HighCostDrugsDemo/IMPLEMENTATION_PLAN.md

# Implementation Plan - Pathway Data Architecture

## Project Overview

Pre-compute patient treatment pathways from Snowflake and store in SQLite for fast Reflex filtering. This replaces the current simplified `prepare_chart_data()` with full pathway hierarchy support.

**Architecture**: Snowflake → Pathway Processing → SQLite (pre-computed) → Reflex (filter & view)

**Key Benefits**:
- Performance: Pathway calculation done once during data refresh, not on every filter
- Simplicity: Reflex filters pre-computed data with simple SQL WHERE clauses
- Full Pathways: Sequential treatment pathways (drug_0 → drug_1 → drug_2...) with statistics

**Design Reference**: See `PATHWAY_DATA_ARCHITECTURE_PLAN.md` for detailed architecture, schema, and data flow.

**Source Code**:
- Existing analysis: `analysis/pathway_analyzer.py`
- Existing visualization: `visualization/plotly_generator.py`
- Existing Reflex app: `pathways_app/app_v2.py`

## Quality Checks

Run after each task:

```bash
# Syntax check for Python files
python -m py_compile <file.py>

# Import verification
python -c "from <module> import <class>"

# For Reflex changes
cd pathways_app && timeout 60 python -m reflex run 2>&1 | head -30
```

## Phase 1: Schema & Data Pipeline Foundation

### 1.1 Extend Database Schema
- [x] Add `pathway_date_filters` table with 6 pre-defined combinations:
  - `all_6mo`, `all_12mo`, `1yr_6mo`, `1yr_12mo`, `2yr_6mo`, `2yr_12mo`
- [x] Add `pathway_nodes` table with:
  - Hierarchy structure (parents, ids, labels, level)
  - Patient counts and costs (value, cost, costpp, cost_pp_pa)
  - Date ranges (first_seen, last_seen, first_seen_parent, last_seen_parent)
  - Treatment statistics (average_spacing, average_administered, avg_days)
  - Denormalized filter columns (trust_name, directory, drug_sequence)
  - Foreign key to date_filter_id
- [x] Add `pathway_refresh_log` table for tracking refresh status
- [x] Create indexes for efficient filtering
- [x] Verify schema with: `python -c "from data_processing.schema import *"`

### 1.2 Create Pathway Pipeline Module
- [ ] Create `data_processing/pathway_pipeline.py` with:
  - `fetch_and_transform_data()` - Snowflake fetch + UPID/drug/directory transformations
  - `process_pathway_for_date_filter(df, date_filter_config)` - Single filter processing
  - `extract_denormalized_fields(ice_df)` - Extract trust, directory, drug_sequence from ids
  - `convert_to_records(ice_df, date_filter_id)` - Convert ice_df to list of dicts for SQLite
- [ ] Integrate with existing `analysis/pathway_analyzer.py` functions
- [ ] Verify: `python -c "from data_processing.pathway_pipeline import *"`

### 1.3 Create Migration Script
- [ ] Create script to set up new tables in existing `data/pathways.db`
- [ ] Pre-populate `pathway_date_filters` with 6 combinations
- [ ] Verify migration runs cleanly on fresh database

## Phase 2: CLI Refresh Command

### 2.1 Create Refresh Command
- [ ] Create `cli/refresh_pathways.py` with:
  - DATE_FILTER_CONFIGS constant (6 combinations)
  - `compute_date_ranges(config, max_date)` - Calculate actual dates from config
  - `refresh_pathways(minimum_patients, provider_codes, ...)` main function
- [ ] Implement refresh flow:
  1. Fetch ALL data from Snowflake (full date range)
  2. Apply transformations (UPID, drug names, directory)
  3. Clear existing pathway_nodes
  4. For each of 6 date filter configs: filter → process → insert
  5. Update pathway_refresh_log
- [ ] Add CLI argument parsing (--minimum-patients, --provider-codes, etc.)
- [ ] Verify: `python -m cli.refresh_pathways --help`

