What actually makes a report structured?

Named sections and fields that appear in the same order every time, holding content in predictable places. Free narrative dictation puts the same information in prose, in whatever order the reader thought of it. The difference is where a fact lives, not how much clinical judgment went into producing it.

Structure runs on a scale, and most reports sit somewhere in the middle rather than at either end. A report with headings for technique, comparison, findings and impression is already structured, loosely. A report that breaks findings into organ systems in a fixed order is structured more tightly. A report where a measurement goes into a labeled numeric field that a downstream system can read without parsing English is structured tightly enough to count as data. Facilities arguing about whether to adopt structured reporting are usually arguing about which point on that scale they want, without saying so, which is why the conversation goes in circles.

The confusion worth clearing up early is that structure says nothing about the quality of the thinking. A tightly templated report can be careless. A dictated paragraph from a reader who knows the case can be excellent. What structure changes is predictability: the referring clinician knows where to look, the next radiologist comparing a prior knows the earlier measurement will be in the same place, and a program that needs to count something has a place to count from. That is a real benefit, and it is a narrower one than the sales pitch usually suggests.

There is a version of this that gets sold as tidiness, and it is worth resisting. Reports that look uniform are pleasant to read and pleasant to show a board. Uniformity by itself buys a facility nothing anyone can measure. The question worth holding onto through every format discussion is what somebody will need to do with the report a year from now, and whether the format makes that thing possible or only makes it look better. A surprising number of format arguments end the moment somebody names the downstream task out loud.

What does a template cost the reader?

Attention. A template asks a fixed set of questions, and a reader working through it can answer all of them correctly while spending less time on the one finding that did not fit any field. The cost is real, it is hard to measure, and vendors rarely mention it.

Radiologists describe the effect in plain terms: the template pulls. When the form walks through the abdominal organs in a fixed order, the eye follows the form. Anything outside that list competes for attention with the next prompt. This is not a failure of discipline; it is how a checklist works, and checklists work partly by narrowing what you attend to. In aviation and in the operating room that narrowing is the point, because the list covers the whole task. In a diagnostic read the list cannot cover the whole task, because the task includes the finding nobody predicted.

The second cost lands on the impression. Templates grow. Every incident, every complaint from a referring physician, every quality committee adds a field, and nobody has the authority or the appetite to remove one. The findings section gets longer, and the impression, which is the part the ordering clinician actually reads, gets shorter and vaguer. That is the trade running backwards. A report can be more complete and less useful at the same time, and the facility will not notice until a referring physician says the reports have gotten harder to use, which that physician will usually frame as a question of tone.

None of this argues against structure. It argues for deciding, on purpose, which parts of a report should be rigid and which should stay open. Services that have thought about it hard tend to land in the same place: a structured findings section, a deliberately unconstrained impression, and an explicit field for something unexpected that belongs nowhere else on the form. That last field does more work than it looks like it should. Without it, the unexpected finding has to be wedged under a heading where it does not belong, or dropped into the impression with no supporting description, and a reader under time pressure will sometimes do neither.

Where does structure clearly earn its keep?

Anywhere the report will be read by something other than a person reading it once. Staging, screening programs, interval surveillance, registry submission, and any workflow where a number has to be compared with the same number from six months ago. In those settings an empty field is a defect the next reader has to work around.

Staging is the clearest case. A staging report has to answer a defined set of questions because the treatment decision downstream depends on all of them, and skipping one is not something the next reader can compensate for. If the primary tumor is described beautifully and nodal status is never addressed, the report is incomplete no matter how well it reads. The same logic covers screening programs, where the assessment category drives what happens next, and where an unassigned category forces somebody to phone the radiologist and ask what they meant. That call is what an optional field costs, paid by two people who both had something else to do that morning.

Registries are the other clear case, and the reason is arithmetic. If a program needs to know how many studies in a year carried a particular assessment category, someone either counts them from a field or reads several thousand reports. Free-text search across narrative reports works better than people expect and worse than a program needs, because it finds words rather than meaning: negation, hedged phrasing, and a sentence quoted from a prior report all defeat it. A discrete field does not have that problem. Where a count has to be defended, that is the entire case for structure, and it is a strong one.

There is a quieter case that matters more day to day. Comparison. A radiologist reading a follow-up study wants the prior measurement, the prior technique, and the prior assessment, and wants them where they were last time. When those live in three different sentences in three different paragraphs written by three different readers, the comparison takes longer and sometimes does not happen at all. Facilities tend to justify templates with registry reporting, then discover that the benefit they actually feel is this one. It shows up in reading time and in the calls they stop having to make about a prior study that was in the system the whole time.

  • Staging reports, where an unaddressed element changes what the treating team can decide.
  • Screening programs that assign an assessment category, since the category drives the follow-up pathway.
  • Interval surveillance, where the previous measurement has to be found quickly and trusted.
  • Anything feeding a registry, an accreditation submission, or a quality count.
  • Studies read by many different radiologists for the same referring service, where predictability is worth more than personal style.
  • Reports a downstream system has to parse, where a labeled numeric field beats a number inside a sentence.

