case-report

name	case-report
description	Convert clinical case data (medical records, IO sheets, lab data, imaging) into a professional, timeline-based PowerPoint presentation (PPTX) and interactive HTML website for M&M conference, case report, or morning conference presentation.
allowed-tools	Bash, Read, Write, Glob, Grep

Case Report Presentation

Overview

When a user provides clinical case data and asks for a case report / M&M / morning conference presentation, use this skill to generate a timeline-based python-pptx presentation and an interactive HTML website. The input is typically a folder containing:

• Markdown medical records (醫師記錄/): Admission note, Progress notes, Discharge note, Surgical records, Objective findings
• IO sheets (IO/): PNG screenshots of fluid intake/output records by shift
• Lab data images (LABS/): PNG screenshots of lab trend charts (CBC, SMAC, CRP, etc.)
• Chest X-rays / imaging (CXR/): PNG images with date-based filenames
• Nursing records (護理紀錄/): PDF files
• Medication records: PDF files at root level

The slide structure should be timeline-based — organized chronologically around the clinical course, with adaptive granularity (finer detail for the main focus period, coarser for secondary periods).

Prerequisites

pip3 install python-pptx pymupdf

Workflow

0a. Ask for Presenter Information & Presentation Type

Before starting, ask the user two things in a concise one-liner:

Presenter info: Who is presenting and who is the supervisor? (e.g., "R2 王大明 / VS 李教授") — press Enter to skip.

Presentation type: (1) M&M Conference (2) Case Report (3) Morning Conference — default: M&M

• M&M Conference → ending section uses "Learning Points / Take-home Messages"
• Case Report → ending section uses "Discussion" (literature comparison & analysis)
• Morning Conference → ending section uses "Summary & Key Points"
• If the user skips → check memory for saved user profile; default to M&M Conference

0b. Identify Input & Create Output Folder

Detect input type

The user provides a folder path containing clinical data organized in subdirectories:

import os, glob

user_input = "..."  # folder path provided by user

# Expected structure
records_dir = os.path.join(user_input, "醫師記錄")
io_dir = os.path.join(user_input, "IO")
labs_dir = os.path.join(user_input, "LABS")
cxr_dir = os.path.join(user_input, "CXR")
nursing_dir = os.path.join(user_input, "護理紀錄")

# Find all files
md_files = sorted(glob.glob(os.path.join(records_dir, "*.md"))) if os.path.isdir(records_dir) else []
io_images = sorted(glob.glob(os.path.join(io_dir, "*.png"))) if os.path.isdir(io_dir) else []
lab_images = sorted(glob.glob(os.path.join(labs_dir, "*.png"))) if os.path.isdir(labs_dir) else []
cxr_images = sorted(glob.glob(os.path.join(cxr_dir, "*.png"))) if os.path.isdir(cxr_dir) else []
nursing_pdfs = sorted(glob.glob(os.path.join(nursing_dir, "*.pdf"))) if os.path.isdir(nursing_dir) else []
root_pdfs = sorted(glob.glob(os.path.join(user_input, "*.pdf")))

Create output folder

{PatientName}_case_report/
├── figures/                ← CXR images, lab images copied here
├── images/                 ← IO sheet images if embedded
├── presentation.pptx       ← PPTX presentation
├── presentation.html       ← HTML website (external images)
└── presentation_portable.html  ← HTML website (self-contained base64)

1. Read & Parse All Clinical Records

Markdown records (primary data source)

Read each .md file using the Read tool and extract structured data:

File	Extract
Admission note.md	Demographics, chief complaint, present illness narrative, past history, allergies, initial vitals, initial labs, physical exam, assessment/impression, initial plan
Progress note.md	Date-stamped entries (split by YYYY-MM-DD HH:MM:SS pattern); per entry: vitals, assessment changes, plan changes, clinical events
手術紀錄.md	Procedure details, TBSA breakdown by region (for burns), operative findings, EBL
Objective finding.md	Serial weights, serial lab data by date (structured), imaging findings — THIS IS THE PRIMARY LAB DATA SOURCE
Discharge note.md	Final diagnoses, course summary, discharge condition, radiology reports, culture results

IO sheets (PNG images)

IO sheets are PNG screenshots from the hospital's electronic medical system. Process them:

1. Visually inspect each IO image using the Read tool to understand the format
2. Extract numerical data by reading the values directly from the image:
   • Per shift: 白班 (07:00-14:59), 小夜 (15:00-22:59), 大夜 (23:00-06:59)
   • Categories: 輸液 (IV fluids), 血品 (blood products), 進食 (enteral), 排尿 (urine output), 總輸入/總排出/差值
3. Structure into tables for the fluid resuscitation slides (Pattern G)
4. If OCR is needed for hard-to-read values, use the macOS Vision OCR script from the notebooklm-to-editable-pptx skill

