How a healthcare IDI or TDI is actually run: discussion guides, moderator craft, and what protects your data

In this guide 6 sections

1. What a discussion guide is actually for
2. IDI or TDI: a methodological choice, not a budget line
3. Moderator craft: the part that doesn’t show up in the deliverable
4. Failure modes that quietly wreck a transcript
5. What good methodology looks like from the outside
6. Why this is worth getting right before fieldwork, not after

Most buyers evaluating a market research partner ask good questions about recruitment: can you find twelve interventional cardiologists in six weeks, what’s the honorarium, how do you verify credentials. Those questions matter. Almost nobody asks the question that determines whether the resulting findings are trustworthy: how is the interview itself actually run?

That gap is understandable — recruitment is visible and interview execution is not. A client sees a recruitment grid, a fielding timeline, and eventually a report. The thirty minutes between “the physician joined the call” and “the physician left the call” happens behind a methodology nobody outside the research team examines closely. It is also where most of the damage to data quality actually occurs.

What a discussion guide is actually for

A discussion guide is not a script. Treating it as one is the single most common methodological failure in commissioned qualitative research, and it is usually invisible to the client reading the final report, because a scripted interview still produces fluent, quotable transcript text. It just produces text that confirms what the guide assumed going in.

The guide’s actual job is narrower and more disciplined: sequence the questions so that the respondent’s own frame of reference is captured before the moderator’s frame of reference is introduced. Every well-built guide follows the same underlying principle, usually called the funnel.

Once the unaided response is captured, the guide can move to aided questions — showing concepts, naming competitors, probing specific attributes — because the respondent’s unprompted priorities are already on record and cannot be retroactively contaminated. Exercises, ranking tasks, and stimulus review come last, after both unaided and aided reasoning exist in the transcript.

A guide that starts with stimulus, or that front-loads the sponsor’s hypothesis as a question (“don’t you think X is important”), has already decided the answer before the respondent speaks. No amount of skilled moderation recovers from that structural error.

IDI or TDI: a methodological choice, not a budget line

In-depth interviews (IDIs) and triad/dyad interviews (TDIs) get treated in a lot of project scopes as interchangeable formats differentiated mainly by cost per respondent. They are not interchangeable. They answer different questions.

An IDI isolates one physician’s decision logic without social influence. It is the right tool when the research question is about individual reasoning: how does this physician personally weigh efficacy against tolerability, what does their actual patient-selection process look like, what would change their current protocol. Anything where group dynamics would distort the answer — compensation-sensitive topics, admissions of uncertainty, departures from guideline-recommended practice — belongs in an IDI. Physicians will say things one-on-one that they will not say in front of peers.

A TDI (or dyad, with two respondents) is the right tool when the research question is about how physicians respond to each other’s reasoning, not just their own. Consensus-testing a clinical statement, pressure-testing a positioning claim, or understanding how a respected colleague’s skepticism changes another physician’s position — these require more than one voice in the room. A TDI run well surfaces disagreement; a TDI run poorly produces groupthink, because a confident first speaker anchors everyone after them.

Choosing IDI vs. TDI based on per-respondent cost rather than the actual research question is how studies end up with technically completed fieldwork and findings that don’t hold up under scrutiny.

Moderator craft: the part that doesn’t show up in the deliverable

The discussion guide sets the structure. The moderator determines whether the structure survives contact with an actual conversation, and this is the part of methodology that is hardest for a client to evaluate from outside, because a skilled moderator’s work is largely invisible — it looks like a natural conversation.

Three disciplines separate a moderator who protects data quality from one who doesn’t:

Probing without leading. “Why do you say that” is a probe. “Is that because of the side effect profile” is a leading question disguised as a probe — it supplies the answer and asks the respondent to confirm it. The distinction sounds small in isolation and compounds across a sixty-minute interview into a transcript that reflects the moderator’s hypothesis more than the respondent’s actual reasoning.

Tolerating silence. Physicians, like most respondents, often need a few seconds to retrieve the actual reason behind a decision rather than the socially easy answer. A moderator uncomfortable with a four-second pause fills it — with a follow-up, a rephrase, a multiple-choice prompt — and in doing so answers their own question for the respondent.

Managing the confident respondent without shutting them down. Some physicians, particularly senior KOLs, will answer a question by reframing it to one they’d rather discuss. A moderator who lacks the clinical fluency to recognize the reframe and redirect back to the actual research question loses the interview’s value without the client ever knowing it happened — the transcript still reads as an engaged, substantive conversation.

Failure modes that quietly wreck a transcript

Most qualitative data quality problems in commissioned healthcare research are not respondent problems. They are moderator and guide problems that never surface, because nothing about a bad transcript looks obviously bad.

Leading-guide transcript: Moderator shows a list of five product attributes in question one, asks the physician to rank them, then asks follow-up questions about the top-ranked item. The transcript reads as focused and on-topic. What’s actually missing: whether the physician would have named any of those five attributes unprompted, and whether a sixth, unlisted attribute matters more.

Funnel-guide transcript: Moderator asks what matters most, unaided, lets the physician struggle and self-correct for ten seconds, then probes the stated reasoning before introducing the same list of five attributes for comparison. The transcript is messier — hedges, false starts, a correction halfway through. What it protects: a defensible record of what the physician actually thinks, independent of what the study was designed to confirm.

The second transcript is harder to read and harder to pull a clean pull-quote from. It is also the one a client can trust when the finding drives a go/no-go decision.

What good methodology looks like from the outside

A buyer evaluating a research partner’s qualitative rigor without sitting in on the fieldwork can still check for specific, verifiable signals:

Questions worth asking before fieldwork starts

• Does the discussion guide ask unaided questions before introducing any list, concept, or stimulus?
• For any TDI or dyad, is there an explicit plan to solicit the second and third respondent’s independent view before group discussion converges?
• Who moderates — someone with enough clinical fluency in the specialty to recognize a reframe and redirect, or a generalist reading a script?
• Will raw transcripts (not just a synthesized summary) be available for review?
• Is there a stated plan for identifying and handling non-committal or socially-easy answers during the interview itself, not just in post-hoc analysis?

If a prospective partner cannot answer these specifically — not generically, but for the actual study design — that is a more reliable signal than any claim about interviewer experience on a credentials slide.

Why this is worth getting right before fieldwork, not after

Discussion guide structure and moderator assignment are decisions made before the first interview happens. Once fieldwork is underway, a guide that leads or a moderator who can’t hold a redirect has already shaped every transcript that follows it — there is no post-hoc analytical fix for a question that was asked in a way that supplied its own answer.

If you’re scoping a qualitative study and want a second opinion on discussion guide structure or IDI-versus-TDI design before fielding starts, talk to us. We’ll tell you plainly where a guide is likely to lead a respondent, before that costs you a finding you can’t use.