[[2 Oct 2024 - HIMSS APAC Day 1]] ![[Pasted image 20241012170803.png]] AI technology has the potential to revolutionize healthcare by automating processes and providing decision support to healthcare professionals. However, there are concerns about the impact on human interaction and the need for careful consideration of ethical and societal implications. Despite these challenges, AI has the potential to improve patient outcomes and reduce healthcare costs. ---- 20241002 - Plenary Talk - Future of AI in healthcare.  **HK speaker shared his view of the deployment of AI is not just about the tool, but the whole processes involves to meet an objective.** - First, Does it work? How to put it in work flow? is it actionable? AI component is not the major component,  - E.g medical safety issues - hepatitis, we built this in medication system - put in rules for safety. How does the patient know that they have hepatitis? Doctor need to do manual flagging in the system, but if doctors does not do it then it will not work. They built in processes to remind doctors when certain medications are prescribed, to check if patients have hepatitis ^994c70 - What is the process/workflow you put in place that support it, not just AI.. AI may be good but there is no workflow to support it then no use.  - Having a problem, using technology to solve the problem. It doesn’t have to be fancy, it just have to work. (Pragmatic)  **How does you bring such a system to clinical practice.**  Difficulties:  1. Hard to get tech to healthcare. Is the thing you trying to do bring benefits? We can build trust.  - **accuracy** and **benefit** assessment (Evaluation), can we see more patients, ROI, capacity.. etc. how does we do benefit assessments for technology?  2. T.E.S.T - Technology Evaluation Safety Test (NHS) for AI. https://transform.england.nhs.uk/key-tools-and-info/digital-technology-assessment-criteria-dtac/ 3. But interestingly, we don’t even do evaluation for non AI technology.  Korea Speaker: - There is a long process to get Regulatory approval. Adoption is slow. Long term to apply to put into clinical practice. hurdle in Korea.  Benefit for healthcare - we have many benefit for AI.  - Speaker shared about the adoption of digital pathology. There is resistance from senior clinicians. - Pathologist - example , government fund, development AI software for pathology data. (Maybe they have many research so far). Concern: you have to push against established doctors/practises, who to push against that. Once they use the digital pathology, they don’t want to go back (Junior doctors yes) - Change management - same everywhere.  UK Speaker: - Public and Private collaboration - private company can register for device and work with public hospital.  - UK, capitalism, approach for AI trial. To do RCT trial (breast cancer).  - How to get good evidence and satisfied benefit assessment for a start up. Collaboration come from University and industry (UK doesn’t have such a framework to support private and public) HK  - open database as a tool for machine learning AI development, 30 years of clinical data in research development basis - Open for commercial now- when someone bring in AI tool, hospital do validation themselves to see if it fit for the purposes (whether it really works) **I want to bring in AI tool, what do you tell hospital CEOs?** 1. Trusted research environment. Created own dedicated research environment. Secure cloud base.  2. Commercial relationships with Private , big pharm  3. Air lock - data is safe, we can interact with AI company to test it out without losing data  4. interface partner (to test and try out) **Private company - how to do contract?**  SG 1. Innovation research relationship - start up taking small amount of equity, codevelopment- we use that product in hospital 2. Consumer trying if it works we buy it.  3. University, IP… hospital don’t retain IP.. in HK - Similar - Co-developed product, they own the IP.  - 43 hospital - per use contract. A strictly commercial arrangement.  Korea - each hospital have own way to manage their own data. - Not easy to share data with each other.  - Each company have to work with each company..  - One of the difficult for the company.  - even though have de-identity data, it is still restrictive to be use by company. - AI company find it hard to access those data for commercial use.  Barriers to adoption. 1. Technical 2. Change management 3. Evaluation **How to overcome these barriers?** 1. HK 1. Barrier = clinical staff imagine poor outcomes, “what if” - - change management  2. How to get pilot in place to demonstrate that what they imagine doesn’t happen. 3. Pilot test - try out at one of his 43 sites 4. Show them, it’s OK 5. The other things, expected benefits — monitoring,looking at outcomes as you implement.  1. SG -  people often resist and doubt 1. Ask them to take the lead of that projects, to propagate it through the organisation.  2. HK: We have tech savvy doctors, I don’t rely on them. Doctors who like technology will say its easy.. but other doctors will not adopt it. It will only be one sided. Get the resistant, but want to improve the clinical result.. and get them to buy it. Look for those who are against me. Change them  UK- - Hearing it from clinical the truth is better, the hard truth first. Have clinican explain to clinicans. - AI have to be better. The tool is a tool, and it’s my clinical judgement making the decisions that is “me”. You don’t have to tell the patient how you do it. Just like you don’t tell patient which textbook you read that knowledge.  - Change management Technical Barrier 4. AI - algorithm - transmission of data outside of hospital to cloud computing. E.g AWS, ..especially LLM, it is outside of Korea. That is not allowed in Korea. Technical barrier.  5. In Korea, they have their LLM - Naver have their own LLM. They are creating some model in hospitals, but the application is still research purpose, if it’s for clinical, then need more investment, funding like GPU, storage.  **Question: Implementation, what excite you in the future of innovations** HK: He left medical because the future… putting in system to help them perform better. He find AI exciting. Opportunity for everyone.. how to get everyone try it out. To learn how this stuff can help, and then build the tools UK: Deep inside. If AI can tell me something i haven’t thought about that will be cool.  Make me a superhero. So much we haven’t thought about. Putting human in the loop, are they putting human all the responsibility? We have MDT and sharing the responsibility.. we may want to think differently.  SG: They are providing surprising output. Etc. read text written by doctor and predict Length of stay. He find it surprising, AI prove him wrong. Second opinion, like an MDT, … balance the pro and cons Korea: Development of software to help pathologist to find report some type of lung cancer. The one patient can have many types.. but we don’t know the % subtypes.. human cannot calculate the exact percentage.. to provide more detail data..cell counting. AI value to take away mandune, repetitive work from human, and leave the valuable work to doctors.  **What is A technology that will grow in the mainstream?** HK 1. AGI — He believe that General Intelligence, “super human”, - if that is better than human, but that will be massive change to everything, not just about medicine.  UK 1. Empathy is hard to reproduce - people are trying. There is a degree of inattention 1. 14% inattention to programme in AI to seems more Human.  2. Like scan like startrek, to diagnose  3. We are like cave man burning meat with fire, not knowing it can bring us to the moon. We still don’t know what AI is at the moment.  3. HK - patient find chatGPT more empathetic then human doctor. ChatGPT can chat with you for longer 4. Korea - We have only work with one aspect of use case, maybe AI can look at comprehensive views of human.. General Pathology tool — Harvard Medical School — already published some thing about. Advancing.  1. **Do we need more pathologist or fewer pathologist?**  1. Can’t answer that. For now, 20 years still need pathologist because data is still analogue. Unlike radiology.  2. But there is not enough pathologist around the world. — SG say some countries don’t have enough, they only do the test after patient died.  3. Currently still need human pathologist to review the test result.  ———