### 2.2 Test Refresh Pipeline
- [ ] Run refresh with Snowflake data
- [ ] Verify all 6 date_filter_ids populated in pathway_nodes
- [ ] Verify pathway structure matches original `generate_icicle_chart()` output
- [ ] Verify patient counts are correct (compare with original app)
- [ ] Document estimated processing time (expect 6-12 minutes for 440K records)

## Phase 3: Reflex Integration

### 3.1 Update AppState
- [ ] Replace date picker state with dropdown state:
  - `selected_initiated: str = "all"` ("all", "1yr", "2yr")
  - `selected_last_seen: str = "6mo"` ("6mo", "12mo")
- [ ] Add `date_filter_id` computed property: `f"{selected_initiated}_{selected_last_seen}"`
- [ ] Rewrite `load_pathway_data()` to query `pathway_nodes` table:
  - Base filter: `WHERE date_filter_id = ?`
  - Trust/directory/drug filters on denormalized columns
- [ ] Add `recalculate_parent_totals()` for filtered hierarchies
- [ ] Update KPI calculations from root node data

### 3.2 Update Icicle Figure
- [ ] Update `icicle_figure` computed property to use all pathway_nodes columns
- [ ] Match original 10-field customdata structure:
  - values, colours, costs, costpp
  - first_seen, last_seen, first_seen_parent, last_seen_parent
  - average_spacing, cost_pp_pa
- [ ] Restore full hover/text templates from `visualization/plotly_generator.py`
- [ ] Verify chart renders correctly with treatment statistics

### 3.3 Update UI Components
- [ ] Replace date pickers with select dropdowns:
  - Initiated: "All years", "Last 2 years", "Last 1 year"
  - Last Seen: "Last 6 months", "Last 12 months"
- [ ] Add "Data refreshed: X ago" indicator from pathway_refresh_log
- [ ] Update filter section layout
- [ ] Verify UI compiles and renders correctly

## Phase 4: Testing & Validation

### 4.1 End-to-End Validation
- [ ] **Pathway hierarchy matches original**: Compare specific pathway ids structure
- [ ] **Patient counts match**: Compare root patient count for same date range
- [ ] **Treatment statistics display correctly**: Verify "Average treatment duration" hover data
- [ ] **Drug filtering works**: Filter to FARICIMAB, verify correct pathways shown
- [ ] **Chart renders with all tooltip data**: Verify 10-field customdata structure

### 4.2 Performance Testing
- [ ] Measure filter change response time (target: <500ms)
- [ ] Measure initial page load (target: <2s including data load)
- [ ] Verify chart interaction (zoom, hover) is smooth with no lag
- [ ] Test with full dataset

### 4.3 Documentation
- [ ] Update CLAUDE.md with new architecture
- [ ] Document CLI usage for `refresh_pathways`
- [ ] Update README with new run instructions
- [ ] Document any breaking changes from original app

## Completion Criteria

All tasks marked `[x]` AND:
- [ ] App compiles without errors (`reflex run` succeeds)
- [ ] All 6 date filter combinations work correctly
- [ ] Drug/directory/trust filters work with instant updates
- [ ] KPIs display correct numbers matching filter state
- [ ] Icicle chart renders with full pathway data and statistics
- [ ] Treatment duration and dosing information displays in tooltips
- [ ] No console errors during normal operation
- [ ] Verified with real patient data from Snowflake

## Reference

### Date Filter Combinations

| ID | Initiated | Last Seen | Default |
|----|-----------|-----------|---------|
| `all_6mo` | All years | Last 6 months | Yes |
| `all_12mo` | All years | Last 12 months | No |
| `1yr_6mo` | Last 1 year | Last 6 months | No |
| `1yr_12mo` | Last 1 year | Last 12 months | No |
| `2yr_6mo` | Last 2 years | Last 6 months | No |
| `2yr_12mo` | Last 2 years | Last 12 months | No |

### Key Files

| File | Purpose |
|------|---------|
| `data_processing/schema.py` | Database schema definitions |
| `data_processing/pathway_pipeline.py` | New pathway processing pipeline |
| `cli/refresh_pathways.py` | CLI refresh command |
| `analysis/pathway_analyzer.py` | Existing pathway analysis logic |
| `visualization/plotly_generator.py` | Existing chart generation |
| `pathways_app/app_v2.py` | Reflex application |