A paediatric dermatologist in Japan has a child with severe eczema in front of her and a parent who has read about biologics online. She asks the rep a precise question: in the long-term data, what proportion of children reached clear or almost clear skin at one year? There is only one acceptable answer, and it is the exact figure from the study. Rounding up is not enthusiasm. It is misrepresenting the evidence to a specialist who will check. The program in this case ran in August 2023.
That was the challenge behind this program; we are writing about it two years later, in 2025. A pharma affiliate in Japan was preparing its dermatology field force, sales representatives and medical science liaisons together, to support a biologic in paediatric atopic dermatitis. The story rested on four clinical studies across different age groups, plus long-term extension data and a body of work on skin barrier function. Over a fifteen-day battle, 241 people answered 92,000 questions. Of 8,080 knowledge gaps identified at first attempt, 74% were verified as closed within the assessment, and readiness rose from 63% to 90%. The remaining quarter tells you where precision is hardest to achieve, and it is not where you might expect.
A paediatric indication carried by dense trial data
Paediatric indications raise the bar for the field. Parents are anxious, specialists are cautious, and every statement about efficacy and safety in children is scrutinised. The evidence base for this kind of product is also unusually detailed. Separate trials cover different age bands, each with its own inclusion criteria, dosing arms and endpoints. Long-term data extends the picture to a year. Safety has to be described with the exact terms the studies used.
For a rep or MSL, that means holding a lot of numbers in the right boxes. Which trial studied which age group? Was the population severe disease only, or moderate to severe? What were the endpoints, EASI-75 or clear or almost clear skin on the IGA scale? Which adverse events were defined as being of special interest? Mix two trials up, and a correct number becomes a misleading one.
The client's starting point was realistic. Readiness at first attempt was 63%. The field had seen the data in training, but the precision required in front of a paediatric dermatologist was not yet there.
The stakes of getting this right early are well documented. McKinsey's analysis of 210 launches between 2003 and 2009, published in 2014, found that about two thirds of new drugs miss pre-launch consensus expectations in their first year, and that most year-one underperformers stay behind in year two. An indication extension into children is not a full launch, but the logic applies. The first months of specialist conversations set how the data is understood, and credibility lost with a KOL over a misquoted figure is slow to rebuild.
What we designed: a long window and a very high volume of answers
We built one game in Japanese with 90 questions, almost all of them on evidence. The content fell into six blocks:
- Long-term extension data to 52 weeks: IGA and EASI trajectories, IgE change, adverse events and weight-based dosing.
- The registration trial in children aged 6 to 11: inclusion criteria, EASI-75 and patient background.
- The adolescent registration trial: design, dosing arms, clear or almost clear skin.
- Preschool and infant data: the trial in the youngest children, carer quality of life and comparators.
- Barrier function: ceramides, transepidermal water loss and the stratum corneum.
- Other supporting modules across the program.
The design choice that mattered most was time. Instead of a three or four-day battle, we opened a fifteen-day window in August 2023, with notifications to bring people back. That gave each question room to return to the same person several times.
The field used it. Players averaged 380 answers each over the fortnight, which is a very high volume for a single game. It is also what makes verified closure measurable. A gap only counts as closed when the same person answers the same question correctly on a later attempt, so the number of repeat encounters sets a ceiling on how much you can fix and prove.
Putting reps and MSLs into the same game was deliberate, and it had a narrow scope. Both roles need to know the same published data accurately, even though they use it in different ways: the rep in promotional conversations within the label, the MSL in non-promotional scientific exchange governed by Medical's own rules. The game tested knowledge of the approved scientific material only. It did not train either role in what to say to a physician, and it did not blur the line between the two. If the two roles understand the trial populations differently, though, the physician notices, and one shared set of questions is a simple way to check that they don't.
What happened: 74% closed, and a clear map of the rest
At first attempt we recorded 8,080 knowledge gaps across 241 players. By the end of the window, 5,941 were verified as closed, which is 74%. The remaining 2,139 were flagged for follow-up coaching. Readiness moved from 63% to 90%. The six blocks below add up to the 8,080.
Closure by block was high everywhere, but not even:
- Other supporting modules: 2,311 gaps, 82% closed.
- Barrier function: 851 gaps, 73% closed.
- Long-term extension data: 2,775 gaps, 71% closed.
- Preschool and infant data: 796 gaps, 71% closed.
- Adolescent registration trial: 694 gaps, 67% closed.
