Healthcare AI Enters the Mainstream
Also published in Radical Reads on November 20th 2022
The pandemic served as a major inflection point for healthcare technology adoption as digital tools leapfrogged into the mainstream.
We anticipated this trend in Radical’s 2020 research thesis, describing two waves of digital healthcare innovation. Wave 1 laid the data, digital infrastructure, and behavioral foundations for the deep tech solutions of Wave 2 with AI as a primary driver. With an acceleration of digitization and available datasets, we anticipated an explosion of
solutions that leverage machine learning, computer vision, natural language processing, and robotics to solve some of our most pressing healthcare challenges.
In Q1 2022, healthcare was the top-funded sector in AI, according to CB Insights’ global State of AI report and, against a backdrop of decreasing investment in healthcare, healthcare AI attracted $4.5B of funding in Q2 2022 and gained 3 new unicorns.
Radical has invested in 10 cutting-edge healthcare AI companies deployed in the clinical setting and the pharma value chain. We believe the healthcare AI revolution has just begun.
Radical’s Healthcare AI thesis
AI will impact every area of our lives, but it will have a disproportionate impact on healthcare and biology, moving healthcare from reactive and generalized, to proactive
and personalized.
Healthcare data today is mostly siloed. However, as this data is unlocked, AI is perfectly positioned to responsibly leverage multi-modal healthcare data and combine it real-time with secure flows of clinically relevant data to dramatically improve health outcomes.
Broadly, there are three primary ways in which AI is improving outcomes:
Decision support
AI helps diagnose and screen various conditions. It is also used for treatment selection, planning, and remote monitoring. We are excited about companies that build across this value chain, from diagnosis to treatment and monitoring, and then close the loop by delivering healthcare services through their platforms.
Fast-tracking medicine-to-market process
After years of AI supporting medical research, AI for drug discovery and development companies have successfully brought AI-discovered drug candidates through the clinical development process to capture the value of the assets. Today, the prediction power of AI is also used extensively to optimize clinical trials, saving significant time and costs for pharmaceutical companies and, by getting drugs to market faster, enabling longer patent-protected revenue cycles for companies and better medications for patients.
Process efficiencies
Many companies also use AI to drive process efficiencies in healthcare institutions by optimizing clinical and operational workflows.
Radical has made many investments across these areas and built a strong portfolio of healthcare and bio-AI companies represented on this map.
These companies use cutting-edge AI to generate tangible impact for patients, payors, providers, and pharma. They are solving problems ranging from using machine learning to detect medical conditions using a baby’s cry to using machine learning models to develop best-in-class drug candidates with pharma companies. We have invested across the spectrum of practice areas including women’s health, pediatric health and diabetes.
Promising Opportunities and Trends in Healthcare AI
AI will play an increasingly important role in driving healthcare innovations and improving outcomes. The following are a few areas across the healthcare sector where AI will generate outsized value.
1. Unlocking healthcare data sets
The true power of healthcare AI will come from unlocking the vast quantity of siloed healthcare datasets.
We generate healthcare data every day from our mobile phones, wearables (e.g., FitBit, Apple Watch, Oura), and at-home devices (e.g., Alexa, Google Home, baby monitors like Owlet & Nanit). That data is augmented by clinically relevant data collected during care events -- patients are diagnosed with medical conditions using imaging, lab tests, genetic sequencing, and doctor-patient conversations. They are then treated with therapeutics, various forms of therapy, and care. More complex cases are monitored at home or in institutional settings using point-of-care devices (e.g., blood pressure monitors, glucose monitors) and through clinical observations. Across the spectrum, demographic data, administrative data, and patient insurance data are stored in electronic medical record systems. Additional data exist in repositories, such as claims data, where billable interactions between patients and the health delivery system are stored, patient/disease repositories for certain chronic conditions, health surveys, clinical trial registries & databases, and clinical research databases. We are also beginning to see a re-patriation of clinical data to patients -- a movement toward patient owned and controlled clinical data, that will slowly emerge as another source of healthcare data.
Collecting clinically validated healthcare data to train algorithms is costly and takes time, thereby impacting the speed to market for healthcare AI companies. So how do we efficiently unlock these available data streams -- a critical ingredient in developing machine learning algorithms for health?
An emerging solution to this challenge is federated learning, enabling machine learning models to train on data stored locally, without transferring that data to cloud servers. Once trained, the model is returned to the server with statistical patterns of the data encoded in numerical patterns for inference. Companies like Owkin have powered a pioneering collaboration between top pharma companies, including Johnson & Johnson, AstraZeneca, and GSK, to train their drug-discovery algorithms on each other’s data. The results of a collaboration like this will open up many more opportunities to unblock and combine data held by institutions toward improving health outcomes for all.
Another approach that is gaining early traction in healthcare is synthetic data. Radical Partner Rob Toews has shared his perspective on synthetic data suggesting synthetic data will one day overtake real-world data in training machine learning algorithms, not unlike how autonomous driving technologies leverage synthetic data to more efficiently train its systems. Many healthcare applications could also use synthetic data, specifically longitudinal data that ensures high accuracy. Radical portfolio company Unlearn is at the forefront of this trend. Unlearn is accelerating the medicine to market process by creating “digital twins” of human patients for the control arms of clinical trials. Unlearn recently announced that the European Medicines Agency (EMA) qualified its proprietary TwinRCT™ solution for Phase 2 and 3 clinical trials, major regulatory validation that the technology is ready to be deployed at scale. Using synthetic data across the board and specifically for clinical AI applications has its limitations but we are hopeful that these will be overcome in the long run.
