For health-system informatics, data science, integration, privacy and governance teams who must pilot, validate, integrate, monitor and contract for clinical AI, one concrete job at a time.
Level
Intermediate
Study time
16 hours
Lessons
24 in 6 modules
Exam
20 questions
About this course
Cedar Junction Health, a fictional four-hospital Ohio system, must recommend a bounded clinical AI pilot, source audit, silent validation, monitoring response, privacy/contract conditions and integration test plan. Build a six-section evidence portfolio from reproducible synthetic cases, six module Decision Labs and a strategic capstone. The 16-hour core uses supplied-output interpretation; optional coding/source extensions add time explicitly listed in the workload plan.
US-focused FDA, HIPAA/Part 2, ONC and CMS teaching is checked November 5, 2026; final rules and future effective dates are distinct. Educational content is not legal or clinical advice. No live patient data, measured pilot outcome, vendor endorsement, clinical authorization or professional credential is supplied. The completion certificate reflects lessons, listed labs and an automated 20-question assessment with 80% pass requirement; it is not licensure, accreditation or demonstrated workplace competence. The portfolio/AI feedback is practice only and can be wrong.
What you will be able to do
Tier a clinical AI intake request by patient impact, autonomy, time-criticality, FDA status and protected-attribute inputs, and select a vendor with pre-signed gates and weights, security attestations, total cost of ownership and an ROI case built on local measures
Design an ambient-documentation pilot with a stepped-wedge arm, a fixed exposure window, pre-set go/no-go thresholds and a sampled note-versus-audio error audit, and write the go/no-go memo
Audit an AI-generated discharge summary or portal reply with an inaccuracy, hallucination and omission taxonomy, and test red-flag routing rules for sensitivity on a held-out message set
Run a silent-mode validation of a vendor deterioration model with physiologic onset and clinician recognition times, and select an alert threshold from sensitivity, PPV, horizon-based calibration and lead time within a unit alert budget
Reproduce a label-choice bias pattern in a risk score, compare enrollment at equal need across groups, and write the 45 CFR 92.210 inventory and mitigation entry
Investigate drift with complete outcome denominators and competing explanations; compare recalibration, manufacturer-led change and retirement, and distinguish operational proxy exposure from actual waiting/capacity evidence
Evaluate a vendor PHI flow, de-identification release method, Part 2 provenance/disclosure packet and contract conditions without treating detector accuracy as legal approval
Specify a CDS Hooks prefetch template and card that meet FDA criterion 4 transparency, and confirm hooks, prefetch, feedback and silent scoring against what your EHR version actually supports
Syllabus
6 modules · 24 lessons · about 16 hours
1
Module 1: The AI Governance Lead: Intake, Device Questions, Risk Tiering and Vendor Selection
Taking a new clinical AI request from informal ask to a tiered, owned and evidence-backed decision, and choosing among vendors on local evidence.4 lessons
Is It a Device? Applying FDA's January 29, 2026 CDS Criteria to a Scribe, a Sepsis Alarm and a Chart Summary
What the EHR Vendor Must Show You: HTI-1 Source Attributes, Feedback Loops and the HTI-5 Final-Rule Transition
Tiering Six Synthetic Intake Requests with the Joint Commission–CHAI RUAIH Guidance and the CHAI Applied Model Card
Selecting a Vendor: A Scored Bake-Off, Security Due Diligence, Total Cost of Ownership and an ROI Case Built on Local Measures
Decision Lab · 12 minThe Friday demo and the unsigned gateCFO wants a Monday demo; W refuses no-training and the model host is outside the supplied report scope. X/Y TCO exceeds the project allocation. No patient-facing activation is authorized.
2
Module 2: The CMIO's Team: Piloting Ambient Scribes, Discharge Summaries and Portal Replies
Running generative documentation and messaging tools as measured pilots with error audits and disclosure rules.4 lessons
Designing a Stepped-Wedge Ambient Scribe Pilot: What the Two 2025 NEJM AI Randomized Trials Teach
Auditing Ambient Notes Against Audio and Writing the Go/No-Go Memo
Auditing an AI Discharge Summary for Hallucinations and Omissions on a Synthetic Admission
Configuring AI-Drafted Portal Replies: Red-Flag Routing, California AB 3030 and AB 489, and Texas HB 149
Decision Lab · 12 minThe missing sentence in the signed noteThe 16-week source-audit illustration has 57% adoption and a descriptive 8% median; the adjusted primary effect/uncertainty against 10% is pending. Two source reviewers are available, but support calls compete for their time.
