Using Artificial Intelligence to Personalize Caring Contact Messages for Recently Discharged Patients: Protocol for a Mixed-Methods Feasibility Study
Bibliographic record
Abstract
Abstract Suicide risk is substantially elevated following discharge from a psychiatric hospitalization. Caring Contact (CC) messages are brief messages of hope, support and information sent post-discharge that can improve mental health outcomes, including suicidal ideation and behaviours. Patients who received CCs in previous studies indicated a desire for increased personalization. To support personalization in an efficient and scalable manner, we will pilot the training of a Large Language Model (LLM) that constructs tailored CCs using information extracted from patients’ electronic health records (EHRs) created during the index psychiatric hospitalization. This is a three-phase, mixed-methods study. In phase 1, psychiatric inpatient clinical staff members will be recruited to generate personalized CC messages from representative, de-identified samples of EHRs. This will be considered the control CCs. Clinical staff focus groups will then identify key components of a successful CC message. In phase 2, the control CCs and focus group feedback will be used to train an LLM on a large number of EHRs to produce personalized CC messages. The LLM will be given the same EHRs used in phase 1 to produce CC messages. In phase 3, we will recruit 30 patients with lived experiences (PWLE) of psychiatric hospitalization to evaluate and compare the control CCs with the LLM-generated messages to determine the acceptability of AI-generated CC messages. We hypothesize that LLM-generated CC messages will achieve acceptability ratings at least as positive as the control CC messages. Since CC messages are modifiable and can be altered to suit the needs of various clinical settings, the findings of this study can potentially be broadly generalizable.
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 | 0.053 | 0.047 |
| Meta-epidemiology (narrow) | 0.003 | 0.004 |
| Meta-epidemiology (broad) | 0.004 | 0.004 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.005 | 0.003 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.005 | 0.007 |
| Insufficient payload (model declined to judge) | 0.051 | 0.009 |
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".