25.9 — How a Trial Actually Runs

A Phase III trial in the middle of enrolment looks like this on any given Tuesday. In Warsaw, a nurse is explaining a twenty-two-page consent form to a patient who does not want to read it. In New Jersey, a data manager is chasing a laboratory value that arrived as 4,500 when the plausible range is 4.5. In Bangalore, a monitor is comparing a hospital's paper notes against what the site typed into the database, line by line. In Tokyo, a shipment of study drug is stuck at customs and the site has two patients due next week. And in a conference room somewhere, a project manager is explaining to the sponsor why the timeline moved again.

Every one of those people is touching a system. Knowing who they are, what they are accountable for, and which system they live in is the difference between building something a trial actually uses and building something that gets worked around.

The cast, by name and accountability

The sponsor is the organisation that owns the trial and is legally responsible for it — usually the pharmaceutical or biotech company. Legal responsibility cannot be outsourced. A sponsor that hands everything to a CRO is still the party a regulator holds accountable.

The CRO — contract research organisation — does the work under contract. Full-service CROs run the entire trial; functional service providers supply one function, such as monitoring or data management, embedded in the sponsor's own process. Which model a client uses changes everything about how your systems must be built and who administers them.

The site is a hospital, clinic or research unit where patients are actually seen. A trial has anywhere from one site to several hundred.

The principal investigator, or PI, is the doctor at each site who is personally responsible for the conduct of the trial there and for the safety of the patients in it. In the United States they sign a Form FDA 1572, a personal commitment to follow the protocol and the regulations. Sub-investigators are the other clinicians they formally delegate work to, recorded on a delegation log that an inspector will ask for.

The study coordinator, often CRC for clinical research coordinator, is the nurse or research professional who does most of the actual work at the site: scheduling patients, taking consent, collecting samples, entering data, answering queries. They are the single most important relationship in trial operations, and almost every complaint about a badly designed system comes from them first.

The monitor, formally the clinical research associate or CRA, works for the sponsor or CRO and visits sites to check that the protocol is being followed, that consent was properly taken, that the data matches the medical record, and that the drug is being stored and accounted for correctly. They are the mechanism by which a sponsor knows what is happening in a hospital it does not control.

The medical monitor is a physician at the sponsor or CRO who answers clinical questions from sites, reviews safety events as they come in, and decides whether a patient may continue.

Behind them: project managers, data managers, biostatisticians, pharmacovigilance staff, regulatory specialists, supply chain planners, and the vendors running the individual systems.

And two independent bodies that the sponsor cannot instruct. The IRB or Ethics Committee approves the study before it starts, approves every amendment and every piece of material shown to patients, and can suspend it. The Data Safety Monitoring Board — covered in Chapter 25.10 — looks at the unblinded results during the trial and can recommend stopping it.

The documents everything hangs on

The protocol is the contract. It states the objectives, who may be enrolled and who may not, the treatment, every visit and every assessment at each visit, the endpoints, the statistical plan in outline, and the safety rules. Anything not in the protocol may not be done, and anything in it that is not done is a deviation.

Changing it requires an amendment, which must be approved by the ethics committee and often by regulators before it takes effect. This is why "can we just add a field?" is never a small question in a clinical system: if the field collects something the protocol does not authorise, the answer is no.

The Investigator's Brochure carries everything known about the drug so far, and defines what side effects are expected — the reference that decides which safety reports are urgent (Chapter 25.22).

The informed consent form must explain, in language the participant can understand, what the study involves, its risks, its potential benefits, the alternatives to taking part, how their data will be used, and their absolute right to withdraw at any time without affecting their care. Consent must be given before any study procedure, freely, and documented. Consent failures are among the most serious findings an inspector can make, because they go to whether the patient agreed at all.

The case report form, or CRF, is the structured record of what happened to each patient — today almost always an electronic form, the eCRF.

And the trial master file, the TMF, now usually an eTMF, is the complete evidence file for the trial. The governing principle is that the TMF must allow an inspector, years later, to reconstruct how the trial was conducted and how the data was produced. A missing document in the TMF is treated as work that was not done.

The rulebook: GCP, and what changed in 2025

Good Clinical Practice is the international standard for designing, running, recording and reporting trials involving humans. It exists to protect participants and to make results credible, and the reference document is ICH E6.

The revision matters because it landed recently and clients are still adapting. ICH adopted E6(R3) on 6 January 2025. The European Medicines Agency made it effective on 23 July 2025. The FDA published it as final guidance on 9 September 2025. Note the difference: an FDA guidance states the agency's expectations and is not itself law, whereas the European route makes it a binding standard — a distinction worth knowing before you assert what is "required" in a client meeting.

What R3 actually changed, in plain terms. It puts quality by design at the front: identify the small number of factors that genuinely determine whether the trial's results and the participants' safety hold up — the critical to quality factors — and concentrate effort there instead of applying equal rigour to everything. It is written to be technology-neutral, so decentralised visits, electronic consent and data flowing directly from devices fit the framework rather than being exceptions. And it strengthens expectations around data governance across the whole life of the data.

