Data reliability and accuracy form the first gate controlling what data may appear in promotional materials for prescription drugs. No matter how favorable a number looks, one question determines whether it may be used: was it the answer to a question decided before the data were collected? The fact that a result has been published, or has passed peer review, is a necessary condition for reliability — but it is not sufficient.

There are two kinds of statistical analysis. Confirmatory analysis tests a hypothesis specified before the study began. Exploratory analysis searches for interesting patterns after the data are in. Chapter 1, Section 2 of the Guide makes this distinction the condition of inclusion itself. The foreword's commitments to "closing off misunderstanding" and to "verifiability" have their roots precisely here.

01Confirmatory vs. exploratory analysis — why the distinction is the condition of inclusion

Clinical trials normally define a statistical analysis plan (SAP) before the study begins: the primary endpoint, secondary endpoints, how multiplicity will be adjusted, and any subgroups considered in advance. This is pre-specification — locking in "what are we testing" before any data are seen.

Without pre-specification, statistics can prove almost anything. Slice the population into twenty subgroups and one will likely cross the 5% threshold by chance alone. Exploratory analysis is the place where hypotheses are generated, not the place where they are confirmed. The Guide encodes this principle as a rule: results may not be recorded unless they come from pre-specified analyses with scientific validity.

Where subgroup analyses stand

A subgroup analysis may be included only if the statistical method for that subgroup was set out in the analysis plan before the study ended. A finding that emerged from dividing patients after the data were collected — "we split the group this way and got significance" — is a post-hoc exploration. It cannot serve as confirmatory evidence. If it is to appear at all, it must follow the rules for post-hoc analyses described below.

02Post-hoc analyses — the conditions for inclusion and required disclosures

Post-hoc analyses are not banned outright. Where the informational value is high — a result suggesting important risk factors for a serious adverse reaction, or data used as the efficacy basis for an orphan drug where few patients make a conventional powered trial impractical — inclusion may be warranted. But conditions apply.

When a post-hoc analysis is included, the opening of that result must state explicitly that it is a post-hoc analysis and give the reason it is being presented. The language used must be restrained — "suggests," "a trend was observed," rather than definitive claims of efficacy.

What this requirement secures is that the reader can tell, on the page itself, whether a result confirmed a hypothesis or searched for one. Placing post-hoc results beside confirmatory results without distinction would give the reader a false sense of certainty — the foreword's duty to "close off misunderstanding" applies directly here.

03Meta-analyses — what it means to be a systematic review

A meta-analysis pools multiple trials statistically, which can appear to provide greater power than any single study. But the quality of the pooling depends entirely on how systematically the literature was collected and selected. Selectively gathering favorable studies makes a meta-analysis a device for amplifying bias rather than reducing it.

The Guide is explicit: a meta-analysis may be included only if it is based on a systematic review. The following information must also be recorded.

Why is this level of detail required? So that the reader — or a later auditor — can trace "where was searched, what was selected, and what was dropped." A meta-analysis that does not disclose its search process offers no way to verify whether studies were excluded by design. This extends the foreword's principle of verifiability all the way into the mechanics of evidence synthesis.

04The trap of "published means pre-planned"

A clinical trial conducted for a regulatory submission may later be written up and published in a journal. The fact that a paper passed peer review raises the apparent standing of the analysis. Yet if the publication involved a re-analysis that departed from the original statistical analysis plan, that published result is treated as post-hoc.

This is an easy blind spot in practice. The reasoning "it is a peer-reviewed original article, so it qualifies" seems sound. But peer review evaluates the internal logic of a paper and the quality of its reporting — it does not verify alignment with the pre-submission analysis plan. A clinical trial conducted for a regulatory application that is later re-analyzed when written up is treated as a post-hoc analysis regardless of publication, and may not be used as confirmatory evidence.

This rule makes visible that two separate evaluation axes exist: external quality assurance (passing peer review) and statistical integrity (answering a pre-specified question). Publication is not a substitute for pre-specification.

05Accuracy: three prohibitions — distortion, cherry-picking, and cropping

If reliability is a question of which data may be used, accuracy is a question of how the data that are used must be shown. The Guide forbids three things.

① Do not intentionally distort the source information

The figures, directions, and context that the trial results show must not be reshaped inside the material. Emphasizing only the relative risk reduction while obscuring a small absolute risk reduction, or making confidence intervals hard to see — these "visual operations" are forms of distortion.

② Do not conduct multiple analyses and present only the results convenient for interpretation

When multiple analyses are run, selecting only the favorable results for presentation is prohibited. This is cherry-picking. A structure that foregrounds a significant subgroup finding while a primary endpoint failed to reach its threshold is a direct violation of this rule.

③ Do not extract convenient portions from figures or tables in original articles

When transferring figures or tables from a publication into a material, it is not permitted to omit the overall trend or the control arm results in order to show only the portion favorable to the drug. Cutting the y-axis away from zero, or showing only a selected segment of a time series, is structurally the same offense.

The three prohibitions look different in form but share the same structure: take scientifically correct facts as raw material, then guide the reader's judgment through arrangement, selection, and truncation. This is the "factually accurate yet misleading" expression the Guide is most alert to.

In closing

What item ⑧ asks is not merely "is this scientifically correct?" The standard is stricter: "does this answer a pre-specified question?" and "has the context that should be disclosed been left in?" Publication, the form of a meta-analysis, the visual authority of a figure — none of these, standing alone, establishes reliability.

Distinguish pre-specified confirmatory analysis from post-hoc exploration; make the search and selection process of a meta-analysis transparent; block the temptation to show "only the favorable part" from three directions. This discipline is how the foreword's commitments to closing off misunderstanding and ensuring verifiability are made concrete at the level of individual data decisions.