🩺MIS 752 — AI and Healthcare
UNLV Lee Business School · Advanced Healthcare Analytics & AI · Fall 2026 · Dr. Richard J. Young
Welcome
This course teaches advanced analytics and AI through healthcare — but the real product is a set of transferable data skills you can carry into any industry. Healthcare is the substrate: readmission ≈ churn, fraud-waste-abuse ≈ gaming fraud, ED arrivals ≈ casino demand. Every lab ships as a live, public demo — your portfolio, not your transcript.
LEARN · Weeks 1–7
Forge one callable tool per week on a shared dataset.
BUILD · Weeks 8–12
LLMs, evaluation, and agents that call your tools.
JUDGE · Weeks 13–15
Causal ROI, fairness, and governance.
Week 16
Final live demos.
Learning goals
- Join messy, multi-table data and engineer one-row-per-person analytic tables (SQL thinking in Python).
- De-identify data properly and measure re-identification risk (k-anonymity, linkage attacks).
- Build, explain, and evaluate predictive models — and know when they fail.
- Use LLMs and agents responsibly: evaluation harnesses, guardrails, and verification.
- Judge AI with causal ROI, fairness audits, and governance — and argue when AI should not be used.
- Ship a public portfolio of working demos, each with an honest account of its limits.
Module index
Lab demos are linked as they ship each week.
| Wk | Act | Topic | Lab demo |
|---|---|---|---|
| 1 | LEARN | Why healthcare AI fails: claims, prior auth & incentives | coming soon |
| 2 | LEARN | Privacy engineering: cohorts & de-identification | Lab 1 — Cohort Finder same data, other takes: De-ID Scanner · Dashboard |
| 3 | LEARN | Cost, SDOH & clinical dashboards | Lab 2 — Cost & SDOH Dashboard |
| 4 | LEARN | Alert fatigue & why AI fails in the field | coming soon |
| 5 | LEARN | Supervised learning: readmission risk + SHAP explainability | coming soon |
| 6 | LEARN | Pretrained vision models & the equity audit | coming soon |
| 7 | LEARN | Clinical operations & ED demand forecasting | coming soon |
| 8 | BUILD | Clinical NLP & the LLM reality check | coming soon |
| 9 | BUILD | LLM failure modes & building an eval harness | coming soon |
| 10 | BUILD | Agentic AI: a prior-auth agent with guardrails | coming soon |
| 11 | BUILD | Patient navigator & RAG over private records | coming soon |
| 12 | BUILD | Multi-agent verification | coming soon |
| 13 | JUDGE | Correlation is not causation: propensity-score ROI | coming soon |
| 14 | JUDGE | Integration week | coming soon |
| 15 | JUDGE | Algorithmic bias, Fairlearn audits & governance | coming soon |
| 16 | JUDGE | The systems that actually work — live demos | coming soon |
One dataset powers everything: richardyoung/synthea-575k-patients — 575k fully synthetic patients (MIT, $0).
⚕️ Data & ethics
All course demos use fully synthetic or properly de-identified data (Synthea — no real patient information exists anywhere in this course) and are built for education only. Nothing here may be used for clinical decision-making about real people. Never upload protected health information, student records, credentials, or API keys to any course Space.