Root cause analysis and quality improvement in the intensive care unit

[1]

Why this matters in intensive care

Critically ill patients are the most vulnerable to harm from the very systems intended to cure them: they receive the most drugs, lines, devices and handovers, cared for by rotating, fatigued teams under time pressure. The Harvard Medical Practice Study established that a substantial proportion of inpatient adverse events are caused by negligence and that the majority are preventable[4]. The ADE Prevention Study showed that most medication errors are systems failures — wrong dose, wrong route, missed allergy — best fixed by redesigning the prescribing and administration system, not by exhorting individuals to "be more careful."[8] The intensive care unit therefore represents the highest-yield environment in the hospital for safety engineering, and the intensivist is expected — in CICM/FFICM/EDIC vivas and in practice — to lead RCA, M&M review, and QI rather than merely attend them.

What root cause analysis is — and is not

RCA is a structured, retrospective, multidisciplinary investigation into an adverse event, sentinel event or near-miss, conducted blame-free, with the aim of identifying the underlying system weaknesses that allowed the event to occur and designing strong, measurable countermeasures so it does not recur. Its essential features are: [1]

• Blame-free and systems-focused. RCA asks "what" and "why," not "who." The first question is never "whose fault was it?" but "what conditions allowed this to happen?"[7]
• Multidisciplinary. The investigation team includes those closest to the event (bedside nurse, junior doctor, pharmacist, technician) plus a facilitator and a senior sponsor. Excluding frontline staff guarantees a sterile, inaccurate report.
• Data-driven. Timeline reconstruction, interview of all involved (using cognitive interview technique), review of charts, monitors, equipment, drug charts and environment — before forming hypotheses.
• Action-oriented. An RCA without implemented, measured actions is a paperwork exercise. The deliverable is not the report; it is the change.[7]

RCA is not: a disciplinary process; a means to allocate liability for litigation; a single-incident "witch-hunt"; or a substitute for proactive risk assessment (FMEA). It is also not research — it does not generate generalisable new knowledge, and it usually does not require ethics committee approval (see audit vs research, below). [1]

The RCA process — step by step

RCA tools

5 Whys

The simplest root-cause technique: state the problem, then ask "why did this happen?" and answer; then ask "why?" of the answer; repeat (typically about five iterations) until you reach an actionable system cause. The number five is a guide, not a rule — stop when the answer is a latent condition you can change, and stop early if you are heading toward a person rather than a system. [1]

Worked example — wrong insulin dose in ICU[8]

1. Why did the patient receive 50 units instead of 5 units of subcutaneous insulin? → The order was written as "50u" and the nurse drew up 50 units.
2. Why was the order written as "50u"? → The prescriber used "u" as an abbreviation for units, and the "u" was read as a zero.
3. Why is the abbreviation "u" used? → There is no enforced ban on dangerous abbreviations in the electronic order.
4. Why is there no enforced ban? → The computerised prescriber order entry (CPOE) system allows free-text doses rather than forcing selection from a dropdown.
5. Why does CPOE allow free text? → Insulin was not configured as a constrained-order-set drug at build. [1]

Root cause: a latent condition in the CPOE build. Countermeasure (strong): configure insulin as a constrained order set with unit-dose dropdowns and hard alerts for doses above a threshold — a forcing function, not a memo to "write more clearly." [1]

Fishbone (Ishikawa / cause-and-effect) diagram

A visual brainstorming tool. The "effect" (the adverse event) sits in the head of the fish on the right; contributing causes are grouped along "bones" under standard categories. For healthcare the common categories are: [1]

• People — fatigue, inexperience, staffing, communication, supervision, human factors (slip, lapse, mistake, violation).
• Process / method — protocol absent, unclear, contradictory, or not followed; workload; time pressure; handover gaps.
• Equipment / technology — unavailable, faulty, poorly designed, look-alike devices, alarm fatigue, missing CPOE safety checks.
• Environment — noise, lighting, interruptions, layout, asepsis, temperature.
• Materials / information — drug labelling, chart design, missing data, unavailable guidelines.
• Management / organisation — culture, leadership, scheduling, training, budget, policy. [1]

