Monitoring Plans: Splitting Central, Remote and On-Site Work
The useful question is not how many on-site visits a study needs. It is what each check requires in order to be possible: data across all sites, a specific document at one site, or someone standing in the pharmacy. Sorting the checks that way produces a monitoring plan. Setting a visit frequency and filling the days produces a schedule.
Most plans are still written the second way, then described as risk-based because the visit interval was lengthened. The reduction is real and the method has not changed, which is the version that generates findings. This post covers the allocation decision; the wider structure sits in the risk-based quality management plan.
Remote Versus On Site Is the Wrong Axis
Framing the plan as a choice between remote and on-site work leaves out the layer that changed monitoring most, which is central review. Central review is not remote monitoring done from a desk. It operates on accumulating data across every site at once, and it sees things that are structurally invisible from inside any single site.
A site whose data is unusually clean, an enrollment pattern that diverges from peers, query aging that clusters in one region or one language: none of these can be found by visiting the site in question, because the comparison is the finding. That is why a reduced-visit plan without a central component has removed effort rather than redistributed it.
The counterpart limitation is worth stating in the plan too. Central review cannot see anything that never reached the data. An adverse event recorded in the chart and not in the case report form is invisible to every dashboard, which is precisely why comparative reporting rates matter: the absence shows up as a site reporting less than its peers.
Allocating a Check to a Layer
The plan becomes usable when each check names its layer and the reason it cannot sit in a cheaper one. On-site time is the scarce resource, so the default is to push a check down until it stops working.
Two rows in that table are worth dwelling on. Adverse event under-reporting is a good illustration of layers working together: the central comparison finds the candidate, a targeted review confirms or dismisses it, and only an unresolved case needs a visit. Running it the other way, by hoping a monitor notices during source review, is how under-reporting persists for a year.
Protocol understanding is the opposite case. It does not come out of records. It comes out of asking a coordinator how they handle a particular situation and hearing an answer that differs from the protocol. No amount of remote document access substitutes for that, which is a reason to keep some presence even at sites whose data looks perfect.
What Remote Review Actually Requires
Remote review is often written into a plan without checking whether it is operationally possible at these sites. It depends on scoped access to site documents, and the permission model is where it succeeds or fails: too restrictive and the monitor cannot do the check, too open and document control is lost.
That tension, and the settings teams commonly get wrong, are covered in giving monitors access without losing document control. The planning consequence is simple: if remote review of a given document type is not actually permitted at a site, the check has to move up to an on-site visit, and the plan should say so rather than assume.
What Pulls a Visit Forward
A calendar interval treats every site and every month as equivalent. Triggers treat them as different, which is the point of the exercise.
Two of these are consistently undervalued. The first-participants visit is the highest-return monitoring activity in most studies, because a misunderstanding caught at the second participant affects two records rather than twenty. The reasoning is the same as for front-loaded source checking, covered in what changes when you stop verifying every field.
Staff turnover is the other. A coordinator leaving takes the site's working knowledge of the protocol with them, and nothing in the data shows it until errors start appearing weeks later. Treating turnover as a visit trigger catches it at the handover instead.
Writing the Plan So It Can Be Followed
Three things make a monitoring plan auditable. Each check names its layer and the reason. The triggers are defined rather than described as judgment. And there is a stated route for what happens when a layer fails to close a question, so escalation is a step rather than an improvisation.
The last of these is where most plans are thin. A central signal that remote review cannot resolve needs a named next step and a named owner, otherwise it stays open in a tracker and surfaces at close-out. The systems carrying visit records, trip reports and follow-up are covered in which system owns which part of the study, with the platform side in Clinera CTMS and Clinera eISF.
References
ICH E6(R3), Good Clinical Practice, which describes monitoring approaches including centralized monitoring and expects the approach to be justified in the monitoring plan. The FDA guidance Oversight of Clinical Investigations, A Risk-Based Approach to Monitoring, August 2013, and its 2019 questions and answers, which address remote and centralized methods directly. The EMA reflection paper on risk-based quality management remains relevant for European studies. ICH E6(R3) Annex 2, covering decentralized elements, reached Step 4 in June 2026 with an effective date of 15 January 2027 and is worth planning against.
Current texts are at www.ich.org, www.fda.gov and www.ema.europa.eu.
Frequently Asked Questions
What is the difference between centralized and remote monitoring?
Centralized monitoring works on accumulating data across every site at once, so its findings are comparative: an outlier site, an unusual distribution, query aging clustered in one region. Remote monitoring works on one site at a time through scoped document access, and it usually follows a central signal or handles document checks. Treating them as one thing loses the layer that finds problems no visit can.
How many on-site visits does a risk-based plan need?
There is no correct number, which is the point. The plan defines what pulls a visit forward: site activation and first participants, a central signal that remote review cannot close, staff turnover, an amendment that changes site workflow, and close-out. A study where every site is visited at the same interval regardless of behavior has a schedule rather than a risk-based plan, whatever the interval.
What can only be checked on site?
Anything requiring physical presence or conversation. Investigational product accountability and storage conditions, paper source that was never digitized, facilities and equipment calibration, and the understanding of the protocol held by site staff. The last one matters most and is easiest to skip, because it comes out of asking a coordinator how they handle a situation rather than out of reviewing any record.
Can centralized monitoring detect under-reported adverse events?
Indirectly, and that is usually the only way it gets found. An event recorded in the chart but not in the case report form is invisible to any dashboard looking at the case report form. What is visible is that one site reports events at a rate well below its peers, which is a candidate signal rather than a finding. A targeted review then confirms or dismisses it, and an unresolved case justifies a visit.
Does remote monitoring need an eISF?
It needs a way to give a monitor scoped access to site documents without losing document control, and an electronic investigator site file is the usual answer. The planning consequence is practical: if remote review of a document type is not actually permitted at a site, that check has to move up to an on-site visit. Writing remote review into a plan without confirming the access model is a common way plans fail in month two.
How should the monitoring plan handle escalation?
As a defined step with a named owner rather than as judgment. When a central signal cannot be closed by remote review, the plan should say what happens next, who decides, and within what period. Where this is thin, unresolved signals sit in a tracker and reappear at close-out, which is both the most expensive point to address them and the hardest position from which to explain the delay.
