← Back

RAPID

Reading Assistance for PTSD Insight and Documentation: a clinical-NLP dashboard that turns long patient records into a scannable overview.

RAPID dashboard
Fig. RAPID dashboard. A: patient info. B: quick summary. C: recent activity. D: symptom timeline. E: flagged psychiatric history.

The problem

For PTSD, electronic health records hold symptoms, triggers, trauma history, treatment plans and progress notes. As those records grow, it gets harder for clinicians to see a patient's whole history in the time they have. Clinical text is also hard for standard NLP: it's unstructured, dense with jargon and acronyms, and spread across many sources. Even state-of-the-art biomedical named-entity recognition reaches only about 80 to 85% precision and recall, so any automated summary needs a way to surface and correct its own errors.

Approach

RAPID builds on the Doccurate clinical-text curation framework. It replaces Doccurate's extraction step with fine-tuned transformer language models for named-entity recognition of PTSD symptoms and organizes the results around the DSM-style PTSD symptom clusters.

NLP pipeline
Fig. NLP pipeline: annotated clinical text goes to a fine-tuned NER model, which extracts symptoms, dates and notes.
PTSD symptom clusters
Fig. PTSD symptom clusters: re-experiencing, avoidance and numbing, and hyperarousal.
Symptom intensity and frequency scale
Fig. Intensity, frequency and severity scale used to code symptoms.

Data

RAPID uses annotated clinical text plus de-identified Clinical Summary Reports from SAMHSA, focused on anxiety and obsessive-compulsive disorders and on trauma- and stressor-related disorders. Preprocessing removes contractions, punctuation and stop words, then lemmatizes and lowercases the text.

Interface

The dashboard is built around how clinicians read: a quick summary first, then a color-coded symptom timeline, then a filterable psychiatric-history list with flags, and the full report one click away.

Symptom timeline
Fig. Symptom timeline, color-coded by symptom.
Psychiatric history list
Fig. Psychiatric history with dates, notes and flags.
Medical history report view
Fig. Expandable medical history report.

Evaluation

I ran a comparative user study with six participants, five college-aged students and one physician. Each completed clinical reading tasks with both the original Doccurate interface and RAPID. The study measured usability and efficiency and gathered perceived strengths and limitations.

A surgeon helped code the clinical data, which grounded the annotations in real clinical reading practice.