanity.com/paper/2603.24704/) (2026)
- [Early Time Classification with Accumulated Accuracy Gap Control](https://mlsanity.com/paper/2402.00857/) (2024)
- [Selective Conformal Risk Control](https://mlsanity.com/paper/2512.12844/) (2025)
- [Conformal Thinking: Risk Control for Reasoning on a Compute Budget](https://mlsanity.com/paper/2602.03814/) (2026)
- [Adaptive Stopping for Multi-Turn LLM Reasoning](https://mlsanity.com/paper/2604.01413/) (2026)
- [Uncertainty-aware abstention in medical diagnosis based on medical texts](https://mlsanity.com/paper/2502.18050/) (2025)
- [Geometry-Calibrated Conformal Abstention for Language Models](https://mlsanity.com/paper/2604.27914/) (2026)
- [Timely Clinical Diagnosis through Active Test Selection](https://mlsanity.com/paper/2510.18988/) (2025)
- [Look Again Before You Abstain:Budgeted Conformal Evidence Acquisition for Reliable Vision-Language Model](https://mlsanity.com/paper/2606.16667/) (2026)
- [Uncertainty-Guided Latent Diagnostic Trajectory Learning for Sequential Clinical Diagnosis](https://mlsanity.com/paper/2604.05116/) (2026)
- [Optimal Stopping in Sequential Clinical Prediction](https://mlsanity.com/paper/2604.22216/) (2026)
- [Thinking Like a Clinician: A Cognitive AI Agent for Clinical Diagnosis via Panoramic Profiling and Adversarial Debate](https://mlsanity.com/paper/2604.23605/) (2026)
- [Ask Before You Diagnose: Safe-Psych, a Sequential Evaluation Benchmark for LLMs in Psychiatry](https://mlsanity.com/paper/2607.13036/) (2026)

## Built on by (most cited first, 1 of 1)

- [What Can Component-Replacement Evidence Establish? A Critical Scoping Review of Local Decisions in LLM Agents](https://mlsanity.com/paper/2609.39989/) (2026)

## Abstract (arXiv)

Clinical diagnosis agents must decide not only what test to request next, but also when to diagnose or defer. Existing agent benchmarks largely evaluate accuracy after fixed or unconstrained interaction, leaving autonomous stopping reliability implicit. We present Cros, a risk-constrained stopping layer combining state-wise error ranking, policy design on disjoint development splits, and LTT-style exact tests of selective diagnostic error and minimum autonomous coverage for complete sequential policies. Its finite-sample guarantee requires the candidate family, testing rule, and any randomization to be frozen before calibration labels are accessed. On a 1,834-episode MIMIC-derived abdominal-pain benchmark, the full ranker achieves exploratory state-error AUROC 0.853, compared with 0.715 for maximum class probability and 0.552 for the backbone's native stop score. On the previously viewed 367-episode evaluation split, analytically averaging over the frozen Cros weights yields 16.9% selective error at 78.8% coverage, cost 5.57, and 0.68 tests, versus 30.8% error at 100% coverage, cost 8.14, and 1.53 tests under native stopping. Forced continuation is non-monotone: error is 28.3% with HPI alone and 34.3% after full workup. However, the uniform-weight mixture ablation is cheaper on this viewed split despite missing the locked development margins, and Cros nominally satisfies the joint criterion in only 6 of 20 development resplits. Because evaluation labels were inspected during earlier development, these findings provide exploratory feasibility and audit evidence, not a confirmatory safety certificate.