|
the future of healthcare is being written in code with AI transforming SaMD from a relatively niche technology into a linchpin of patient care. SaMD was still in its formative stage, with global regulators cautiously defining its scope and implications, when AI leapfrogged its application into uncharted territories and ushering in an era of rapid innovation.
We are now witnessing AI-powered real time imaging diagnosis to predictive health analytics. But the adaptive learning nature of AI and relative opacity of the closely guarded decision trees challenges the traditional regulatory frameworks.
Global regulators are scrambling to set systems in place to ensure patient safety and fair use without stifling innovation. For the OEM behemoths of the world along with the upstart AI-first trailblazers, navigating this complex regulatory landscape is not just a legal hurdle, but a strategic imperative.
AI upends core assumptions
AI upends some of the core assumptions of medical device regulations- that the device once manufactured, validated and deployed remains static in its functioning. The addition of adaptive AI means continuous learning and the evolving element of algorithm drift means the old regulatory playbook don’t apply post-market.
This has forced regulators to adopt a Total product Life Cycle (TPLC) approach to SaMD. The IMDRF (International Medical Devices Regulators Forum) has set forth principles to this regard and the global regulators are following suit.
The US FDA focused on Predetermined Change Control Plan (PCCP) which starts with the manufacturers defining the intended use, the roadmap of modifications including the changes to algorithms and data sourcing, along with methods for change control and validation before implementation.
This way, the manufacturer can continue improving without needing a new clearance for every update. Risk is clearly identified and accounted for- the traditional design inputs, V&V protocols and DHFs document capture it all.
FDA has structured it around guiding principles of AI/ML SaMD Action Plan, Good Machine Learning Practices (GMLP), Transparency for Machine Learning-Enabled Medical Devices, and created the latest draft guidelines on Lifecycle Management and Marketing recommendations.
The European Union’s AI Act read along with MDR (Annex VIII, Rule 11) introduces risk-based classification requirements for SaMD. The AI Act reclassifies AI medical systems as high risk as their impact is direct and tangible.
There is stringent QMS documentation, emphasis on strict clinical evidence and post-market surveillance along with re-evaluation requirements for algorithm updates. The AI systems must comply to ‘lawfulness, ethical alignment and robustness’ along their lifecycles. Transparency and traceability are to be built in.
Considering the roadblocks to innovation, European Notified Bodies (NB) have developed joint guidance in 2024 for AI in medical devices, so that NBs accredited in both MDR and AI Act can undertake a single conformity assessment for the manufacturers.
Harkening to changing winds, Japan’s PMDA (Pharmaceutical and Medical Devices Agency) had set up AI Subcommittee in late 2017 and based on its recommendations, enacted the revised PMD Act in 2019.
The Act introduces conditional early approval and a regulatory framework analogous to FDA’s PCCP, PACMP- Post Approval Change Management Protocol. The PMDA is cautious but accommodating and requires iterative scientific consultation with PMDA reviewers and external expert evaluations before final approval.
To offset algorithmic bias, a local dataset and a subgroup analysis comparing Japanese population to global cohorts is a must. Japanese government has created unique programs like Dash for SaMD which allows for both PMDA approval and a formal pathway for inclusion into public health insurance for reimbursement.
SaMD with demonstrable clinical outcomes especially for chronic diseases can apply for both in a single dual track process, which is an enabler for the SaMD OEMs in early revenue realisation.
Other regulators like Australia’s TGA, India’s CDSCO and Health Canada largely follow IMDRF guidelines which help streamline GTM strategies but has its local nuances. ISO 13485 (QMS), IEC 62304 (software lifecycle), IEC 82304 (health software safety) and AI Act readiness are helping pave the way. ISO is actively charting AI management system standards and ethical AI modules all in a bid to ensure the software evolution is safe and controlled.
OEMs need to rethink many aspects
To successfully adapt to these converging frameworks, OEMs must anchor regulatory rigour into the entire AI lifecycle as an end-to-end design principle. There are some real operational dilemmas looming.
OEMs need to rethink many aspects like product validation to ensure the future versions remain safe; design controls to include algorithm training, data quality, retraining plans and drift mitigation; QMS to factor in and document AI validation practices; risk classification to mitigate bias; PMS to include feedback loops, testing for algorithmic bias across diverse demographics and model retraining etc.
Given the ‘black box’ nature of deep learning models, and the ambiguity around accountability, researchers are pointing to concerns of racial and gender bias and threats to health equity. Corollary to this, SaMD also needs thorough cybersecurity testing against attacks, strict data privacy enforcement and a Human-in-the Loop system to mandate clinical accountability. In this context, patient- centric ethical data governance, traceability and transparency are of paramount urgency.
Navigating this turbulent regulatory environment requires a paradigm shift in the way compliance is approached. It begins with upfront risk classification and building practical roadmaps which anticipates future updates.
This dictates the depth of clinical evaluation and clinical evidence needed to ensure AI’s accuracy metrics and ultimately the clinical utility. Robust design control measures and Real-World Evidence (RWE) frameworks need to be in place along with early proactive engagement with regulators and cross industry coalitions.
Enormous promise
AI/ML enabled SaMD brings with it enormous promise. The regulators worldwide are trying to put in effective checks for patient safety and fairness in place. It is up to the OEMs to leverage the new compliance strategy- planning early for lifecycle changes, integrating risk management with AI transparency documentation and embracing region-specific expectations instead of fighting them.
The biggest takeaway is that regulatory readiness isn’t about paperwork- it is about trust and continuous improvement. It is about OEMs incorporating these principles into their culture and quality systems to gain a decisive market advantage and spearheading responsible innovation.
(The author is Delivery Manager and
Regulatory Practice Head, Tata Elxsi)
|