(5) Clinical results | JPMA Guide Commentary | Advertising Regulation in Japan (Promotional Material Review) and Artificial Intelligence

Clinical results tend to be the longest section in any promotional material for prescription medicines, and at the same time the easiest to misread. The numbers a trial produces are neutral in themselves, but the impression they leave swings widely depending on which trials are chosen, which endpoints are put forward, and how the figures are drawn. The spirit set out in the foreword — that the electronic package insert is the original and these materials only complement it, that one must avoid not only falsehood but misunderstanding, and that areas without explicit rules are still governed by higher principles — is tested most severely here. What follows explains not only what may be shown, but why, tying each requirement to the hierarchy of evidence and the discipline of statistics.

01Where the trial data may come from

There is a gatekeeping question at the entrance. Not every result on hand may be used. Clinical trial data cited in materials must originate from one of the following.

Data evaluated during the domestic approval review
Material assessed as the basis for a revision of the electronic package insert
Original papers that have passed rigorous peer review (conference-presentation data and review articles do not count)
Material evaluated as part of a re-examination application

What these four share is that each has passed at least once through a checkpoint of third-party verification. Peer review, approval review, re-examination — all are mechanisms that curb the temptation to line up only the convenient results. Conference data and reviews are excluded because they carry the marks of early reporting and an author's interpretation, and differ in nature from settled primary evidence. The idea here is that evidence comes in a hierarchy, and the denser the verification, the more fitting the data is as a foundation for materials.

Comparative trials and real-world evidence

What may be cited as a clinical comparative trial is a double-blind trial, a randomized trial, or material assessed during approval review as a substitute for double-blinding. Blinding and randomization are design defenses against the expectations of observers and patients leaking into the data; they are the backbone of a result's reliability.

For real-world evidence (RWE), whose weight has been growing, use it only when all four of the following conditions are met.

Four conditions for citing RWE

It is positioned to reinforce or complement the approved scope.
Fairness is preserved in the choice of controls, concomitant therapy, and reference drugs.
It is presented together with the confirmatory trial that served as the main basis for approval (most often a randomized controlled trial).
Bias, as a material limitation, is stated prominently.

RWE reflects actual practice, but it is prone to imbalances in patient background and to confounding. That is precisely why a double brake is required: read it alongside the confirmatory trial that forms the main basis, and put its limitations up front rather than hiding them. Do not let an observational study appear to be decisive evidence on the same footing as a randomized trial — that is the heart of these conditions.

02The presentable range — do not step beyond approval

The principle is plain: results are presented within the approved range. Even for a drug whose dose may be adjusted as appropriate, efficacy data above the approved dose are not shown. Nor are trial data that conflict with the starting dose or the method of adjustment. Figures that do not match the approved usage, however favorable, would mislead actual clinical use.

Annotations when data include off-label elements

Where off-label information cannot be avoided, a discipline of explicit labeling applies.

A dose-finding study that includes an off-label dosing arm must be labeled clearly as a "dose-finding study", with the approved usage annotated.
When presenting results that include off-label elements, place at the head a note stating that part is off-label and the reason for inclusion, and annotate the corresponding indication and usage.
For data that include deviating cases, add a note that the data have been limited to cases within the approved range and partly modified, together with the reason for deletion.
Results whose dosing arms contain off-label cases must not be re-analyzed.

The line "must not be re-analyzed" matters most. To remove off-label cases and rebuild the numbers is to produce not the result the trial showed, but a different result made by the author of the material. The wariness toward altering, after the fact, an analysis plan that was specified in advance is what shows through here.

Control-drug usage and subgroups

Confirm that the dosage of the control drug falls within the approved range of every country that took part in the trial. If domestically off-label information is included, annotate the domestic approved usage; if the drug is not approved domestically, say so. Readers read in the domestic clinical setting, so the care taken is to keep them from misreading numbers based on foreign standards.

Subgroup analyses are not shown, except those included in the original plan and those for which scientific validity is recognized. Subsets carved out conveniently after the fact are the classic pitfall of making a chance difference look like a real effect.

03The opening caution and the required items

At the top of the first page of clinical results, in a point size larger than the body text, place "For warnings, contraindications, etc., see page ○○." Before reading favorable results, the reader is turned toward the premise of safety — a concrete expression of the stance that these materials complement the electronic package insert as the original.

Items each trial cannot do without

For every trial, show the following elements, neither too little nor too much.

Item	Key points of description
Trial title	Indicate the phase (I/II/III, etc.). Name the control drug by its generic name. For comparisons among the company's own products, the brand names may also be shown.
Type of trial	For overseas data or international joint trials, state this explicitly.
Trial method	Objective, subjects, number of cases, dosing method, endpoints, analysis plan, criteria for judgment. Distinguish the positioning of endpoints and make confirmatory analysis items clear.
Source	State on the first page of each data set. For approval-time evaluation material, say so; for original papers, give the bibliographic details.
Conflict of interest	If the company has a COI, record it together with the bibliographic details.