The fishbone generates the candidate causes; you then test each against the timeline and the 5 Whys to confirm it contributed. It is a thinking aid, not evidence on its own. [1]

Pareto chart

A bar chart of contributing causes ranked by frequency, with a cumulative-percentage line. Based on the Pareto (80/20) principle: a minority of causes usually account for the majority of incidents. Use it to prioritise — attack the few vital causes (typically the leftmost bars) rather than scattering effort across the "trivial many." In an ICU medication-error series, for example, dosing and omission errors often dominate and should be fixed first. [1]

Process mapping (flowchart / swim-lane)

A diagram of every step in the care process as it actually happens (not as it is supposed to happen), showing decision points, handoffs, waits, and rework. Process mapping reveals non-value-adding steps, bottlenecks, duplication, and failure-prone handoffs that are invisible to people who own only their slice of the pathway. Swim-lanes show who performs each step, making handover failures explicit. It is the natural starting point for any Lean redesign and for RCA of process-of-care events (e.g., delayed antibiotics in sepsis). [1]

Systems models of error — the conceptual backbone

The Swiss cheese model (James Reason)

Reason's model portrays an organisation's defences against harm as multiple slices of Swiss cheese: procedures, training, supervision, technology, alarms, double-checks. Each slice has holes — weaknesses such as a missing checklist, a fatigued nurse, a muted alarm, a look-alike drug. An adverse event occurs when the holes in every slice momentarily align, allowing a trajectory of accident opportunity to pass right through.[7]

Reason classifies the failures that create the holes into two kinds, and this distinction is central to blame-free analysis: [1]

• Active failures (the sharp end) — the unsafe act immediately preceding the event: a slip (action not as planned — wrong button pushed), a lapse (memory failure — omitted dose), a mistake (wrong plan from misapplied knowledge or misdiagnosis), or a violation (deliberate deviation from a rule). Active failures are usually committed by frontline staff and are the visible trigger of the event. They are the last and least interesting layer.
• Latent conditions (the blunt end) — the resident, built-in weaknesses created by decisions of designers, managers, and policymakers long before the event: noisy design, understaffing, impossible schedules, look-alike packaging, absent CPOE guardrails, a punitive culture. Latent conditions are the root. They are present every day and silently set the holes in the cheese; the active failure merely lines them up.[7]

The corollary: blaming and removing the person who committed the active failure changes nothing — the latent conditions remain, the holes persist, and the next operator will line them up. Fix the system, not the person. [1]

The London Protocol (Charles Vincent and colleagues)

The London Protocol operationalises Reason's model into a structured framework for analysing clinical incidents, organised under seven domains of contributory factors — a checklist that ensures the team does not fixate on the individual and forget the system:[5][6]

1. Patient factors — condition (complexity, acuity, anatomy), personality, social and language factors.
2. Task and technology factors — protocol design and availability, test results availability, task complexity, the design and usability of equipment and IT.
3. Individual (staff) factors — knowledge, skill, experience, health, fatigue.
4. Team factors — communication, leadership, supervision, team structure, culture.
5. Work and environmental factors — staffing, workload, noise, lighting, interruptions, layout.
6. Organisational and management factors — financial resourcing, structure, policy, culture of safety, priorities.
7. Institutional context — regulatory, political, economic, external pressures. [1]

Each event is scored against all seven domains; the domains with the most contributory factors become the targets for change. The London Protocol is now in a newer edition, refined for systems-level analysis of clinical incidents.[6]

The hierarchy of countermeasures (Reason)

Once root causes are identified, countermeasures are not all equal. Ranked from strongest to weakest: [1]

1. Forcing functions and constraints — physical or hard-logical barriers that make the error impossible (a gas-pin index that prevents connecting nitrous oxide to an oxygen port; a CPOE that will not accept an unsafe dose; removal of concentrated potassium from ward stock).
2. Simplification and standardisation — reduce steps, standardise equipment (one type of CVC, one standard concentration of adrenaline infusion) and processes so the correct path is the easy path.
3. Reminders, checklists and double-checks — catch errors before harm (Pronovost central-line checklist; independent double-check of high-risk infusions).
4. Rules and policies — written standards; moderate strength because they rely on human compliance.
5. Education and information — weakest; essential but never sufficient on its own. "Educate the staff" is the default lazy action plan — almost always inadequate as a root-cause fix. [1]