The practical consequence for engineers is that "we check everything" is no longer the expected answer, and neither is "we check nothing". The expected answer is a documented, risk-based reason for what is checked, how much, and why — which is exactly the kind of argument a well-instrumented system can support and a spreadsheet cannot.

The life of a site, start to finish

Feasibility. Before anything, the sponsor asks candidate sites whether they have the patients, the equipment and the staff. Sites systematically overestimate how many patients they can enrol; experienced sponsors discount the answer heavily.

Selection and contracting. Budget negotiation, the clinical trial agreement, and confirmation of ethics approval. This step is slow and is a frequent cause of delay.

Site initiation. A formal visit or meeting where the site team is trained on the protocol, the systems and their responsibilities, and it is documented that they were. Nobody may work on the study before they are trained and delegated, and the delegation log proves it.

Enrolment. Patients are screened against the eligibility criteria, consented, and randomised. First patient in is a milestone the whole industry tracks; last patient out is the one that determines when results exist.

Monitoring during conduct. Visits at defined intervals, plus remote review. Historically monitors compared every field in the database against the hospital's records — source data verification — which is enormously expensive. The modern approach, now explicitly supported by GCP, is risk-based: verify the critical data completely, sample the rest, and use centralised statistical monitoring to spot sites whose data looks unusual.

Close-out. All queries resolved, all drug accounted for and returned or destroyed with records, all documents filed, and the site formally closed. Records must then be retained for years — commonly a minimum of two years after the last approval in any region, and in practice far longer, with some jurisdictions requiring twenty-five years.

How the data actually moves

Follow one blood pressure reading from a patient's arm to a regulatory submission, because every clinical data system exists somewhere on this path.

The nurse measures it and writes it in the hospital record. That record is the source data — the first place it was recorded, and the thing a monitor will check against. Source data may not be altered to match the database; the database is corrected to match the source.

The coordinator enters it into the electronic data capture system, the EDC. As it is entered, automatic edit checks fire: is it within a plausible range, is it consistent with the previous visit, is a required field missing.

Anything questionable becomes a query — a formal question back to the site, tracked, answered and closed, with the whole exchange kept. Query volume is a real operational metric; a badly designed form generates thousands of unnecessary queries and burns site goodwill.

Meanwhile other data arrives from elsewhere: laboratory results transferred electronically from central labs, imaging read centrally, patient-reported outcomes entered by patients on their own devices, and readings from wearables. Each of these transfers is governed by a data transfer agreement specifying format, frequency, and how the two sides reconcile counts.

Adverse events and medications are coded onto standard dictionaries — MedDRA for medical terms and WHO Drug for medicines — so that the same event described in five languages counts as one thing (Chapter 25.22).

When enrolment is complete and every query is closed, the database is locked. After lock, nothing changes without a formal, documented, approved unlock. Then the treatment assignments are revealed, the pre-specified analyses are run, and the results become the report described in Chapter 25.10.

The systems, by their industry names

You will be handed these acronyms on day one of any clinical project.

System	What it does
EDC	Captures patient data from sites
CTMS	Tracks sites, visits, milestones, payments
eTMF	Holds the trial's evidence documents
IRT / RTSM	Randomises patients, manages drug supply
eCOA / ePRO	Collects data from patients directly
eConsent	Presents and records informed consent
Safety database	Holds adverse event cases and reporting
CDR	Combines all study data for review

The randomisation system deserves a note because it is the one most often misunderstood. IRT — interactive response technology, also called RTSM for randomisation and trial supply management — does two jobs at once. It assigns each patient to a treatment group according to the randomisation scheme, and it tells the site which numbered medication kit to dispense. It therefore holds the blind. It also predicts and triggers drug resupply to each site, which is why a fault in it can simultaneously break the science and empty the pharmacy shelf.

All of these are GxP systems. All of them need validation (Chapter 25.21), audit trails and controlled access (Chapter 25.20). A change to any of them mid-study is a controlled change, because the study data must remain interpretable across the change.

Where trials go wrong, and what an inspector looks for

Regulators inspect trial sites and sponsors, and the findings repeat with remarkable consistency across decades.

Consent problems — taken after a procedure, taken on an outdated form, not re-taken after an amendment that changed the risks.

Protocol deviations — patients enrolled who did not meet the criteria, visits outside their allowed windows, prohibited medications taken.

Data that does not match the source, or source data that does not exist because it was only ever entered into the database.

Drug accountability gaps — kits unaccounted for, storage temperatures out of range without an assessment of whether the product was still usable.

Delegation and training gaps — work done by someone not trained or not delegated to do it.

And record-keeping failures, which in modern trials usually means an incomplete eTMF or an audit trail that cannot show who changed what and when.

Notice that none of these are scientific errors. Almost every serious inspection finding is a process and documentation failure, which is precisely why this industry buys so much software and why the software is judged on evidence rather than on features.

Next: Chapter 25.10, the science of deciding whether the trial worked — endpoints, randomisation, blinding, statistics, and the committee that can stop a study before it finishes.