Part of VibeRounds — Socratic learning & Guided Discovery · Clinical Cognition Operating System Sibling course: Evidence-Based Medicine for Techies →
07 Clinical Cognition, From First Principles

Safety & Systems Thinking

A single wrong diagnosis is an error to explain. A pattern of similar errors across similar cases is a system to fix — and this lesson is about learning to tell the difference, then acting on it.

Lesson 7 of 9 Builds on Lesson 6 — Analytics at Scale Feeds into Lesson 8 — Evidence & Calibration Source modules M29, M38, M42, M49
VibeRounds This course is built in the spirit of VibeRounds — Socratic learning (AI that questions rather than answers) and Guided Discovery, part of the wider Clinical Cognition Operating System.

Lesson 4 taught bias and failure modes at the level of one reasoner's cognition — anchoring, premature closure, the traps a single mind falls into mid-case. This lesson asks a different question: when the same kind of error keeps happening across different clinicians, different shifts, different patients, at what point does it stop being a string of individual mistakes and become evidence of a system built to produce that error? Systems thinking is the discipline of noticing that shift and responding to the system instead of just the individual.

This is also where the course's engineering parallel is most literal rather than metaphorical. Failure Mode and Effects Analysis (FMEA) was borrowed by patient-safety work directly from manufacturing and aerospace engineering, with almost no adaptation to the underlying method — which makes it the cleanest bridge in the whole course between clinical and technical reasoning.

🩺 For clinicians & students

A root-cause analysis after a bad outcome and a near-miss report before one both ask the same question — not "who made the mistake" but "what about this process makes the mistake easy to make, and how many other people are one bad day away from making it too?"

💻 For techies

This is a post-mortem culture, but run proactively instead of only after an incident — FMEA is the engineering discipline of walking every step of a process and asking "how could this step fail, how would we know, and how bad would it be" before the failure happens, not after.

Walking the Process Before It Fails: Module 29

Module 29 imports FMEA into clinical workflow almost unchanged from its engineering original: pick a process (medication reconciliation, handoff at shift change, discharge planning), break it into discrete steps, and for each step ask three questions — how could this step fail, how severe would that failure be, and how likely is it to happen and go undetected. The three answers combine into a risk score that ranks steps by where attention is actually needed, rather than by which failure happened most recently or most memorably.

The module's discipline is in the ordering: severity, likelihood, and detectability are scored independently before they're multiplied together, because collapsing them too early lets one alarming-sounding failure dominate the ranking even if it's rare and easily caught.

#VibeRounds prompt — Module 29, Step 29.3 (score before you rank)
For each step in this process, score independently: severity of failure (1–10), likelihood of occurrence (1–10), and likelihood the failure goes undetected before reaching the patient (1–10). Do not combine these into a single risk score until all three are scored separately.
🩺 Clinical framing

A low-severity failure that happens constantly and goes unnoticed every time can carry more real risk than a dramatic failure that's rare and always caught — FMEA's whole value is making that comparison visible instead of letting the more vivid failure win the attention by default.

💻 Techie framing

This is a risk matrix straight out of a systems engineering review — severity × likelihood × detectability is the same triage logic used to prioritize which bugs get fixed before a release, applied to a clinical workflow instead of a codebase.

The Near-Miss That Didn't Happen: Module 38

Module 38 addresses the failure mode that never shows up in outcome data at all: the error that was caught before it reached the patient. A near-miss report is structurally identical to an incident report except for the ending, and the module argues that treating them differently — investigating the incident, filing away the near-miss — throws away most of the available signal, since near-misses vastly outnumber actual incidents and share the same root causes.

Its central design constraint is about psychological safety rather than analysis method, and it's worth stating plainly because it determines whether any of this data gets reported at all:

#VibeRounds prompt — Module 38, the just-culture framing
Draft this near-miss report in language that describes the process condition that allowed the error to become possible, not the individual action that happened to trigger it this time. Avoid language that assigns blame to a specific person — the same condition will produce the same near-miss for someone else if it isn't fixed.

The module's underlying claim is uncomfortable but well-supported in the safety literature it draws from: a reporting culture that punishes the person who reports a near-miss will simply stop receiving near-miss reports, and the system keeps its blind spot exactly where it was.