Other approaches to unlock healthcare data include health data marketplaces and patient-first APIs to access health data. Radical portfolio company PocketHealth gives patients digital access and control of their health imaging data and wider healthcare data riding on the trend of the consumerization of healthcare data. PocketHealth recently launched their 2022 Patient Pulse survey, which has many useful insights on the changing relationship between providers and patients in controlling healthcare data.
2. Genomics AI and Precision Medicine
Precision medicine is primed to use AI to deliver personalized healthcare solutions -- the holy grail of healthcare AI. Gene sequencing is at the center of this trend.
Today, doctors recommend genetic testing sparingly relative to common blood tests and imaging. When they do, it is to diagnose complex medical conditions or identify specific diseases & syndromes. Prenatal genetic screening has emerged as a common test during pregnancy, with a third of pregnant women in the U.S. undergoing this test. And, in critical cases, it has triggered life-saving therapeutic interventions for genetic disorders.
The main reason genetic testing has lagged behind in widespread adoption is cost. Until recently, the cost of gene sequencing for consumers made it prohibitive to use at scale. And, even when sequencing was part of a care journey, it was tough to collect and store this data because the infrastructure for capturing and recording standardized, computer-readable genetics data in real-time didn’t exist in care settings.
However, we are now at an inflection point for genomics. In September this year, Illumina, the leader in the gene sequencing market, unveiled a new line of instruments that cut the cost of genome sequencing. This drop in price for whole genome and exome sequencing by 5 - 10x will be a massive tailwind for the space embedding genetic data into care delivery.
Applying the future of health framework described earlier, we are now squarely at the center of Wave 1 for precision medicine. Genetic data will become mainstream in this wave. Illumina’s pricing move will drive volume and push competitors to follow suit. Wider populations will be sequenced, avoiding bias in treatments and therapeutics yet to be built with this data. And, more startups will build infrastructure at clinics and hospitals to collect this data at scale.
In Wave 2, precision medicine will become a standard of care, instead of a form of novel care accessible to a few. We will see higher quality sequencing tests emerge that improve the standard of diagnostics. Genomic data combined with clinical data, demographic data and behavioral data will usher in a new era of medicine and change the course of disease progression across indications. Higher quality and quantity of genetic data will also enable researchers to find solutions for rare diseases, and will be the basis of next generation gene therapies.
Finally, Wave 2 will also power the generative biology revolution that will transform medicine forever.
3. Generative AI applied to biology
Generative AI is all the rage today, reflecting tremendous potential that goes well beyond text-to-image applications. Last year, Radical Partner, Rob Toews, claimed that the generative AI breakthrough, Alphafold, is the most important achievement in AI ever.
He was spot on.
Generative AI will have its most transformative impact on biology. Simply put, generative biology uses machine learning methods to generate or design proteins
that are tested rigorously in the lab, with the goal of developing better drug targets to
cure diseases.
Since AlphaFold, innovation in this category is moving at breakneck speed. Earlier this year, Deep-Mind researchers used AlphaFold to predict the structures of 200 million proteins from 1 million known organisms or species. At the time, DeepMind Chief Executive, Demis Hassabis, called this “the beginning of a new era of digital biology.” Just a few weeks ago, researchers at Meta AI predicted the structures of 600 million proteins from bacteria, viruses, and other micro-organisms that haven’t been characterized. And in October, NVIDIA released its BioNeMo LLM service and framework to help biology researchers generate, predict and understand biomolecular data.
We are entering an era of transformational change to biology as we know it. The technology precedent has been set by innovations described above. The tooling is available for researchers to build. And, solutions are being developed by teams that combine machine learning expertise with expertise on taking drugs to market to power a new generation of AI for drug discovery and development companies.
4. AI enabled Sleep solutions
70 million Americans suffer from chronic sleep problems, and the CDC has described sleep deprivation as a public health epidemic. Sleep deprivation, in turn, contributes to other chronic conditions and is a looming crisis, as we try to keep up with the distractions of modern life.
However, it’s not all bad. Sleep data has become mainstream, and the category is ripe for disruption. Consumers are becoming aware of the problems that lack of sleep poses to their health. As they consume the sleep data at their fingertips, they are looking for the next step -- how do they optimize a good night of sleep, and how is their sleep impacting other medical conditions.
Applying the future of health framework described earlier, we are now entering Wave 2 in this category. In Wave 1, we saw the sleep data collection market grow rapidly. Sleep data moved from being measured in the clinic to being measured wherever consumers were. A large number of companies collecting sleep data at scale through different approaches emerged. This included multi-sensor technologies in wearables and consumer devices, as well as integrated approaches leveraging EEG, radar, and wireless systems in the home. Today, the scale and quality of sleep data collected has improved drastically. Companies collecting sleep data are fine-tuning their algorithms to improve the quality of their sleep dataset, and the scientists are calling for higher standards being applied to the use of this data, in the right drection.
In Wave 2, we hope to see therapeutic interventions, built on the continuously improving and clinically validated sleep datasets collected in Wave 1. We believe a new category of startups will emerge that incorporate sleep science and insights to solve multi-billion dollar problems like sleep apnea and insomnia. We are also encouraged by solutions that are leveraging deep technology and AI to build end-to-end sleep management solutions.
Conclusion
It is clear that Wave 2 in the future of digital health has arrived. The founders quoted above are responsibly deploying healthcare AI solutions as you read this. We shared their insights to spur more innovation in the space.
Healthcare AI has entered the mainstream, and we are at an inflection point for the widespread use of AI in healthcare. As we said before, AI will impact every area of our lives, but it will have a disproportionate impact on healthcare and biology, moving healthcare from reactive and generalized to proactive and personalized.
We are very excited to back teams building solutions in the areas we have noted.
If you are building a business in healthcare AI, have thoughts that resonate or have a different point of view, please reach out.
Stay tuned for more insights, and continued work on our De-mystifying Healthcare AI series.