"Building Clinical AI: How We Became a Class IIa Medical Device"

If you are an entrepreneur developing AI capabilities for mental health, your input is needed before Nov 6, 2025! The @Food and Drug Administration has announced a public meeting and opened a public docket requesting comments regarding Generative Artificial Intelligence-Enabled Digital Mental Health Medical Devices. This is a huge opportunity for companies, including from Europe, and stakeholders developing GenAI capabilities for mental health to impact regulatory approaches. These new devices have the potential to address the significant demand for mental health services and the insufficient access to care providers in the U.S. The Digital Health Advisory Committee will discuss the benefits, risks to health, and potential risk mitigations for these devices, including considerations for premarket evidence and postmarket monitoring. Details and how to apply: Companies involved in the development of Generative AI-Enabled Digital Mental Health Medical Devices are strongly encouraged to submit their suggestions and comments on the safety, effectiveness, and regulatory pathways for this evolving technology. Meeting Date: The Digital Health Advisory Committee will hold a virtual meeting on November 6, 2025, from 9 a.m. to 6 p.m. ET. Committee Review Deadline: Comments received on or before October 17, 2025, will be provided to the Committee for consideration during the meeting. More details and submission guidelines here: https://lnkd.in/dJK-cteJ #DigitalHealth #GenAI #MentalHealthTech

Today I was reminded how difficult this MD Regulation business is, after giving incorrect information to a start-up. The guidance I provided was correct, based on MHRA guidance published February 2025. I think it's hilarious, if not slightly disconcerting that the guidance is already on version 1.2, having been updated for a second time in July 2025. The latest update provides guidance on product classification for DMHT’s (Digital Mental Health Technologies) and will place many Class I devices into Class IIa and some into Class IIb, further aligning the UK regulations with those of Europe! The full text can be found here: https://lnkd.in/eiQcnZ9U Below are some examples of how to apply the device classification rules, as dictated by Annex IX, Rule 10, of the MDD 93/42/EEC (as referred to by the UK MDR 2002 No. 618, as amended) (1) - General mental wellbeing support SW that processes patient data, designed to educate, and inform the clinical care pathway: Class I, but may not be considered a DMHT due to its simple to understand algorithms. (2) - Symptom tracking for anxiety / depression, Class IIa as the device is used to monitor a disease and so would trigger Rule 10. (3) - Clinical dashboard for DMHT that aggregates patient data and flags deterioration of the patient. Class IIa, based on the fact the device monitors, however the classification will depend on the clinical impact and so see (4) below… (4) - Suicide risk prediction tool, AI driven triage tool to alert clinicians, Class IIb, as the device is linked to immediate danger of the patient. You can see from these examples just how important it is to publish the correct and appropriate Intended Use Statement and how this can impact your device classification! Trying to correctly interpret the regulations on your own is next to impossible, as my blunder proves. If you need help applying the regulations to your MD / IVD / SW technology, please remember I hold a free Regulatory Clinic every Wednesday. Please direct message me to gain access and you never know, the information I provide might even match the latest version of the regulations 😅

AI‑Enhanced Medication Adherence Programs Non‑adherence undermines therapy outcomes. Community pharmacies implementing AI programs reported a 40 % increase in adherence and 55 % reduction in missed refills. Walgreens, for example, uses an AI‑driven patient engagement platform that combines refill reminders, personalized education and behavioral analytics; the program increased adherence by 40 %, reduced missed refills by 55 % and improved patient satisfaction by 62 %, yielding annual savings of $3.2 million. Reference attached. Another success story comes from Cleveland Clinic’s medication therapy management program. Their AI‑assisted platform achieved a 42 % reduction in readmissions, improved adherence by 35 % and detected 58 % more drug interactions. These results demonstrate that AI can prioritize high‑risk patients and tailor interventions. Critical to implementation are robust data privacy measures, integration with pharmacy management software and cultural sensitivity when delivering reminders. Question: How can AI solutions be tailored to improve adherence among populations with limited digital literacy?

MDR’s regulatory burden is disproportionate to clinical risks and killing innovation. Avoiding MDR approval makes market access easier Or so many believe. I am not surprised many digital health apps choose to under the radar as a “wellness app”. But is that the best decision? The need for MDR compliance for wellness tools would lead to unnecessary redesigns that is not aligned with the tool’s original purpose. First of all, if a tool would fall under the definition of a medical device by the MDR, they should take necessary steps for MDR certification. But, let’s face it, MDR certification is time consuming and expensive.  It would involve clinical investigations, which most startups don’t have a budget for. And some borderline products could be unnecessarily classified as medical devices. Risking over-classification, and burdensome MDR processes, not well-suited for innovative software or apps. Not only could MDR increase costs, but it could also make startups abandon entering the EU market, or scaling down features to avoid classification.  Impacting innovation and patient access. However, choosing the "wellness label" to dodge MDR risks creating a dangerous public health gap. With many unregistered health apps available in Europe,  There is no clinical validation available. Being able to show effectiveness and other clinical evidence can help getting the tools reimbursed. So, instead of avoiding MDR, startups should consider regulatory pathways as strategic path. But they should avoid these beginner MDR compliance mistakes: 1. Underestimating clinical evidence requirements early on 2. Ignoring quality management system expertise needs 3. Overlooking full financial impact of certification delays 4. Failing to budget for necessary regulatory consultants 5. Relying on marketplace approval as a safety net So, early on they should focus on: 1. Building MDR compliance into roadmap 2. Hiring or partnering for regulatory expertise 3. Planning clinical validation cost and timelines 4. Engaging early with notified bodies and consultants 5. Exploring reimbursement pathways like Germany’s DiGA It’s not all black and white.  Sometimes, it’s decisions that need to be made.  Which direction to go. Whether you want market credibility and trust.  Or access to clinical markets. MDR is a long road to take, but could be a strategic road. Interested in the conversation about going through MDR or not? Join us on the 16th of October for a conversation on where the EU regulations are going. Sign up here: https://lnkd.in/egrCv9c4 

Great take. Too often compliance is seen as a blocker, when in reality early MDR planning can become a real strategic advantage for startups.

