Reflections from ISPOR 2024: AI's Transformative Impact on HEOR and beyond

From computational biology to 20+ years in health economics and outcomes research - who would've thought? At #ISPOR2024, watching thirteen AI panels felt like coming full circle. We're entering an "autonomous era" where humans will focus on asking the right questions and setting guardrails for AI. Here are my personal takeaways from watching ISPOR24 AI sessions:

1. AI-powered HTA simulations are coming. #SvenKlijin from BMS showed how virtual committee members, modeled on real-world parameters, like “pharma fondness”, “willingness to argue”, “job”, etc, produce outcomes similar to human committees.

?2. Large Language Model (LLM)s are unlocking EMR goldmines. ~80% EMR data are unstructured. #AnitaLoughlin from Moderna demonstrated how they're finding more rare disease patients and mapping complex patient journeys by applying LLM onto Truveta EMR unstructured data.?

3. Wearable sensors are entering the EHR world. Emory University's work on "freezing of gait" in Parkinson's patients hints at a future of continuous, real-time health monitoring.

?4. Clinical AI is growing up fast. We're "schooling" our AI babies to be specialists, balancing creativity with accuracy. I can see customizable "hallucination thresholds" for different use cases in the future.

5. "Human in the loop" is still key, but AI's catching up quick. #XiaoyanWang shared two mind-blowing cases:

???a) #Google Deepmind's LLM outperformed both unaided and AI-assisted clinicians on complex NEJM cases. The diagnosis accuracy: clinician unaided 36%, clinician aided with research 44%, clinician aided by LLM 52%, LLM alone 59% (https://arxiv.org/pdf/2312.00164)

???b) A 4-year-old's rare condition, missed by 17 doctors, was diagnosed by #ChatGPT. Mom for the win! (https://www.businessinsider.com/chatgpt-diagnose-child-disease-tethered-cord-syndrome-doctors-2023-9)

6. Social media: 25% health noise or goldmine? #AbeedSarker from Emory University showed how to curate social media info and analyze patient sediments associated with specific therapies – talking about getting instant market feedback, here is the potential

7. New Patient Reported Outcomes Measurement (PROM) and Patient Reported Experience Measurement (PREM) are emerging from AI-analyzed EHRs and social media. Real patient voices, amplified by tech. PRO will finally become critical!

8. AI for causal inference? Could methods like Targeted Learning become a bridge to traditional stats methods for regulatory and HTA approval? TLME (Targeted Maximum Likelihood Estimation) was used in a NICE submission for Multiple Myeloma but challenged for the rationale. #DaliaDawoud from NICE says they are open to exploring AI's value.

9. #FDA's #AskFDALabel framework is published and paving the way for regulatory LLMs. Transparency and trustworthiness are the name of the game.

11. #NEJM.AI is pro-LLM for scientific writing, “as long as authors take complete responsibility for the content and properly acknowledge the use of LLMs”. Will "prompt protocols" become the new methodology sections?

10. Health equity in the AI era is a growing concern. We're developing fairness outcomes metrics at individual and population levels. Access drives data, drives benefits - a cycle we need to watch and not allow AI to amply the existing inequity.?

12. LLMs are reshaping trial design. #ErginSoysal from Eli Lilly showed how AI can spot novel endpoints in protocols. Next stop: AI-curated EHR screening for trial design.

13. Decentralized health data use is accelerating. Seeing #NancyDreyer join #PicnicHealth is exciting. It's also why I co-founded #EtheroHealthData Foundation - to empower individuals to own, share, and benefit from their health data.

?We're living in the future of HEOR, powered by AI. As we navigate this new world, let's keep ethics, patient-centricity, and human expertise at the forefront. I'm excited to dive deeper into these topics and explore how we can leverage AI in healthcare data and drug development. If you're interested in learning more or discussing potential collaborations, please reach out.?

#ISPOR2024 #AI #HEOR #HealthcareInnovation #DecentralizedData #PatientEmpowerment