Ask a hospital to produce its last twenty root cause analyses and read the conclusions. In most organisations, a clear majority will end at some version of the same finding: the staff member did not follow the procedure. Retraining was arranged. The case was closed.
That is not a root cause. It is where an investigation stops when it has run out of time, authority or method — and it is the single most common reason the same incident recurs eleven months later in a different ward.
Why investigations stop early
Three forces push an RCA toward the first plausible human error, and none of them is malice.
The first is comfort. “The nurse did not check the label” is a finding that can be actioned this week. “Our medication storage layout makes look-alike vials adjacent in seventeen clinical areas” is a finding that requires capital, a project and someone senior to own it. Investigations drift toward the conclusion the organisation can afford to act on.
The second is authority. An RCA led by a department manager can recommend changes inside that department. It cannot recommend changes to procurement, staffing ratios, or the electronic prescribing configuration. So it does not look there — and the causes that live there stay invisible.
The third is method. “Five whys” is a useful discipline, but only when each “why” is answered with evidence rather than with a plausible narrative. Without a structured contributing-factor framework, an investigation team will converge on whatever explanation the most confident person in the room offers first.
What a system cause actually looks like
A useful investigation asks a different question. Not “who made the error?” but “what made the error likely, and what made it hard to catch?” That reframing usually surfaces causes in five places:
- Task and technology. Ambiguous forms, look-alike packaging, an interface that defaults to the wrong dose unit, an alert that fires so often it is dismissed reflexively.
- Team and communication. Handover that carries no structured content, escalation paths that depend on knowing whom to call, hierarchies that make challenge socially expensive.
- Staffing and workload. Skill mix on the shift in question, consecutive shifts worked, the number of concurrent tasks at the moment of the event.
- Environment. Lighting, interruption rates, storage layout, physical distance between the point of preparation and the point of administration.
- Organisational. Policy that contradicts practice, procurement decisions made without clinical input, a training programme that exists on paper but was last delivered two years ago.
None of these are discoverable by asking the individual involved what went wrong. They are discoverable by reconstructing the conditions that existed at the moment of the event — which requires data the organisation has to have been collecting already.
The evidence problem
This is where most RCAs quietly fail before they begin. To establish that interruption rate was a contributing factor, you need observational data. To establish that the same near miss has occurred four times in eighteen months across three wards, you need an incident register that classifies consistently enough for the pattern to be visible. To establish that a policy was not followed because it was not read, you need read-and-understood records rather than a signature sheet.
Organisations running incident reporting on paper or in email cannot produce any of this. Their investigation is confined to what the people in the room remember, which is precisely how an investigation converges on the most memorable element — the human at the sharp end.
An incident register that cannot show you the pattern is not a safety system. It is an archive.
Four changes that raise the quality of an RCA
1. Classify at capture, consistently
If classification happens after the fact, by different people, using different judgement, the register cannot be aggregated. System-enabled classification at the point of reporting is what makes trend analysis possible at all — and trend analysis is what tells an investigation team that they are looking at the fourth instance of a systemic problem rather than the first instance of an unusual one.
2. Score harm separately from cause
Harm score and root cause are different axes, and conflating them distorts prioritisation. A near miss with no harm can carry the most important system finding in the register. Scoring harm at capture, and severity of the underlying condition separately, keeps low-harm signals from being filtered out.
3. Give the investigation a risk matrix, not an instinct
Positioning each incident on a likelihood-and-consequence matrix does two things: it forces an explicit judgement about recurrence probability, and it produces a defensible basis for deciding which findings justify capital expenditure. Without it, prioritisation follows whoever escalated most loudly.
4. Track the corrective action to evidenced closure
The most common failure in the entire cycle is not the analysis. It is the action. A CAPA agreed in committee, assigned to a busy clinical lead, with no due date and no escalation, has a low probability of completion — and a near-zero probability of effectiveness review. Every action needs an owner, a date, an escalation path and a scheduled check that it actually worked.
What good looks like
A mature incident system produces, without anyone assembling it manually: reporting rates by department and category; harm score distribution over time; the proportion of incidents with completed root cause analysis; median time from report to closure; CAPA completion rate and overdue count; and the recurrence rate of previously actioned incident types.
That last measure is the one that matters. Recurrence rate is the only honest test of whether an organisation's investigations are finding real causes — and it is the measure almost nobody tracks, because tracking it requires the register to be consistent enough to recognise a repeat.