For a look at the real-world impact of health tech, check out this article. It explains how AI and analytics are transforming patient care and drug development. #HealthTech #AIinHealthcare #PatientCare #DrugDevelopment

#GenAI #HealthcareInnovation #ClinicalAI #EthicalTech #SystemicChange #ThoughtLeadership In this narrative review led by Professor Ian Scott and just published in the Medical Journal of Australia, we offer a roadmap for integrating GenAI into healthcare without compromising safety, trust, or clinical rigour. Key Takeaways: >GenAI is reshaping workflows from documentation to diagnostics. >A 5-phase framework guides implementation, starting with clerical efficiency and scaling to patient self-management. >Risks like hallucinations, bias, and over-reliance are real—but mitigation strategies are evolving fast. >Adoption hinges on interdisciplinary collaboration, local validation, and transparent governance. We call for co-design involving clinicians, consumers, and technologists working together to ensure GenAI augments care, not replaces it. As we transition from ambient scribes to decision support, the question isn’t whether GenAI belongs in clinical practice, but how we deploy it responsibly. https://lnkd.in/gruJgMY9

𝐃𝐞𝐩𝐫𝐞𝐬𝐜𝐫𝐢𝐛𝐢𝐧𝐠 𝐆𝐨𝐞𝐬 𝐃𝐢𝐠𝐢𝐭𝐚𝐥: 𝐖𝐡𝐲 𝐂𝐚𝐥𝐚𝐛𝐚𝐫 𝐌𝐞𝐝𝐢𝐱𝐓𝐫𝐚𝐜𝐤 𝐈𝐬 𝐁𝐮𝐢𝐥𝐝𝐢𝐧𝐠 𝐟𝐨𝐫 𝐒𝐚𝐟𝐞𝐫, 𝐒𝐢𝐦𝐩𝐥𝐞𝐫 𝐌𝐞𝐝𝐬 Polypharmacy isn’t just inconvenient—it’s dangerous. The next wave of innovation is digital deprescribing: clinical decision support that flags potentially inappropriate medications and streamlines “stop, swap, or taper” conversations. Recent studies show that electronic decision support embedded in routine workflows can make deprescribing scalable and effective, rather than an ad-hoc best effort. That matters for people living with cognitive challenges, where complexity and risk stack up fast. At Calabar MedixTrack, we design for this reality. Our platform already focuses on the essentials—interaction alerts, personalized reminders, care-team coordination, and refill prediction—all aimed at reducing avoidable harm while keeping patients and caregivers in sync. By extending that foundation with deprescribing-aware features—med lists that surface risk signals, context cards that summarize why a drug is a candidate to review, and shared notes that travel with the care team—we aim to translate evidence into everyday practice. What’s changing in the field: ✨ Clinical protocols and guidelines for deprescribing are being formalized and digitized, improving consistency across settings. Read more about it here - https://lnkd.in/grZfyd7b. ✨ Research programs are exploring AI-assisted deprescribing, from risk stratification to dialogue supports for shared decisions. Read more about it here – Read more about it here - https://lnkd.in/dwfgQqfN. Our north star is simple: fewer risky meds, fewer adverse events, and more confident caregivers. If your organization wants to pilot deprescribing workflows inside a modern medication platform, Calabar MedixTrack is ready to collaborate. #DigitalHealth #Deprescribing #MedicationSafety #HealthTech #AIinHealthcare #CaregiverSupport #ClinicalDecisionSupport #MedixTrack #CalabarMedixTrack

𝐒𝐜𝐚𝐥𝐢𝐧𝐠 𝐌𝐞𝐝𝐢𝐜𝐚𝐭𝐢𝐨𝐧-𝐀𝐝𝐡𝐞𝐫𝐞𝐧𝐜𝐞 𝐎𝐮𝐭𝐫𝐞𝐚𝐜𝐡 𝐰𝐢𝐭𝐡𝐨𝐮𝐭 𝐄𝐌𝐑 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐢𝐨𝐧𝐬 An operational case study on scaling medication-adherence outreach without EMR integrations. We looked at how this work actually runs today – call lists, spreadsheets, payer files – and how one vendor (Rivvi AI, Inc) treats that “last mile” as the system of action. The flow: ingest unstructured lists, normalize, run rules-bound conversational outreach (refills, side effects, intent to schedule), and only return actionable items to staff. The design target is the outreach team, not physicians. Why it matters: nonadherence remains at ~50% globally; U.S. plans are graded on pharmacy proxies like Proportion of Days Covered, with ≥80% as the threshold for key classes (CMS 2025 Star Ratings Technical Notes). Cost estimates for nonadherence vary widely ($5,271–$52,341 per patient/year; PMC11700791) but make performance impacts measurable. Rivvi reports over 1 million patient interactions, 1.5. million automated workflows, typical 2–3% PDC lift, and one pharmacist managing ~30k patients at ~90% adherence. These are self-reported; methods and context are in our article: https://lnkd.in/d4_mFtsJ.

1M+ patient conversations, zero EMR integrations. Turns out healthcare teams just want their spreadsheets to actually work. Full story here from Andersen Med.