Skip to content

slide_deck_outline.md

Latest commit

 

History

History
107 lines (66 loc) · 5.72 KB

slide_deck_outline.md

File metadata and controls

107 lines (66 loc) · 5.72 KB
Raw

Slide Deck Outline: Healthcare Budget and Wait Times

Format: 5 slides | Audience: Executive / policy | Timing: 12–15 minutes

Slide 1: The Problem

Title: "1 million Canadians waited in 2018. Where is the funding going wrong?"

Visual: Single large number — 21 weeks — centred on slide. Below it: "Average wait time after GP referral (Fraser Institute, 2018)." Simple Canada map with provincial wait time variation indicated by colour gradient.

Key points:

  • Wait times are measurable, consequential, and politically salient
  • Provincial variation is extreme — suggesting structural, not uniformly national, causes
  • The question is not whether to invest — it is whether increased investment is translating into outcomes

Speaker note: Open with the Inez Rudderham case (two-year undiagnosed cancer due to lack of GP access). This is not a statistics story — it is a system design story. "Before we recommend more spending, we need to know whether more spending is working."

Slide 2: The Data

Title: "What we measured: 10 provinces, 6 years, 10 procedures"

Visual: Clean data provenance diagram (left to right): CIHI Budget Data → SQLite → Python Pipeline → 3-Model Analysis → Decision Output Below: small table showing data dimensions (54 observations, 2013–2018, procedure list).

Key points:

  • CIHI: Canada's authoritative health data source (not-for-profit, government-affiliated, founded 1994)
  • Budget: total health expenditure per province, millions CAD
  • Wait time: median (50th percentile) averaged across 10 priority procedures
  • Stack: Python, SQLite, scikit-learn, XGBoost — fully reproducible

Speaker note: Address limitations proactively here, not defensively at the end. "n = 54 is small. We'll show you what that means for interpretation. Small N does not mean the findings are wrong — it means we are careful about what we claim."

Slide 3: The Finding

Title: "Budget matters — but explains only 1 in 5 days of wait time variation"

Visual: Two-panel chart (side by side):

  • Left panel: Scatter, all 54 observations pooled. Trendline slopes downward (r = −0.50). Title: "National view: more spending = shorter waits"
  • Right panel: Same data, coloured by province, with per-province trend lines. Lines are flat or slightly upward. Title: "Provincial view: higher-spending provinces often wait longer"

Key points:

  • Left panel: statistically significant (p < 0.001) — this is not noise
  • Right panel: the same data, disaggregated — the relationship reverses
  • This is a classic Simpson's Paradox. The aggregate trend obscures the structural story.
  • R² = 0.205: budget explains 20.5%. The other 79.5% is the more important signal.

Speaker note: Spend 60% of slide time on the right panel. "Most analyses stop at the left panel and conclude 'spend more.' This is why disaggregation matters. The right panel is what a system planner needs to see."

Slide 4: The Interpretation

Title: "The funding signal is real. Why isn't it translating?"

Visual: Causal diagram — two competing hypotheses, side by side:

Assumed model (left): More Budget → Increased Capacity → Shorter Waits

Observed pattern (right): Structural Constraints (aging, geography, staffing) → Long Waits → Political Pressure → Budget Increase → Structural Constraints Persist → Waits Unchanged

Key points:

  • Reverse causality: provinces with the worst wait times receive the most budget — but structural constraints absorb the investment without producing outcomes
  • Omitted variables: aging population, physician-to-patient ratio, urban/rural facility distribution are likely the dominant factors
  • This does not mean spending doesn't help — it means the signal is obscured by reactive funding into structurally constrained systems
  • XGBoost feature importance: budget_rank (structural position) and budget_per_capita outperform raw budget as predictors — confirms the structural hypothesis

Speaker note: "This analysis cannot prove reverse causality — that would require a controlled study. What it shows is that the pattern is consistent with reactive funding, and inconsistent with a simple 'more money = shorter waits' story."

Slide 5: What Should Be Done

Title: "Three decisions this analysis informs today"

Visual: Three-column action card layout (consulting-grade, minimal text)

Immediate 90 Days Next Fiscal
Redefine the KPI: add wait-days per $M spent to provincial performance scorecards Commission multi-variable analysis: add % population 65+, active physicians per 10K, urban/rural index Shift investment mix: prioritise structural capacity (staffing, facilities) in high-variance provinces (BC, NB, PEI) over aggregate budget increases

Below the table:

This analysis does not establish causality. All recommendations are directional hypotheses requiring quasi-experimental validation before implementation at scale.

Speaker note: End with a question for the room: "Which of these three is most valuable to validate with your operational data?" This opens a conversation rather than closing it — which is the right posture for correlational analysis being presented to decision-makers.

Appendix Slides (if needed)

A1: Full model comparison table (Baseline OLS / Ridge / Lasso / XGBoost — R², RMSE, CV R²)

A2: Feature importance table (XGBoost) with interpretation column

A3: Diminishing returns curve (PDP on budget_per_capita) with annotated threshold zone

A4: Provincial wait time trend lines 2013–2018 (highlighting BC, NB, PEI divergence)

A5: Data methodology detail (CIHI sourcing, cleaning steps, assumptions)