The point of "making confirmatory analysis items clear" connects directly to the efficacy section below. Planting a flag at the outset for what the trial concluded, and on what basis, is what prevents later exaggeration.

04Efficacy — separate confirmatory results from nominal p-values

Efficacy is shown according to the trial design, with the positioning of endpoints stated. The core discipline is to distinguish confirmatory results from results that rest on nominal p-values alone. Only an analysis specified in advance as a primary endpoint, with statistical error controlled, can be called "confirmed." A p-value computed exploratorily carries no more meaning than proposing a hypothesis, even if it falls below 0.05. Mixing the two leads the reader to mistake a secondary finding for an established effect.

Concrete brakes to avoid situations that breed "same data, different impression":

Do not pull out and emphasize only the favorable parts.
Do not make results look larger than they are with figures or arrows.
When the number of cases is fewer than ten, do not use percentages or graphs; show actual counts (proportions from small samples are unstable and mislead the impression).
When hazard ratios and the like are placed within a figure, keep them from being taken as emphasis. If there is no significant difference, do not state a risk-reduction rate.

That a difference is statistically significant and that the difference carries clinical meaning are two different things. To hold up "significant" alone, without the size of the difference, the width of the confidence interval, and the weight of the endpoint, is an attitude that keeps the numbers from speaking rather than letting them speak.

05Safety — the requirements change by column

Safety descriptions, even for the same drug, demand different things depending on the column. In the safety column of clinical results, show, for both the drug and the control (including placebo), the event names, the number of cases, and the incidence rates. For serious adverse reactions including death, and for those leading to discontinuation, record the event names and the number of cases.

Do not write "there were no serious adverse reactions." Absence within the observed range is not proof that none exist. Safety information is disclosed even when unfavorable — this principle forbids the easy assertion of reassurance.

Operations that "emphasize" safety are also forbidden. One does not stage safety by, for instance, presenting a significance test where there is no difference from placebo. For the control drug, present only the facts, adding no evaluation or commentary. Furthermore, unless an item is both a primary endpoint and confirmatory, do not present between-group significance tests or confidence intervals (showing the estimate for each group is acceptable).

Contrast with the product-features column

Even for the same safety topic, the column describing a product's features does not state the control drug's adverse reactions. This is the mirror image of the clinical-results column, which lays out the events of both drugs side by side for fairness. When the column's purpose differs, so does what is shown and what is held back. Borrowing content across columns without noticing this contrast steps outside the discipline.

06Reference information and case presentations

Reference information

Results obtained secondarily within the approved range are set apart from the proper efficacy evaluation as "reference information." For each trial result, state explicitly that it is reference information, and do not let the indication or effect be misunderstood. Keep the title to an expression that does not assert and emphasize an action, such as "Effect on ○○." This is a partition that keeps a secondary finding from being read, before one notices, as if it were the main effect.

Case presentations

Individual case presentations, because they readily lead to the emphasis of exceptional data, are as a rule not created. The exceptions allowed are limited to the following.

Cautions regarding adverse reactions
Rare diseases and other settings where only a small number of cases can be obtained
Contrast media and the like, where matters can be shown only by images (fictional model cases are treated the same way)

Even when created as an exception, state the source and name other companies' products by generic name. Do not emphasize or guarantee efficacy or safety. Do not evaluate or comment on the drug as a whole from that one case. And at the head, in a point size larger than the body text, place a statement that the presented cases are only a part and that not all cases show the same result, and always record any adverse reactions. It is a structural device to keep the vividness of a single case from governing the impression of the whole.

In closing

What runs through the clinical-results chapter is a set of brakes against the force that wants to make numbers look bigger. Confine sources to verified evidence, do not step beyond the approved range, separate confirmatory results from nominal p-values, and disclose safety even when unfavorable. Each is drawn from the foreword's spirit of "avoiding misunderstanding too" and from the scientific discipline of respecting the hierarchy of evidence and pre-specification.

Good clinical-results material is not a curated selection of the most convenient facet, but a set of context — positioning, limitations, controls, safety — sufficient for readers to judge soundly on their own. Precisely because the chapter is thick, the restraint of its editing becomes, directly, the thickness of trust.

Case studies — the "it isn't written, so…" mindset, and how each lands

The clinical data section carries the most content and is the most susceptible to editorial distortion. Which trial to include, which endpoint to lead with, how to frame the graph — each choice is a place where the "it isn't written" logic can quietly tilt the reader's impression.

GRAY 1 — Straying from the spirit (untouched by the wording, but against the foreword's spirit)

The scheme: "The source rules limit which trials can be cited. Within that eligible pool, selecting the two most favourable trials and omitting the three less favourable ones is not a rule violation — every included trial meets the source requirement."

