Device & equipment dashboard

Preview — under active development. This module is still being refined. Patterns, mappings, and category boundaries should be validated against your own corpus before being relied on for analysis or publication.

device_dashboard.py is the visualization layer for the device-extraction outputs. It reads the two CSVs produced by device_extraction.py and emits a single, self-contained HTML report (no external dependencies, no network calls at view time, no server required) that you can open in any browser.

Live demo

Live demo — pre-loaded against a synthetic ALS cohort Open in new tab ↗

Inputs

File	Produced by	Description
device_codes.csv	device_extraction.py	Structured matches against DEVICE_CODES (HCPCS / CPT-4 / SNOMED-CT) including ALS-care-indicator procedures.
device_extractions.csv	device_extraction.py	Regex matches against notes[].narrative_text and documents[].plain_text for PEG, BiPAP, tracheostomy, cough assist, power wheelchair, AAC/SGD brand names, etc.

Output

A single HTML file — typical name device_dashboard.html — with five tabs:

1. Overview by category — bar chart of patient counts per device category (e.g. als_care_indicator_respiratory, peg, niv, wheelchair_power, aac_device) with record counts and a marker for whether matches came from structured codes, free-text regex, or both.

2. All devices / indicators — filterable per-label table. Type in the filter box to narrow by name or category.

3. Patient × category matrix — heatmap-style grid showing which PT-NNNN-pseudonymized patients have records in which categories. Cell shading scales with record count: light blue (1–4), medium (5–19), deep purple (20+).

4. Source-record snippets — up to three representative text excerpts per regex pattern, sampled across unique patients. For chart-review preparation and source verification — not for analysis.

5. About — privacy controls and the SMART-on-FHIR-doesn't-capture-DME caveat (see below).

Privacy controls (baked in, cannot be disabled)

• Patient identifiers replaced with PT-NNNN pseudonyms, stable within one run.
• Calendar dates reduced to year only.
• Snippet text truncated to 200 characters.
• Devices, patterns, and categories with fewer than k=2 unique exposed patients are suppressed entirely (k-anonymity at the device level).
• Resource UUIDs are never emitted.

Usage

In a notebook:

import device_dashboard

device_dashboard.main(
    codes_csv_path       = './device_codes.csv',
    extractions_csv_path = './device_extractions.csv',
    out_path             = './device_dashboard.html',
    cohort_name          = 'Your cohort name',
    k                    = 2,           # k-anonymity threshold
)

From the command line:

python -c "import device_dashboard; device_dashboard.main()"

(defaults read from the working directory).

Important caveat about ARC data

SMART on FHIR pulls from the clinical EHR rarely capture DME-supplier records, so direct device-procurement codes (E0470 BiPAP, K0813 power wheelchair, B4034 gastrostomy supplies) are typically absent from structured procedure records. The structured side of this dashboard surfaces mostly ALS-care-indicator procedures — spirometry, sleep studies, EMG/NCS, speech screening, botulinum toxin, PT/OT re-evaluations, and mobility/self-care status reporting — that signal where the patient is on the typical ALS-care pathway. Direct device presence is typically captured by the regex side from clinical notes.

This is documented in the dashboard's About tab so users don't over-interpret the absence of structured DME codes as absence of devices.

Visual design

The dashboard uses a navy → purple gradient header that distinguishes it from the exposure dashboard (teal → navy → purple gradient). Both follow the same privacy-safe template so you can view either of them in the same context without confusion.