the health strategist
institute for
strategic health transformation
& digital health
Joaquim Cardoso MSc.
Chief Research and Strategy Officer (CRSO),
Chief Editor and Senior Advisor
August 28, 2023
What is the message?
The article highlights that large language models (LLMs) are a transformative innovation similar in impact to the computer mouse, allowing intuitive interaction with AI through natural language.
The introduction of LLMs, like PatientGPT™ by Semalytix, signifies a breakthrough in AI technology for healthcare.
PatientGPT™, a generative AI tool, effectively processes and interprets patient-generated data, rapidly generating valuable insights from millions of patients’ experiences.
This technology empowers pharmaceutical companies to enhance research, optimize clinical trial design, and ultimately improve patient outcomes.
By harnessing real-time patient experience data, this tool has the potential to revolutionize drug development and patient-centered healthcare strategies.
DEEP DIVE
PatientGPT Reduces Months of Pharma Research To Seconds
August 23, 2023
Large language models are as profound an innovation as the computer mouse. Just as the mouse makes it possible for humans to interact with computers in an intuitive way, LLMs make it possible for humans to interact with AI using natural language. Scientists are beginning to develop practical applications using this powerful technology.
This month German AI company Semalytix launched a new patient-centric generative AI tool called PatientGPT™. This new tool is able to recognize, summarize, and translate patients’ own words about quality-of-life, symptoms experienced, medication side effects, treatment experiences, and unmet needs. Semalytix collects real world patient data and uses a branch of AI called natural language processing to identify unrecognized issues associated with a disease. This tool generates multidimensional insights from millions of patients instantly, reducing months of research into a few seconds. These insights can help pharmaceutical companies accelerate research, de-risk clinical trial design, and improve patient outcomes. This is a milestone for AI in healthcare.
PatientGPT is exclusively tuned to supervised patient experience data and always acts with conceptual disease models in the background.
Semalytix has collected over 50 million patient experience statements in 26 languages to make the world’s largest real-time patient experience data stream. Using powerful AI algorithms to interpret it, combined with a proprietary analytics platform, PatientGPT produces clear summaries of patients’ experiences in dozens of disease areas. Semalytix’s AI platform has already successfully been used to generate evidence and patient insights for Crohn’s disease, breast cancer, psoriasis, obesity, ulcerative colitis, melanoma, lupus, and diabetes, and the company’s pipeline includes 25 additional indications. Several top pharma companies are already using the platform.
Generating synthetic patients with global disease memory
There’s huge potential to use PatientGPT technology to generate synthetic patients with global disease memory. PatientGPT has a ChatGPT-style interface that has been exclusively tuned to supervised patient experience data. These synthetic patients can be interviewed interactively or answer questions in batch mode. This novel way of extracting and condensing information from a massive dataset has only recently become possible due to advances in AI. By restricting the training set to specific cohorts of patients, the synthetic patient may be configured to assist with exploring particular questions of interest that match specific demographic variables. This can help drug developers identify new clinical endpoints and massively scale how they achieve real-world value for patients.
“Since most AI companies analyze data from electronic health records and patient registries, the largest source of unstructured patient data is being overlooked. Semalytix has collected over 50 million patient experiences and has the world’s largest real-time patient experience data stream. This data holds huge potential to address patient needs, discover new markets for existing drugs, identify new therapeutic opportunities, inform future clinical trial development, and even help accelerate the development of novel therapies for rare conditions. PatientGPT can interrogate, analyze and interpret this data in a matter of seconds. We are really excited by the possibilities”
Janik Jaskolski, Co-founder and Chief Product Officer, Semalytix
Using AI to extract and condense massive datasets
Over the last 20 years people on blogs, forums, and social media have compiled the world’s largest patient experience dataset. Health ranks among the top 5 topics discussed on social media and over 50% of patients and families share their health experiences online. However, most pharma companies don’t collect this data and lack sophisticated tools to create consistent value from it. Most AI companies that focus on real-world evidence, collect data from electronic health records in hospitals.
Since 2015 Semalytix has been collecting public data that people share on blogs, forums, and social media about their experiences living with diseases and taking medications. The data sets that Semalytix collects contain different types of information than found in electronic health records. Since patients don’t report all of their side effects and experiences to their doctors, electronic health records are incomplete and don’t provide an accurate picture of a patient’s daily life. Without access to authentic patient-experiences, pharma companies risk spending years developing treatments that don’t address the true needs of patients.
How many side effects do prescription drugs cause?
- 69% of drugs have between 10 and 100 side effects
- 22% of drugs have more than 100 side effects
- 9% of drugs have fewer than 10 side-effects
Source: Healthcare Analytics Research Group, IBM, 2013
Patient-centric drug development
New guidance from the FDA and the EMA acknowledges that online patient experience research is a powerful tool for collecting comprehensive and representative input for patient-focused drug development. Semalytix is actively working with a community of experts at Pistoia Alliance to develop a framework proposal for the use of social media insights in a regulatory context. The community is developing recommendations to leverage real-world data from social media to support patient-centric drug development. The initiative, which was originally introduced to the Pistoia Alliance by Semalytix, is supported by AstraZeneca, Bayer, Biogen, Boehringer Ingelheim, Chiesi, Johnson & Johnson, EMD Serono, and Roche.
PatientGPT Highlights
- Semalytix has collected over 50 million patient experiences in 26 languages.
- The Semalytix AI has been developed to accurately recognize over a million medical concepts and entities in free-form text.
- The platform produces AI-generated Conceptual Disease Models to give context to insights.
- PatientGPT always acts with Conceptual Disease Models in the background and within the context of what supervised AI-models have previously predicted.
- PatientGPT has a ChatGPT-style interface but it is tuned to and exclusively limited to supervised patient experience data.
- PatientGPT operates on top of patient experience data that are cleaned and enriched by Semalytix’s supervised machine learning framework without drawing information for responses from other sources.
- Use-cases include safety signal monitoring, automated patient-preference studies, bottom-up patient journey construction, conceptual disease models, competitor analysis, brand health monitoring, and unmet need analysis.
- The tool, which interprets the effects of medications in seconds, will change how patient insights and evidence are generated, accessed, and used from early research and development to commercialization.
- PatientGPT could help pharma companies accelerate time consuming research in the drug development process.
- This new tool could help improve patient outcomes in the medium and long-term.
PatientGPT Sample Searches
PatientGPT answers qualitative research questions in seconds using AI-generated Conceptual Disease Models that provide crucial quantitative context. Below are examples of potential focus areas, prompts, and sample searches.
Tamoxifen for breast cancer: patient reported side effects
Metformin in type-2 diabetes: patient reported advantages and disadvantages
Crohn’s Disease: patient complaints that may lead to non-adherence
Hypothetical patient persona based on current segmentation
Summarized findings in patient-reported outcome measure format
About the Author
Margaretta Colangelo is a leading AI Analyst tracking AI milestones in healthcare.
Originally published at linkedin.com
