This is a "news"letter today 🦙
What have we covered so far and what were some headlines in the past month
In the past couple of months, we’ve used this newsletter to look at what might be seen as the basics and foundation of healthcare and some applications of AI in healthcare. I have thoroughly enjoyed learning so much about these different topics and having a library of digital health that I am starting to put together. I hope you’ve found these topics interesting and engaging.
Once every few weeks, we will use a post to highlight the latest headlines in AI and healthcare so the newsletter truly lives up to its name. Here is a quick recap of what we’ve covered so far before we jump into the news. I hope this helps add more context to the earlier posts if you joined us halfway through or later in the journey.
Recap of the newsletter so far
The State of AI in Healthcare PART I: The first in the series on AI in healthcare, this one covers some hot topics and companies in the healthcare administration, LLM, and software as a medical device spaces.
The State of AI in Healthcare PART II: The second in the series on AI in healthcare, this post goes over the AI applications for payor and provider solutions, real-world data, and real-world evidence. Real-world evidence is an exciting space as clinical solutions demand more truth and granularity in healthcare data insights and analytics. It is likely one of the biggest upcoming markets in digital health.
The State of AI in Healthcare PART III: The third and final post in the AI in healthcare series was probably the most fun one. We looked at five niche AI applications through the lens of five exciting companies: Cleerly in medical devices, Absci in de novo antibody development, Insitro in clinical trial acceleration, Behold AI in rapid cancer detection, and Overjet in dental AI.
Top Trends in Digital Health: This post covered some of the spaces in Healthtech that are growing and up and coming. We went over interesting trends such as Fintech + Healthtech, Fertility tech, and Social Determinants of Health.
The Different Payer Models in Healthcare: This series of posts helped us jump into the payer space. This first post went over the foundations of health insurance in the US along with some history. We covered the payer models, payment types, and some of the large/notable companies on a high level.
Options for and Innovation in health insurance: The second post in the payer series helped us dive deeper into how people obtain health insurance and where the innovation is happening today. Some of the interesting trends we saw were companies acting as third parties to expedite the payment process and companies innovating in the self-coverage space.
Value-Based Care: The last post focused on a more specific type of payment model - value-based care. Value-based care has definitely become a buzzword and hot topic in the last few years but there is a lot of value in the concept. However, there are challenges that new startups are trying to overcome such as overcoming physician resistance and focusing on primary care or specific specialties.
Let’s jump into the new now.
Healthcare
Top VCs name the most promising healthcare startups
A Business Insider report collected data from multiple VCs (Pear, Greycroft, Norwest, and Inspired to name some) to list the most exciting startups in healthcare right now.
Octave is a company we’ve discussed before. Octave provides virtual and in-person behavioral-health services in over six states. The cool thing about Octave is that it also helps with growing the value-based care model. Octave has already established value-based contractions with multiple plans and also used their data to analyze the efficacy of care in this space. Their latest funding round was this past month, a 52 million Series C led by Cigna Ventures.
We covered Insitro and its impact on the efficiency and speed of clinical trials. Another company, Entos used AI to speed up the drug discovery process. Entos combines AI and quantum computing, to cut down on the time it takes to discover a compound or a treatment for a particular condition. Existing machine learning techniques are challenged by the scarcity of training data when exploring unknown chemical spaces. Entos’ OrbNet overcomes this barrier by systematically incorporating knowledge of molecular electronic structures into deep learning. By creating a physics-inspired neural network, the method learns molecular representations based on the electronic structures that simulate an environment similar to molecular structures. While still in its early stages, Entos’ models have shown promise and are much faster than traditional machine learning models based on molecular data.
Another interesting startup, Xilis is building cancer tumor models to replicate how tumors grow and understand their environment. Clinicians and researchers can then test different cancer treatments on those tumor models to see which treatment the patient might respond to best. Xilis’ collaboration with academic centers will support research and development of new cancer therapies, according to the organizations.
Pear VC releases an awesome healthcare AI playbook
Pear VC put together a collection of companies and AI applications in healthcare broken up by clinical workflows, bio and pharma, and non-clinical workflows.
Specifically, non-clinical workflows include diagnostics and clinical education, remote patient monitoring, and robotic surgery. Some of the most exciting companies in this space include ones we’ve covered before such as Cohere Health, Science IO, and Adonis.
Clinical workflows include diagnostics, hospital operations, and remote patient monitoring, including companies such as Cleerly, Caption Health, and Olive.
Finally, Bio and Pharma include companies such as Benchling, Concert AI, Komodo Health, and Veeva.
The post includes a voice-over presentation along with a full list of companies, read more here.
AI
Meta releases Llama 2, the next-gen open-source LLM.
Meta released its latest version of the Llama model. Llama is a large language model that was released by Meta earlier this year to compete with Open AI’s chat GPT. The primary use of these models is for organizations that are serving generative use cases; hence, the pre-training that it comes with on large volumes of data all over the internet. If you have a generic use case you are set, the model is trained on up to 70B parameters. The whole model is open-source which means anyone can use it and there are plugins that easily integrate into LLama 2 for managing and organizing datasets. Llama 2 is also a foundational model, a model that can be used as a starting point and fine-tuned to deliver a more specific use case or work in a certain domain. While the foundational and open-source pieces are impressive, Llama 2 is behind in comparison to GPT4 and Google’s PalM line of AI models when it comes to creativity and spontaneity.
Llama 2 will be fine-tuned to serve more specific use case. Its use in a discriminative fashion remains to be seen. Will the model’s pre-trained foundation make fine-tuning in a particular domain easier or will users find it trivial to add to their model pipelines that are trained comprehensively on specific datasets? Even in healthcare, healthcare data has specific formats and text, it will be interesting to see how well the model performs on healthcare claims, EHR, and other types of datasets.
It seems like AI is what everyone wants a piece of right now, including Databricks. Databricks acquired Mosaic ML, an infrastructure platform for companies to train and develop generative/large language models. Why is this a big deal? Mosaic not only provides a foundation model which is quite powerful, MPT-30B, its also ways ahead in providing a platform for other companies to train and build their own models in a cost-effective way. The fine-tuning, training, and data organization are all taken care of; all companies need to do is bring their data with them.
Similar to Databricks, Snowflake, a data warehousing and querying platform also picked up Neeva, a company specializing in generative AI for search. The AI marketplace is growing and time will tell how the giants compete and acquire smaller companies.