644. CUSTOMIZING A LARGE LANGUAGE MODEL TO PROVIDE CLINICALLY TAILORED ADVICE FOR EMERGENCY ESOPHAGEAL IMPACTIONS
Bibliographic record
Abstract
Abstract Background There is interest in applying Large Language Models (LLMsclinically, but they are not trained for medical purposes. Equally, the development of LLMsrequires significant time and financial resources. Customizing an existing LLM for clinical purposes may overcome these limitations and would be of interest to clinicians, patients, hospital managers, and government/policy makers. Methods Thirty patient cases with nonbony esophageal impactions were developed using guidance from UpToDate. Standardized prompts were engineered for physicians as the end user. The customization tool was applied to ChatGPT-4.0 to create the Esophageal Impaction Tool (GPT-EIT), which provides advice for managing patients with esophageal impactions. Both GPT-EIT and the generic ChatGPT-4.0 models were queried February 28th, 2024. Their performance was evaluated based on guidelines for the removal of foreign bodies from the European Society of Gastrointestinal Endoscopy. Outcome data was presented using descriptive statistics in the form of counts and percentages. Results The GPT-EIT provided accurate recommendations for 30/30 (100.0%) cases. Each case was managed appropriately, recommending that stable patients received endoscopy within 24 hours, while patients with drooling were recognized as a surgical emergency and advice was given to provide endoscopy within 2 hours, but no more than 6 hours. GPT-EIT also advised to consider the appropriateness of treatment for patients presenting at 97 years of age based on their comorbidities and health status. The generic ChatGPT-4.0 model produced accurate guidance for 14/30 (46.7%) of cases. Patients with signs of significant obstruction would not have received timely endoscopy (Table 1). Conclusions ChatGPT can be customized to overcome limitations with its training dataset to develop clinically relevant, accurate advice for esophageal impactions. Clinicians, patients, hospital managers, and policymakers should take note that this methodology may be applied for other clinical purposes to avoid limitations related to resource constraints with developing LLMsfrom scratch.
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 | 0.003 | 0.014 |
| 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 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.012 | 0.005 |
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".