The exam-worthy principle: design out the opportunity for error rather than rely on the operator to remember. When asked "what is the best countermeasure?" the answer is the forcing function. [1]

FMEA — Failure Mode and Effects Analysis (proactive risk assessment)

FMEA is a prospective, team-based technique that anticipates how a process could fail before anyone is harmed, ranks the risks, and redesigns the process to prevent them. It is the proactive mirror of the reactive RCA — a mature safety system performs both. In healthcare the healthcare-adapted variant is HFMEA.[12]

[1] [1]

Incident reporting systems

A no-blame (or fair/blame-free) reporting system is the sensory nervous system of a safety culture: it surfaces the active failures and near-misses that would otherwise stay invisible. Principles:[7]

• Report everything that matters — especially near-misses. Near-misses (incidents that could have caused harm but did not) outnumber actual-harm events by an order of magnitude yet share the same system causes; they are richer, cheaper learning material and reporting them is not optional decoration.
• Make reporting easy, quick, and protected. Electronic, mobile, two minutes; clear legal protection that the report will not be used for disciplinary action except in cases of reckless/criminal behaviour (the just-culture compact).
• Close the loop. Staff who report must see that action follows; otherwise reporting collapses. Feedback may be aggregated and de-identified but must be visible.
• Triangulate with other data. Reported incidents are a biased sample (under-reporting is universal). Combine them with case-note review (trigger tools), M&M findings, complaints, mortality data, and SPC trends to see the whole picture.
• National/regional aggregation (e.g., ANZICS Safety Surveillance, the NHS National Reporting and Learning System) lets units benchmark and spreads lessons across organisations. [1]

Morbidity and mortality (M&M) conferences

M&M is the recurring, case-based forum in which adverse outcomes and deaths are reviewed for learning. Done well it is a powerful engine of culture change and QI; done poorly it is a ritualised blame-fest that suppresses disclosure.[10][11]

Audit cycle vs research

Trainees are frequently asked to distinguish audit from research. The distinction has regulatory consequences (research requires ethics approval; audit generally does not). [1]

[1]

The audit cycle (or audit spiral): (1) choose a topic and an evidence-based standard; (2) measure current practice; (3) compare practice against the standard and identify the gap; (4) implement change (often via PDSA); (5) re-audit to confirm improvement and detect drift; (6) embed and re-audit periodically. Re-audit is what converts a one-off measurement into a quality cycle. [1]

Quality improvement methodology

The PDSA cycle

The engine of iterative improvement — small, rapid, sequential tests of change rather than a single grand rollout:[2]

• Plan — define the aim, the prediction ("we believe X will happen"), the measure, the population, and the small-scale test (often one patient, one shift, one unit).
• Do — run the test; document what happens, including problems.
• Study — compare results with the prediction; what worked, what did not, what was learned?
• Act — adopt (standardise and spread), adapt (revise and run another cycle), or abandon. [1]

PDSA cycles are deliberately small: test on one patient first, then a bay, then a shift, then the whole unit. Failure is cheap and learning is fast. Most successful QI projects run many cycles, refining the intervention as they go. [1]

Lean — eliminate waste

Lean, derived from the Toyota Production System, defines value from the patient's perspective and relentlessly removes everything else. The eight wastes (TIMWOODS): [1]

• Transport (moving patients, samples, equipment unnecessarily)
• Inventory (overstocked, expired drugs; unused lines)
• Motion (staff walking long distances to fetch drugs or notes)
• Waiting (patient waiting for scan, antibiotic, bed, ward round)
• Overproduction (routine bloods that no one acts on)
• Over-processing (duplicate charting, redundant sign-offs)
• Defects (rework — re-doing a line, re-taking bloods)
• Skills (using senior staff for jobs juniors could do, and vice versa) [1]

ICU applications: one-piece flow on rounds, 5S organisation of bedside carts, visual management (whiteboards, supply par levels), removing waits in the sepsis pathway. [1]

