An interventional study of Conventional Discharge Documentation Workflow and LLM-Assisted Discharge Documentation Workflow in Health Communication and Patient Discharge, sponsored by University Hospital, Montpellier. Completed at 2 sites in France. Open to participants aged 0 Years to 18 Years. Per ClinicalTrials.gov, last updated 2026-10-15.
Sponsored by University Hospital, Montpellier · Not applicable, Interventional, and Health services research
This study compared two ways of preparing discharge documents for children and families leaving a pediatric emergency department. In the usual workflow, information from the electronic health record was assembled into a fixed document template. In the alternative workflow, ERIOS Assistant, an artificial intelligence system hosted within the hospital, drafted documents for health professionals and for families or children. A clinician reviewed and validated the documents before they were handed to the family in both workflows. According to the completed study report, assignment concerned document production after the decision to discharge and did not determine diagnosis, treatment or disposition. Families were contacted by telephone after discharge to assess how their account of the care and instructions agreed with the medical record, as well as their perceived understanding, satisfaction and use of the documents. The protocol also planned assessment of document quality, readability, clinician experience, usability and human supervision. This study is being registered retrospectively after the randomized component was completed.
The study was conducted in the pediatric emergency department of Montpellier University Hospital, France. The protocol described a randomized parallel comparison of standard documentation and documentation assisted by ERIOS Assistant, with quantitative assessments and complementary qualitative observations and interviews. Its operational enrollment target was 120 children, 60 per main workflow. The completed study report records 153 randomized participants between September 4 and November 27, 2025.
In the conventional workflow, selected electronic health record fields were transferred by a deterministic office-automation module into a fixed word-processing template. In the ERIOS Assistant workflow, a structured questionnaire populated from the clinical record was provided to an on-premises conversational assistant using the open-weight GPT-OSS 120B model, August 2025 version. According to the completed study report, no patient data left the hospital infrastructure during operation of the evaluated assistant. This statement concerns the intervention system and not any subsequent processing used in preparing the report.
Participants were assigned to the two main workflows in a 1:1 ratio using an advance randomization list generated in R. Within the assistant workflow, a second randomization used a 1:1:1 ratio in blocks of nine to assign a physician liaison letter plus a family-oriented document, a physician liaison letter plus a child-oriented document, or a single integrated document for clinicians and families. Children younger than eight years or unable to read who were assigned to the child-oriented strategy were re-randomized between the other two assistant strategies. The primary comparison pooled the three assistant strategies against conventional documentation.
A clinician reviewed, could edit, validated and signed the discharge documentation before delivery in both workflows. The clinician triggering production also validated it, without a systematically interposed second reader. The completed study report states that workflow allocation occurred after the clinical discharge decision and did not determine diagnosis, treatment or disposition. The supplied protocol described eligibility in terms of anticipated discharge home; the timing of screening, inclusion, randomization and the final discharge decision is to be documented in the study records.
Family understanding was assessed by comparing the family account with the source clinical record. The initial protocol scheduled the telephone interview for day 1, or day 2 after a non-working day; the completed study report describes actual follow-up between 24 hours and seven days. Its participant-level understanding measure averages six equally weighted binary concordance domains. These domains correspond to questions in the protocol interview schedule. The detailed scoring rules and their date of adoption should be distinguished from the broader outcome specified in the protocol.
Clinical-document safety assessment was planned as expert identification of residual errors, omissions and hallucinations, initially classified as factual, medical or serious severity. The completed study report uses a revised distinction between information domain and potential clinical consequence, and restricts fidelity assessment to physician-facing documents. Separate family-facing documents and the family-directed paragraph of integrated documents were not included in that fidelity assessment. Readability, perceived understanding, satisfaction and perceived usefulness were additional planned domains. Other protocol components included production and supervision times, clinician ratings, usability and acceptability questionnaires, and qualitative observations and interviews; the completed manuscript does not report all of these components.
The completed analysis is based on an evaluable population rather than a strict intention-to-treat population. The study report describes post-randomization exclusions and replacement within the same allocation slot, exclusion of two participants assigned to the assistant workflow who received conventional documents, and one further exclusion for an unscorable primary outcome. The reported analysis population is 59 conventional and 58 assistant-assisted episodes. Missing data were not imputed. The primary unadjusted comparison used a Mann-Whitney test, whereas the supplied protocol specified generalized linear models and intention-to-treat and per-protocol analyses.
Retrospective registration and outcome history: the protocol introduction and its dedicated outcome section identify effective family understanding as the primary outcome, while an overview paragraph retains overall satisfaction as the primary outcome. The completed study report identifies understanding as primary and satisfaction as secondary, and reports that this prioritization was decided before recruitment. That claim requires the dated decision record to be retained with the study documentation. The record is being registered after completion and does not constitute prospective registration.
