Synthetic patient and interview transcript creator: an essential tool for LLMs in mental health
Bibliographic record
Abstract
Developing high-quality training data is essential for tailoring large language models (LLMs) to specialized applications like mental health. To address privacy and legal constraints associated with real patient data, we designed a synthetic patient and interview generation framework that can be tailored to regional patient demographics. This system employs two locally run instances of Llama 3.3:70B: one as the interviewer and the other as the patient. These models produce contextually rich interview transcripts, structured by a customizable question bank, with lexical diversity similar to normal human conversation. We calculate median Distinct-1 scores of 0.44 and 0.33 for the patient and interview assistant model outputs respectively compared to 0.50 ± 0.11 as the average for 10,000 episodes of a radio program dialog. Central to this approach is the patient generation process, which begins with a locally run Llama 3.3:70B model. Given the full question bank, the model generates a detailed profile template, combining predefined variables (e.g., demographic data or specific conditions) with LLM-generated content to fill in contextual details. This hybrid method ensures that each patient profile is both diverse and realistic, providing a strong foundation for generating dynamic interactions. Demographic distributions of generated patient profiles were not significantly different from real-world population data and exhibited expected variability. Additionally, for the patient profiles we assessed LLM metrics and found an average Distinct-1 score of 0.8 (max = 1) indicating diverse word usage. By integrating detailed patient generation with dynamic interviewing, the framework produces synthetic datasets that may aid the adoption and deployment of LLMs in mental health settings.
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 | 0.009 | 0.047 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.027 | 0.016 |
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".