Retrospective evaluation of a rule-constrained GPT-4-based LLM system for neuropsychological lateralization and localization in presurgical epilepsy diagnostics

Retrospective evaluation of a rule-constrained GPT-4-based LLM system for neuropsychological lateralization and localization in presurgical epilepsy diagnostics - REPRO-EP

Status

Active, not recruiting

Phases

Unknown

Study type

Observational

Source

DRKS

Registry ID

DRKS00039480

Enrollment

Registered

2026-07-21

Start date

2026-02-02

Completion date

Unknown

Last updated

2026-08-03

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Drug-resistant epilepsy in the context of presurgical epilepsy diagnostics G40.1 G40.2 G40.6 G40.9

Interventions

Group 1: Retrospective observational cohort: Patients undergoing presurgical epilepsy diagnostics at the University Hospital Bonn in 2024 with a complete neuropsychological report. No intervention is

pseudonymized routine clinical data are analyzed.

Eligibility

Sex/Gender

All

Age

18 Years to No maximum

Inclusion criteria

Inclusion criteria: Presurgical epilepsy diagnostics at the University Hospital Bonn during calendar year 2024 Availability of a neuropsychological report with structured test documentation Availability of a documented neuropsychological conclusion on lateralization and/or localization Sufficient documentation to extract study-relevant variables Pseudonymized availability of routine clinical data for retrospective analysis

Exclusion criteria

Exclusion criteria: Prematurely terminated neuropsychological assessment (e.g., insufficient cooperation, fatigue, or medical reasons) Documented intellectual disability with no meaningful domain-specific interpretation of the neuropsychological profile Insufficient documentation to extract structured neuropsychological test values and/or the routine neuropsychological conclusion

Design outcomes

Primary

Measure	Time frame
Agreement between lateralization (left/right/bilateral/unclear) and coarse localization (temporal/frontal/fronto-temporal/unclear) labels generated by a rule-constrained GPT-4-based LLM system (UKB-GPT) from structured neuropsychological test data and the routinely documented neuropsychological conclusion. When: One-time retrospective assessment at the presurgical work-up timepoint (calendar year 2024). How: Codebook-based extraction/coding of categories from the neuropsychological report and comparison with UKB-GPT outputs; analysis using percent agreement and Cohen’s kappa.	—

Measure

Time frame

Agreement between lateralization (left/right/bilateral/unclear) and coarse localization (temporal/frontal/fronto-temporal/unclear) labels generated by a rule-constrained GPT-4-based LLM system (UKB-GPT) from structured neuropsychological test data and the routinely documented neuropsychological conclusion. When: One-time retrospective assessment at the presurgical work-up timepoint (calendar year 2024). How: Codebook-based extraction/coding of categories from the neuropsychological report and comparison with UKB-GPT outputs; analysis using percent agreement and Cohen’s kappa.

—

Secondary

Measure	Time frame
(1) Distribution of the model-reported confidence rating (low/medium/high); (2) proportion of “unclear” outputs in routine neuropsychology and UKB-GPT (lateralization/localization); (3) agreement of UKB-GPT and routine neuropsychology with the presurgical summary assessment documented in the discharge/physician letter (lateralization/localization); (4) exploratory qualitative categorization of discordant cases based on model “evidence” statements (e.g., cutoff/rule errors, evidence–label mismatch, context-related divergence). When: One-time retrospective assessment (calendar year 2024). How: Descriptive statistics, contingency tables, Cohen’s kappa, and qualitative categorization.	—

Measure

Time frame

(1) Distribution of the model-reported confidence rating (low/medium/high); (2) proportion of “unclear” outputs in routine neuropsychology and UKB-GPT (lateralization/localization); (3) agreement of UKB-GPT and routine neuropsychology with the presurgical summary assessment documented in the discharge/physician letter (lateralization/localization); (4) exploratory qualitative categorization of discordant cases based on model “evidence” statements (e.g., cutoff/rule errors, evidence–label mismatch, context-related divergence). When: One-time retrospective assessment (calendar year 2024). How: Descriptive statistics, contingency tables, Cohen’s kappa, and qualitative categorization.

—

Countries

Germany

Contacts

Public ContactSarah Al-Haj Mustafa

Universitätsklinikum Bonn (AöR), Klinik und Poliklinik für Epileptologie

Sarah.Al-Haj-Mustafa@ukbonn.de+4922828713638

Outcome results

None listed

Source: DRKS (via WHO ICTRP) · Data processed: Aug 10, 2026