Competency: To gain acceptance, AI doesn’t need to be perfect, but it must meaningfully increase efficiency for the user, fit into the current workflow and do so at scale. Examples abound of algorithms that fail to gain traction because they have been forced into a process without sufficient humancentred design. In addition, to ensure AI is working as advertised, companies developing and deploying algorithms must use real-world evidence to validate (and re-validate) their effectiveness in the clinical setting. Several years ago, an algorithm employed by a healthcare payer to flag patients for high-risk care management was found to be systematically biased against Black patients. Governments must ensure AI meets standardized accuracy thresholds for all groups before allowing technology to be deployed at scale. They should consider using appropriate policy levers for high-risk AI use cases and taking steps to ensure developers prioritize equity when designing AI solutions. Developers also should be made to demonstrate high standards of documentation and transparency for algorithms and be encouraged to be future regulation ready in case auditing and certification requirements are introduced. Payers must continue to encourage the use of AI through greater reimbursement incentives, which will ultimately help ensure algorithms are adopted. Transparency: Doctors and patients must be fully apprised of the risks and benefits of any AI-based application to guide their care. Mayo Clinic, for example, has recommended that governments institute AI “nutrition labels” similar to those the FDA employs on packaged foods to provide transparency on ingredients and nutrition for greater transparency.40 Government policy can drive transparency standards around algorithmic fitness across demographic groups. Still, these standards may only be extended to highly regulated software as a medical device (SaMD) applications – potentially leaving the much broader, less-regulated category of healthtech tools to police themselves. Nevertheless, governments, payers, healthcare providers, technology and life sciences companies must equip leaders with a baseline level of AI fluency to ensure appropriate safeguards and hold their organizations accountable in the event of shortfalls. Partnerships: Many of the world’s most transformative innovations – the personal computer, to name one – came about only with government support and assistance. Governments must use the power of the purse to spur private investments into healthcare AI. In the US, the National Institutes of Health plans to invest $130 million over four years to accelerate the use of AI in healthcare,41 and the UK plans to put in a similar amount over the same timeframe through its AI in Health and Care Award.42 This is a great step, but much more investment is needed. Countries like Saudi Arabia, Singapore, the United Arab Emirates and the UK have created centralized AI authorities to help their countries maximize the value of healthcare AI. Other countries should follow suit by creating mechanisms to fund or incentivize partnerships to encourage AI innovation. Transferability: Great ideas should not be bottled up inside high-income countries, as they often are. Successful AI solutions must be scaled across organizations and borders. This will ensure a faster, more efficient scale-up of impact and more equitable transformation as low- and middle-income countries benefit from AI in health. Governments shouldn’t stop at funding pilots; they must provide funding to scale AI that’s already working in other countries. Of course, data must be trained on local nuances, but advances in generative AI will help overcome some of these barriers. More favourable reimbursement schemes may also help accelerate scaling. At least eight AI tools already receive payer reimbursement in the US, but many public health authorities outside the US are playing “catch up” rather than proactively creating incentives that spur innovation. Great ideas should not be bottled up inside high-income countries, as they often are. Successful AI solutions must be scaled across organizations and borders. 5.3 Building scale Scaling Smart Solutions with AI in Health: Unlocking Impact on High-Potential Use Cases 31
RkJQdWJsaXNoZXIy NzQwMjQ=