Policy Innovations in Health Technology

1. How do you envision the role of AI, particularly bioacoustics-based models like HeAR, in addressing diagnostic challenges for marginalised populations and underserved regions?

AI holds transformative potential for addressing diagnostic challenges in marginalised and underserved regions. Bioacoustics-based models like HeAR (Health Acoustic Representations) are designed to detect early signs of diseases such as tuberculosis (TB) by analysing sounds like coughs. Given TB is the leading infectious killer globally, these models offer a scalable, affordable diagnostic solution.

By utilising smartphones, AI-powered tools can provide remote diagnostics where access to radiologists or X-rays is limited. Health workers with minimal training could use these tools to triage patients, improving healthcare outcomes in underserved areas. HeAR, trained on 300 million diverse audio samples, can identify 100 million cough sounds, showcasing its ability to scale.

However, to maximise impact, these models must be rigorously validated across diverse populations. Ensuring the model's robustness in distinguishing TB from other respiratory diseases, especially in patients with comorbidities, is critical. To avoid exclusion and bias, diverse, real-world clinical data must underpin model training. Generating high-quality clinical evidence and evaluating cost-effectiveness in various regions will also be key to integrating such tools into routine care.

2. What specific policies should governments implement to ensure AI-based healthcare technologies, like digital biomarkers, are accessible to low-resource health workers in developing regions?

For AI in healthcare to have a transformational impact in developing regions, several challenges must be addressed through regulatory clarity, ethical governance, and global collaboration. Digital biomarkers, digital phenotyping (DP), and digital endpoints (DEP) offer the potential to shift healthcare from a reactive to a proactive model, but their integration is hindered by gaps in governance, data standards, and infrastructure limitations.

First, governments must promote collaborative governance by fostering partnerships between global and local research initiatives. These collaborations are vital to ensuring that AI tools are contextually relevant and tailored to the specific needs of diverse populations, particularly in resource-poor settings. An important aspect of this is ensuring that novel diagnostic data and tools such as DEP and DP can be securely repurposed and shared across networks, maximising their utility and that of limited clinical data, and improving overall healthcare access.

Second, regulatory clarity is essential. Many AI tools operate in a grey area between wellness applications and medical devices (MD), creating uncertainty about how they should be regulated. To address this, governments should collaborate internationally to develop clear and consistent guidelines, particularly for the emerging types of general-purpose AI-as-a-MD (AIaMD) and clinical decision support (CDS) tools. These frameworks must consider the needs of low-resource health workers and ensure that AI tools are accessible, safe, and effective in environments with limited resources.

Lastly, financial incentives and digital public infrastructure development are crucial. Governments must provide subsidies and financial support for the deployment of AI technologies in underserved areas, helping to overcome cost barriers that often prevent adoption. Alongside this, there must be a concerted effort to improve health literacy and digital skills, ensuring that frontline health workers can engage with AI-based tools effectively.

By bridging the gap between technological advancement and healthcare delivery, AI can truly drive better health outcomes in developing regions. By addressing these interconnected areas AI can be harnessed to transform healthcare in developing regions, creating more equitable, accessible, and effective healthcare systems.

3. What are the most significant ethical considerations that must be addressed when implementing AI-based diagnostic tools in diverse clinical environments?

The deployment of AI-based diagnostic tools in diverse clinical settings raises important ethical considerations, which must be carefully managed to foster trust and ensure equitable outcomes.

Equity, Diversity, and Inclusion (EDI) is central to ethical AI development. AI systems often inherit biases from their training data, disproportionately affecting marginalised or underrepresented groups. To prevent discriminatory outcomes, EDI must be embedded into AI governance frameworks, ensuring diverse representation in datasets and algorithm development. This mitigates the risk of digital redlining, where algorithmic bias could lead to exclusion from care or suboptimal treatment.

Another key consideration is transparency and accountability. AI systems must be explainable and transparent, especially in healthcare. Clinicians and patients should understand how decisions are made, which builds trust and allows for proper accountability. Disclosing data sources, model limitations, and potential biases is crucial for aligning AI development with ethical standards and regulatory expectations.

Patient-centred risk management is equally vital. Ethical data governance frameworks should include patient input, ensuring that AI systems reflect real-world complexities. Additionally, synthetic data from generative AI should be used cautiously and anchored in real-world data to avoid amplifying biases. Including Patients with Lived Experience (PWLE) as contributors ensures that tools are sensitive to diverse patient needs.

