Why aren’t we using the data we know we have to improve patient outcomes?

© Suriyo - stock.adobe.com

“Even internally, collaborating [on data] can be tough with privacy and protection requirements.” — Alexander Gusev, PhD

Efficient and scalable data flows are critical to every industry, and health care is no different. Data are one of the biggest challenges in improving patient outcomes in health care. The efficiency and scalability of any data flow is impacted by data security, privacy, governance and the overall maturity of data operations of those involved. Of course, personal health information is a protected data type, and the flow of these data is neither efficient nor scalable due to the bulky and manual processes required to satisfy security and privacy requirements. To simplify the conversion, we’ll divide data flows into two categories: direct data flows when providing direct health care services and indirect data flows to research groups identifying breakthroughs to be used by care providers in delivering better services.

Direct care: Sharing data between facilities

Adi Hirschtein
_{© Duality Technologies}

Direct care is when a care provider needs data from another care provider to treat a common patient.

Let’s say a patient has a heart monitor installed at a provider in California. While traveling in New York, this patient experiences chest pains and goes to the emergency room, as instructed. The facility in New York will most likely have to manually reach out to the California facility to coordinate access to the data. In some cases, it may be faster for the New York facility to rerun the same tests rather than wait for the process required for data sharing to occur.

Many facilities transfer sensitive data via fax, and requests are manual rather than self-service. Delays or failures in getting answers to critical questions can reduce the quality of patient outcomes.

Direct care: Delivering living health plans and care

CVS and other notable health care giants have been working toward a goal of a new type of living health care. The idea is that we can keep people healthier for longer if patients engage with health professionals before issues become problems. This type of service requires access to a wide range of personal and health data that we have but are strewn across various services, organizations and infrastructures: lab tests, genomic data, Internet of Things devices such as wearables, sleep trackers, scales and more. To accomplish such a goal, these health giants must figure out how to efficiently store, share and use these highly sensitive data while meeting compliance requirements.

Advanced models will be used to process all these data to come up with care instructions to improve the lifespan and health of patients. Living health plans could include preventive lifestyle changes such as a focus on high-quality sleep, which requires attention to gut health, diet, exercise and mental health. Living health would also benefit by incorporating recent breakthroughs in early detection of issues as we see with the study of pathogens, cancers and diseases at a genomic level as studied at Dana-Farber Cancer Institute, the Atlanta Genome Project and others. The problem at every step is not that the data do not exist but that they are very difficult to gain access to. This access typically takes time and requires processes like deidentification that reduce data quality and, therefore, insight quality and accuracy.

Indirect care: Discovering medical breakthroughs to improve patient outcomes

Indirect care is when medical researchers, pharmaceuticals and more need patient data (often, real-world data) to improve everything from early detection to treatments and cures. Naturally, care providers benefit from and want to support these breakthroughs but are challenged by the time, resources and regulations involved in such data sharing. In addition, requirements around data sharing for indirect care are far more stringent than for direct.

Essentially, improving areas like disease detection, prevention and treatment requires that care providers share data from real-world patients (as opposed to screened clinical trial participants); this is called real-world data (RWD). RWD must be analyzed for researchers and pharmaceutical organizations to discover medical improvements and breakthroughs that can then be passed to the care providers.

In talking with various researchers, the common challenge isn’t the value of what will be done with the data but getting care providers to trust that we can and will protect the data.

Improving patient outcomes with PETs-enabled solutions

There is a segment of technologies called privacy enhancing technologies (PETs). Such technologies span advanced cryptographic methods as well as hardware solutions. The various PETs are not in competition with each other but serve different purposes. In many cases, combining multiple PETs is the solution. Raw technologies themselves require deep expertise and time to leverage. The health care ecosystem needs a solution that operationalizes these technologies in a software layer that institutions can easily deploy and use.

“Now, [with PETs-enabled solutions] we can start recruiting other collaborators to show that it can work in a plug-and-play way for these hospitals…that will lead us to participation from far more institutions,” says Alexander Gusev, PhD, lead researcher at Dana-Farber Cancer Institute.

For direct care, PETs-enabled solutions allow providers to obtain information about patients in a secure and more self-service way. Another example would be to address the difficulty in using and sharing unstructured data: e.g., MRIs, X-rays, doctors’ notes fields. With PETs-enabled workflows, doctors at any facility will submit an MRI to an artificial intelligence model trained on tens of thousands of MRIs from many facilities to quickly receive a diagnosis based on the most recent medical discoveries. This could require a combination of PETs such as confidential computing infrastructure and federated learning to create a confidential federated learning flow to support the many disparate data sources while protecting both the model and data while these insights are generated.

Providers can use this new confidential federated learning for indirect care to train new models across data sets from multiple medical centers. This approach optimizes models for disease identification (e.g., cancer identification) while ensuring that data never leave their premises, thus remaining protected. This combination of PETs protects the model weights sent to the global server for aggregation, providing a robust, secure solution.

Every industry in the world is data-driven. The speed and scale of data operations must not sacrifice security or privacy. Traditional methods of using data leave companies vulnerable to breaches, like those experienced by UnitedHealth, 23andMe and many more. These breaches give pause to people being asked for consent for additional use and distribution of their data. PETs are specifically designed to protect data while in use, which not only closes major vulnerabilities and allows people to confidently provide consent but accelerates the work we already do while unlocking the value of data not previously accessible.

Adi Hirschtein is the vice president of product at Duality Technologies. He brings more than 20 years of experience as an executive, product manager and entrepreneur building and driving innovation in technology companies primarily focused on B2B startups in the field of data and AI.