Physician Recruitment · September 1, 2026 · KOLgroups

How to tell whether a physician respondent is real

Every healthcare study rests on an assumption that is rarely stated out loud: that the person answering is the physician the screener says they are. Most of the time that assumption holds. When it doesn’t, nothing in the data announces it. The responses look plausible, the crosstabs run, the report gets written, and a decision gets made on the strength of people who were never qualified to inform it.

This is not usually fraud in a dramatic sense. It is a chain of small, ordinary failures: a screener that gives away its own answers, an incentive large enough to be worth pursuing, a panel that has been asked the same questions too many times, and no step anywhere that checks a name against a public record. Each is minor. Together they decide whether a study is worth anything.

What “verified” should actually mean

The word gets used loosely. In practice there are four separate things to establish, and they fail independently.

Identity — the person is who they claim. This is the only one with a public source of truth. In the United States every practising clinician has an NPI number, and state licensure records are searchable. A name, a specialty and a practice location can be checked against those records before anyone is invited.

Licensure status — the licence is current and unrestricted. A physician who retired three years ago is still a physician in the colloquial sense, and will still answer confidently. Whether they belong in a study about current prescribing behaviour is a different question.

Specialty — the registered taxonomy matches the study’s target. Self-reported specialty drifts. A general internist who reads a lot of cardiology may sincerely describe themselves as cardiology-adjacent, and a screener asking “do you treat heart failure patients?” will let them through.

Current practice — they are actively seeing the patients the study is about. This is the one no public record answers, and the one a well-built screener has to carry.

Only the first two can be verified from outside. The last two depend entirely on how the screener is written, which is why screener design is a verification problem and not a questionnaire problem.

How a screener gives itself away

The most common failure is a screener that contains its own answer key. Consider:

Do you personally prescribe biologics for patients with moderate-to-severe atopic dermatitis at least 10 times per month?

A respondent who wants to qualify now knows the target: biologics, atopic dermatitis, ten or more, personally prescribing. The question has taught them what to say. Someone answering honestly and someone answering strategically produce identical data, and nothing downstream can separate them.

The same question, written so it cannot be reverse-engineered:

Approximately how many patients did you personally initiate on a systemic therapy in the past month? (open numeric)

Which conditions did you treat with systemic therapy in that period? (list of twelve conditions, several irrelevant)

Now the target is hidden in a distribution. There is nothing to aim at, and the qualifying combination only emerges after the fact.

Three principles follow from that:

1. Never reveal the threshold. Ask for the number, then apply the cut-off in analysis. The moment a respondent sees “at least 10,” ten becomes the answer.
2. Bury the target among plausible neighbours. A specialty list containing only the target specialty is not a question. Include adjacent specialties a genuine respondent would pick correctly and an imposter would not.
3. Include options that should never be selected. A drug that does not exist, or an indication the drug is not approved for, catches people selecting everything to maximise their chance of qualifying. Used sparingly, it is the single cheapest fraud check available.

The questions imposters fail

The most reliable screening questions are open-ended and ask about the ordinary texture of clinical work — the things that are obvious to anyone who does the job and unguessable to anyone who doesn’t.

Useful shapes:

• Workflow rather than knowledge. “Walk me through what happens between a patient’s referral and their first infusion at your site.” A practising clinician answers with logistics, scheduling frictions and staff. Someone who has read about the therapy answers with mechanism of action.
• Constraints rather than preferences. “What most often delays starting this therapy for a patient you have already decided to treat?” Real answers are administrative: prior authorisation, insurance, infusion chair availability. Imposters describe clinical hesitancy, because that is what the literature discusses.
• Local specifics. Which formulary they work under, what their institution requires, who else signs off. These are unglamorous, highly specific, and very hard to fabricate convincingly.

None of these are graded for correctness. They are graded for texture. A response can be short and still be obviously real.

The incentive is part of the design

Honoraria and screening rigour have to move together, and this is where studies most often get themselves into trouble. A high incentive attached to a transparent screener is an invitation. The same incentive attached to a screener that cannot be reverse-engineered is simply fair payment for a specialist’s time.

Physicians should be paid properly for their expertise, and underpaying is its own recruitment problem.

Where repeat respondents quietly distort things

A respondent can be entirely genuine — correct specialty, current licence, real practice — and still be the wrong person for a study, because they answered three similar surveys last quarter. They have now heard the concept, formed a view, and possibly encountered the competitor’s positioning too. Their response is no longer naive, and naive response is usually the thing being measured.

This is the least visible quality problem in healthcare market research, because nothing about the data looks wrong. It only shows up as a strange consistency across studies that should have produced different answers.

The controls are unglamorous: track who has been fielded recently and on what, and do not go back to the same people because they are easy to reach. A panel that returns respondents quickly is often doing so because it is returning the same respondents.

What to ask a recruitment partner

Whether recruitment is in-house or outsourced, four questions surface most of what matters.

Four questions worth asking before you commission

• How is identity verified, and against what source? “We check their email domain” is not verification. NPI or licence records are.
• What percentage of candidates do you reject, and for what reasons? A partner who cannot answer is not screening. A very low rejection rate means the screener is not doing any work.
• How recently were these respondents in another study, and on what topic? If the answer is unknown, freshness cannot be claimed.
• Who wrote the screener? If the people writing it have never run a study, it will read like a questionnaire rather than a filter.

The honest summary

Verification is not a single step that happens once. It is identity checked against a public record, a screener written so it cannot be reverse-engineered, open questions that expose the difference between reading and practising, and knowing who you have spoken to recently.

None of it is technically difficult. It is mostly a matter of whether anyone is accountable for doing it — which is exactly why we keep recruitment, screening and fielding with the same team rather than passing them between vendors. When one team owns the whole chain, there is nowhere for an unqualified respondent to slip through unnoticed, and no one to point at when they do.