Site selection is the most consequential decision in any IVD study, and one of the least rigorously made. Most CROs treat it as a regulatory and logistical exercise — does the site have IRB capacity, does it have the right equipment, does the PI have the right credentials, has the feasibility form come back affirmative — and then activate the sites that pass those filters. The result is an enrollment plan that depends on whether the activated sites can convert paper qualification into actual subjects, which is a question no one has yet answered when the contract is signed.
The pattern is consistent enough across rescue engagements that it is worth stating plainly. The single biggest source of avoidable rescue work, on the studies we inherit, is bad site selection at study start. Sites that qualified were not the sites that enrolled. The CRO that activated them did the regulatory work correctly and the predictive work not at all. By the time the gap became visible — typically three to six months in — the contract structure made it expensive to swap sites, the timelines had already slipped, and the sponsor had paid for activation work that produced no subjects.
This framework describes how to avoid that pattern. None of the four principles below replace qualification work; qualification remains a hard filter. They describe what to do after the qualification filter has reduced the candidate list, when the actual selection happens.
Qualification is a filter, not a criterion.
Every site under consideration must qualify. The ones that will enroll are a much smaller subset, and qualification does not predict which subset.
The most basic site-selection error is to treat qualification as the answer rather than the precondition. A site that has IRB capacity, the right equipment, the right credentials, and a signed feasibility form has cleared a series of hurdles that disqualify other candidates. It has not, by clearing those hurdles, demonstrated that it will produce subjects. Qualification predicts the absence of disqualifying conditions, not the presence of enrolling conditions.
This sounds tautological, but the consequences are not. CROs that treat qualification as the criterion produce site lists optimized for paperwork completion — the sites that respond fastest to feasibility outreach, the sites with the most polished regulatory operations, the sites embedded in the SMO networks the CRO already works with. None of these characteristics correlate strongly with enrollment performance on any specific study, and several of them correlate negatively, because polished regulatory operations are often a sign of a site that runs many studies and has finite enrollment capacity to allocate across them.
Treat the qualification filter as removing the lower 30–50% of candidates. Treat the remaining candidates as the population from which the site list will be selected on substantive criteria — the next three principles.
Validate eligible pool against observed care, not feasibility forms.
Feasibility forms are sales documents. Observed care tells you what the practice actually sees. The two systematically diverge.
The feasibility form is a useful tool for confirming logistics, equipment, and PI willingness. It is not a useful tool for predicting eligible patient counts, especially for cohorts the site does not routinely see. Sites overstate eligible counts as a matter of course — not because they are dishonest, but because the question is asked of staff who don't have the data and answered through the lens of wanting the work. The forms come back optimistic. The reality, when enrollment begins, is consistently below what the forms predicted.
The corrective is not to interrogate forms harder. It is to validate every candidate site against independent evidence of what that practice actually does in routine care — who walks in, what gets diagnosed, what gets ordered — not against what the site imagines it could recruit if the right study came along. The gap between feasibility-form predictions and observed reality is often large enough to decide whether a study enrolls.
On rescue engagements we have repeatedly inherited site lists where the prior CRO had relied on feasibility-form responses. Independent review of those same sites against observed care consistently shows actual eligible-patient volumes well below the feasibility predictions, particularly for cohorts that diverge from the practice's primary patient population. The feasibility forms are not wrong about whether the site can run the study. They are wrong about how many subjects the site can find.
Before any candidate site enters the active list, validate its claimed eligible pool against independent evidence of routine care for the target indication. If that evidence is materially below the feasibility-form prediction, treat the lower number as the planning input.
PI engagement is observable before the contract is signed.
A PI who is not personally responsive during pre-contract diligence will not be personally responsive once the contract is in place.
The strongest single behavioral signal we have found for site enrollment performance is the PI's personal engagement during the pre-contract phase. Sites where the PI takes the diligence call directly, asks substantive questions about the protocol, and follows up promptly on outstanding items consistently outperform sites where the PI is represented entirely by site-management staff during diligence and only appears at the SIV. The pre-contract phase is the audition. The behavior the PI demonstrates during diligence is the behavior the site will demonstrate during enrollment.
This is not a paperwork test. It is an attention test. PIs who run multiple studies allocate their attention deliberately, and the studies that get their attention are the ones they are personally interested in. A PI who delegates diligence entirely to staff is signaling that this study is not in the top tier of their attention budget — and the enrollment performance, six months in, will reflect that allocation. Attention is the rate-limiting resource at the site, not capacity.
Require a direct conversation with the PI as part of every site diligence process. If the PI is unavailable, require their personal written engagement with at least two substantive protocol questions before the site advances. Sites where the PI cannot or will not engage personally at this stage should be filtered out before contract.
Drop non-enrolling sites at 30 days, not 90.
A site that hasn't enrolled in the first 30 days under active engagement will not enroll later. The cost of waiting is paid in months.
Site selection does not end at activation. It continues through the first 30 days of active enrollment, during which the activated sites either produce subjects or do not. The CRO's discipline at the 30-day mark — what to do about sites that have produced nothing — is the difference between a study that recovers and one that drifts.
The conventional response is to wait. Sites that have not yet enrolled at 30 days might still enroll; activation is a process, the enrollment campaign is just ramping up, the recruitment vendors are just hitting their stride. Sometimes this is true. More often, a site that has not enrolled in the first 30 days under active engagement is signaling something durable — the patient panel doesn't actually contain the eligible population the feasibility form predicted, the PI is over-allocated to other studies, the staff isn't equipped to manage the consent flow. The behaviors that produce zero enrollment in month one are the behaviors that produce zero enrollment in month four.
Acting on this signal at 30 days is operationally hard. Contracts are signed, activation costs are sunk, and the conventional wisdom is that more time will fix the problem. The right response is to move faster than the conventional wisdom suggests. Replace non-enrolling sites at the 30-day mark with sites that meet the criteria above, and absorb the activation cost as the price of an enrolling study. Studies that do this hit timelines. Studies that wait until 90 days or beyond rarely recover.
Build the 30-day site review into the study calendar at contract. At day 30 of active enrollment per site, sites with zero subjects enrolled trigger a structured review: PI conversation, eligible-pool re-check, recruitment vendor alignment. Sites that don't produce a credible recovery plan within two weeks of that review should be replaced.
What this framework rules out.
The four principles describe how to select sites that actually enroll. They also rule out a few things the industry takes for granted.
They rule out large pre-emptive site networks as a competitive advantage. CROs that brag about thousands of qualified sites are advertising a filter, not a criterion. The same site lists are available to multiple CROs through the same site management organizations; the differentiator is the selection discipline applied on top of the network, not the network itself.
They rule out feasibility-form-based site lists as the primary planning input. Feasibility forms confirm logistics. Observed care and PI engagement determine performance. A site list compiled from feasibility forms alone is unvalidated against the only metrics that matter.
They rule out passive activation as a path to enrollment. Activating ten sites and waiting to see which ones produce is a way of paying for nine sites' worth of activation cost while waiting for one or two to do the work. The 30-day discipline replaces passivity with active selection during the first month of execution.
The framework is not closed. When the study outcome matters, you call RDI. Site selection is where the study is won or lost — well before the first subject enrolls.