Lab data strategy

• Primary source: Objective finding.md — already has parsed, structured lab values by date
• Lab images (PNG): Use as visual verification and can be embedded as supplementary reference
• DO NOT recreate charts — use the structured data to build formatted tables

Medication records (PDF — 急診用藥紀錄, 住院後用藥簽用記錄)

Extract using PyMuPDF (fitz.open()). Key data to parse:

• IV fluids: LR, NS, D5W — with execution times (not order times). Multiple execution records under the same order number may represent either separate bags or multi-nurse sign-offs — cross-reference with nursing records to disambiguate.
• Blood products: FFP, FP, RBC — doses and execution times
• Antibiotics: Drug name, dose, frequency, start/stop dates, route
• Supportive medications: NaHCO₃, Albumin, diuretics, vasopressors, sedatives

CRITICAL: Use execution timestamps (執行時間), not order timestamps (開立時間), for fluid volume calculations. The same order number with multiple execution records may be the same bag signed by different nurses — verify with nursing records and IO sheets.

CXR and imaging

• Copy all CXR images to figures/ directory with date-based filenames
• These will be used in serial comparison slides (Pattern I)
• Label by actual imaging location — check PACS header (institution name) in the image. Do NOT assume the first CXR is from an outside hospital.

Wound / clinical photos

• Check for existing PPTX files (e.g., trauma team presentations) that may contain clinical photos
• Extract images via shape.image.blob from python-pptx
• Resize large images (>3MB) using sips -Z 2000 -s format jpeg -s formatOptions 80 before embedding in PPTX

2. Build Clinical Timeline

Construct a chronological timeline from all parsed data. The timeline granularity should be adaptive:

Timeline structure (acute/critical care default):
├── Pre-hospital: mechanism, rescue, initial hospital management
├── ED / Arrival (Hour 0): vital signs, primary survey, initial labs
├── Hour 0–8: resuscitation phase 1 (MAIN FOCUS — finest granularity)
├── Hour 8–24: resuscitation phase 2 (MAIN FOCUS)
├── Day 2–3: early ICU course
├── Day 4–7: continued course
└── Day 7+: later course / outcome

For each time point, record:

• Events: procedures, interventions, clinical changes
• Vitals: BP, HR, RR, Temp, SpO2
• Labs: key values with abnormal flags
• Fluid balance: intake by category, output (UOP), cumulative balance
• Medications: new starts, dose changes

3. Design Slides Based on Clinical Timeline

3.5. Text Formatting Rules (CRITICAL)

Same rules as journal-reading skill — apply to ALL python-pptx text:

Never use p.text = "..."

Always use set_run() or p.add_run(). The p.text = pattern does not guarantee formatting.

Never use \n in text strings

Each line must be a separate paragraph with its own run and explicit formatting.

Every text element must have explicit formatting

Every run must set: font.size, font.color.rgb, font.name. Never rely on defaults.

Every paragraph must have explicit alignment (CRITICAL)

Always set p.alignment = PP_ALIGN.LEFT (or CENTER/RIGHT) on every paragraph. Never rely on PowerPoint defaults — they are inconsistent across text boxes, tables, and shapes. Without explicit alignment, text may render LEFT in some boxes and CENTER in others.

# CORRECT — explicit alignment on every paragraph
p = tf.paragraphs[0]
p.alignment = PP_ALIGN.LEFT  # ← MUST SET
set_run(p, "text", font_size, color)

# WRONG — missing alignment → inconsistent rendering
p = tf.paragraphs[0]
set_run(p, "text", font_size, color)  # ← alignment undefined

Alignment conventions:

• Bullets, captions, section numbers, highlight/key-point box text: PP_ALIGN.LEFT
• Vital signs cards, table headers, table data (non-first-column): PP_ALIGN.CENTER
• Slide numbers: PP_ALIGN.RIGHT
• Title/ending slide (Thank You): PP_ALIGN.CENTER
• Content slide titles: PP_ALIGN.LEFT

Prevent text overlap between elements (CRITICAL)

When a slide has multiple vertically stacked elements (bullets + key_point box, timeline + bullets, etc.), calculate the available vertical space before adding content. If content exceeds the space, either:

1. Split into two slides — preferred for dense content
2. Reduce font size — minimum Pt(12) for readability
3. Move the emphasis box lower — but never below top=Inches(6.2) (slide numbers at 7.05)

Rule of thumb for vertical spacing:

• Mini-timeline occupies y=0.92 to y=1.75 (~0.83 inches)
• Content area after timeline: top=Inches(1.85) to top=Inches(5.5) max
• Key_point/highlight box: calculate top = bullets_top + (n_lines × line_height)
• Never stack more than 12–14 bullet lines + 1 emphasis box on a single slide

4. Slide Structure

Do NOT force a fixed slide count. Use as many slides as needed. Prefer more slides with less content each over fewer dense slides.