Is a template the same thing as standardized language?

No, and conflating them is the most common mistake in this area. A template governs the shape of the report: which sections exist and in what order. A lexicon governs the words: that a given structure is called one thing and a given assessment level means one thing. You can have either without the other.

A perfectly structured report full of idiosyncratic wording is still hard to use. Two readers can fill the same field with 'no acute abnormality', 'unremarkable', and 'within normal limits'. A human understands all three; a query counts three different answers. That is a vocabulary problem, and form design does not touch it. Radiology developed shared vocabularies for exactly this reason. BI-RADS is the standard assessment framework for mammography reporting, and its value comes from a category meaning the same thing across facilities, not from the layout of the page it sits on. RSNA has published shared report templates for years, and the American College of Radiology maintains assessment and lexicon systems used in breast imaging and elsewhere.

The DICOM standard carries structured reporting objects, which are a third thing again: a machine-readable container for measurements and observations, distinct from the prose a clinician reads and distinct from the words chosen to write it. Facilities sometimes buy one of the three and expect the benefits of all three, and a good deal of disappointment starts there. For most facilities the practical order is lexicon first, layout second. Agreeing that a particular assessment scale is used, and used the same way by every reader, delivers most of the retrievability benefit and costs almost nothing in workflow. Doing it the other way round produces an immaculate form filled with phrases that do not agree with each other, while the counts underneath stay exactly as unreliable as they were.

Why do template rollouts fail so often?

Because the people who have to use the template were not in the room when it was written. A committee, an informatics team, or a vendor builds it, the readers see it the week it goes live, and the ones who can bypass it start bypassing it almost immediately. The template still exists. Nobody uses it.

The pattern repeats often enough to predict. Somebody identifies a reporting problem, usually a real one. A template is designed, often carefully, by people who understand what the data will be used for later and who do not read studies for a living. It is presented finished. The radiologists find that it adds keystrokes, asks for elements the study cannot answer, and forces a sentence order they consider wrong. A few object, are told the template is required, and comply for a while. The rest dictate over it or push everything into the free-text field. By the time anyone audits the output, the reports look structured and the data behind them is unusable.

The fix is not enthusiasm or training. It is authorship. Readers write the template, or at minimum sit in the room with a veto, and the first version ships knowing it is wrong somewhere. Then there has to be a way to change it that does not depend on a committee cycle: a named owner, a route for a reader to flag a field as unanswerable, and a scheduled review that actually removes fields rather than only adding them. Removal is the part everybody skips. Templates only grow, and a template that only grows eventually becomes the thing readers work around.

What should a facility ask a reading service about report format?

Three things, in this order: what the report looks like today and whether you can see real examples, whether specific elements come out as data or only as text, and who decides when the template changes. The third question is the one that gets skipped and the one that causes arguments later.

Ask for de-identified sample reports across several study types rather than one polished example, and read them the way a referring physician will: fast, at the end of a clinic, looking for the answer to one question. A format that reads well on a slide can read badly at four in the afternoon. Ask specifically what happens to the impression, because that is the part your clinicians will judge the service on, and it is the part most affected by template pressure. If every sample impression runs to most of a page, you have learned something useful before the first contract call.

On extraction, everyone says yes to the question of whether they support structured reporting, so the question is worth very little. Ask instead what happens to the specific element you care about. Pick one, an assessment category, a measurement, a recommendation for interval follow-up, and ask how it arrives: as a discrete coded value, as a labeled field in the report, or as a sentence somebody would have to parse. The answers differ a great deal, and they determine whether your quality reporting is a query or a chart review. Get the answer in writing, because the sales answer and the implementation answer often differ.

Change governance sounds like paperwork and is the thing that saves the relationship. Templates will need to change, because requirements change and because the first version was wrong somewhere. Settle in advance who can request a change, who approves it, how much notice the reading radiologists get, and what happens to reports already in flight when a new version lands. Settle also what happens when the facility wants an element the readers consider clinically unanswerable, because that conversation will happen and it goes better when there is an agreed route for it.

  • De-identified sample reports across several study types, not one showcase example.
  • Whether the impression stays free text, and who is responsible for its length.
  • How a specific element reaches you: coded value, labeled field, or a sentence to parse.
  • Which lexicons or assessment scales the readers use, and whether that is documented anywhere.
  • Who owns the template, who approves changes, and how much notice readers receive.
  • What happens to the format during a system outage or a change of reporting platform.

Sources and scope

  • Radiological Society of North America, professional work on radiology report templates and reporting practice
  • American College of Radiology, reporting and data system assessment categories and practice guidance on communication of diagnostic imaging findings
  • DICOM Standards Committee, the DICOM standard including structured reporting objects
  • HL7 International, standards for clinical document and observation exchange
  • IHE International, profiles for imaging reporting and results distribution workflow