TRANXACT.AI Aiuto

Notes taken during the conversation: Record visit and Record session

You talk with the patient or the student. TRANXACT turns the conversation into words as it is spoken, sorts them into a draft and adds a short summary. You check the draft against the transcript and change anything that is wrong. Nothing is written to the chart or the student record until you sign.

Cosa fa

You talk with the patient or the student. TRANXACT turns the conversation into words as it is spoken, sorts them into a draft and adds a short summary. You check the draft against the transcript and change anything that is wrong. Nothing is written to the chart or the student record until you sign.

The work is done with speech recognition and machine learning. Generative AI is not used. Every word in the draft is a word that was said, and every line points back to where it was said in the transcript. Each time someone signs, their corrections train the next version, so the drafts get better over time.

Dove trovarlo

  • EHR › Clinical › Clinical Documentation › Record visit: draft sections are Subjective, Objective, Assessment and Plan.
  • Student Lifecycle › Faculty & Advising › Academic Advising › Record session: draft sections are Concern raised, Discussion, Agreed actions and Referrals. You can also use Record session on a scheduled advising session's row.
  • EHR › Clinical › Clinical Quality › Visit notes accuracy and Student Lifecycle › Early Alerts & Analytics › Student Success & Retention › Session notes accuracy: how well each version of the model sorts sentences.

Step by step

  1. Who and consent. Choose the patient or student. You can also choose the encounter or advising session. Then choose the kind of conversation; ordinary; substance-use treatment under 42 CFR Part 2; confidential counseling; FERPA. Confirm that the person agreed to the recording, and say how they agreed. Recording cannot start without this.
  2. The conversation. Press Start recording and talk normally: The words appear as they are said. Each finished line is labelled with who said it, guessed from the voice. To set the speaker yourself, press the speaker's button while they talk; automatic goes back to the guess; You can type a line at any time, for example something said before the recording started; If no speech recogniser is set up, or the microphone is not available, a message says so. You can type the whole conversation instead.
  3. Verify and sign. Press Stop and make the draft: The draft opens beside the transcript. Hover over a line of the draft to highlight the part of the transcript it came from: You can move a sentence to another section, edit it, or leave it out with ×: If a word was misheard, correct it in the transcript and choose Redo the draft: Tick what should go into the record. For a visit this is problems (with ICD-10 codes), medications (with dose and frequency) and allergies (with reaction). For a session it is early alerts and referrals: Things that were said not to be present, such as "denies chest pain", stay in the note as pertinent negatives. They are never offered for the chart: If the draft missed something, use The draft missed something to add it and the words used for it. It will be recognised in every later conversation: Check the summary. Choose the signing provider (visit) or the advisor (session), then sign.

The visit is signed as a clinical note through the same Sign that a typed note uses. After signing, it can only be addended. The session is written to Academic Advising.

Privacy

  • Where the audio goes. The sound goes only to this installation's own speech recogniser. It does not go to the browser maker's speech service or to any AI vendor. The browser's built-in voice typing is deliberately not used for these screens.
  • Audio retention. By default the audio is not stored at all. If Platform › Configuration › Core Settings › Runtime Settings › Keep the audio of a recorded visit (or advising session) until it is signed is switched on, the audio is kept until signing or discarding, and then deleted.
  • Voices. The voiceprints used to tell the two voices apart are deleted when the recording is signed or discarded.
  • Who can open a recording. Before it is signed, a recording can be opened only by the person who made it. Every time a patient's recording is viewed, the view is written to the PHI access log.
  • Sessions that never teach the model. These recordings are never used for learning; Part 2 recordings; confidential counseling sessions; FERPA-only sessions.

A confidential counseling session written to Academic Advising can be seen by its author only.

  • Discarded recordings. A discarded recording is removed after 30 days. A recording left unfinished is discarded after a day.

How it gets better

Every signature is a lesson:

  • each sentence you kept or moved;
  • each word you corrected;
  • each concept you added.

Each night, and whenever someone chooses Train now on the accuracy panel, a new version is trained and tested on sentences it did not learn from. The new version is used only if it scores better than the version in use. Otherwise it is recorded, with the reason, and the current version stays in use.

Who may use it

Grant these on Platform › Access › Roles:

  • Record visit (conversation notes) in EHR. Clinicians, physicians and nurses have it by default.
  • Record session (advising conversation notes) in Student Lifecycle. Advisors and counselors have it by default.

Signing also needs the grant for each thing it writes:

  • Clinical Documentation add, for the note;
  • Problems, Medications and Allergies add, for each ticked item;
  • Academic Advising add or change, for the session;
  • Early Alerts add, for an alert.

For the administrator

The speech recogniser is a separate service that the installation runs itself (see core/deploy/ambient-asr). It is set up with AMBIENT_ASR_URL, and optionally AMBIENT_ASR_KEY. In the Helm chart, set ambientAsr.enabled. Without it, both screens still work, and the conversation is typed.

<!-- from help-ehrclinic.md -→