Disrupting HTA and HEOR: the role of AI

The relentless progress of generative AI in healthcare has now reached Health Technology Assessment (HTA). In this month’s issue of Value in Health, an ISPOR working group has published its report exploring how AI can be used in HTA. It gives an excellent summary of the progress to date, as well as the current limitations of its use (1). It's well worth reading, especially for all those in market access and HEOR roles. Below, I give a summary of their key findings. 

Before getting to the specifics, it is worth highlighting that the authors stress the benefits that AI brings: greater efficiency and the potential to accelerate the process. Personally, I also believe that it can bring greater rigour and could help address human biases. The authors balance their enthusiasm for the technology with its current limitations which can lead to inappropriate conclusions. Therefore, they argue that AI should be seen as a tool for use by people, requiring human oversight and verification.

Three Key Areas Where AI Can Enhance HTA Processes

Systematic Literature Reviews and Meta-Analyses

Systematic literature reviews (SLRs) and meta-analyses are fundamental components of HTA, providing the evidence base for evaluating new health technologies. Traditionally, these tasks are time-consuming and labour-intensive, requiring extensive manual screening of articles, data extraction, and synthesis.

How AI Can Help:

• Propose search terms, screen abstracts, and extract relevant data
• Code generation for meta-analyses, conduct statistical evaluations
• Draft SLR reports

Real-World Evidence (RWE) Generation

Real-world evidence (RWE) plays an increasingly important role in HTA, leveraging patient data from sources such as electronic health records (EHRs), insurance claims, and registries.

How AI Can Help:

• Process and analyse large datasets more efficiently than traditional methods
• Extract valuable insights from unstructured data e.g., clinical notes, images, other types of data

Health Economic and Outcomes Research (HEOR) Models

HEOR models help assess the economic impact of new medical technologies, modelling of disease progression etc

How AI Can Help:

• Model conceptualisation, parameterisation, and validation

Additional AI Applications in HTA and Market Access

In addition to the areas called out by the ISPOR working group, there remains significant potential to deploy AI in other areas of HTA and market access, including: 

• Dossier Drafting: writing HTA submissions
• Market Access Landscape Assessments: analysis of market dynamics such as helping companies define HTA submission scope, including Population, Intervention, Comparator, and Outcomes (PICO) criteria
• Modelling Pricing Scenarios: modelling dynamic pricing strategies by simulating various reimbursement outcomes e.g., labelling scenarios

Challenges and Considerations for Companies Adopting AI in HTA

In their article the authors give a good overview of the limitations of AI in HTA, highlighting in particular: 

Scientific Validity and Reliability: 

AI models can produce hallucinations (fabricated or incorrect information). For example errors in data extraction and classification can undermine SLRs. Furthermore, can results be reproduced accurately between models? If not, this may call into questions around the validity of the approach.

Bias and Equity Issues: 

There is a significant risk that AI models are trained on datasets that do not reflect the target populations, resulting in bias. For example in HEOR modelling this could result in in appropriate parameterisation and model design.

Regulatory and Ethical Considerations: 

In particular, compliance with data privacy regulations such as HIPPA and GDPR. There is a significant that risk AI models may inadvertently memorise and reproduce protected health information, or programmers expose large language models during training to such data. Furthermore, AI programs may be able to identify (and act on) protected characteristics of patients that may be difficult to discern by a human analysing the data alone. It may also be necessary to obtain informed consent for certain types of data for specific applications e.g., with electronic health records.

These issues are common not just to HTA, but also to healthcare more broadly. I refer the reader to the paper for ways that these issues can be addressed. 

Key Questions for Companies Considering AI in HTA

For biotechnology companies preparing for HTA, or planning their HEOR and RWE programs, the following questions may help:

• Which of our HTA processes would benefit most from AI?
• How will requirements be met while using AI in HTA submissions?
• How do we establish trust in our results? 
• Should we develop AI tools in-house or contract these capabilities from an external provider? If external, who is best to deliver this?
• What risk management should we deploy to manage the technology? (2)

Conclusion

Despite the challenges of applying AI in HTA, there remains a significant opportunity to improve efficiency, speed, accuracy and validity. I expect that within the next couple of years AI will have already become the standard way of how we perform many of the tasks highlighted here. Therefore, health technology developers seeking market access for their products should be thinking about this now and working out how best to deploy this for their products. 

Contact

If you are interested in discussing any of the issues above for your company/drug development program, please contact me through my email address dniven@nivenbiopharma.com. Feel free to also visit my website at www.nivenbiopharma.com for more information. 

Sources

1. Generative Artificial Intelligence for Health Technology Assessment: Opportunities, Challenges, and Policy Considerations: An ISPOR Working Group Report [Editor's Choice], Value In Health, Fleurence et al, February 2025
2. How AI is enhancing HTA in Precision Oncology, Andree Bates, Niamh Boyle, April 24, AI for Pharma Growth Podcast