Apple Health Deep Analysis — Clinical-Grade Report Engine

This skill produces a comprehensive health assessment report at the standard of a top-tier academic medical center. It applies 20+ peer-reviewed statistical methods to Apple Health export data, contextualizes every finding against evidence-based clinical reference ranges, grades the confidence of each conclusion, and handles missing or insufficient data transparently.

Medical Disclaimer: This analysis is for informational and educational purposes only. It does not constitute medical advice, diagnosis, or treatment. Consult a qualified healthcare provider for medical decisions.

Workflow

Step 1: Locate the Export

Find the Apple Health XML file inside

apple_health_export/

导出.xml

export.xml

Export.xml

find . -path "*/apple_health_export/*.xml" -size +1M | head -5

Step 2: Run Both Analysis Engines

# Base analysis — comprehensive health metrics
python3 <skill-path>/scripts/analyze_health.py <xml-path> --output json 2>/dev/null > /tmp/health_base.json

# Advanced analysis — 20 peer-reviewed statistical methods
python3 <skill-path>/scripts/advanced_analytics.py <xml-path> 2>/dev/null > /tmp/health_advanced.json

Both scripts: streaming XML parsing (handles 1GB+), Python 3.6+ only, no external deps. If >500MB, warn about 2-3 min runtime.

Step 3: Generate HTML Report Shell

Generate the visual HTML report with charts, gauges, and data tables:

python3 <skill-path>/scripts/generate_report_html.py /tmp/health_base.json /tmp/health_advanced.json -o /tmp/health_report.html --lang zh

This produces a self-contained HTML (~160KB) with all visualizations but EMPTY narrative sections. The narrative sections are

<div class="ai-narrative" id="narrative-XXX">

Step 4: Assess Data Quality & Read Clinical Reference

Before interpreting ANY results, read the

data_quality

Data sufficiency rules:

• If a metric has

  reliability: "insufficient"

  → do not present that section. Mention it in the Data Availability section as "not enough data."
• If

  reliability: "low"

  → present with a prominent caveat: "Limited data (X days of Y); interpret with caution."
• If

  reliability: "moderate"

  → present normally with a brief note on coverage.
• If

  reliability: "high"

  → present with full confidence.

From the advanced analysis, check

data_requirements

Read

references/clinical_interpretation.md

Step 5: Fill Narrative Sections

Read both JSON outputs. For each section in the HTML, use the Edit tool to insert your clinical interpretation directly into the corresponding

<div class="ai-narrative">

The narrative IDs and what to write:

Each narrative should:

• Be written as HTML (use

  <p>

  ,

  <ul>

  ,

  <li>

  ,

  <strong>

  ,

  <h4>

  tags)
• Reference specific numbers from the data
• Cite clinical reference ranges from

  references/clinical_interpretation.md

• Include evidence grades where applicable
• Be in the user's language
• Be direct and clinically informative, not generic

Example Edit for body composition:

old_string: <div class="ai-narrative" id="narrative-body-composition"></div>
new_string: <div class="ai-narrative" id="narrative-body-composition"><h4>Clinical Assessment</h4><p>Your current weight of <strong>105.0 kg</strong> (BMI 30.5) places you in <strong>Obese Class I</strong>...</p></div>

Step 6: Open Final Report

After all narrative sections are filled:

open /tmp/health_report.html  # macOS

Key Principles

Never Simplify the Analysis

Present the full statistical results. Do not round away precision, omit p-values, or skip advanced metrics to "keep it simple." If a method was computed, present it. Users who export their Apple Health data and ask for analysis want depth, not dumbed-down summaries. Let the report structure (executive summary first, deep dives later) handle the complexity gradient — not omission.

Handle Missing Data Explicitly

Every health data export is different. Some users have CGM data; most don't. Some have years of Apple Watch data; others have weeks. The report must adapt gracefully:

• Present only what exists. Never fabricate or guess at missing data.
• State what's missing. In the Data Availability section, list which data types are present and which are absent, so the user knows what they're NOT seeing.
• Explain impact. If CGM data is absent, note that glucose analysis, MAGE/MODD/LBGI/HBGI, and glucose-exercise correlations cannot be performed. If VO2 Max is missing, note that fitness age cannot be estimated.
• Adjust composite scores. Only score dimensions that have data. If only 4 of 8 dimensions have data, the overall score is the mean of those 4 — don't penalize for missing data.

Evidence Grading

Every major finding should carry a confidence indicator:

Example: "Your resting heart rate has been rising over the past 3 months (Mann-Kendall tau=0.24, p=0.03; Evidence: B)"

Report Language — MUST Match User Language

The entire report — including the HTML visual report AND the conversational narrative — must be in the user's language. Detection priority:

1. The language of the user's message (highest priority)
2. The XML locale attribute (e.g.,

   zh-Hans_US

   → Chinese)
3. Data source names (Chinese app names → Chinese)

When generating the HTML report, pass the

--lang

python3 <skill-path>/scripts/generate_report_html.py base.json adv.json -o report.html --lang zh  # Chinese
python3 <skill-path>/scripts/generate_report_html.py base.json adv.json -o report.html --lang en  # English

For Chinese: use standard medical terminology (静息心率, 心率变异性, 血糖时间达标率, 最大摄氧量, 自主神经功能, 生物年龄, etc.). For the conversational analysis, write entirely in the detected language — do not mix languages unless quoting metric names that have no standard translation.

What the Scripts Output

Base Analysis (

analyze_health.py

)

Outputs JSON with keys:

personal_info

data_quality

data_overview

circadian_rhythm

heart_rate

heart_rate_variability

glucose

sleep

activity

body_composition

workouts

respiratory

correlations

lagged_correlations

change_points

composite_scores

trend_momentum

audio_exposure

risk_stratification

Advanced Analysis (

advanced_analytics.py

)

Outputs JSON with keys:

methods_applied

references

data_requirements

trend_tests

causal_inference

nonlinear_dynamics

advanced_glucose

circadian_quantification

statistical_rigor

bayesian_changepoints

biological_age_models

disease_risk_screening

The

disease_risk_screening

{condition, risk_level, score, max_score, risk_pct, factors, recommendation, references}

references/wearable_screening_literature.md

Each section is

null

Reference Files

references/clinical_interpretation.md

references/report_template.md

references/wearable_screening_literature.md