Finally, a sustainable AI ecosystem must be developed, with collaboration across stakeholders. Governance frameworks should ensure equitable distribution of AI’s benefits, especially to underserved populations, and support continuous improvement through shared data and resources. This collaborative approach will help maintain the long-term ethical use of AI in healthcare.

4. What are the key challenges in generating robust clinical evidence for AI-powered medical devices, and how can regulatory frameworks evolve to accommodate these new technologies?

Generating robust clinical evidence for AIaMD presents distinct challenges, largely due to the constraints of existing regulatory frameworks, which often struggle to incorporate the diverse standards needed for evaluating AI tools. These frameworks, especially in the AIaMD sector, tend to prioritise safety and performance, while falling short in assessing the efficacy and real-world clinical effectiveness of these technologies.

One of the primary challenges is the absence of consistent evidence standards for AI MDs. While pharmaceutical regulations typically require randomised controlled trials (RCTs) to demonstrate efficacy, MDs, including AI-driven tools, often rely on less rigorous clinical investigations. This discrepancy makes it difficult for regulatory bodies and health technology assessment (HTA) agencies to effectively compare different health technologies, which in turn delays their integration into clinical practice.

Additionally, manufacturers may be reluctant to invest in the extensive clinical integration studies necessary to prove long-term effectiveness. Without sufficient evidence, HTA bodies are unable to make informed decisions on pricing, reimbursement, and widespread adoption of these AI technologies.

To address these challenges, regulatory frameworks must evolve towards more adaptive models. For instance, conditional coverage strategies could allow promising AI tools to enter the market earlier, with post-approval commitments to ongoing evidence generation. This approach would enable AI-powered devices to reach patients sooner, while still ensuring that long-term safety and efficacy are continuously assessed.

Additionally, establishing high-quality MD registries is crucial. These registries would serve to standardise data collection and tracking across the industry, facilitating consistent reporting on device performance and patient outcomes. By providing a structured, transparent system for monitoring AI tools in real-world settings, these registries would help improve transparency and support more informed HTA evaluations.

Moreover, the use of evidence sandboxes—controlled environments where AI models can be tested—would significantly accelerate both evidence generation and regulatory alignment. These sandboxes would allow AI tools to be tested in real-world scenarios under the supervision of regulatory bodies, ensuring their safe and effective integration into healthcare systems.

Evidence sandbox facilities (ESFs) could play a key role in fostering innovation in evidence generation and evaluation. These controlled environments would allow for testing new approaches to data collection, model development, and AI integration. By bringing together regulators, industry experts, healthcare providers, and patients, ESFs would enable a more dynamic and responsive regulatory process that can keep pace with the rapid advancements in AI. ESFs would also support the integration of AI-powered MDs into clinical research and HTA pathways, ensuring that both real-world applications and long-term sustainability are considered.

Finally, well-designed target and marketing strategies are crucial. Regulatory bodies should implement governance frameworks that link biomarkers and disease phenotypes with clinical endpoints, and now digital endpoints, to reflect more meaningful proximal targets like readmission rates, quality of life, and reductions in financial toxicity. Moving beyond traditional, narrowly defined metrics, marketing strategies should focus on patient-centred outcomes, enhancing the overall value proposition for AIaMD. By aligning marketing with well-defined health metrics, AIaMDs can better resonate with regulatory bodies, healthcare providers, and end-users, ultimately emphasising improved patient outcomes and cost-effectiveness.

Ultimately, such measures would streamline the process of integrating AI tools into clinical practice, while maintaining high standards of safety and effectiveness.

5. How do you foresee the role of data governance evolving with the integration of AI in healthcare, and what standards need to be in place to ensure patient data security and system accountability?

As AI becomes more integrated into healthcare, data governance will be crucial for balancing innovation with patient security and system accountability. Governance frameworks must prioritise equity, security, and transparency to foster trust and facilitate global collaboration.

A key principle to support this is digital health autonomy (DHA). DHA emphasises patient-centred governance, civic engagement, and a participatory regulatory ecosystem. By allowing patients and communities to actively participate in data governance, DHA promotes an iterative system that supports continuous improvement and evidence generation, rather than relying on one-time approvals. This ongoing feedback loop fosters innovation in AI and digital health while maintaining accountability and inclusiveness.

An important framework in this context is healthcare data recycling (HDR). HDR focuses on repurposing and updating clinical data throughout the patient care continuum. By enabling AI systems to access real-world, contextually relevant data, HDR improves the accuracy and inclusiveness of AI interventions. It also facilitates continuous learning and model enhancement while ensuring patient privacy and data integrity are upheld.

