How to read an incident pattern report

The most common misreading in a pattern report is not a hard one. It is reading a numerator as though it were a fact about your operation. Backing appears in 54 means nothing at all until you know 54 of what — and the answer changes the size of the finding by an order of magnitude depending on whether the denominator is 214 incidents or 1,847.

So every count in a well-built report is printed over the population it was scored in, and the habit that protects you is mechanical: find the denominator first, then read the numerator. If a number appears without one, treat it as decoration until somebody produces it.

It is normal, not a defect, for one document to carry several populations. A condition count may be scored across the whole log while a pattern is scored inside a single stratum, because a pattern that only occurs on vehicle incidents should be measured against vehicle incidents rather than diluted across everything else. The report should say which is which on every figure.

What is a defect is comparing across them. A count scored over 604 vehicle incidents and a count scored over 1,847 total incidents are not comparable, even when they are counts of the same condition and even when both are correct. Putting them side by side produces a ratio that means nothing, and it is an easy thing to do by accident when two numbers describe the same-sounding thing. If you take one habit from this guide, take that one.

Findings carry one of three grades, and each has a distinct shape as well as a word, so the distinction survives being printed in greyscale or read by someone who is colorblind.

Confident means the finding cleared the statistical bar: enough supporting incidents, and worse outcomes associated with it. Established means the condition is real and pervasive — there are plenty of incidents — but severity was never recorded, so it is a coverage finding and must never be read as a severity claim. Directional means thin recurrence, below the support bar: worth knowing about, not worth acting on alone.

Here is the part that gets misread. A grade describes how much your record supports the finding. It says nothing whatsoever about how dangerous the underlying thing is. A directional finding can concern something that will eventually kill somebody, and a confident finding can concern something that costs you a wing mirror. Confidence and importance are separate axes, and a report that keeps them separate is doing so deliberately rather than hedging.

One consequence is worth anticipating before you open your own report. If your record carries no recorded severity — and many EMS records do not — then nothing in it can reach Confident on severity grounds, and a document full of Established findings is not a weak analysis. It is an accurate description of what your instrument is currently able to see.

The condition map draws a condition once it reaches a display floor — a minimum number of incidents, not a percentage. Below that count it is left out of the figure, with two qualifications worth knowing: a condition that crosses several groups is drawn regardless, because a floor able to hide a crossing would be a display choice overruling the finding the figure exists to show, and the full inventory appears in the table underneath either way.

That floor is a drawing rule and nothing more. It exists because a chart with every condition on it is unreadable, and an unreadable chart teaches nobody anything. It is not a judgment about risk, it was not derived from any safety standard, and it does not mark a boundary between conditions that matter and conditions that do not.

A condition below the floor has not been assessed and found wanting. It has not been assessed. Treating the threshold as a safety boundary — below the line, therefore not a concern — is the single most consequential misreading available in the document, because it converts a decision about legibility into a decision about attention.

This is also why a good report states the total number of conditions found rather than only the number drawn. Twenty-one conditions of which eight are plotted is an honest sentence. Quietly reporting eight would shrink the denominator to whatever happened to fit on screen, which makes every proportion in the section wrong in the flattering direction.

The same caution applies to how a node looks. A larger or more connected node is not automatically more dangerous. It is more frequent, or more entangled, which are facts about the record rather than facts about harm.

A condition repeats when it reaches a minimum share of the incidents — five per cent — in more than one group: several locations, say, rather than concentrating in one. This is a different threshold from the one that decides what gets drawn, which counts incidents rather than proportions. The report marks repeats because a condition that turns up across contexts is telling you something structural rather than local.

A condition that concentrates in a single location is not a lesser finding. It is a different one, and can be the more actionable of the two, because a condition confined to one station or one maneuver has a boundary you can put a control around. Do not read the absence of a repeat mark as a downgrade.

Separately from repetition, conditions travel together: two or three of them turning up in the same incidents more often than they turn up apart. This is genuinely useful — it is how you find out that time pressure and solo work are not two problems but one situation described twice.

It is also where the reader is most tempted to write a causal sentence that the report did not make. The finding is that these arrive together in the record. It is not that one produces the other, and it is not that removing one removes the other. The most useful question to ask of any co-occurrence is whether there is a third thing that would explain both — a staffing pattern, a scheduling change, a layout — because if there is one, acting on either condition you can see may do nothing about it.

Most people read it last, if at all. It is the section that tells you what the record could not settle: where the evidence ran out, where the population was too small to support a conclusion, where a finding is an artefact of how the form is worded rather than what happened in the field.

Three limits deserve reading slowly every time. Recurrence is not cause: the report describes what repeats, and repetition on its own never establishes why. Absence is not proof: a condition that does not appear may genuinely be rare, or the form may simply have no way to record it, and the record cannot distinguish those two. And a quiet record is not a safe system — a period with few reports is at least as likely to describe a reporting problem as a safety improvement.

A proposed control is not an effective control either. Everything in the recommendations section is a candidate to be tested, not a conclusion that has been earned.

The reason this section is worth more than the findings is straightforward: a document that only makes claims cannot be checked, and a document that states its own boundaries can be. When a report tells you precisely where it stops, the parts before that line become usable — you know what you are standing on. A pattern report with no limits section is not a more confident analysis. It is an unfinished one, and that is the first finding.

The point of reading carefully is to arrive in a meeting able to say which findings are ready for a decision and which need one more question first. Both categories are useful; conflating them is what produces action plans that collapse under their first challenge.

Say the grade out loud with the finding. This condition is established across the whole log, severity unrecorded is a sentence that survives being repeated by somebody who was not in the room, and it prevents the quiet upgrade that otherwise happens between a report and a slide.

When a finding is ready, convert it rather than restating it. A candidate control needs five things attached at the moment it is decided: a named owner, the constraint that should hold, the operational step that makes it hold, an observable check that would show it was actually available and used, and a description of what its absence would look like. A recommendation with none of those is a sentiment.

And decide what you would need to see before you find out what you got. Agreeing the evidence in advance is what makes the ninety-day conversation honest — including in the case, which is common and worth planning for, where the answer turns out to be that nothing moved.