Hospitals are increasingly motivated to drive digital transformation in order to improve patient outcomes, reduce costs, meet regulatory requirements and stay competitive. Additionally, digital transformation can support medical research and drive innovation in healthcare, as well as generate new revenue streams. An increasingly important tool as hospitals undergo these transformations is federated learning, a technology that we will expand on later. Federated learning is a machine learning approach that enables the collaboration and sharing of data between multiple institutions while preserving patient privacy and security.
Going back to the general topic of digital transformation of hospitals, we have seen several trends, some of which have been long-standing and some less so:
- Adopting electronic health records (EHRs) has already been encouraged by the Health Information Technology for Economic and Clinical Health (HITECH) Act’s ‘Meaningful Use’ clause, and EHRs have been used over the last decade to improve communication and coordination among healthcare providers. Additionally, digitized health records have been used to make patient information more easily accessible, first to meet Health Insurance Portability and Accountability Act (HIPAA) and HITECH mandates, and now, increasingly, to meet the 21st Century Cures Act ‘Final Rule’ mandate, requiring timely and standardized access to all patient data.
- Implementing telemedicine and remote patient monitoring have grown in popularity for years now, and were especially accelerated during the COVID-19 pandemic, in order to increase access to care and reduce the need for in-person visits. They were also enabled by regulatory changes that allowed cross-state patient care and other ways to relax the burden on provider organizations and patients alike.
- Data analytics and population health management tools are often layered on top of the medical records in order to identify high-risk patients and improve care coordination, as well as adopt value-based care models that focus on outcomes and cost-effectiveness (such as the MACRA-MIPS reforms, encouraging the formation of ‘Accountable Care Organizations’).
- Increasing the use of technology such as AI and ‘Robotic Automation’ to improve administrative efficiency, diagnostic accuracy and treatment planning. This is the newest trend, reaching broad adoption over the last 2 years (or so), and we will later focus on that as a driving force and component of digital transformation.
As one could assume, the implementation of hospital digital transformation creates significant demands on data availability and interoperability. These demands include:
- Data Availability: Hospitals need to have access to complete, accurate, and up-to-date patient data in order to provide effective care. This includes both structured data (such as lab results and medications) and unstructured data (such as notes from doctors and nurses).
- Data Interoperability: Hospitals need to be able to share data with other healthcare providers and organizations in order to coordinate care and avoid duplicate tests and treatments. This requires the use of common data standards and protocols, as well as the ability to securely exchange data across different systems.
- Data Security: Hospitals need to protect patient data from unauthorized access, use, and disclosure. This requires the use of robust security measures such as encryption, multi-factor authentication, and regular security audits.
- Data Governance: Hospitals need to have effective policies, procedures, and systems in place to manage and control access to patient data. This includes managing data access controls and permissions, monitoring data usage, and implementing data quality and integrity checks.
- Data Analytics: Hospitals need to be able to analyze patient data in order to identify trends, patterns, and insights that can inform care decisions and improve patient outcomes. This requires the use of advanced computational and analytics tools and techniques, as well as the ability to integrate data from multiple sources.
These demands are increasingly being met with additional technological innovations entering the conservative hospital world. Despite advances in data interoperability, such as the adoption of FHIR APIs and increasing leverage of Common Data Models (CDMs), the hospital IT world is still highly fragmented, and even the advent of Cloud Compute has not drastically reduced that fragmentation. Moreover, in order to meet these increasing needs, hospital IT must now implement a record number of new technologies in a short period of time, and despite a fairly consolidated EHR market, the rapid cycles of innovation and massive capital investments pouring into the Healthcare IT market have resulted in more vendors than ever before.
How could Federated Learning assist in clinical transformation?
First of all, let’s define federated learning. Federated learning is a powerful machine learning technique that allows for the training of machine learning ‘models’ on multiple disparate datasets. That means that data does not need to be shared/centralized in order to leverage it for creating powerful analytics and algorithms. Thus, federated learning could assist in clinical transformation by allowing hospitals to collaborate and share data in a secure and privacy-preserving way. If provided alongside a comprehensive platform that supports integrating with different infrastructures found in different hospitals, federated learning can assist in, alleviating many risks around data sharing such as compliance risks (e.g., HIPAA, GDPR regulations), data privacy risks (ie, that extend beyond the regulations), reputational risks (e.g., in the case of ‘data leaks’ and malicious use of data by a 3rd party) and financial risks (e.g., the investments required to setup massive repositories and leverage centralized ‘data lakes’). Some specific ways it could be used include:
- Eliminating the need to create multiple ‘data flows’ to the cloud, and keep all data under the ‘sovereignty’ of the hospital at all times.
- Reduce the need to integrate multiple technologies one-at-a-time, by having a standard communication method to connect (often) multiple internal ‘data silos’ with external data consumers.
- Facilitating foundational and clinical research by allowing hospitals to share data insights for research purposes, without driving an ever-growing amount of labor-intensive efforts (e.g., anonymizing, certifying, contracting and governing). This in turn drives translation of ‘better baked’ medical AI models into the clinical workflow, as well as better commercialization of inventions that have been afforded with external data validation (ie, the model works elsewhere) and market validation (users wish to use the model elsewhere).
- Capture value by increasing the value of data, and enabling hospitals to tap into new revenue streams such as the ones from drug development. Federated learning allows pharmaceutical companies to access sought-after ‘multi-modal’ data from multiple hospitals in order to identify potential drug candidates and evaluate drug safety and efficacy. In addition, this enables hospitals that implement new technologies to ‘leapfrog’ many traditional players that already capture value from such efforts, and ‘take out the middleman’ in the form of data intermediaries that erode the hospital’s value.
- Enhancing disease surveillance by sharing data insights across multiple jurisdictions to track the spread of diseases in real-time, identify outbreaks quickly and plan interventions accordingly.
A specific area that is mentioned above and is dear to my heart, is supporting the translation of algorithms created by researchers in hospitals. Federated learning supports this, often grueling and difficult, process by reducing the barriers for researchers and clinicians who would like to collaborate, and thus enabling what could often remain as a good publication, turn into a product used to improve patient care, and let hospitals once again take the lead on innovation. The sad reality today is that without being able to translate ‘homegrown’ models into clinical impact, the ability to capture value is reduced, and few models have been licensed to date. Ultimately, I believe ‘Medical AI’ will follow pathology in providing ‘lab developed tests’ as diagnostics. Lab-developed tests (LDT) are tests that are developed, performed and analyzed in-house by a laboratory. One of the key things keeping medical AI behind, is the need for much bigger datasets and need for much broader diversity in order to create and validate a performant model. These needs are not supported by the current paradigm of single-site research and ‘once in a blue moon’ massive collaborations (that usually use redacted data or small amounts of patient data). Federated learning can fix this.
Alongside the many promises, it is important to note that Federated Learning is a complex technology, it requires a well-designed infrastructure and a platform that supports a complex workflow, strong data governance and robust security protocols, a vendor that can take on liability and proper certifications in order to be implemented. FL also requires hospitals and researchers to work closely together to ensure that data is shared in a way that is both secure and compliant with regulations. I am a staunch believer in this approach, and believe that it will lead to (finally) the actual ‘democratization’ of healthcare data and innovation, drive adoption of responsible AI and ultimately lead to improved outcomes for patients and cost savings for the entire industry.
This article was written by Ittai Dayan, CEO of Rhino Health and co-founder from HIT Consultant and was legally licensed through the Industry Dive Content Marketplace. Please direct all licensing questions to legal@industrydive.com.
