Supervision That Survives an Inspection
Part of Principal Investigator (PI) Training: Delegation and Supervision.
About 4 minutes. Free, in your language, no account needed.
Why this matters
"Failure to supervise" is one of the few findings written against the investigator personally rather than against the site. It is also, in a sense, unfalsifiable by assertion: you cannot answer it by saying you supervised. You answer it by showing what supervision consisted of and where the evidence is.
This lesson is about designing supervision that leaves that evidence behind as a by-product of doing it.
Proportionate to what
Section 2.3.2 sets the calibration: supervision proportionate to the importance of the data and the risk to participants. Those are two independent axes, and mapping your site's activities onto them takes about half an hour once.
| Low participant risk | High participant risk | |
|---|---|---|
| Low data importance | Spot-check occasionally — appointment scheduling, non-critical questionnaires | Direct oversight of the risk — a routine procedure in a frail participant |
| High data importance | Systematic check — primary endpoint entry, eligibility data | Investigator involvement close to real time — consent, dosing decisions, event assessment |
The point of the grid is not the categories. It is that supervision should be unequal. A site that reviews everything equally is spending its attention in the wrong place and will still miss the primary endpoint transcription error.
Four routines that constitute supervision
Supervision at a busy site is not vigilance. It is a small number of scheduled habits.
- The eligibility review. No participant is enrolled until you have personally reviewed the eligibility source and recorded your decision. This is Module 4 and it is the highest-value routine you have, because an ineligible enrollment cannot be undone.
- The weekly medical review. A fixed slot — thirty to sixty minutes — in which you go through new events, out-of-range laboratory values, concomitant medication changes and anything the coordinator has flagged. Fixed, so it happens; short, so it survives a bad week.
- The consent review. Within a defined window after each consent, you check the form version, the dates, the signatures, and the consent note. Ten minutes per participant.
- The data endorsement milestone. §2.12.5 — you review and endorse reported data at points agreed with the sponsor. Treat it as a real review, not a batch signature. Module 9.
None of these requires you to have been in the room. All of them leave a dated, attributable record.
Make the evidence a by-product
The classic mistake is to supervise diligently and prove nothing, because the supervision lived in conversations. Corridor oversight is real oversight and it is invisible six months later.
Cheap ways to make it visible:
- Initial and date the eligibility checklist yourself, with a one-line reason where any criterion needed judgment.
- Keep a dated note of the weekly medical review — even four lines listing what was reviewed and what action followed.
- Sign and date consent reviews on the source note rather than mentioning them verbally.
- Minute team meetings in three sentences: date, attendees, decisions.
Each takes under a minute and each converts an activity you were doing anyway into evidence.
When supervision finds something
It will, and that is the system working. A deviation found by your own review, documented, explained, and followed by a change that prevents recurrence, is exactly the pattern §2.5.3 describes. A site with zero self-identified deviations across a two-year trial is not a clean site; it is a site that is not looking.
Say that to your team explicitly, because the instinct to conceal is strong and it is the one habit that turns a small finding into a serious one.
Key takeaways
- §2.3.2 calibrates supervision to data importance and participant risk — so supervision should be deliberately unequal across activities.
- Four routines carry most of the weight: eligibility review, a weekly medical review, consent review, and data endorsement.
- Supervision that leaves no dated record cannot be shown; make the evidence a by-product with initials, short notes and brief minutes.
- Self-identified deviations are evidence of oversight, not of failure. Zero of them is the warning sign.