Skip to main content

DeepPhe Visualizer — quick guide

Build a cohort, review patient results, then explore an individual patient.

The Cohort Explorer

1. Build a cohort​

  1. Find a filter card and select a value.
  2. Add values in the same card as alternatives (OR).
  3. Add selections from different cards to narrow the cohort (AND).
  • The toolbar shows All N patients, then N of total patients selected.
  • A value that can no longer add a matching patient is disabled; a value you already selected stays removable.
  • Select a value again to remove it. Reset filters clears all selections.
  • After criteria are active, a value may read included / total — how many of that value's patients remain in the cohort.

A selected filter value

2. Low-count patient dots​

When a value represents 20 or fewer patients, each patient is a dot. Hover a dot to preview the patient; click a dot on a card to open that patient as a drawer tab (this never changes the cohort). Bars behind dots adds a faint bar for scale.

3. Review patient results​

Open the Selected Patients drawer. It loads 10 patients per page (40 when maximized).

  • Search and sort apply to the loaded page only.
  • Click a row to expand it; only one row is expanded at a time.
  • Indicators: negated, historic, uncertain, conflicted, and source.

The Selected Patients drawer

4. Explore a patient​

Open a patient from a dot or from Show in Document Viewer in an expanded row.

  • Patient Summary items that resolve to notes are links: a single source opens directly; a multi-source item opens a confidence-ranked picker.
  • On the timeline, click a document to open it; the open note has a larger ringed marker, and fact-linked notes have dashed outlines.
  • In the Document Viewer, filter highlights with Concept List, Group Filter (CHECK ALL / UNCHECK ALL), and Confidence Filter (By Mention / By Concept).

5. Export​

CSV export includes the currently loaded page only, after its search and sort, with visible columns.

Important limitations​

  • Available filters depend on the loaded dataset and services.
  • Extracted concepts and structured findings may require validation against source notes.
  • An absent value is not a confirmed negative clinical fact.
  • Exported files may contain patient-identifying information; handle them accordingly.