Verdict: GRAY 1. Selectively presenting only favourable trials from an eligible pool recreates publication bias within a technically compliant format. The foreword prohibits not only false statements but also impressions that diverge from the full factual picture. Cherry-picking among approved trials overstates the strength of the evidence base. The correct approach is to disclose the selection criteria and not omit eligible trials on the basis of unfavourable results.

GRAY 2 — Exploiting the gap (deliberately working a loophole to invite misreading)

The scheme: "The RWE citation rule says biases must be 'prominently stated.' The rules don't specify where. A single line in a footnote — 'potential biases exist in this dataset' — satisfies the letter of the requirement while keeping the main numbers front and centre."

Verdict: GRAY 2. "Prominently" means the bias disclosure reaches the reader, not merely appears on the page. A footnote buried below the primary RWE figures fails that standard: readers encounter the data first and the caveat never, or too late to recalibrate. The result is RWE treated as equivalent to an RCT — which is the specific misreading the rule aims to prevent. The correct approach is to place the bias disclosure at the same visual level as the RWE figures, not beneath them.

GRAY 3 — Violating higher norms (arguably "not in the text," yet almost certainly out under the PMD Act, Fair Advertising Standards, and the JPMA Code)

The scheme: "A post-approval peer-reviewed paper shows a striking ORR in a specific genotype subgroup. The paper passes the source test. Describing the subgroup result within the approved patient population — it's the same disease — stays within the approved indication."

Verdict: GRAY 3. A subgroup analysis is an exploratory finding with a nominal p-value, not a confirmatory result. Presenting it in the clinical data section in a way that implies a specific biomarker-selected patient group has a confirmed superior outcome implies an unapproved use pattern. PMD Act Article 66 and the Fair Advertising Standard's prohibition on implying unapproved efficacy apply regardless of whether the subgroup is technically within the approved disease population. The correct approach is to label subgroup findings explicitly as exploratory and to keep them separate from primary endpoint data.

BLACK — Breaching the explicit text (a plain violation of the provision itself)

The scheme: "Conference presentation data is prohibited as a source. But if we re-aggregate the raw numbers ourselves and label the result 'in-house re-analysis,' we've created a new dataset from an admissible source — the data has been transformed."

Verdict: BLACK. The permitted source categories are: data evaluated during domestic regulatory review; material used as the basis for a package insert revision; peer-reviewed primary publications; and re-examination application data. Re-aggregating conference presentation data does not change the underlying source; the origin of the numbers remains outside all four categories. Relabelling the output does not change where the input came from.

Improper intent by implication — three scopes of "insinuating" without stating it

Clinical data are numerically neutral, yet the reader's impression can swing dramatically based on which trials are selected, which endpoints come first, and how figures are drawn. The three examples below cite individual data points accurately while using editorial structure to create an improper impression.

Implication 1 — Local: carried in a single word or sentence

How it is built: "In an exploratory subgroup analysis, a similar response rate was observed in the elderly population" — placed in the same paragraph immediately after the primary-analysis result.

Intent that can be read: The word "similar" equates the confirmatory primary result with an exploratory subgroup result. Exploratory analyses are hypothesis-generating only, but "similar" makes the subgroup result read as a co-equal conclusion.

Verdict: Exploratory analysis results must be clearly distinguished from confirmatory evidence. The evaluative word "similar" erases the difference in evidential weight. Whether an analysis was pre-registered as a primary endpoint or conducted post hoc must be stated explicitly.

Implication 2 — Adjacent: read across neighboring clauses

How it is built: "Response rate in the approved Indication A trial: XX% (primary endpoint, p<0.001). In a trial including the unapproved Indication B, a response rate of YY% was reported" — the off-label cohort placed directly after the approved cohort.

Intent that can be read: Placing the off-label cohort immediately after the approved cohort makes both appear to be a continuous record of results. Comparing the numerical values implies that the off-label use performs at least as well.

Verdict: Trial results for off-label indications must not appear in the product information summary. Presenting the approved and off-label cohorts consecutively treats them as equivalent evidence, which is not permitted. The guidelines explicitly prohibit presenting approved and unapproved cohorts in the same unbroken narrative.

Implication 3 — Whole-context: arising from the document as a whole

How it is built: The entire Clinical Results chapter is built from the three trials that met their primary endpoints; negative pivotal trials and discontinued trials with safety signals are omitted entirely. Efficacy figures use forest plots in which all subgroup estimates align in the positive direction.

Intent that can be read: By selecting only positive trials at the curation stage, the document as a whole implies "it worked in every trial." The consistent positive orientation of every figure ensures that any negative results remain imperceptible.

Verdict: Supplementing the package insert requires an objective presentation of the entire clinical program that supported approval. Selective citation of positive trials only is an editorial bias that distorts the evidentiary picture and constitutes misleading presentation.