Context-Dependent FHIR Serialisation Strategies for Clinical LLM Deployment: A Multi-Layer Benchmark on UK Core Data
- 1The Hong Kong Polytechnic University
Abstract
The choice of FHIR-to-text serialisation format significantly impacts clinical LLM quality (Kruskal-Wallis H=163.86, p< 10-33, {Delta}=0.24 on a 5-point scale), yet remains unstudied as a clinical deployment variable. We present FHIRBench-UK, evaluating five large language models across six serialisation formats and three clinical tasks on 100 UK Core FHIR patient bundles (18,000 scored prompts across clean and perturbed cohorts). The optimal format is context-dependent: raw_json dominates for clinical QA, hybrid_adaptive for clinical reasoning, and structured_markdown for summarisation. In 58% of model-task-complexity scenarios, raw_json is suboptimal. Model capability moderates format sensitivity: Claude Sonnet 4.5 shows 0.10-point sensitivity versus Llama 3.3s 0.39, making adaptive serialisation most valuable for budget-constrained deployments using mid-tier models. All findings replicate under clinically realistic data perturbation. The study additionally confirms a complete ranking inversion between token-level F1 and clinical quality ({rho}=-0.90), replicating US findings across UK Core profiles, and converges with independent work on open-weight models [1] to establish serialisation strategy as a replicable determinant of clinical LLM performance. We recommend task-aware serialisation routing as a zero-cost quality intervention for NHS FHIR-based LLM deployments.
Related
- Co-designing deployment strategies for disease-resistant cultivars: Insights from a participatory landscape-scale modeling study with a wine cooperative
- Medico-legal integration of molnupiravir rapid deployment during the COVID-19 pandemic: Emergency use authorization, clinical evidence, and risk of viral mutagenesis
- From Algorithm Performance to Laboratory Deployment: A Critical Review of Artificial Intelligence in Clinical Chemistry
- Clinical Validation, Bias, and Ethical Deployment of Artificial Intelligence in Imaging.
- Feature-Engineering Strategies for EEG-Based ADHD Classification in Children: A Controlled Benchmark of Expert, LLM-Guided and Full-Space Comparators
BibTeX
@misc{chong2026contextdependent,
title = {{Context-Dependent FHIR Serialisation Strategies for Clinical LLM Deployment: A Multi-Layer Benchmark on UK Core Data}},
author = {J. Chong},
howpublished = {medRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.05.26359794},
url = {https://www.medrxiv.org/content/10.64898/2026.08.05.26359794v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Context-Dependent FHIR Serialisation Strategies for Clinical LLM Deployment: A Multi-Layer Benchmark on UK Core Data
AU - J. Chong
PY - 2026
DA - 2026/08/05
PB - medRxiv
DO - 10.64898/2026.08.05.26359794
UR - https://www.medrxiv.org/content/10.64898/2026.08.05.26359794v2
AB - The choice of FHIR-to-text serialisation format significantly impacts clinical LLM quality (Kruskal-Wallis H=163.86, p< 10-33, {Delta}=0.24 on a 5-point scale), yet remains unstudied as a clinical deployment variable. We present FHIRBench-UK, evaluating five large language models across six serialisation formats and three clinical tasks on 100 UK Core FHIR patient bundles (18,000 scored prompts across clean and perturbed cohorts). The optimal format is context-dependent: raw_json dominates for clinical QA, hybrid_adaptive for clinical reasoning, and structured_markdown for summarisation. In 58% of model-task-complexity scenarios, raw_json is suboptimal. Model capability moderates format sensitivity: Claude Sonnet 4.5 shows 0.10-point sensitivity versus Llama 3.3s 0.39, making adaptive serialisation most valuable for budget-constrained deployments using mid-tier models. All findings replicate under clinically realistic data perturbation. The study additionally confirms a complete ranking inversion between token-level F1 and clinical quality ({rho}=-0.90), replicating US findings across UK Core profiles, and converges with independent work on open-weight models [1] to establish serialisation strategy as a replicable determinant of clinical LLM performance. We recommend task-aware serialisation routing as a zero-cost quality intervention for NHS FHIR-based LLM deployments.
ER -