MSE433-Module4

Prototype workflow for AF ablation surgical variability: transcript-style process segments, mock multi-clinic cases, dashboards, and statistical analyses on real timing data.

---

Repository layout

Path	Purpose
apps/	Runnable applications (Streamlit UI).
data/	Inputs only — representative procedure transcript; place course Excel here (see below).
notebooks/	Jupyter analyses (same_doctor_same_abl_analysis.ipynb, explaining_variation.ipynb).
outputs/	Generated CSV, JSON, and HTML (mock variability tables, ANOVA outputs, dashboard build).
scripts/	Python modules and command-line pipelines.

---

Setup

From the repository root (MSE433-Module4-1/):

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Data you must provide

• data/MSE433_M4_Data.xlsx — cath-lab timing spreadsheet used by the doctor / #ABL analyses and the “explaining variation” notebook.
  scripts/data_utils.py loads this path by default. Add your own copy under data/ (it is not committed here).

• data/Procedure_Transcript.txt — representative narrative transcript (included in the repo).

---

Run the analysis scripts

All commands assume the repository root as the current directory.

Mock variability dataset + static HTML dashboard

Regenerates mock tables, variability_schema.json, dashboard JSON payload, and standalone HTML:

python scripts/build_variability_dashboard.py

open outputs/variability_dashboard.html

Open outputs/variability_dashboard.html in a browser.

Mock data only (no dashboard)

python scripts/variability_data.py

Transcript phase summaries (from data/Procedure_Transcript.txt)

python scripts/transcript_macro_analysis.py

Writes outputs/transcript_phase_summary.csv and outputs/transcript_micro_step_coding.csv.

Real data: same-doctor, same–#ABL style stats

Requires data/MSE433_M4_Data.xlsx.

Baseline ANOVA / within-cell summaries:

python scripts/doctor_abl_variation_analysis.py

Extended models (additional ANCOVA tables, e.g. extended and pre-map controls):

python scripts/doctor_abl_variation_analysis_anova.py

Outputs include outputs/phase_effect_anova.csv, outputs/case_time_ancova.csv, outputs/doctor_abl_cell_summary.csv, outputs/within_cell_phase_variation.csv, and related CSVs documented in each script’s main printout.

Regenerate the large same-doctor notebook (optional)

Requires nbformat:

python scripts/build_same_doctor_same_abl_notebook.py

Overwrites notebooks/same_doctor_same_abl_analysis.ipynb.

---

Run the Jupyter notebooks

source .venv/bin/activate
jupyter lab

Open notebooks from notebooks/. The first cells resolve the repo root whether Jupyter’s working directory is the repo root or notebooks/.

notebooks/same_doctor_same_abl_analysis.ipynb

Data: data/MSE433_M4_Data.xlsx plus functions from scripts/ (e.g. data_utils, doctor_abl_variation_analysis, transcript_macro_analysis).

Kind of work	What this notebook does
Framing	States the hypothesis, assumptions (representative transcript vs timed cases), and how the transcript was coded.
EDA / cohort definition (Step 1)	Loads cath-lab timing data, defines the analysis sample, and summarizes the comparison set.
ANOVA / ANCOVA (Step 2)	Fits models to see which factors matter for phase-level outcomes overall (omnibus-style inference).
Core solution design (Step 3)	Holds physician and #ABL fixed and quantifies residual (within-cell) variation across procedure phases.
Derived metrics / EDA (Step 4)	Examines hidden-time proxies (e.g. non-energy ablation time, post-ablation LA dwell) as direct outcomes of interest.
Transcript integration (Step 5)	Merges transcript phase summaries and micro-step coding with measured variability; compares transcript complexity to residual variation by phase.
Synthesis	Interprets category mix, complexity vs. variability, and a short final conclusion tying timing patterns to the process narrative.

notebooks/explaining_variation.ipynb

Data: data/MSE433_M4_Data.xlsx. The notebook re-fits ANCOVA-style models via scripts/doctor_abl_variation_analysis_anova.py helpers (same logic as the CSVs written to outputs/). Narrative in the notebook points to files such as outputs/case_time_ancova.csv and outputs/within_cell_phase_variation.csv for consistency with the batch scripts.

Kind of work	What this notebook does
Setup	Repo path resolution and imports.
EDA (Section 1)	Describes what the dataset contains (columns, structure) and includes a macro-phase visualization (schematic timeline view).
ANCOVA / inference (Section 2)	Case-time ANCOVA: what explains overall case time (Type II ANOVA, effect summaries). Includes extended predictor variants (e.g. extra terms) as noted in the notebook.
Residual / solution logic (Section 3)	Same doctor, same #ABL: characterizes residual variation after fixing those factors (tables and focused comparisons, e.g. ablation window vs pulse-on time).
EDA / comparison (Section 4)	Case time spread by physician — distributional comparison across doctors.

---

What the mock layer represents

The mock pipeline encodes a transcript-to-process layer (stages, microsteps, times, actors, delay drivers) and a cross-case variability layer (doctors, clinics, benchmarks). Generated artifacts include:

• variability_cases.csv, variability_stage_summary.csv, variability_transcript_annotations.csv
• variability_process_benchmarks.csv, variability_step_rankings.csv, variability_schema.json

This is synthetic, not production transcription; the structure matches how a real multi-clinic implementation could store annotations and drive benchmarking.

---

Assumption

The solution assumes every procedure can be segmented into timed transcript annotations joined with patient, provider, and clinic context. The mock generator illustrates that design; clinical validation is separate from this repository layout.