When a multi-site CBT trial sets up its training programme, the standard pattern is recognisable. The trial team adapts a manual, builds a two-day or five-day workshop, runs it through the available trainers, certifies the trial therapists, opens recruitment. The training is finished by the time the first participant is randomised. It is treated as a one-off step in the protocol, like ethics approval or site initiation visits — completed, ticked off, behind the team.
This works adequately when the intervention is well-rehearsed and the trial therapists are already fluent in the modality. It works less well when the intervention is novel, when therapists vary widely in baseline competence, when the manual is mid-revision, when the trial spans multiple sites with different supervisory cultures, or when the trial needs to retrain mid-recruitment because something has been learned in early sessions.
The training programme finalised in the planning phase and treated as closed is, in those conditions, a structural liability. It cannot adapt. It cannot deliver the manual update at pace. It cannot re-certify therapists efficiently when a competency drift is detected at one site. The trial absorbs the consequences as measurement contamination, and the field reads the consequences in the published effect size.
What "pivoting on a dime" actually means in trial training
The phrase gets used loosely. In trial-training terms, it has a specific operational meaning. The training programme needs to be able to respond, within weeks rather than quarters, to several distinct triggers.
A manual revision. The trial protocol committee has approved an update to the manual — a clarified description of an intervention component, a revised dose specification, a tightened criterion for moving between phases of the protocol. Trial therapists need to be re-trained against the revised version, re-certified, and demonstrably aware that the v1.1 protocol now supersedes the v1.0 protocol they were originally certified against. The re-training cannot wait three months for the next training cycle.
A drift signal at site or therapist level. The fidelity monitoring layer — the subject of the continuous fidelity infrastructure piece — has detected that a site, or a specific subset of therapists, is delivering one intervention component at lower intensity than the protocol specifies. The response is targeted booster training for the affected therapists, plus supervision input at the site, plus re-rating of those therapists' subsequent sessions to confirm correction. The standard training programme has no mechanism for this. The targeted intervention has to be built ad hoc, often with the trial therapists' time competing against ongoing recruitment.
A new cohort of trial therapists arriving mid-recruitment. Trials are not closed systems. Therapists leave (parental leave, role changes, attrition). New therapists need to be trained to the same standard as the original cohort, against the current version of the manual, and certified before they take on trial participants. If the standard training programme was a five-day in-person workshop delivered once at the start of the trial, the trial team is now in the position of either running a bespoke abbreviated version for the new cohort, or asking the new therapists to wait until the next centrally-organised training event, which may be six months away. Neither option is good.
A learning from the early treatment cohort. Sometimes the protocol-design choices that looked sensible in the planning phase produce something unexpected in the first treatment cohort. A homework component is being under-delivered everywhere because it conflicts with the session structure. The exposure sequence in week three is producing higher dropout than the trial team expected. The trial team meets, reviews, and decides that the manual needs adjustment and the training needs to incorporate the lesson. The training programme that was finalised pre-recruitment has no mechanism for absorbing this. The adjustment lands as an emergency.
None of these triggers are unusual. Most multi-site CBT trials encounter at least one of them during recruitment. The standard training-programme architecture treats each as an exceptional event to be managed by the trial coordinator on top of their existing role. A training architecture designed for the contemporary trial design treats them as routine operations that the training infrastructure absorbs without delay.
What that requires structurally
The infrastructure needed to support adaptive trial training is not particularly exotic. It is, however, different in kind from a standard CPD training portal.
Modular content that can be re-sequenced and updated. Not a five-day workshop as a single artefact, but a modular training architecture where each protocol component has its own training module, each module is versioned, and the training delivered to any cohort of trial therapists is the current version of each component module. When a manual revision affects one component, only that module needs updating, and the re-training to be delivered is the updated module rather than the whole workshop.
Certification gates that distinguish manual versions. "Trained" is not a sufficient certification state. The relevant state is "trained and certified against version X.Y of the protocol manual." When the manual is revised, therapists certified against v1.0 are no longer current; they need re-certification against v1.1, and the trial's fidelity rating capability needs to know which version of the manual to apply when rating any given session.
