Data Before Conclusions — Standing as a Custodian of Evidence

When building a promotional piece, somewhere in the back of your mind sits the feeling that this drug should work. That feeling is genuine. But the moment it starts guiding which data you reach for, the material shifts from a record of evidence into a post-hoc rationalization of a conclusion already held. The third conviction is noticing that slide. It means holding your initial judgment as a hypothesis — provisional, open to refutation — and deliberately searching for evidence that breaks it. Conscience lives in the willingness to face disconfirming findings head-on.

So What / So Why — Why This Conviction Is Necessary

From the March 2019 report through the March 2025 report, the Ministry of Health, Labour and Welfare's monitoring program for pharmaceutical sales information activities recorded seven years of data-manipulation and data-repurposing incidents. A single structural pattern runs through all of them. The numbers used were real. The source trials existed. Yet the conclusions physicians received were wrong.

The structure is this: the conclusion came first, and the data went looking for it.

In the case documented in the March 2019 report, involving a bronchial asthma treatment, the primary endpoint results from a Phase III international joint trial — roughly 250 patients per arm in the full analysis — were omitted from slides and brochures. Only the Japanese subgroup, roughly 15 patients per arm, appeared. When asked why, the responsible person said physicians want Japanese data. It also happened to be true that the Japanese subgroup results were better than the full-analysis results. The explanation — "physicians want it" — was most likely a rationale that came to look coherent only after the favorable data had already been selected.

The case in the March 2024 report, involving a dyslipidemia treatment, went further. Survey data purporting to show that patients prefer in-facility administration was presented at product information meetings. The survey population, however, consisted of patients with severe asthma — a condition outside the product's indication — from a country with a substantially different healthcare system. The original publication compared five endpoints; the meeting materials extracted only the two that favored the product. The starting point was a desire to show competitive advantage on convenience, and from that starting point a literature search pulled in data from a different disease, a different country, and a different healthcare structure.

In 1990, social psychologist Ziva Kunda named this cognitive mechanism "motivated reasoning": the motivation to reach a desired conclusion unconsciously distorts which evidence is gathered and how it is weighted. Those involved do not feel biased. As long as an internal narrative holds — "the numbers are real," "physicians want this," "the conditions are similar" — awareness of the deviation rarely surfaces.

This conviction is necessary because human cognition defaults to moving in this direction. "Evidence first, conclusion second" does not happen unless you choose it deliberately. (See also: Past Incidents vol-01: Data Use and Manipulation That Risk Factual Misrepresentation and Under Pressure vol-03: Conclusion First, Data Second — Motivated Reasoning)

The Conviction — Who We Want to Be

I want to define my role not as a persuader but as a custodian of evidence. A persuader has every reason to lead with favorable data and bury the unfavorable. A custodian works the other way around: put the initial conclusion on a shelf as a hypothesis and search for evidence that breaks it first.

When subgroup analysis results come out better than the full analysis, the first question I want to ask myself is: why is there a difference? Is it patient background, trial design, or chance? I don't put the result in the material until I can answer that. The fact that the full analysis was less favorable still needs to reach the audience, even after I've decided to show the subgroup.

There will always be moments of uncertainty. I know the thought will come — "I can cover this verbally" or "the reader can check the original paper." But I want to honestly ask whether I am actually providing that verbal supplement every time. If the habit of adding caveats out loud isn't there, "verbal supplementation" becomes a retroactive excuse.

I don't think I can do this perfectly. But when I encounter unfavorable data, I want to be the kind of person who stops and asks, "How should I handle this?" That pause is the substance of this conviction.

Daily Practice — Concrete Actions and Checks

• Write down why you selected this data before you build the slide. Before creating slides or materials, make a one-line note explaining why you chose each trial dataset. If all you can write is "because it's favorable," revisit the selection.
• Order primary and secondary endpoints in your own mind first. When secondary endpoint results are strong, place them next to primary endpoint results and compare. Check that the order of presentation hasn't been reversed.
• State differences in preconditions in the body of the material. If the comparison trial has a different design, patient population, or healthcare system, write those differences into the document. Don't rely on verbal asides to cover the gap.
• Search for at least one piece of disconfirming evidence. When a draft is complete, set aside time to look for data that would contradict the conclusion. If you find any, make a conscious decision about how to handle it.
• Hold safety data to the same standard as efficacy data. For adverse events, contraindications, and risk management plan (RMP) key risks, ask yourself: "Am I communicating this with the same weight, time, and space as the efficacy claims?" The final few seconds and the smallest font are hard to call genuine disclosure.
• Don't treat "I wasn't asked" as an exemption. The guidelines require providing negative information proactively. Waiting to be asked does not fulfill the duty to provide information.

The Test — Under Pressure

This conviction erodes most during the six to twelve months after launch. Evidence is thin. The pressure to generate prescriptions is at its peak. The question "how do we present limited data effectively" is itself the entry point for conclusion-first thinking.

The day after a regional manager says your appeal to physicians is weak, you go back to the materials. The number that catches your eye is the subgroup figure, better than the full analysis. The moment you think "this will land better with Japanese physicians," the direction of the selection is already set. The question is whether the thought simultaneously surfaces: "Why does this subgroup differ from the full analysis?" In the bronchial asthma case from the March 2019 report, that question did not surface.

The same structure emerges right after a competitor gains a new indication. Searching the literature under pressure to "find a differentiator" pulls in data from different diseases, different countries, and different systems — because the numbers are real and that feeling is enough. Checking the differences in preconditions looks like unnecessary hesitation when pressure is running high. The dyslipidemia case from the March 2024 report maps that structure precisely.

If I had to draw a line, it would be this question: "Can I explain to the physician, honestly and including all preconditions, why I'm using this data?" Any use I cannot explain that way, I don't make. That is the minimum.

For a detailed analysis of how pressure operates and what it looks like from the inside, see Under Pressure vol-03: Conclusion First, Data Second — Motivated Reasoning.

The Anchor — Something to Return to When You're Close to Breaking

When in doubt, I return to the Guidelines on Sales Information Activities for Prescription Drugs. The principles — "must be based on scientific and objective evidence, and must be accurate and fair," "provide information on both efficacy and safety," "do not excerpt, modify, or combine data when citing" — convert what would otherwise be a question of personal resolve into a question of procedure.

"I'm not sure whether I can use this" is actually a moment to apply the guidelines. The uncertainty itself is a signal that you have sensed a gap in preconditions or an omission somewhere. Before dismissing that sense as overthinking, try mapping it against the guideline language. In many cases, the question "does this use constitute excerpting, modifying, or combining data?" is enough to reach a decision.

I think conscience is located in the habit of asking yourself that question. What seven years of recorded incidents show repeatedly is how many deviations happen because the question was never raised. Cases where a prescribing physician checked the package insert later and discovered the explanation didn't match are on record because that physician noticed. The number of prescriptions written by physicians who didn't notice cannot be known.

(See also: Past Incidents vol-01: Data Use and Manipulation That Risk Factual Misrepresentation — all incidents from the March 2019 report through the March 2025 report)