Every clinical study report carries a set of case narratives: short prose accounts of each death, each serious adverse event, and certain other significant adverse events. In a small Phase 1 study there may be three. In a large Phase 3 study there can be several hundred, and they are often the last thing standing between database lock and a finished report.

This guide covers the CSR-specific mechanics: what ICH E3 actually requires, how to decide which events need a narrative, where the narratives sit in the report, and how to write them consistently when there are many. For the craft of the individual narrative (structure, chronology, causality wording), see our companion guide on how to write a patient safety narrative.

What ICH E3 says

Section 12.3.2 of ICH E3 is short, and worth reading in full. The core sentence is:

"There should be brief narratives describing each death, each other serious adverse event, and those of the other significant adverse events that are judged to be of special interest because of clinical importance."

Three things follow from that sentence.

Deaths and other serious adverse events always get a narrative. "Each" is not qualified. The seriousness criteria come from ICH E2A: an event that results in death, is life-threatening, requires or prolongs hospitalisation, results in persistent or significant disability or incapacity, is a congenital anomaly or birth defect, or is otherwise medically important.

Other significant adverse events get one only if they are of special interest. E3 section 12.3.1.3 defines "other significant adverse events" as marked haematological and other laboratory abnormalities that are not serious, and any events that led to an intervention: withdrawal of study drug, dose reduction, or significant additional concomitant therapy. Not all of them need a narrative, only those "judged to be of special interest because of clinical importance." That judgement belongs to the medical monitor or safety physician, and it should be written down before drafting starts.

Clearly unrelated events can be short. E3 allows that "events that were clearly unrelated to the test drug/investigational product may be omitted or described very briefly." Most sponsors still write a brief narrative for every SAE, because the line between "clearly unrelated" and "probably unrelated" is exactly where reviewers push back. If you do omit any, record the rule you used.

Where the narratives go

E3 gives two locations: "These narratives can be placed either in the text of the report or in section 14.3.3, depending on their number."

In practice:

  • A handful of narratives can sit in section 12.3.2 of the report body, next to the safety discussion they support.
  • More than that go in section 14.3.3 ("Narratives of Deaths, Other Serious and Certain Other Significant Adverse Events"), which sits outside the text of the report. Section 12.3.2 then becomes a short pointer: how many narratives there are, which categories they cover, and where to find them.

Whichever you choose, the deaths listing in 14.3.2 has to reconcile with the narratives. The E3 Questions and Answers (R1) clarify a common source of miscounting: a fatal serious adverse event may appear in the 12.3.1.2 "other serious adverse events" listing, but "all deaths should be captured in the listing for section 12.3.1.1." The same subject should never be counted twice, or missed.

What each narrative must contain

E3 lists the content in two parts. The clinical story should describe:

  • the nature and intensity of the event
  • the clinical course leading up to it, with timing relative to study drug administration
  • relevant laboratory measurements
  • whether the drug was stopped, and when
  • countermeasures
  • post-mortem findings, where there are any
  • the investigator's opinion on causality, and the sponsor's, if appropriate

And each narrative should include these identifiers and context:

  • patient identifier
  • age and sex, and general clinical condition if relevant
  • disease being treated, with duration of the current episode (not needed if it is the same for every patient)
  • relevant concomitant or previous illnesses
  • relevant concomitant or previous medication, with dosage
  • study drug, dose (if it varied), and length of time administered

A narrative that covers all of these is complete by E3's standard. Most quality problems come from the clinical course, not the identifiers: a timeline that does not say when the last dose was taken, or an outcome that is never stated.

Deciding the narrative set before you write

The most expensive mistake with CSR narratives is starting to write before the population is fixed. Agree the following with the safety physician and the statistician first, and record it in the CSR writing plan:

  1. The data cut. Narratives are written from the locked database. Drafting from a pre-lock snapshot is fine for planning, but every narrative must be re-checked against the final data.
  2. The trigger list. All deaths, all SAEs, and an explicit definition of which "other significant" events qualify. For example: adverse events leading to discontinuation, Grade 4 laboratory abnormalities, and adverse events of special interest named in the protocol.
  3. The time window. E3 asks for deaths "including the post treatment follow-up period, and deaths that resulted from a process that began during the study." Say how you are handling events after the last dose.
  4. Treatment of blinded and placebo cases. Narratives are written after unblinding, and placebo subjects get narratives under the same triggers as active-arm subjects.
  5. The source hierarchy. Usually the clinical database first, the safety database for SAE follow-up detail, and CIOMS or MedWatch forms for the investigator's description. Decide what wins when they disagree, and how the discrepancy is flagged.

That list becomes the reconciliation check at the end: every subject in the death and SAE listings has a narrative, and every narrative maps to a listing row.

A working order for writing at volume

When there are dozens or hundreds of narratives, consistency matters more than eloquence. A reviewer reading the fortieth narrative should find information in the same place as in the first.

  1. Build a template from E3's list. One paragraph for patient background, one for the event course, one for treatment and outcome, one for causality. Use the same order every time.
  2. Generate the factual skeleton from data. Demographics, dosing dates, event onset and resolution dates, seriousness criteria, and outcome come straight from the datasets. Writing them by hand is where transcription errors creep in.
  3. Write the clinical course by hand, or review it carefully. This is where judgement lives: which laboratory values matter, which concomitant medications are relevant, what the sequence of events was.
  4. State study day, not just calendar date. Reviewers think in days relative to first dose. Give both, or give study day consistently.
  5. Keep causality attributed. "The investigator assessed the event as not related to study drug" is a fact. "The event was not related to study drug" is a claim the CSR author cannot make.
  6. Reconcile. Check the final narrative count against the 14.3.2 listings, and check every date and dose in each narrative against the datasets.

Common reviewer findings

  • A narrative says the patient "recovered", but the listing shows the event as ongoing at data cut-off.
  • The last dose before the event is missing, so the temporal relationship cannot be judged.
  • Concomitant medications are pasted in full, burying the two that matter.
  • The sponsor's causality opinion contradicts the investigator's without saying so.
  • Deaths during follow-up are missing because the narrative trigger was applied only to the treatment period.

Where automation helps, and where it does not

The factual skeleton of a narrative, and the reconciliation against the listings, are mechanical. They are also where most errors come from when done by hand. Asthra's CSR writing tool drafts from your listings and source documents in a fixed structure and cites each fact to where it came from, so a medical writer can spend the time on the clinical course and the causality rather than on retyping dates. The judgement about which events are of special interest, and what the clinical course means, stays with the safety physician and the writer.

FAQ

Do non-serious adverse events need narratives in a CSR? Not as a rule. E3 asks for narratives of deaths, other SAEs, and only those other significant adverse events judged to be of special interest because of clinical importance. A non-serious event that led to discontinuation may qualify if the safety physician judges it important.

Can narratives go in an appendix instead of the report body? Yes. E3 allows them in section 12.3.2 of the text or in section 14.3.3, "depending on their number." Large studies almost always use 14.3.3 and keep a short summary in 12.3.2.

Do placebo patients need narratives? Yes, if they meet the same triggers. The narrative set is defined by the event, not the treatment arm.

Is ICH E3 a fixed template for narratives? No. The E3 Q&A (R1) confirms that E3 "is a guidance, not a set of rigid requirements or a template." The content list in 12.3.2 is what reviewers expect to find, but the layout is yours.

Sources