Self sampling in a clinical trial is a protocol decision, not a kit decision. Most teams run it the other way. They pick a collection device, then work out how it fits the study, then spend six months rewriting the protocol around a kit they already bought. Build it in the right order and most of the hard problems resolve before anyone opens a purchase order.
Start with what the sample has to prove
The first question is not which device. It is what the data is for. A sample supporting a primary efficacy endpoint carries a different burden than one supporting a safety assessment or an exploratory objective, and the controls should scale to match.
ICH E6(R3), adopted by the ICH Assembly on 6 January 2025, makes this explicit. It asks sponsors to identify the factors critical to the quality of the trial during planning, and to manage risks to those factors in proportion to how much they matter. Lighter controls where the risk is low. Concentrated effort where the data actually decides something.
Apply that to self sampling honestly. If a variable collection time or an occasional insufficient volume would compromise the endpoint, self sampling may be the wrong instrument for that endpoint no matter what it does for recruitment. Better to find that out in protocol design than in a data review meeting.
Write self sampling into the protocol before you source a kit
FDA’s final guidance on conducting clinical trials with decentralized elements, issued in September 2024, is the practical starting point. It expects the protocol to state which activities happen at a traditional trial site and which happen elsewhere, whether the participant may choose, and which personnel perform which tasks. It also asks for detailed instructions covering activities performed outside a site, specifically to reduce variability in the data those activities produce.
All of that can be written before a kit exists, and writing it first means the protocol constrains the kit instead of the kit constraining the protocol. It forces answers to the questions that otherwise surface late. What is the collection window. What happens when a participant misses it. Who does the participant call. What does the investigator review, and when.
Sponsors also have to keep records of the service providers doing decentralized work and the roles assigned to each. Investigators still supervise everything they delegate, review what comes back from remote personnel and local providers, and follow up on abnormal findings. Self sampling moves the collection. It does not move the responsibility.
The person collecting is untrained, and that is a design input
Participants are not phlebotomists and should not need to be. Published work suggests that is workable when the kit is built for it. A 2024 study in Sexually Transmitted Infections gave self sampling kits to people with no healthcare training and compared their samples against professionally collected samples from the same participants. Agreement between the two was high, about one percent of samples came back invalid or unusable, and more than ninety four percent of participants reported that they understood the kit instructions.
Read that as a finding about instructions, not about devices. The usable sample rate traveled with comprehension. Instruction design is the intervention, it is the cheapest element in the entire program to get right, and it is the one most often left until the end.
Three consequences worth building into the plan:
- Write the instructions against the kit as packed, in the order the participant will actually touch the components.
- Usability test with people who resemble the enrolled population, not with staff who already know the answer.
- Version the instructions as a controlled document, because changing them mid study changes the collection process.
Capture the data around the sample, not only the sample
A tube arriving at a laboratory is not the record. ICH E6(R3) expects data to be attributable, legible, contemporaneous, original and accurate across the full lifecycle, including data originating with participants and with third parties, and it expects the systems collecting that data to be fit for purpose and validated in proportion to risk.
For self sampling, that means deciding in advance where the collection timestamp comes from. A participant typing a time into an app six hours later is not contemporaneous. The realistic options are an app entry made at the moment of collection, a scan of a kit barcode, or a timestamp generated when the return label is inducted. They differ in reliability, and one of them has to be named in the protocol rather than assumed.
The same applies to identity. The link between a kit and a participant has to hold without depending on anyone writing legibly on a small label in their own kitchen.
Name the receiving laboratory early
FDA’s guidance separates routine, well standardized testing, which local clinical laboratory facilities can reasonably handle, from specialized testing, which belongs at a designated laboratory. Investigators have to document which facility was used, including the local provider’s name and the date of the activity.
Settle this before kit design, because the receiving laboratory sets the tube, the fill volume, the preservation approach, the acceptable transit window and the rejection criteria. A kit designed before that conversation is a kit that gets redesigned after it.
Whether a given collection device or assay is acceptable for your regulatory pathway is a question for your own regulatory lead. It is not answered by a packaging supplier, and it is not answered by a kit manufacturer.
Define the deviations before you enroll
Self sampling creates a class of event that site collection does not. The participant who collects on the wrong day. The participant who collects twice. The participant who collects too little. The participant who collects perfectly and then ships four days later.
Decide in advance which of those is a protocol deviation, which is a data point carrying a flag, and which is nothing at all. Put the tolerance around collection timing in the protocol instead of adjudicating it case by case. The risk proportionate framing in ICH E6(R3) supports pre specified acceptable ranges, and pre specifying them costs far less than a retrospective argument over whether a sample counts.
Practical takeaways
- Decide what the sample has to prove, then scale the controls to that. Not every sample in a trial needs the same rigor.
- Write the decentralized elements into the protocol before sourcing a kit, so the protocol constrains the kit.
- Name the receiving laboratory before kit design, because it sets the tube, the volume, the transit window and the rejection criteria.
- Treat participant instructions as a controlled document and usability test them with the enrolled population.
- Name the source of the collection timestamp in the protocol. Do not rely on participant recall.
- Pre specify the collection timing tolerance and what counts as a deviation, before the first participant enrolls.
- Keep provider records and the delegation log current, because moving the collection does not move investigator responsibility.
Related reading: What patient centric sampling actually means and You cannot validate a process you have not developed.
