Considering the massive potential in healthcare, these somewhat modest numbers aren’t a result of a shortage of companies entering the space. Rather, it shows the challenging nature of it, driven by regulatory hurdles, stagnating adoption and reimbursement rates, and poor data interoperability.
While the outlook might not seem the most inviting for a VC and the industry has its inherent challenges, here at Inventure, we prefer to embrace the complexity of the space. We are actively seeking entrepreneurs who are willing to take on the challenge and significantly impact the future of healthcare.
In the 2010s, we witnessed a monumental shift in the industry, primarily fuelled by the extensive integration of cloud technologies, creating a huge upsurge in healthtech innovations. Now, we believe we’re once again at the forefront of a significant shift. This time driven by regulatory changes, improved data interoperability, and the extensive development of LLM and LMM-based services.
AI as a catalyst for change
We’ve all heard about it: an aging population, an increase in chronic diseases, and systemic staff shortages keep our people from receiving the best care possible and are causing an increase in mortality rates. One of the most impactful ways to increase efficiency in healthcare is through digital solutions. Yet, the current level of digitization presents significant challenges. The use of outdated legacy systems, originating from the 1990s, has resulted in poor integrations and data quality. This, in turn, has led to weak data interoperability and fragmented data, constraining the ability to build upon this data.
Despite the need for digital solutions, their adoption remains sluggish. Most often, the trade-off between adoption costs and efficiency isn’t enough to convince hospital decision-makers. AI-based solutions, however, offer a more attractive trade-off. They not only deliver superior efficiency but are often more user-friendly, reducing the overall cost of implementing new solutions by needing less staff training. For example, in imaging, one of the earliest medical areas to be digitized on a large scale, AI-based clinical decision support systems are well on their path to becoming standard.
Anticipating a profound enhancement in the foundational data layer, driven by advancements in data interoperability and thorough data cleansing processes, we expect a fertile environment for accelerating innovation. This will likely encompass the development of new record systems, optimization of non-clinical administrative workflows, and a massive shift in medical search capabilities and diagnostic procedures.
— Kevin Lösch, Inventure
Opportunities rising with LLMs, LMMs, and LVMs
The development of LLMs and LMMs (large multimodal models) will significantly streamline many non-clinical tasks, including patient communication, clinical transcription, and report generation. For example, LLMs can be used by providers to collect and clean the data, making it possible to assess symptoms already before admission. This way, providers don’t need to spend any time on a patient until the actual delivery of care. Companies like Google are already working on healthcare-specific multimodal models that will be able to work with in- and output modalities like electronic health records or scans. Using those models for administrative tasks would make, e.g., information retrieval much faster.
The most intriguing prospects lie in the aggregation and innovative reconfiguration of existing data. With hundreds of millions of electronic health records, diagnostic reports, and detailed PDFs chronicling patient and hospital conditions, these can be synergistically combined with crucial life science data using multimodal models. This integration has the potential to unlock entirely novel shifts, generating profound insights and revealing patterns. It effectively aligns the current state of research with real-time patient data, offering a more dynamic and comprehensive view of healthcare analytics.
— Kevin Lösch, Inventure
Additionally, we are on the cusp of a breakthrough regarding LVMs (large visual models), which will be able to handle a vast array of medical tasks. Clinical imaging is currently the most important application area for AI models, but they rely on supervised learning, which requires the laborious procurement of suitable data sets for each specific application. Furthermore, these models are susceptible to data shift, where the training data often differs from the deployment data. By being trained on a massive amount of data and then being fine-tuned for a specific use case, LVMs could accelerate the development of medical imaging tools and improve their robustness.
Moreover, LVMs differ from LLMs in that images can exhibit considerable variability compared to language, implying that there will likely be a more substantial number of players involved in the LVM field, as domain-specific LVMs will need to be developed. While companies in the LLM domain will primarily operate at the application level, we anticipate ample opportunities within the LVM model layer for healthcare applications.
Regulation paving the way for innovation
We often like to think about regulation as an inhibitor of innovation. However, in the field of healthcare, correct regulation can accelerate development significantly. The EU is currently working on a couple of acts that will allow for a higher degree of digitization, laying a solid foundation for further innovation and giving companies more avenues to explore.
The EU Data Act, which entered into force now in January 2024, mandates that manufacturers share collected data with third parties upon receiving user consent. This provision applies to the medical sector as well. One of its focuses is data portability, which will allow individuals to seamlessly transfer their data between different controllers, enabling patients to switch healthcare providers without losing access to their health data. This provision puts pressure on EHR providers to enable interoperability between systems while paving the way for new entrants. For example, our portfolio company, Livv Health, is building a decentralized, patient-owned EHR system that gives users full control over their data. Patients will be able to take their medical records with them wherever they go and receive AI-based recommendations to promote a prevention-focused lifestyle.
Another key aspect of the EU Data Act is non-discriminatory access, which prevents data holders from discriminating against potential users based on size or business model. This levels the playing field, making it easier for smaller healthtech companies to access data from larger entities. The third aspect of the Data Act involves fair and reasonable terms, obliging data holders to provide equitable conditions for accessing and using data. This aims to prevent the imposition of excessive fees or restrictions on data sharing, again fostering accessibility for early-stage startups.
Expected to be put into force in 2025, the European Health Data Space is another substantial regulatory influence on healthcare. It establishes a comprehensive framework for the secure and interoperable exchange of health data across the EU. While the extent to which data must be shared for secondary use cases is not yet clear, making it mandatory to share data in a standardized format, with the patient’s approval, could open up new possibilities for a variety of AI-enabled applications.
As is often the case, regulations are playing catch-up with technological advancements, trailing behind by years. While regulations might damper innovation in some sectors, we’re counting on regulations to lay the groundwork for a functional data landscape in healthcare.
Although the healthcare industry comes with complexities, we firmly believe that now is the time to be building in the space. The developments we are seeing both on the technological and regulatory front are setting the stage for new, transformative companies that could truly change the course of healthcare.
If you’re as hooked on the topic as we are or are building something in the space — let’s talk!