Layout density principles:

• Max 4–5 bullet points per slide
• Max 1 table + 1–2 figures per slide
• Font sizes for projection: body text ≥ Pt(18), titles ≥ Pt(28), table cells ≥ Pt(14), captions ≥ Pt(12)
• Leave breathing room — generous padding, whitespace between elements

Required slides (always present):

1. Title slide — Case type label (M&M / Case Report / Morning Conference), chief complaint as title, patient demographics summary, presenter info
2. Outline slide — Numbered TOC matching subsequent sections. Use uniform 0.5"×0.5" rounded-square badges (not circles) so 2-digit numbers fit. Badge font: 16pt for 1-digit, 13pt for 2-digit, bold white, vertically + horizontally centered, word_wrap=False, zero margins. Row step 0.6", label font 18pt vertically aligned with badge center. (Implemented in add_outline_slide() in generate_aesthetic_pptx.py.)
3. Patient Profile — Demographics, comorbidities, baseline functional status, social history
4. Mechanism / Presentation — How the injury/illness occurred, pre-hospital care, include outside hospital fluids/medications with execution times
5. Initial Assessment — Arrival vitals (Pattern K cards), primary/secondary survey
6. Injury/Disease Assessment — e.g., TBSA by region (Pattern J) for burns, staging for cancer
7. Wound / Injury Photos — Clinical photos from ER and procedures (Pattern I). Source from existing PPTX (add_picture extraction) or image folder. Always include if available.
8. Initial Labs — Key lab results in table (Pattern H) with abnormal highlighting
9. Initial Imaging — CXR/CT with interpretation. Label by actual source (e.g., "ER Arrival" if taken at receiving hospital, not "Outside Hospital")
10. Treatment Focus slides — Detailed slides on the main focus area (e.g., fluid resuscitation for burns). Use PBD (Post-Burn Day) or post-event timeline, not calendar dates, for time reference
11. Procedure slides — Operative details, clinical photos
12. Daily Course slides (PBD-based) — Each slide MUST include: Weight (with Δ), Labs, IO balance, Rx (daily management), Abx (antibiotics), Vitals from nursing records. Add a mini-timeline indicator (Pattern L) on each slide
13. Lab Trends — Serial lab tables (Pattern H) grouped by system (Renal, Hematology, Inflammatory)
14. Fluid/UOP Trend — PBD-based summary table: In / UOP / mL·kg⁻¹·hr⁻¹ / Balance / Cr / Weight
15. Serial Imaging — Side-by-side comparison (Pattern I)

Case-specific slide templates:

For burns / fluid resuscitation:

• Weight: Use pre-burn weight from op note (not admission weight which includes pre-hospital fluids). Flag the discrepancy explicitly.
• TBSA: Report both ED assessment and post-op reassessment if different.
• Parkland formula: Calculate from burn time (not hospital admission). Account for late presentation — if first 8h window passed, note explicitly. Show mL/kg/%TBSA actual vs expected (4 mL).
• PBD-based fluid slides: Break down by Phase (outside hospital → ER → BU shifts). Show fluid type detail (LR, NS, NaHCO₃, Albumin, blood products) with execution times from 急診用藥紀錄/住院後用藥簽用記錄.
• UOP tracking: Calculate mL/kg/hr using pre-burn weight per PBD phase. Red-highlight when below target.
• Outside hospital fluids: Include in PBD1 total. Note Foley bladder drainage vs sustained UOP.
• Wound/escharotomy photos: Extract from existing PPTX if available (e.g., trauma team PPTX). Use slide.shapes → shape.image.blob to export images.
• Weight trend: Daily weights with Δ from baseline and % change.

For surgical complications:

• Pre-operative assessment slide
• Intra-operative events timeline
• Post-operative complication timeline

For infection / sepsis:

• Antibiotic timeline per PBD — show ER empiric → BU definitive → changes
• Culture results with sensitivity data
• Inflammatory marker trends (CRP, PCT, WBC)

Slide Layout Patterns (16:9, coordinates in Inches)

Reuse patterns A–E from journal-reading skill, plus these new patterns:

Pattern	Layout	Coordinates	When to use
A: Figure + Analysis	Image left, bullets right	Image: (0.6, 1.4, w=6.0), Bullets: (7.0, 1.4, w=5.7)	Single figure with interpretation
B: Figure + Analysis (rev)	Bullets left, image right	Bullets: (0.6, 1.4, w=5.7), Image: (6.8, 1.4, w=6.0)	Alternating visual flow
C: Table + Key Takeaway	Table top, highlight box bottom	Table: (0.6, 1.4, w=12.0, h=3.5), Box: (0.6, 5.2)	Results table with summary
D: Side-by-Side	Two images side by side	Img1: (0.6, 1.4, w=5.8), Img2: (6.8, 1.4, w=5.8)	Before/after, comparison
E: Pure Content	Full-width bullets	Bullets: (0.8, 1.4, w=11.7, h=5.5)	Text-only slides
F: Timeline	Horizontal bar + event cards	Bar: (0.8–12.5, y=1.8), Cards below	Key events overview
G: Fluid Balance	I/O table + totals highlight	Table: (0.6, 1.4, w=12.0), Box: (0.6, 5.5)	Resuscitation monitoring
H: Lab Trend	Serial values table	Table: (0.6, 1.4, w=12.0)	Lab progression
I: Image Comparison	2–4 dated images side by side	Dynamic widths, date labels below	Serial CXR, wound photos
J: Body Region	Assessment table by region	Table: (0.6, 1.4, w=12.0)	TBSA, staging
K: Vital Signs	Card boxes with key vitals	Cards: 4–6 across, (0.6, 1.6)	Admission vitals
L: Mini-Timeline	Horizontal dots at top of slide, inside a container box	Container: L=0.5", W=12.33", H=0.7" (must contain edge labels); track: x=1.1–11.65; ovals 0.2"; labels 0.9" wide, 10pt centered	Daily course slides — use add_pbd_timeline_header(slide, days, current_idx); current day highlighted, past blue, future gray. Container width is mandatory so first/last day labels never overflow.

Emphasis Box Policy

Same as journal-reading:

• add_key_point() — deep blue box, white text — for KEY FINDING, most important takeaway
• add_highlight_box() — warm yellow box — for supporting context, targets, criteria
• Max 1 emphasis box per slide
• Default font: 14pt, left-aligned (both helpers). Override font_size= only when a slide truly needs larger emphasis. Never use centered alignment for multi-line summary text.

5. Generate the PPTX

Write a self-contained Python script to /tmp/create_case_report.py that inlines all helper functions from both:

• scripts/generate_aesthetic_pptx.py (base helpers)
• scripts/generate_case_report_pptx.py (case-report-specific helpers)

Do NOT import from the helper file paths — copy all function definitions directly.

Script structure:

1. Inline all helpers at the top (base + case-report-specific)
2. Build slides using layout patterns by name in comments
3. Call add_slide_numbers(prs) as the final step before prs.save()
4. Save to {output_dir}/presentation.pptx

Execute the script:

python3 /tmp/create_case_report.py

6. Generate HTML Website

After the PPTX is generated, create an interactive HTML presentation website following the presentation-website skill patterns:

• Use the same content and slide structure as the PPTX
• Follow the HTML template conventions from presentation-website/templates/presentation_template.html
• Include all mandatory features:
  • Dot navigation (right sidebar)
  • Progress bar (top)
  • Keyboard navigation (arrows, space, page up/down, presentation clickers)
  • scroll-snap-type: y mandatory
  • Card-based layout with left accent borders
  • Inline editing mode
  • Double-click source text popup — every content card must have hidden <div class="source-text"> with original record excerpt
  • Presentation mode toggle (P or F5)
  • PDF download via html2canvas + jsPDF (screenshot approach)
  • PPTX download via html2canvas + PptxGenJS
  • Slide counters on every slide ("X / N")
  • Fade-in animations

Generate two versions:

• presentation.html — external images referenced from figures/
• presentation_portable.html — self-contained with base64-encoded images

7. Deliver

Notify the user:

• Output folder path
• Number of PPTX slides and section breakdown
• Number of figures/images embedded
• That both PPTX and HTML are available
• HTML features summary (editing, download, navigation)

Content Guidelines

• Default language: English — use standard medical terminology
• If the user requests Chinese or bilingual, switch accordingly
• De-identify patient data in the presentation — use initials or generic identifiers, not full names
• Preserve exact lab values, vital signs, and clinical data — accuracy is critical
• Use red/bold for abnormal or critical values
• Use green for values returning to normal
• Significant findings should use add_key_point() boxes
• Target values (e.g., UOP target) should use add_highlight_box() boxes

When to Use

This skill applies when a user:

• Provides clinical case data (medical records, IO sheets, lab data, imaging)
• Requests a "case report" / "M&M" / "case presentation" / "mortality and morbidity"
• Mentions "case discussion" / "morning conference case" / "晨會 case"
• Wants a timeline-based presentation of a clinical course
• Provides a folder with 醫師記錄/, IO/, LABS/, CXR/ subdirectories