Six Sigma — reduce variation

Six Sigma attacks variation in processes using the DMAIC sequence (Define, Measure, Analyse, Improve, Control) and statistical tools to drive defect rates down. ICU applications: standardising antibiotic timing, ventilation weaning protocols, glucose targets, and discharge criteria so that every patient reliably receives the same evidence-based care. [1]

Checklists and care bundles

• Checklists are cognitive forcing functions that ensure critical steps are not omitted under stress. The Pronovost central-line checklist (with a stocked cart and nurse empowerment to stop the procedure) drove CLABSI down by ~66% and toward zero across Michigan ICUs.[1][13] The WHO surgical safety checklist halved perioperative mortality and complications in a global study.[3]
• Care bundles are small groups (typically three to five) of evidence-based interventions delivered together and reliably — the SSC Hour-1 sepsis bundle, the ventilator bundle (head-of-bed elevation, sedation vacation, peptic-ulcer prophylaxis, DVT prophylaxis, daily readiness-to-wean). The defining rule is all-or-none: every component must be delivered on every eligible patient, every time — partial delivery is not a bundle. Bundle compliance is measured and reported as a process quality metric.

QI measurement tools — statistical process control

"What gets measured gets improved," but a single before/after comparison is easily misled by chance. Statistical process control (SPC) distinguishes common-cause (inherent, random) variation from special-cause (assignable, real) variation, so you know whether a change has actually moved the system or whether you are over-reacting to noise.[9]

Run charts

A line chart of a metric over time with a median centre line. Simple, robust, and the first SPC tool to reach for. Apply the run-chart rules to detect special-cause variation: [1]

• Shift — six or more consecutive points all above or all below the median.
• Trend — five or more consecutively increasing or decreasing points.
• Astronomical point — an obviously extreme outlier.
• Too few or too many runs — a non-random pattern. [1]

A signal triggers investigation; absence of a signal means keep collecting data — do not claim improvement. [1]

Control (Shewhart) charts

A run chart enhanced with a calculated mean centre line and upper and lower control limits (usually ±3 sigma). Different chart types suit different data: p-chart (proportion/percentage, e.g., bundle compliance), u-chart (rate per unit of exposure, e.g., CLABSI per 1000 line-days), c-chart (count of events), Xbar-S / I-chart (continuous variables such as time-to-antibiotic). The Western Electric / Nelson rules flag special cause: a single point beyond 3 sigma; two of three points beyond 2 sigma on one side; eight consecutive points on one side of the mean. Control limits are the system's voice — do not redraw them just because you dislike a point.[9]

[1]

Just culture and the culture of safety

A just culture (Reason, Marx, Dekker) is the ethical and operational core that makes everything else work. Without it, staff hide errors, near-misses go unreported, and RCA runs on a fraction of the truth.[7]

Just culture distinguishes three kinds of behaviour and matches the response to each: [1]

• Human error (a slip, lapse, or honest mistake) → console the person, fix the system that allowed it. Never blame.
• At-risk behaviour (a rule-bend or shortcut that seemed reasonable at the time — "everyone does it") → coach, remove the incentive to bend, redesign the rule if it is unrealistic.
• Reckless behaviour (knowing, conscious disregard of a substantial and unjustifiable risk) → disciplinary action. Recklessness is the boundary of the blame-free compact. [1]

The exam-critical point: human error is to be expected and managed; reckless behaviour is not. A blame-free culture is not a blameless culture. [1]

The broader culture of safety rests on five pillars: (1) blame-free (just) reporting; (2) visible leadership commitment; (3) staff involvement at every level; (4) continuous learning from errors, near-misses, and best practice; and (5) transparent communication about safety performance. The high-reliability organisation (HRO) literature adds five properties of organisations that operate hazardously yet safely — preoccupation with failure, reluctance to simplify, sensitivity to operations, commitment to resilience, and deference to expertise over hierarchy. [1]

Exam practice — SAQs

Clinical pearls

Worked RCA examples in the ICU

Example 1 — CVC insertion error (wrong vessel / arterial puncture)

Event. A junior registrar attempts an internal jugular CVC; the guidewire and dilator are advanced; brisk pulsatile blood returns — the carotid artery has been cannulated. The error is recognised, the wire removed, pressure applied, no long-term harm (a near-miss by outcome, a sentinel event by potential). [1]

Timeline → contributory factors (London Protocol).