- Paediatric registration trial (6 to 11 years): 653 gaps, 64% closed.
The pattern is worth noticing. The blocks that closed least were the trial design questions, which patients were included and on which dose. Numbers and safety terms closed faster than definitions of the population. That is counterintuitive at first. It makes sense once you look at the individual questions.
The readiness figure deserves a word too. A move from 63% to 90% on a single cohort, measured on the same questions, is a large shift in how accurately the field answered questions about the data. It is a result inside the assessment. It does not show what anyone said in a clinic, and one later right answer does not rule out a lucky guess. But readiness is an index across everyone and every question. It rewards the many easy fixes and can hide the few hard ones. For a paediatric indication, the few hard ones are where the risk sits, so we read the index as the headline and the question-level data as the work plan. The managers who received that work plan knew which people to check on which study, which is a far more useful brief for a ride-along than a percentage.
Three questions that show how precision fails
Safety language closed almost completely
One question asked which event the long-term study defined as an adverse event of special interest. At first attempt, 97 people chose a plausible but wrong safety event. By the last attempt, that was down to 7. Of 123 people who were wrong at first, 113 corrected themselves through repetition, a 92% fix rate. Safety terminology is exact and binary, and once people had seen the right term a few times, most of them chose it again.
A long-term efficacy figure kept being overstated
Another question asked for the proportion of patients reaching clear or almost clear skin at week 52 in the continuous treatment arm. At first attempt, 106 people chose a figure well above the real result. By the end, 79 still did. Only 28 of 123 people who were wrong at first fixed it through repetition. The wrong option sounded better, and it kept being chosen on later attempts.
A trial population kept getting wider
A third question asked who was studied in the trial in children aged 6 to 11. At first attempt, 111 people chose an option that widened the population to younger children and to moderate as well as severe disease. By the end, 59 still did, and only 57 of 147 people who were wrong fixed it. Widening the population in a conversation with a specialist isn't a small slip. It misstates what the evidence actually supports.
Precise figures are either quotable or they are noise. The gaps that resisted repetition were the ones where the wrong answer sounded better than the truth.
Four lessons for evidence-heavy programs
1. Volume is what makes precision verifiable
Three hundred and eighty answers per player in fifteen days sounds like a lot, and it is. It is also the reason we can measure 74% at all: a gap can only be verified as closed if the question comes back to the same person. We can't tell you how much of the result is due to the long window itself. This was one cohort with no comparison group, and the people, the content and the exposure would all differ in a shorter program. What we can say is that short windows give questions fewer chances to return, so for evidence-dense content a longer window with steady notifications is worth testing.
2. Terms close faster than boundaries
Safety terms and single facts closed quickly. Population definitions and trial design closed more slowly, because they require people to keep several studies separate in their heads. Programs should spend more questions, and more comparative questions, on the boundaries between trials.
3. Watch for the flattering wrong answer
The most stubborn misconceptions were the ones that made the product look better: a higher efficacy figure, a broader population. They are also the ones with the greatest compliance risk. They deserve to be flagged by name to Medical Affairs and first-line managers, not left to repetition.
4. Put reps and MSLs on the same page, literally
Running both roles through the same questions made differences visible and gave Medical a clear view of where the field's understanding diverged from the data. Follow-up still has to respect each role's rules. Internal training of reps by Medical on the published data is common, but it should run through the company's approved process, not as informal coaching in a team meeting.
What we would do next
The 2,139 open gaps were not spread evenly. They clustered on a small number of questions, and a handful of misconceptions accounted for a large share of them. Our next step would be a targeted round aimed at those, with a short fix for each person built around the exact point of confusion: which trial, which age band, which severity, which figure.
For the stubborn questions, like the long-term efficacy figure and the population of the trial in 6 to 11 year olds, we would go beyond the platform. A ten-minute internal session led by Medical through the approved training route, comparing the trials side by side, is likely to do more than another week of repetition. Managers would receive the names of the people still holding the wrong answer, so the next ride-along could include a quick check.
And we would repeat a short version of the game ahead of the next major dermatology congress, when the same specialists will be hearing new data and asking the field to put it in context.
Seventy-four per cent is a strong result for dense paediatric data. The more useful part, for anyone preparing a similar indication, is the shape of what stayed open: not the obscure numbers, but the ones that sounded better than the truth. If your field is about to carry a complex evidence story to specialists, that is the list Atrivity is built to find, and it is worth finding before the first call rather than after it.