🩺 Clinical framing

Every clinician has a near-miss story they've never written down, precisely because writing it down felt like confessing to something. A just-culture reporting process exists to make that story valuable data instead of a personal liability.

💻 Techie framing

This is a blameless post-mortem — the same cultural shift that separates teams that fix root causes from teams that just find someone to blame and ship the same bug again next quarter under a different name.

Layers, Not a Single Cause: Module 42

Module 42 pushes back against the instinct to stop at the first plausible explanation for an error — the nurse who gave the wrong dose, the resident who missed the allergy flag — by walking the case through the Swiss cheese model: multiple layers of defense, each with holes, where a bad outcome only occurs when the holes in several layers happen to line up at once. The module's repeated prompt is "and what let that happen" — asked again after every answer, until the chain runs out of new layers to blame.

#VibeRounds prompt — Module 42, Step 42.5 (the five-whys extension)
You've identified one contributing factor. Now ask: what upstream condition allowed that factor to exist? Continue this chain until you reach a system-level condition (staffing, workflow design, tooling, training) rather than stopping at the first individual action.

The module is explicit that this isn't about diluting accountability — it's about recognizing that an error which required five separate layers of defense to fail simultaneously is a different kind of event than one where a single, easily-preventable check was simply missing, and the two call for very different fixes.

🩺 Clinical framing

This is Lesson 4's bias audit run on the system instead of the individual reasoner — premature closure on "the nurse made a mistake" is the systems-level version of premature closure on a diagnosis, and it's just as costly to stop there.

💻 Techie framing

This is a root-cause analysis five-whys chain, and the Swiss cheese model is the clinical-safety equivalent of defense-in-depth in security architecture — no single layer is assumed sufficient, and an incident review that stops at the first broken layer misses which other layers also need hardening.

Designing the Fix: Module 49

Module 49 closes the loop from analysis to redesign — taking an FMEA or a root-cause chain and turning it into a concrete process change, with an explicit bias toward fixes that remove the opportunity for the error rather than fixes that just ask people to be more careful. The module ranks candidate interventions on a hierarchy borrowed directly from safety engineering: forcing functions and constraints at the top, standardization and checklists in the middle, and training or reminders — the weakest and most commonly reached-for fix — at the bottom.

#VibeRounds prompt — Module 49, the intervention-hierarchy check
For this failure mode, propose interventions at each level of the hierarchy: elimination or forcing function, standardization or checklist, and training or awareness. If the only proposed fix is training or a reminder, explain why a stronger-tier intervention isn't feasible before accepting it as the solution.

The module's insistence on justifying training-only fixes is deliberate — "we retrained the staff" is the easiest intervention to propose and the least durable one, because it depends on every future person remembering under pressure what the last person forgot.

🩺 Clinical framing

A checklist that makes the wrong dose physically hard to select beats a memo asking everyone to double-check doses — the fix that survives someone's worst day is the fix that doesn't depend on that day going well.

💻 Techie framing

This is the same hierarchy as fixing a bug class with a type system instead of a code-review comment — a forcing function that makes the error unrepresentable beats a reminder that relies on someone remembering to check, every single time, forever.

Homework for Lesson 7

  1. Take a process from the case you've carried since Lesson 2 (medication reconciliation, a handoff, a discharge step). Break it into 4–6 discrete steps and score each one's severity, likelihood, and detectability (1–10) if it were to fail (Module 29).
  2. Write one near-miss you've witnessed or heard about — real or plausible — in process-condition language rather than blame language, following Module 38's framing.
  3. Pick your highest-risk step from question 1. Propose one intervention at each level of Module 49's hierarchy — a forcing function, a checklist or standardization, and a training-based fix — and state which one you'd actually implement first, and why.

This lesson draws directly on Module 29 — Failure Mode & Effects Analysis, Module 38 — Near-Miss & Just-Culture Reporting, Module 42 — Root-Cause & Swiss Cheese Analysis, and Module 49 — Systems-Level Intervention Design, all from the VibeRounds Prompt Directory. FMEA, the Swiss cheese model, and the intervention hierarchy are all established patient-safety frameworks borrowed largely intact from engineering and safety science — this lesson applies them, it doesn't originate them. If you're coming from the evidence side, the companion Evidence-Based Medicine for Techies course pairs well with this one. Neither course is a clinical decision tool; see the VibeRounds disclosure statement for full terms.