• Patient: obese, short neck, difficult landmarks.
• Staff (individual): registrar on hour 14 of a busy shift, had performed fewer than ten unsupervised IJ lines.
• Task/technology: ultrasound available but not switched on; landmark technique used out of habit; no checklist run.
• Team: senior unavailable; bedside nurse did not feel empowered to stop the procedure.
• Environment: bay cramped, monitor alarms loud, frequent interruptions.
• Organisation: no unit-wide ultrasound-mandatory policy; no CVC insertion checklist at the bedside; no stocked CVC cart. [1]

5 Whys to root cause. Why arterial puncture? → ultrasound not used. → Why not used? → not switched on, habit of landmark technique. → Why the habit? → no enforced unit policy mandating ultrasound. → Why no policy? → safety governance had not prioritised line safety. → Root cause (latent): absence of a unit-level forcing function (ultrasound-mandatory policy + stocked cart + checklist + empowered nurse). [1]

Countermeasures (ranked).

• Forcing function: ultrasound-mandatory policy (cannot proceed without image); stocked CVC cart with integrated checklist; nurse empowered to halt the procedure (stop-the-line authority).
• Standardisation: one type of CVC; standard insertion kit; dedicated line team.
• Education: simulator training and sign-off before unsupervised insertion.[1][14]

Measure. Track ultrasound use at insertion, checklist compliance, and arterial cannulation / CLABSI per 1000 line-days on a control chart, before and after.[13]

Example 2 — Medication error (insulin ten-fold overdose)

Event. A patient receives 50 units of subcutaneous insulin instead of 5 units; severe hypoglycaemia requires 50% dextrose and ICU monitoring; full recovery. [1]

Timeline → contributory factors. The ADE Prevention Study established that medication errors are overwhelmingly systems failures — wrong dose, wrong drug, wrong route, missed allergy — best fixed by redesigning the prescribing and administration system, not by exhorting individuals.[8]

5 Whys (worked above) → Root cause: CPOE allows free-text insulin dosing and the dangerous abbreviation "u." [1]

Countermeasures (ranked).

• Forcing function: configure insulin as a constrained CPOE order set with unit-dose dropdowns and hard alerts above a threshold dose; ban the abbreviation "u" system-wide.
• Standardisation: standard insulin infusion concentrations; independent double-check of all subcutaneous insulin before administration.
• Education: high-alert-medication training (weakest, adjunct only).[8]

Measure. Insulin-error rate per 1000 doses and hypoglycaemia (glucose less than 4 mmol/L) episodes per 1000 patient-days, before and after. [1]

Example 3 — Failed intubation / unrecognised oesophageal intubation

Event. An out-of-hours intubation for septic shock; the tube is placed in the oesophagus; this is recognised only after several minutes of desaturation; the patient suffers a cardiac arrest and is resuscitated with hypoxic brain injury. [1]

Timeline → contributory factors (fishbone).

• People: most junior operator available; fatigued; recent rotation, limited airway experience.
• Process: no difficult-airway algorithm visible; no mandatory waveform capnography check; no pre-procedure checklist.
• Equipment: capnography monitor present but tubing missing; bougie not at bedside.
• Environment: out-of-hours, single nurse, poor lighting, unfamiliar bay.
• Management: no intubation bundle or checklist standardised across the unit; capnography not mandated. [1]

5 Whys to root cause. Why unrecognised oesophageal intubation? → no waveform capnography confirmation. → Why none? → capnography tubing missing. → Why missing? → no pre-intubation equipment check. → Why no check? → no intubation checklist / bundle. → Root cause (latent): no standardised intubation bundle with mandatory waveform capnography and pre-procedure equipment check. [1]

Countermeasures (ranked).