The paired-document-state framework, detailed transformation and discrepancy taxonomy, stage-of-origin attribution, phenotype analyses and class-specific proportions of discrepancies resolved before document delivery were developed after data collection. They are exploratory post hoc analyses, not retrospectively designated prespecified outcomes. The investigators report that the study was initially handled under the French framework for research not involving the human person and CNIL reference methodology MR-004, and was not entered in a trial registry before enrollment. This account of delayed registration is subject to confirmation by the institution.
278 studies on the registry are indexed under Patient Satisfaction; 130 are open to participants now.
This study's enrollment of 153 is above the median of 93 across 212 interventional studies indexed under Patient Satisfaction.
Browse Patient Satisfaction studies →University Hospital, Montpellier is the lead sponsor of 1,261 studies on the registry; 227 are open to participants now.
Counted across the registry records on this site, refreshed daily.
Inclusion Criteria:
Inclusion Criteria in the Supplied Protocol:
Exclusion Criteria:
Additional Operational Restrictions Reported in the Completed Study:
The completed study report states that randomization occurred after the decision to discharge home. The protocol used anticipated home discharge in its eligibility criteria. These differences are disclosed in this retrospective registration rather than silently combined into an amended initial protocol.
Participants received the physician liaison letter produced through the routine fixed-template workflow and reviewed, edited when necessary, validated and signed by the discharging clinician.
Other: Conventional Discharge Documentation Workflow
Assistant Erios generated a physician liaison letter and a separate child-oriented discharge document from structured clinical information derived from the electronic health record. The generated documents were reviewed and could be edited by the clinician before validation, signature, and delivery.
Other: LLM-Assisted Discharge Documentation Workflow
Assistant Erios generated a single integrated discharge document intended for both clinicians and families from structured clinical information derived from the electronic health record. The generated document was reviewed and could be edited by the clinician before validation, signature, and delivery.
Other: LLM-Assisted Discharge Documentation Workflow
Assistant Erios generated a physician liaison letter and a separate child-oriented discharge document from structured clinical information derived from the electronic health record. The generated documents were reviewed and could be edited by the clinician before validation, signature, and delivery.
Other: LLM-Assisted Discharge Documentation Workflow
Routine deterministic discharge-document production workflow used in the pediatric emergency department. An office-automation module transferred selected electronic health record fields into a predefined document template. The resulting document was reviewed, edited if necessary, validated, signed, and delivered by the clinician.
Discharge documentation was generated by Assistant Erios, a hospital-developed on-premises conversational assistant based on the open-weight GPT-OSS 120B language model. The assistant received structured clinical information derived from the electronic health record and generated discharge documentation according to the assigned document strategy. All generated documents were presented to a clinician for review, editing if necessary, validation, signature, and delivery. The intervention affected only document production and did not determine or modify diagnosis, treatment, or disposition.
Record-Concordant Family Understanding After Discharge
Family responses collected through a semi-structured telephone interview are compared with the clinical record across six domains: diagnosis, care/investigations performed, investigation results, treatments, follow-up plan, and conditions for returning to the emergency department. Each domain is coded as concordant or discordant, and the overall score is the mean of the six equally weighted domains. Scores range from 0 to 1, with higher values indicating greater concordance. All responding families are assessed, regardless of whether they reported reading the discharge document. The measure reflects understanding of the entire discharge process and is not a pre-post assessment of learning from the document alone.
Time frame: 24 hours to 7 days after discharge (actual follow-up; originally planned for day 1, or day 2 after a non-working day).
Family-Reported Perceived Understanding
The legal representative rates their understanding of the child's medical situation after the emergency department visit using the five-category scale in the study questionnaire: 1, not understood at all; 2, understood a little; 3, understood fairly well; 4, understood well; and 5, understood very well. Higher scores indicate greater perceived understanding. This is a post-discharge rating, not a change from a baseline score.
Time frame: 24 hours to 7 days after discharge (actual follow-up; protocol schedule: day 1 or day 2 after a non-working day).
Family Satisfaction With Information and Explanations Received
The legal representative rates overall satisfaction with the information and explanations provided during the child's emergency department care, irrespective of whether information was oral or written. The study questionnaire uses five categories: 1, not at all satisfied; 2, slightly satisfied; 3, neither satisfied nor dissatisfied; 4, rather satisfied; and 5, very satisfied. Higher scores indicate greater satisfaction. This item does not rate overall hospital service quality or waiting times.
Time frame: 24 hours to 7 days after discharge (actual follow-up; protocol schedule: day 1 or day 2 after a non-working day).
Family-Reported Usefulness of the Main Discharge Document
The legal representative is asked whether the main discharge document is useful for them and other members of the household. The interviewer codes the response as useful or not useful, while retaining optional free-text comments. The outcome is the distribution of these two response categories among respondents with an interpretable answer; missing or unclassifiable responses are not automatically coded as not useful. This item is not a five-point usefulness scale.
