# Compositional Reasoning Depth Predicts Clinical AI Failure: Empirical Evidence Consistent with Transformer Compositionality Limits in Electronic Health Record Question Answering

arXiv: https://arxiv.org/abs/2606.16890 · HTML: https://mlsanity.com/paper/2606.16890/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Clinical and biomedical applications, one of 17 directions in LLM reasoning.
- Builds on 16 of its 28 references that are in the map.
- First posted on arXiv 15 Jun 2026.

## Placed in

- LLM reasoning / [Clinical and biomedical applications](https://mlsanity.com/lr/clinical-and-biomedical-applications/)

## Builds on (most cited first, 16 of 16)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Large Language Models are Zero-Shot Reasoners](https://mlsanity.com/paper/2205.11916/) (2022)
- [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://mlsanity.com/paper/1809.09600/) (2018)
- [What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams](https://mlsanity.com/paper/2009.13081/) (2020)
- [Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Steps](https://mlsanity.com/paper/2011.01060/) (2020)
- [OpenAI o1 System Card](https://mlsanity.com/paper/2412.16720/) (2024)
- [MuSiQue: Multihop Questions via Single-hop Question Composition](https://mlsanity.com/paper/2108.00573/) (2021)
- [Show Your Work: Scratchpads for Intermediate Computation with Language Models](https://mlsanity.com/paper/2112.00114/) (2021)
- [On Limitations of the Transformer Architecture](https://mlsanity.com/paper/2402.08164/) (2024)
- [MedAlign: A Clinician-Generated Dataset for Instruction Following with Electronic Medical Records](https://mlsanity.com/paper/2308.14089/) (2023)
- [Brainformers: Trading Simplicity for Efficiency](https://mlsanity.com/paper/2306.00008/) (2023)
- [Compositional Reasoning with Transformers, RNNs, and Chain of Thought](https://mlsanity.com/paper/2503.01544/) (2025)
- [First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations](https://mlsanity.com/paper/2512.01241/) (2025)
- [FDARxBench: Benchmarking Regulatory and Clinical Reasoning on FDA Generic Drug Assessment](https://mlsanity.com/paper/2603.19539/) (2026)

## Abstract (arXiv)

Aggregate accuracy benchmarks conceal a systematic structure in how large language models fail at electronic health record (EHR) question answering: questions requiring more inferential steps produce disproportionately more errors. Motivated by theoretical results on transformer compositionality limits, we introduce a pre-specified hop-count taxonomy -- the number of distinct reasoning steps required to answer a clinical question from an EHR -- as a principled predictor of model failure. We annotate 313 clinician-generated MedAlign EHR question-answer pairs across four hop levels and evaluate 301 questions in a within-model ablation (claude-sonnet-4-6, zero-shot vs. extended thinking) and cross-architecture replications (gpt-4o and gpt-5.4-2026-03-05, zero-shot). All three models, spanning two providers and two OpenAI generations (GPT-4 and GPT-5), show monotone accuracy decline with hop count: Claude Sonnet zero-shot falls from 30.6% (hop=1) to 17.6% (hop=4) (Cochran-Armitage z=-2.30, p=0.011; OR per hop 0.72, 95% CI [0.56,0.92], p=0.008); GPT-4o replicates this (37.8% to 14.7%; OR 0.58 [0.45,0.75], p<0.001); and gpt-5.4-2026-03-05 confirms it (37.8% to 23.5%; OR 0.80 [0.66,0.98], p=0.027). A pre-specified context-sufficiency audit shows higher-hop questions are not differentially disadvantaged by EHR truncation (answerability 93-95% at hops 2-4 vs. 79% at hop=1), so the decline reflects compositional reasoning difficulty. Extended thinking did not significantly flatten the accuracy-depth curve across three reasoning conditions, and thinking-token usage scaled with hop count (r=0.31, p<0.0001), consistent with the predicted O(k) computational requirement. Hop count is thus a theory-motivated, cross-architecture predictor of large-language-model error on EHR question answering, with direct implications for deployment risk stratification of clinical AI.