Rapid delivery channels. Not only live workshops. A training architecture that depends entirely on synchronous in-person delivery cannot pivot at the pace the trial requires. The standard mix that works is asynchronous video content for the didactic component (delivered when the therapist can take it, with completion tracking), supplemented by short live sessions for the skills-rehearsal component that requires interactive practice, and individual case-based feedback within the trial's supervisory structure. Borrelli and colleagues' (2005) framework for treatment fidelity identifies training as one of the five domains that need active monitoring throughout the trial, not just at the start.
CTS-R rating capability tied to the specific manual version. The fidelity rating layer cannot be version-agnostic. When the manual changes, the rating rubric needs to change with it — or, more usefully, the rating instrument needs to know which version of the manual is in effect for the session being rated, so that "this is not in the manual" assessments are made against the correct manual. Bearman and colleagues' (2022) work on adherence measurement has demonstrated that the rating instrument matters as much as the rater for producing accurate adherence data, and version-mismatch between manual and rating instrument is one of the cleanest ways to produce noise in the fidelity signal.
Supervision infrastructure that can absorb the re-certification cohort. When a re-training event produces a group of therapists who all need re-certification supervision in the same fortnight, the trial's supervisory structure has to absorb the load without breaking. Site supervisors are typically the bottleneck. The training architecture either supports a temporary reallocation of supervisory capacity, or the re-certification queues up and the trial absorbs the delay. The supervision infrastructure that the trial uses for routine fidelity work — the same infrastructure that Liness and colleagues' (2019) supervision-improves-CTS-R findings support — needs to be capable of absorbing these bursts.
What trials usually have instead
The standard trial training arrangement is structurally different from the above in several ways.
It is built as a single training package, delivered once, in a workshop format that is difficult to re-sequence or update. The certification state is binary: trained or not. Re-training is treated as an unbudgeted emergency rather than a routine operation. The fidelity rating layer is set up against the manual version at trial launch and is rarely updated when the manual is revised. The supervisory structure is whatever the participating sites already had, with the trial's bespoke supervisory requirements added on top.
The result, when the trial encounters one of the adaptive triggers above, is that the response is improvised. The trial coordinator builds an ad hoc re-training event, often on a short timeline, with limited capacity to certify the participants in any rigorous way. The therapists' re-certification, if it happens at all, is a checkbox exercise rather than a substantive demonstration of competency against the revised protocol. The fidelity rating continues against the old rubric because updating it would require coordination the trial does not have time for. Drift goes uncorrected because the corrective infrastructure was not built into the trial design.
This is not, generally, a failure of the trial team's competence. It is a failure of the underlying training infrastructure to match the demands the contemporary trial design places on it. The trial team is being asked to run a training programme that needs to be adaptive, version-tracked, and integrated with fidelity monitoring, using infrastructure designed for a standard one-off workshop. The infrastructure cannot do the job.
Why repurposing a generic CPD portal does not work
The natural temptation, when a trial team realises they need a more capable training architecture, is to use the off-the-shelf CPD or learning-management infrastructure that the host institution already has. This is rarely a good fit, for reasons that are worth being explicit about.
Reuse pattern is different. A generic CPD portal is designed for content that is used across many learners over long time horizons. The investment in producing a module is amortised across years of use. Trial-specific training is the opposite: high investment in module production, used by a small cohort of therapists for the duration of the trial, then archived. The economics that make a generic CPD portal sustainable do not apply.
Certification model is different. Generic CPD certification is typically of the form "the learner completed the module and passed the assessment." Trial certification is of the form "the therapist is certified to deliver version X.Y of the protocol's intervention in the context of trial N, with re-certification triggered by any of the following events." The state model is more complex and is tied to the trial's protocol architecture rather than to the institution's CPD framework.
