P.184 Synthetic neurosurgical data generation using large language models
Bibliographic record
Abstract
Background: Use of neurosurgical data for research and machine learning model development is often constrained by privacy regulations, small sample sizes, and resource-intensive data preprocessing. We explored the feasibility of using the large language model (LLM) GPT-4o to generate synthetic neurosurgical data. Methods: A plain-language prompt instructed GPT-4o to generate synthetic data based on univariate and bivariate statistical properties of 12 perioperative parameters from a real-world open-access neurosurgical dataset (n = 139). The prompt was input over independent trials to generate 10 datasets matching the reference size (n = 139), followed by an additional dataset representing a ten-fold amplification (n = 1390). Fidelity was assessed using t-tests, two-sample proportion tests, Jensen-Shannon divergence, two-sample Kolmogorov-Smirnov, and Pearson’s product-moment correlation. Results: Generated data preserved distributional characteristics and relationships between desired parameters. In all generations, at least 11/12 (91.67%) parameters showed no statistically significant differences in means and proportions from real data, including the amplified dataset. Five of the synthetic datasets showed no significant differences in all 12 parameters. Conclusions: The findings demonstrate that a zero-shot prompting approach can generate synthetic neurosurgical data and amplify sample sizes with consistent high fidelity compared to real-world data. This underscores LLMs’ potential in addressing data availability challenges for neurosurgical research.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.026 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.006 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".