Furthermore, integrating social determinants of health (SDOH) into AI data governance is essential. By incorporating data on social, environmental, and economic factors, AI systems can shift care upstream, addressing the root causes of health disparities, particularly in underserved populations. This ensures AI-driven interventions are proactive, focusing on broader social factors that affect health outcomes.

To safeguard patient data, AI systems must adhere to strict privacy standards. Governments should enforce robust frameworks that guarantee data anonymisation and restrict unauthorised access while allowing secure data sharing for research and clinical purposes. These standards are critical for building public trust and supporting the scaling of AI in healthcare.

The IEEE P3493.1™ working group, focused on secure and compliant healthcare data recycling in cancer care, presents a promising model for global standards development. It emphasises coordinated governance, inclusivity, and the ethical reuse of healthcare data.

Additionally, the IEEE Global Mobile Health App Standard Registry will serve as a directory of apps that have met the criteria required to attain an IEEE standard identifier, offering a key regulatory benchmarking tool in the medical device industry.

As AI technologies become more widespread in healthcare, frameworks like these will be instrumental in ensuring that AI's use is both ethical and impactful. By aligning data governance with global health priorities, AI can help address healthcare disparities and improve outcomes for diverse populations.

6. What role should international bodies like IEEE play in developing standards for AI as a medical device, and how can these standards ensure global compliance?

International bodies like IEEE play a pivotal role in shaping global standards for AI in healthcare, ensuring alignment with pressing global challenges such as climate change, rising comorbidities, and healthcare disparities, particularly in marginalised communities. IEEE's initiatives address the intersection of AI, social determinants, and health equity, ensuring that AI technologies are designed to meet the needs of diverse and vulnerable populations. By setting standards that prioritise inclusivity, IEEE helps to ensure that AI can responsibly and equitably tackle global health challenges.

To ensure global compliance, IEEE collaborates closely with policymakers to harmonise AI standards and codes of practice across different regions. This not only promotes consistency in AI regulation but also facilitates access to best practices and standardised assessments. By doing so, healthcare providers and AI developers can more easily navigate the complex regulatory landscapes that vary across countries. Through fostering trust, transparency, and accountability, IEEE's standards encourage the widespread adoption of AI-driven medical devices and health technologies.

Looking ahead, IEEE and other international bodies can play an even larger role in creating collaborative innovation ecosystems that bring together academia, industry, and policymakers. By engaging diverse stakeholders in the co-development of AI solutions, these ecosystems can address complex, cross-cutting healthcare challenges, including those exacerbated by climate change, while maintaining high ethical and technical standards. This collaboration will drive AI innovation in a way that is adaptable, ethical, and aligned with the global mission to improve healthcare outcomes for all populations.

7. How can health technology policies be designed to foster continuous improvement and iteration of AI systems, ensuring that they adapt to changing patient needs and technological advancements?

Fragmented healthcare delivery is a significant barrier to improving global health system performance. Siloed care, disjointed work processes, and fragmented innovation arise from the complex governance structures and disparate financing models that characterise corporatism in healthcare. This fragmentation has hindered progress toward open, collaborative innovation and value-sharing, preventing integrated care models from achieving the patient-centred transformation and preventative care that modern healthcare requires.

To enable continual learning in AI systems, health technology policies, frameworks, and innovation ecosystems must embrace a culture of open innovation. The objective is to shift this paradigm by bringing together multi-stakeholders—including healthcare providers, policymakers, patients, and industry leaders—to tackle the adaptive challenges healthcare systems face.

This shift requires rethinking how clinical practice and health governance can integrate DHTs to prospectively study and address complex health phenomena while building resilient and adaptive healthcare systems that can dynamically respond to evolving community needs. This is the core goal of systems medicine and a critical element in fostering community engagement in the innovation process. By focusing on bottom-up approaches, we can not only address healthcare equity globally but also ensure that AI tools evolve sustainably and responsibly to meet the needs of diverse patient populations.

To achieve this, embracing the concept of digitally integrated learning health systems (DLHS) is essential. These are digital twins of physical LHS that implement clinical translation safeguards, optimise patient navigation, and enhance precision treatment. By leveraging collaborative DLHS, healthcare systems can accelerate the responsible development and deployment of AI, leading to the creation of evidence-based guidelines, biomarkers, and outcomes that significantly improve the quality of care and support the quintuple aims of healthcare. Moreover, this approach enables healthcare systems to adapt in near real-time to technological advancements and shifting patient needs, advancing both patient outcomes and healthcare delivery.

--Issue 05--