Integration with fidelity monitoring is different. A generic CPD portal does not, generally, integrate with a CTS-R rating workflow, a site-level fidelity dashboard, or a supervision-records system. Trial-specific training is operationally meaningless without those integrations — the training is the upstream input to the fidelity monitoring loop, and disconnecting them produces a training programme that satisfies the workshop step but has no corrective loop with what is actually being delivered in session.
Versioning model is different. Generic CPD modules are versioned occasionally, with the latest version superseding the previous one for all learners. Trial protocols are versioned on the trial's timeline, with therapists certified against specific versions, sessions rated against specific versions, and the version applicable to any given session determined by when that session occurred relative to the protocol-revision dates. The bookkeeping is materially different.
Trying to repurpose a generic training portal generally produces a training programme that satisfies the workshop step (therapists have been trained, and there is a record of it) but does not satisfy the ongoing operational requirements (mid-trial re-certification, version tracking, fidelity-rating integration, drift response). The trial team ends up running the substantive operational work outside the portal, which means the portal is doing the easy part and the team is doing the hard part by hand.
The strategic point
Trial-specific training infrastructure is not the same problem as standard CPD infrastructure, and not the same problem as one-off workshop delivery. It is a distinct infrastructure problem that the trial-design community has been managing without adequate tooling for some time. The trials that build adaptive training architecture into the design from the protocol stage produce trials that can absorb manual revisions, respond to drift signals, and re-certify cohorts mid-recruitment without delay. The trials that do not absorb the cost as measurement contamination — drift that goes uncorrected, fidelity that decays through the trial window, effect sizes that underrepresent the intervention's actual efficacy.
The connection to the drift detection in multi-site trials piece is direct. Drift detection produces a signal; the response to the signal is training. If the training architecture cannot deliver a response, the signal is not actionable, and the detection layer becomes a documenting-the-problem layer rather than a correcting-the-problem one. The two infrastructures only work together.
The investment case is straightforward when laid out. Trials cost millions per arm. The marginal cost of building adaptive training infrastructure as part of the trial design is small relative to the total. The cost of not building it — drift uncorrected, effect sizes contaminated, trials that under-report the intervention they are testing — is large and durable, both for the trial team and for the evidence base that subsequent meta-analyses will inherit.
The standard pattern — single training package, delivered once, treated as closed — is not adequate to contemporary trial designs. The pattern that is adequate looks different in shape, and the difference is not stylistic. It is operational, and it is the difference between a trial that can pivot and a trial that absorbs the consequences of not being able to.
Supervisia Research builds bespoke trial training programmes.
Research delivers content modules matched to the trial's manual, with version-tracked certification gates so the trial knows which therapists are current against which protocol version. The training architecture integrates with the fidelity-monitoring layer the same platform provides — CTS-R rating against the specific manual version, drift dashboards that surface signals at therapist and site level, supervision infrastructure that absorbs the re-certification load when a manual revision lands. The training programme stays adaptive across the recruitment window rather than freezing at trial launch.
References
- Borrelli, B., Sepinwall, D., Ernst, D., Bellg, A. J., Czajkowski, S., Breger, R., DeFrancesco, C., Levesque, C., Sharp, D. L., Ogedegbe, G., Resnick, B. & Orwig, D. (2005). A new tool to assess treatment fidelity and evaluation of treatment fidelity across 10 years of health behavior research. Journal of Consulting and Clinical Psychology, 73(5), 852–860. DOI: 10.1037/0022-006X.73.5.852. PubMed: 16287385.
- Bearman, S. K. et al. (2022). A randomized trial to identify accurate measurement methods for adherence to cognitive-behavioral therapy. PubMed: 36229116.
- Liness, S. et al. (2019). Clinical supervision in cognitive behavior therapy improves therapists' competence: A single-case experimental pilot study. PubMed: 32213046.
- Waller, G. & Turner, H. (2016). Therapist drift redux: Why well-meaning clinicians fail to deliver evidence-based therapy, and how to get back on track. Behaviour Research and Therapy, 77, 129–137. DOI: 10.1016/j.brat.2016.01.007. PubMed: 26752326.
Last updated: May 2026