• Forcing function: mandatory continuous waveform capnography for every intubated patient (alarm cannot be silenced for "no waveform"); pre-intubation checklist and stocked difficult-airway bag.
• Standardisation: unit intubation bundle (plan, prepare, drugs, equipment, capnography, post-intubation care); difficult-airway trolley.
• Double-check: second operator confirms tube position before securing.
• Education: simulator-based airway training and annual competency sign-off (adjunct). [1]

Measure. Capnography use at intubation (proportion), intubation-related hypoxia events, and cardiac arrest within 30 minutes of intubation, on a p-chart, before and after.[7]

Key trials and evidence

[1] [3] [13] [4] [7] [5] [6] [8] [9] [10] [11] [14]

Red flags

[1]

Take-home

A safe ICU is built from systems, not from careful individuals. Every adverse event is a window into the latent conditions that allowed it; RCA is the structured, blame-free method that turns that window into strong, measured countermeasures; FMEA does the same work prospectively; the audit cycle and M&M forum embed learning into the routine; and PDSA, Lean, Six Sigma, checklists, bundles, and statistical process control are the tools that deliver and prove the improvement. Underpinning all of it is a just culture in which human error is consoled and designed out, and only reckless behaviour is punished. The intensivist who can lead this work — and articulate it in the CICM/FFICM/EDIC exam — is the one who actually makes the unit safer, not merely the one who attends the meetings. [1]

Examiner densify anchors

[1]

Practical ICU checklist (densify)

[1] [1]

Extended fellowship notes (densify)

[1] [1]

Topic-specific densify anchors — Root cause analysis and quality improvement in ICU

Line-fill densify notes

[1]

References

1. [1]Pronovost P, Needham D, Berenholtz S, et al. An intervention to decrease catheter-related bloodstream infections in the ICU N Engl J Med, 2006.PMID 17192537
2. [2]Boonyasai RT, Windish DM, Chakraborti C, Feldman LS, Rubin HR, Bass EB. Teaching quality improvement and patient safety to trainees: a systematic review Acad Med, 2010.PMID 20543652
3. [3]Haynes AB, Weiser TG, Berry WR, et al. A surgical safety checklist to reduce morbidity and mortality in a global population N Engl J Med, 2009.PMID 19144931
4. [4]Brennan TA, Leape LL, Laird NM, et al. Incidence of adverse events and negligence in hospitalized patients. Results of the Harvard Medical Practice Study I N Engl J Med, 1991.PMID 1987460
5. [5]Vincent C, Taylor-Adams S, Chapman EJ, et al. Framework for analysing risk and safety in clinical medicine BMJ, 1998.PMID 9552960
6. [6]Vincent C, Amalberti R, Vincent M, Edwards M. Systems analysis of clinical incidents: development of a new edition of the London Protocol BMJ Qual Saf, 2025.PMID 39986680
7. [7]Reason J. Human error: models and management BMJ, 2000.PMID 10720363
8. [8]Leape LL, Bates DW, Cullen DJ, et al. Systems analysis of adverse drug events. ADE Prevention Study Group JAMA, 1995.PMID 7791256
9. [9]Thor J, Lundberg J, Ask J, et al. Application of statistical process control in healthcare improvement: systematic review Qual Saf Health Care, 2007.PMID 17913782
10. [10]Bechtold ML, Scott S, Nelson K, Cox KR, Dellsperger KC, Hall LW. Educational quality improvement report: outcomes from a revised morbidity and mortality format that emphasised patient safety Qual Saf Health Care, 2007.PMID 18055885
11. [11]Marang-van de Mheen PJ, van Haperen M, Kievit J. Barriers and facilitators to learn and improve through morbidity and mortality conferences: a qualitative study BMJ Open, 2017.PMID 29133335
12. [12]DeRosier J, Stalhandske E, Bagian JP, Nudell T. How to make the most of failure mode and effect analysis Biomed Instrum Technol, 2003.PMID 12677747
13. [13]Berenholtz SM, Pronovost PJ, Lipsett PA, et al. Eliminating catheter-related bloodstream infections in the intensive care unit Crit Care Med, 2004.PMID 15483409
14. [14]O'Horo JC, Kierepka EM, Kallstrom PA, Schmitz M, Brown DL, Safdar N. Enhancing Quality and Safety in Critical Care-Challenges and Strategies for Central Venous Catheters JAMA Intern Med, 2024.PMID 38436978