How structured reporting cut our PET/CT turnaround by 25%

Turnaround time is not a vanity metric in oncologic imaging. A scan reported on Thursday instead of Monday can be the difference between a patient starting treatment this week or next, and between a tumour board discussing a case with the images or deferring it seven days.

When we set up Prismaa we treated reporting as a system to be designed rather than a habit to be inherited. This note describes what we did, because the changes were not complicated and are worth copying.

The starting problem

Free-text narrative reporting has two costs that compound.

The first is variability. Two competent radiologists describing the same study will organise it differently, mention different negatives and use different phrasing for the same degree of certainty. The referring oncologist then has to extract the answer from prose.

The second is comparison failure. Response assessment depends on measuring the same lesions the same way across studies. If the prior report describes "a few enlarged nodes in the mediastinum" without designating target lesions, the follow-up cannot be quantitative, and the assessment collapses into impression.

Both of these also cost time. Composing prose from scratch for every study is slower than it feels.

What we changed

Templates per indication, not per modality. A staging FDG PET/CT, a lymphoma interim study, a PSMA scan for biochemical recurrence and a cardiac viability study ask different questions and need different structures. One "PET/CT template" would have been almost useless.

Designated target lesions recorded explicitly at baseline, with location, series and image number, and measurements in a fixed format. This single change did more for follow-up quality than anything else. Whoever reports the next study can find the same lesion in seconds.

Response criteria named in the impression. RECIST 1.1, PERCIST, Deauville, mRECIST, Choi or RANO as appropriate, with the category stated, not implied. Oncologists write their protocols in these terms; reports should answer in the same language.

A fixed impression structure. Three parts: the direct answer to the clinical question, the change from the previous study, and anything incidental that needs action. Referrers learn where to look, and nothing important gets buried in the fourth paragraph.

Standardised hedging vocabulary. A small controlled set of certainty terms, used consistently, rather than an idiosyncratic gradient of "possible / probable / likely / suspicious / concerning" that means different things to different readers.

Prior imaging chased before the patient arrives, not at reporting time. This was a scheduling change, not a reporting one, and it removed the most common single cause of delay.

What it did

Overall diagnostic turnaround fell by about 25%. That figure is less interesting than where the time came from, which was mostly not the dictation itself:

• Less time spent hunting for prior studies mid-report.
• Less time reconstructing what the previous reader had measured.
• Fewer clarification calls from referrers after the report was issued.
• Fewer addenda.

Reporting the first structured study of a given type is slower. The gain is entirely in the follow-ups, which in oncology is most of the workload.

What did not work

Over-templating. Our first version of the staging template had too many mandatory fields, several of which were irrelevant to most studies. Mandatory fields that do not apply get filled with boilerplate, and boilerplate makes reports longer and less readable, which is the opposite of the goal. We cut roughly a third of the fields in the second revision.

Assuming referrers would read the structure. Some did immediately; others continued to read the report as prose and phone anyway. Sitting down with the higher-volume referring practices and walking them through the format did more than any amount of internal refinement.

Rigid quantitation everywhere. Not every study needs SUV tabulated for every lesion. Numbers that nobody will act on are noise, and they lend spurious precision to studies where the honest answer is qualitative.

The part templates cannot fix

Structured reporting improves consistency and speed. It does not improve perception, and it does not substitute for subspecialty knowledge. A missed peritoneal deposit is missed whatever template it is not described in.

It also does not, by itself, help the patient. A structurally impeccable report is still a page of technical language to the person it is about. We keep time for patients to come back and have their scan explained with the images on screen, and no template replaces that conversation.

If you are setting this up

Three suggestions from doing it:

1. Start with your two highest-volume indications, not with a comprehensive library. You will revise them, and revising two is easy.
2. Write the impression structure first. It is the part that gets read, and designing it first disciplines everything above it.
3. Measure something before you change anything. We nearly did not, and the 25% figure would have been an anecdote instead of a measurement.

I am happy to share our current templates with colleagues setting up similar workflows — write to prismaasurat@gmail.com.