3
Module 3: The Clinical Data Scientist: Silent-Mode Sepsis Validation and Risk-Score Bias Audits
Validating a vendor predictive model on local data and auditing a care-management score under Section 1557.4 lessons
Setting Up a Silent-Mode Run: Outcome Labels, Onset Times and Lessons from the Epic Sepsis Model Study
Choosing a Threshold Under a Nursing Alert Budget: AUROC, Calibration, PPV and Lead Time on 2,000 Synthetic Encounters
Subgroup Checks and the 45 CFR 92.210 Inventory Entry for a Decision Support Tool
Reproducing the Cost-as-Proxy Bias Pattern and Re-Labeling a Care-Management Risk Score
Decision Lab · 12 minA feasible threshold is not a deployment decisionImani limits alert budget to 0.6/shift. C has 180 events, D has 221 independent events;0.30 fits the printed screen. Existing tables do not establish benefit or subgroup acceptance.
4
Module 4: The MLOps and Operations Analytics Engineer: Drift Monitoring, Retirement and Operational Models
Keeping deployed predictive and generative tools honest after go-live, knowing when to stop using them, and holding operational models to a baseline and an equity check.4 lessons
Building a Monitoring Plan: Input, Output and Outcome-Linked Metrics for Predictive and Generative Tools
Reading an 18-Month Control Chart: Attributing Shifts to an Assay Change and a Seasonal Surge
Recalibrate, Request a PCCP Update or Retire: Decision Rules for Hospital-Built and FDA-Cleared Models
Operational AI: Testing a No-Show Overbooking Model Against a Naive Baseline and Checking Proxy Exposure
Decision Lab · 12 minThe quiet chart with missing outcomesThe alert chart falls while a new lactate code appears. True positives and PPV alone cannot identify incidence or sensitivity. A clinical continuity workflow is already running.
5
Module 5: The Privacy Officer and Contracting Lead: PHI Flows, De-identification and AI Contracts
Approving exactly what data reaches an AI vendor and locking the safeguards into the agreement.4 lessons
Approving a PHI Flow to an LLM Vendor: BAA Terms, Minimum Necessary, Hosting Region and Log Retention
Measuring De-identification Recall on 25 Synthetic Notes: Safe Harbor, Expert Determination and Limited Data Sets
Substance Use Disorder Records in AI Pipelines: 42 CFR Part 2 After the February 16, 2026 Compliance Date
Redlining a Clinical AI Agreement: No Training on PHI, Update Notice, Model-Card Delivery and Subgroup Reporting
Decision Lab · 12 minOne consent checkbox is not the outbound packetThe summary feed mixes independent non-Part 2 provider documentation, copiedOTPmaterial and unknown origin. A BA log subprocessor has uncertain retention; its contract permits secondary training.
6
Module 6: The Integration Engineer: Wiring an AI CDS Service into the EHR
Connecting a third-party AI service through SMART on FHIR and CDS Hooks with minimum data and visible reasoning, then confirming what your EHR version actually supports.4 lessons
SMART on FHIR or CDS Hooks 2.0.1? Choosing the Launch Pattern and Working with EHR Developer Programs
Writing a Minimum-Necessary Prefetch Template Against a Synthetic FHIR R4 Fixture
Designing the Card: Rationale and Source Links for FDA Criterion 4, Override Capture and Failure Behavior
Confirming What Your EHR Supports: Epic and Oracle Health Integration Paths, Silent Scoring Inside the EHR and CMS-0057-F Readiness
Decision Lab · 12 minA412 response nobody seesA card fixture passes its partial checker, but real site/version/auth tests are pending. Required patient identity and optional lipids have different contracts.
Capstone simulation · 30 minThe governance board wants one honest recommendationTwelve-week internal review is tomorrow; $120k is uncommitted. The six evidence strands contain intentionally missing real clinical/legal/site approvals. Funding, clinical benefit and actual EHR support remain conditional.
Exam and certificate
The exam unlocks once you have completed every lesson and played each Decision Lab to an ending. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by Microsoft, Epic, Oracle Health, Abridge, Nabla, Bayesian Health, LeanTaaS, Qventus, or Johns Hopkins ACG System. Product and company names are trademarks of their respective owners.
A free introduction built on eight healthcare jobs: auditing an ambient-scribe note, answering a portal message from an AI draft, acting on a sepsis alert, reading a triage-flagged CT worklist, reviewing a prior-authorization request, checking AI-suggested codes, and triaging drug and device safety reports. Each lesson shows where AI helps, where it must not decide, and the rule that governs that exact task.
Use a synthetic US care-team case to check AI drafts against source records, preserve nursing judgment and coordinate accountable follow-up. Build a practical evidence packet without patient data or a model account.