EUDAMED’s vigilance module won’t be mandatory before 2028, but MDR post-market obligations already apply, and the hard part is keeping in touch with patients for years. This post explains where conversational AI can help with that, and what to check before relying on it.
4-minute read
Most manufacturers have a post-market clinical follow-up plan. Far fewer have a way to keep reaching the same patients for five or ten years.
Getting a device CE-marked is where the evidence work starts, not where it ends. Under the EU Medical Device Regulation (MDR), manufacturers must keep updating their clinical evaluation with real-world data for as long as the device is on the market. For devices that stay in use for years, from heart valves and joint replacements to insulin pumps and hearing implants, that can mean a decade or more.
Writing the post-market clinical follow-up (PMCF) plan is the easy part. Carrying it out is harder: reaching the same patients year after year and getting usable answers from them. That is where evidence tends to get thin, and thin PMCF data is a common sticking point with notified bodies.
Where EUDAMED actually stands
Four EUDAMED modules became mandatory on 28 May 2026: actor registration, UDI/device registration, notified bodies and certificates, and market surveillance (TÜV Nord).
The module that matters most for post-market work isn't live yet. In October 2026, the European Commission said the vigilance and post-market surveillance module will not become mandatory before 2028, after its first functional audit found problems (Emergo by UL). Until then, serious incidents are still reported through national authorities.
The delay doesn't buy any time. MDR post-market obligations already apply. When the module goes live, vigilance data and, for higher-risk devices, periodic safety update reports (PSURs) will go through one central system, and gaps will be easier to spot. The best use of the time between now and then is to fix how the data gets collected.
The hard part is year five, not year one
Picture a PMCF study following 1,000 patients for five years, with check-ins at 30 days, at one year and every year after that. That comes to about 6,000 scheduled contacts. Each completed contact can take several call attempts, a reschedule and data entry, so the real workload is several times larger.
Meanwhile, follow-up erodes. Patients move, change phone numbers, or feel fine and stop answering. Coordinators change jobs. Every patient lost to follow-up weakens the long-term picture the study was meant to produce.
A well-designed PMCF plan is not the same as the capacity to run it for five years. Manual outreach gets harder to sustain the longer the follow-up runs.
Where conversational AI helps
An AI phone assistant can take on the repetitive part of this work. It calls patients on the protocol schedule, asks the same questions the same way every time, and organises the answers for the study team.
Three things make this especially useful for PMCF in Europe:
Consistency over years. The schedule doesn't depend on who happens to be on staff in year four.
Language. A pan-European study can reach each patient in their own language, without a separate call team for each country.
Catching what patients mention in passing. A patient may bring up a new symptom or a hospital stay during a routine check-in. If it turns out to be a serious incident, MDR reporting deadlines are short. Reports are due within 15 days of the manufacturer becoming aware, 10 days for a death or serious deterioration in health, and 2 days for a serious public health threat (MDR Article 87). Flagging those remarks for human review right away helps them reach the vigilance team in time.
Check-ins can also happen between formal visits. Symptoms don't follow the visit schedule, and a short call can pick up a change that would otherwise wait months to be recorded.
None of this replaces clinical exams, imaging or the manufacturer's judgement. The AI collects and routes information. People decide what it means.
Five questions to answer first
Data collected by AI still has to hold up in your PMCF evaluation report. Before you build it into a study, get clear answers on these:
Validated instruments. If your protocol uses a validated questionnaire such as the EQ-5D, or a device-specific score such as the KCCQ (cardiac) or the Oxford Hip and Knee Scores (orthopaedic), confirm that a phone-administered version is valid for your purpose.
Data protection. Health data is special-category data under GDPR. Define consent, the controller and processor roles, and where the data is stored.
The tool's regulatory status. A system that only collects and forwards answers is in a different position from one that interprets symptoms or triages patients. Know which one you're using.
Escalation. Decide who reviews flagged responses, how quickly, and how that review feeds into your vigilance process.
Accountability. The manufacturer stays responsible for the evidence and for every report built on it.
Approval is where the evidence starts
EUDAMED's timeline will keep shifting. The obligation to know how your device performs in real patients, years after launch, will not. Manufacturers who build reliable follow-up now will have stronger PSURs, fewer surprises in notified body reviews, and better answers when a safety question comes up.
Calico Care builds Cali, an AI assistant that makes proactive phone check-ins with patients, collects patient-reported information and organises it for clinical and research teams. Patients don't need an app or a portal. Learn more at calico.care.
