19 AUG 2026 · PAPER INTO PRACTICE · AI & DATA

Can an LLM Choose the Right Flow Panel?

In a 100-case study, an EHR-integrated LLM matched staff performance for extracting prior diagnoses but performed worse at direct panel selection. The distinction is central to clinical implementation.

By Juan Manuel Ojeda · CFCM

PRIMARY EVIDENCE · PEER-REVIEWED · CORRECTION

Primary evidence

Rojansky R, Keyes T, Oak J. Automating clinical history extraction for flow cytometry panel selection using an EHR-integrated large language model. Cytometry Part B. Published 10 July 2026. DOI: 10.1002/cyto.b.70051.

What it actually demonstrates

This study evaluated an EHR-integrated LLM across 100 cases to automate clinical-history extraction and support panel selection against a standardised institutional decision-tree workflow. It is more precise to describe this as LLM-assisted clinical-context extraction and panel-selection support, rather than an unconstrained LLM independently designing a flow panel.

Evidence boundary

Editorial correction: CFCM will use the narrower description above when interpreting this study. The distinction matters because deterministic institutional rules remain part of the workflow.

Primary record →

The useful distinction is extraction versus rule execution: ChatEHR and laboratory staff each classified prior diagnoses correctly in 78% of the 100 cases, while direct panel-selection accuracy was 47% versus 78%. This supports testing a hybrid extraction-and-rules workflow rather than handing panel selection to the LLM.

Processing averaged 20.3 seconds with ChatEHR versus 41 seconds manually. Faster extraction did not establish reliable execution of institutional selection rules; the annual time saving reported by the authors is an estimate.

The evaluation concerns selection among established panels at one institution. It does not test fluorochrome assignment, de novo panel design or autonomous clinical deployment.

CFCM view

The important question is not whether the technology or marker is interesting. It is whether it changes a validated clinical workflow in a measurable, reproducible and explainable way.

Editorial note

CFCM is an independent publication. References to manufacturers, instruments, reagents, software or therapies do not constitute endorsement. Technical content should be interpreted in the context of local validation, applicable regulation and manufacturer instructions for use.

Juan Manuel Ojeda — Founder & Editor, CFCM

Professional background includes Sysmex, Newcastle University and the NHS. CFCM is Juan Manuel Ojeda's independent editorial project and represents the author alone. Manufacturer references do not imply endorsement. Relevant professional or financial relationships are disclosed alongside the affected coverage.

Original CFCM report: the August date shown above. Expanded web version and source-check pass: 19 September 2026. This was not independent scientific review. Source publication dates are separate; this is not retrospective web publication.

Resources →Clinical Notes →Related: evidence, validation and clinical AI →

References

Rojansky R, Keyes T, Oak J. Automating clinical history extraction for flow cytometry panel selection using an EHR-integrated large language model. Cytometry Part B. Published 10 July 2026. DOI: 10.1002/cyto.b.70051.

Primary record →