Post a comment
Print articleEarly adopter's playbook: Deploying AI in healthcare
February, 2024 - Shane Evans, Vanessa Mellis
Those in the healthcare industry are driven by a commitment to patient care and improving lives. AI presents a powerful tool that can accelerate treatments, enhance care quality, reduce costs, and get healthcare to more people. Adopting this technology is not just an opportunity; it is a necessity. The life sciences sector in particular are already realising the benefits of AI. Their shared experiences offer other health organisations the opportunity to pre-empt some of the challenges associated with AI.
Use cases for AI in healthcare
Large language models such as ChatGPT are already revolutionising text-based operations, accomplishing feats that were considered impossible until recently. This breakthrough is poised to transform healthcare, particularly in areas that heavily rely on text, such as doctor's notes and research articles. The true constraint lies not in the technology itself, but in our ability to envision how we can utilise it to its full potential.
While there are significant AI related opportunities across healthcare, other prominent use cases currently being explored and implemented include:
- healthcare administration and workflow optimisation
- clinical decision support systems
- medical imaging analysis
- telemedicine and virtual health assistants
- mental health and therapy
- personalised medicine
- health monitoring and wearables
AI is a truly transformative tool, offering key benefits including speed, scale, consistency, quality and cost-efficiency. Notwithstanding this, there are challenges to overcome to ensure the safe, ethical and successful implementation of AI.
Life sciences sector leading the way
With AI use booming across industries, within healthcare the life sciences sector is leading the way. Biotechnology, pharmaceuticals and medical device companies are already reimagining their models and re-establishing themselves as 'technology' companies. They are leveraging AI at an enterprise-wide level, from early research through clinical development and into commercialisation. AI has been successfully used to accelerate and streamline business processes as well as to improve the design and development of therapeutics.
One breakthrough example is the use of mRNA. mRNA is a programmable information molecule and it has reached a stage where researchers are able to programme into the nucleotide sequence of mRNA whatever code they want for whatever protein they want to produce. This platform model means researchers have a single way of doing things that will then work for countless different medicines. Enter the power, capability and scale of AI. Utilising AI to create an algorithm to design better mRNA can lead to an improvement in the entire platform of medicines.
One thing is clear, other healthcare sectors have the opportunity to learn from the challenges these early adopters have faced and how they have overcome them.
Practical strategies for successful AI adoption in healthcare
Whether you are at the start of your AI journey, or have already begun implementing AI, here are our top tips from early adopters' playbooks, to help health organisations safely, ethically and successfully adopt and operationalise AI.
Start with data
The foundation of AI lies in data and its effectiveness hinges on the quality of that data. Organisations should start by undertaking a data mapping exercise to understand what data they hold, where that data is held and assess its quality.
Unstructured or inconsistent data, such as free text or variable information, often pose a challenge for AI systems. Data needs to be unified and structured in a way that AI algorithms, which primarily work with numerical information, can process efficiently. Organisations should prioritise developing tailored workflow tools that encapsulate the structured processes adopted by scientists and other professionals. Investment in a robust data management system, congregating information into a centralised data lake, can provide invaluable support.
If the data is not stored in a warehouse, establishing pipelines to channel the data is essential. When dealing with unstructured, low-quality, or free text data, modifications may be required in the source system to facilitate structured use for the algorithm. If these challenges remain unresolved and data is not captured, organisations might need to explore alternative, more feasible use cases, and develop a roadmap to build their data infrastructure over time.
Enlist the right mix of expertise
Organisations must ensure their technical teams not only understand the AI models but also have a solid grasp of the business context in which they're applied. The healthcare industry has seen numerous instances of external technology specialists struggling due to a lack of industry-specific knowledge.
The ideal Steering Committee or Project Team comprises experts who not only understand the technology but also the health industry and the organisation, as well as the cultural transformation required for successful AI integration.
