Adds a substantial layer of data-backed statistics (all grounded in verified,
clean source data — no fabricated metrics).
Backend (new routers/statistics.py, prefix /api/stats; +106 pytest still green):
- /inpatient-clinical: LOS dist + by-disease quartiles, cost dist + by-disease +
cost-vs-LOS, outcome counts, admission-route counts, BMI-by-age, KPIs
(5822 admissions, median LOS 4d, mean ¥6294, cure 99.1%, emergency 47%)
- /symptoms: 主诉 keyword frequencies (发热/咳嗽/肺炎…) + revisit ratio (36%)
- /incidence-rate: per-10k-population standardized rate by district (cases ÷ pop)
- /env-correlation: pollutant×cases Pearson + 7×7 pairwise matrix + PM2.5 scatter
- /temporal: weekday distribution (+ month/yoy returned but UI omits them — data
is December-only, so seasonality/YoY would be misleading)
Frontend:
- NEW 住院临床分析 page (/analysis/clinical, nav 临床分析): 9 charts + KPI row —
LOS histogram + box-by-disease, cost histogram + scatter + by-disease, outcome
donut (severity-colored), admission-route donut, age-band BMI box
- DiseaseAnalysis: 主诉症状词频 horizontal bar + revisit ratio
- DistrictComparison: 标化发病率（每万人）with 病例数↔发病率 toggle (rate is
epidemiologically correct; raw counts mislead by population)
- EnvironmentalHealth: pollutant-cases correlation bar + 7×7 correlation heatmap +
PM2.5×cases scatter with least-squares regression line
- TrendAnalysis: 星期就诊分布 + honest "data is December-only" note
- statsApi client + types
Gates: tsc 0 · build ok · functional e2e 43/43 (incl 2 new clinical) · verified
live against real backend data via dev proxy
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- Add vitest config and unit tests for components, api, stores
- Add Playwright e2e test for user flows
- Add backend test files
- Update risk.py with LOD grid KDTree optimization
Add 3 new data-driven insight cards (daily cases, district risk
comparison, weather impact) with real parquet data. Fix season
card to use current date instead of data date. Expand to 11 cards.
Add POST /api/chat endpoint proxying to ai.2890.ltd with JWT auth.
Create ChatBot frontend component with collapsible chat panel,
message bubbles, and auto-scroll. Chat API key stored in .env only.
Clean up duplicate typing imports in insights.py, export cachedPost.
Add /api/insights/cards endpoint with proper card format matching
frontend expectations. Fixes "Cannot read properties of undefined
(reading 'map')" crash. Switch frontend to use new endpoint.
Replace alert marker rectangles with circleMarkers so they don't
look like a second grid. Default showAlertMarkers to false.
Add || [] guards on data.features.map() and alerts.map().
Reduce RiskMap grid count 3000→1500, debounce 150ms→300ms.
P0: Fix KeyError in 3 analysis endpoints. geojson.py stores 1d risk
as "risk_value" but analysis.py accessed "risk_1d" — always crashed.
Backend: Add lru_cache to GeoJSON/CSV/Parquet loaders, date helpers,
and district loader. Add try/except and FileNotFoundError guards.
Frontend: Debounce riskRange, merge counts into useMemo, stabilize
handleGridClick with ref, memoize nearest-grid scan, wrap AlertMap
in React.memo, switch useLodGrid from fetch to cachedGet.
Context: Build a spatial risk assessment system correlating air quality
data with children's respiratory disease incidence across Wuhan.
Approach: FastAPI backend serving PostGIS spatial queries, React
frontend with Deck.gl maps, and a PyTorch SpatialTemporalGCN pipeline
for multi-day (1d/3d/7d) risk prediction.
Changes:
- backend/ — FastAPI API with auth (JWT), alerts, risk analysis,
geocoded case data, grid statistics, and report endpoints
- frontend/ — React dashboard with interactive risk maps, alert
monitoring, district comparison charts, and timeline player
- models/ — SpatialTemporalGCN model with trained weights and ONNX
export for inference
- scripts/ — ETL pipeline for weather + medical data, grid generation,
feature engineering, training, and daily inference
- deploy/ — Docker Compose configs for backend, frontend, and MLflow
- docs/ — API docs, deployment guide, user guide, and code review
Impact: Enables spatial risk visualization, alert monitoring, and
ML-driven health risk forecasting for environmental health teams.