Time frame: 24 hours to 7 days after discharge (actual follow-up; protocol schedule: day 1 or day 2 after a non-working day).
Lexical Readability of Discharge Documents: Standard Frequency Index
The completed study report operationalizes lexical readability as the Standard Frequency Index of open-class words represented in the French Manulex lexical database. Higher values indicate more frequent vocabulary. Document-level values are summarized separately for family-facing documents and for physician liaison letters. The protocol planned readability assessment but did not name the Standard Frequency Index or Manulex; the date of selection of this specific metric is not established by the supplied protocol. This metric is not a reader comprehension score and is not represented as a bounded 0-100 clinical scale.
Time frame: Index emergency department visit (day 0), using discharge documents collected for retrospective assessment.
Number of Residual Errors in Clinician-Validated Discharge Documents
The initial protocol planned medical-expert identification and counting of residual errors after clinician validation, with results distinguished by document type. The protocol planned separate factual, medical and serious severity categories: factual errors without an effect on management; medical errors affecting follow-up without a real patient risk; and serious errors potentially creating a health risk. The completed report revised this classification to distinguish information domain from potential clinical consequence and restricted fidelity assessment to physician-facing documents. Protocol categories and revised categories must not be treated as identical.
Time frame: Index emergency department visit (day 0), using discharge documents collected for retrospective assessment.
Number of Residual Omissions in Clinician-Validated Discharge Documents
The initial protocol planned medical-expert identification and counting of residual omissions after clinician validation, with results distinguished by document type. The protocol planned separate factual, medical and serious severity categories: factual errors without an effect on management; medical errors affecting follow-up without a real patient risk; and serious errors potentially creating a health risk. The completed report revised this classification to distinguish information domain from potential clinical consequence and restricted fidelity assessment to physician-facing documents. Protocol categories and revised categories must not be treated as identical.
Time frame: Index emergency department visit (day 0), using discharge documents collected for retrospective assessment.
Number of Residual Hallucinations in Clinician-Validated Discharge Documents
The initial protocol planned medical-expert identification and counting of residual hallucinations or fabricated content after clinician validation, with results distinguished by document type. The protocol planned separate factual, medical and serious severity categories: factual errors without an effect on management; medical errors affecting follow-up without a real patient risk; and serious errors potentially creating a health risk. The completed report revised this classification to distinguish information domain from potential clinical consequence and restricted fidelity assessment to physician-facing documents. Protocol categories and revised categories must not be treated as identical.
Time frame: Index emergency department visit (day 0), using discharge documents collected for retrospective assessment.
Number of Delivered Physician-Facing Documents With a Discrepancy Judged to Pose a Serious Health Risk
The completed study report counts physician-facing discharge documents with at least one discrepancy judged liable to create a serious health risk in retrospective medical assessment. Each document contributes once to the affected-document count. This is an assessment of potential consequence in documentation, not a count of clinical serious adverse events or observed patient harm. Separate family-facing documents and family-directed paragraphs of integrated documents are outside the assessed fidelity perimeter. The initial protocol included a serious severity category, but the final document-level operationalization and revised classification must be distinguished from that original plan.
Time frame: Index emergency department visit (day 0), using discharge documents collected for retrospective assessment.
Documents are hosted by the registry — open the source record to download them.
Plan to share: Yes — De-identified individual participant data underlying the reported results may be made available to qualified researchers upon reasonable request, subject to review and approval by the governance committee of Montpellier University Hospital and to applicable data-protection requirements. Data cannot be made publicly available because the regulatory framework under which the study was conducted and the information provided to participants did not permit unrestricted secondary sharing. Where access is approved, it will be restricted to the authorised research purpose and governed by an appropriate data use agreement.
Supporting information: Study protocol, Statistical analysis plan, Analytic code
Time frame: Beginning after publication of the primary study report and for the duration of data retention under Montpellier University Hospital institutional policy.
Access criteria: Access may be granted to qualified researchers for a scientifically justified research purpose, subject to review and approval by the governance committee of Montpellier University Hospital. Only derived, de-identified analytic datasets necessary for the approved purpose may be made available. Access will require an appropriate data use agreement and compliance with applicable data-protection and institutional governance requirements. Direct identifiers, unrestricted electronic health record data, and unrestricted public access will not be provided.
Ran from Sep 2025 to primary completion in Nov 2025, 2 months.
Of 137 non-industry interventional studies with no phase under Patient Satisfaction, half reached primary completion within 12 months of starting (the median), counting those still running; a quarter within 6 and three quarters within 23. This study's 2 months is below that.
Counting every sponsor, half of 139 studies reached it within 12 months; a quarter within 6 and three quarters within 23.
How these figures are worked out, and which studies are left out of them.
From the registry record's own update history. This site started tracking changes on Sep 25, 2026; for anything earlier, see the record history on ClinicalTrials.gov ↗
This study is completed, as verified in Oct 2026. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.
University Hospital, Montpellier →