Identify and prioritise use cases
Identify what challenges your organisation is facing or what big business value you are trying to solve for. Ask yourself: are there particular operational bottlenecks or patient care concerns? Then analyse how AI can make a difference, however small, in these areas. This analysis will provide organisations with a feasibility overview, differentiating between easy and complex problems, and indicating which can be solved with the data at hand. It will assist with prioritisation.
Start small and focus on value
While it's tempting to tackle the big problems, it's often best to start small. Tackling smaller projects will help to build momentum, strengthen credibility and provide valuable learning experiences. Organisations have found that balancing early quick wins that build credibility, help learning and de-risk other projects, and longer term, larger projects ultimately deliver greater value for the organisation.
With all of this in mind, AI adoption should be seen as a strategic investment rather than a novelty. It's important to choose projects that deliver tangible benefits to your organisation and align with your business objectives.
Embrace imperfection
Don't shy away from projects where a lower accuracy rate is acceptable. An AI solution that's less than 100% accurate can still deliver value, especially if its performance exceeds human capabilities or complements human effort. Consider projects where AI can handle scalable, low risk, simpler tasks, leaving complex problems for humans.
Anticipate risk
There are various risks to consider when deploying AI in a healthcare setting including legal, regulatory and clinical risks. AI also carries ethical considerations, particularly in relation to biases. Organisations that fail to scrutinise the data used in their algorithms, or the ways these algorithms are applied, risk systematising and amplifying biases. These biases can manifest in both technical and social forms, underscoring the importance of proactive bias control in AI use cases.
Establishing AI and data ethics and governance frameworks, AI guiding principles as well as mapping potential risks and establishing controls to mitigate or manage these risks is critical to ensure the safe and ethical use of AI in healthcare.
Don't underestimate cultural change
Cultural change management - whether related to employees or consumers - is difficult and shouldn't be understated. The more ingrained the process is, the more difficult it is for people to shift their behaviour and attitudes. Consider establishing a dedicated team focused on AI transformation. This team should be well-versed in both digital and cultural transformation, helping the organisation or consumers understand and adopt AI. Additionally, find your early adopters and use them as change agents to influence others as well as help you to learn.
Educate users
For frontline workers using AI, they need to understand not only the capabilities of AI but also its limitations, particularly in relation to the models being used. For example, understanding that some models may 'hallucinate' is crucial to scrutinising outputs and avoiding adverse outcomes.
To address this, some organisations have launched in-house AI training programs or academies. These initiatives aim to enhance the overall AI literacy within the organisation, equipping employees with basic AI knowledge and facilitating the development of use cases that can improve operations.
Follow through
Finally, don't stop at just building an AI solution; ensure it's deployed in the real world. Taking an AI project through to deployment reveals challenges related to data management, deployment, and monitoring in production.
The future of AI in healthcare
Regulation and collaboration
Currently in Australia, there is limited specific regulation of AI. On 17th January 2024, the Australian Government published its Safe and responsible AI in Australia consultation interim response, which acknowledges the current regulatory framework does not adequately address known risks of AI, as covered in our recent article. Globally, this is changing quickly. Both the FDA and the MMA recently released position documents on the appropriate use of AI within drug development, a trend likely to continue. It is probable we will start to see more regulations around how we monitor and control for bias and transparency on when and how algorithms are being used, particularly in the healthcare or the criminal justice systems. This emphasises the importance of establishing frameworks, policies and procedures for the safe and ethical use of AI with regard for the rapidly changing landscape.
Collaboration between healthcare providers, researchers and tech companies
The healthcare industry is vast and it will take a lot of parties to come together to solve some of these problems. We expect to see more strategic collaborations where healthcare providers, who have interesting data and problems to work on, form partnerships with technology companies and academic institutions to solve these more complex challenges.
Given all the pressures placed on the health system, providers, leaders and workforce, it's inevitable that AI will become part of standard practice, whether that's clinical service delivery or back of house operations. Health organisations must seize the opportunity so they are not left behind.
