As published in HealthIT Analytics, August 13th, 2018
Every healthcare organization across the care continuum is currently racing to gain an edge over their peers by using big data analytics to provide the best solution to the puzzle of the Triple Aim.
There are countless impactful strategies to help organizations reduce costs and improve outcomes by leveraging data, but the past few months have seen two intriguing concepts rise above the rest.
Artificial intelligence and blockchain have quickly become the tools of choice for developers, providers, and payers looking to bulk up their health IT infrastructure with innovative, effective data management capabilities.
Both purport to offer speedy and substantial returns on investment, and both have been lauded as game-changing for providers and payers looking to get ahead.
However, both are admittedly near to the peak of their respective hype curves, leaving some providers and payers wary of diving in too deep and too soon.
Concerns over security, utility, and return on investment are common, leading many organizations to put one or the other on the back burner until someone more adventurous has taken the time to work out all the kinks.
There may be some merit to this approach, especially for organizations that find themselves close to foundering in the midst regulatory uncertainty, value-based care initiatives, and the challenges of the modern cybersecurity landscape.
But those who take a bold approach to machine learning and blockchain are likely to discover that combining these two emerging technologies can help them overcome some of the fundamental challenges of securely exchanging the big data sets that support innovative analytics.
“Machine learning is very data-hungry, and some of the limitations of deploying machine learning into the healthcare environment revolve around the problems with accessing and drawing data from one silo to another,” explained David Houlding, MSc, CISSP, CIPP, Principal Healthcare Program Manager at Microsoft Corporation.
“With limited data, you can only produce limited models that tend to have higher error rates. That is simply not acceptable in healthcare, especially if you are trying to make a diagnosis or predict a clinical event.”
While federal, academic, and research organizations are beginning to open up the doors to their big data treasure houses and allow broader access to curated data sets, the vast majority of data resources remain behind lock and key in individual institutions.
There are plenty of good reasons for this, Houlding said, many of which revolve around the fundamental requirement for healthcare organizations to keep patient data private and secure.
“HIPAA says to protect the confidentiality, integrity, and availability of data, which leads a lot of organizations to err on the side of caution and say that they aren’t going to share what they have with the community, just in case,” he said.
“There is certainly some logic to that – no one wants to be the subject of the next data breach headline. But that’s where blockchain can come in. It has the potential to solve a large number of the issues that are stunting the deployment of artificial intelligence for healthcare purposes.”
BLOCKCHAIN AS A CATALYST FOR TRUSTED DATA DISCOVERY
Blockchain is a shared digital ledger that creates an immutable record of transactions taking place between participating members of the community. Each member of the community holds his or her own copy of the ledger, which allows the community to come to a consensus regarding the validity of any new transaction.
If enough of the members cryptographically agree that the transaction is valid, it will be recorded on the blockchain and turned into an unchangeable “fact” with a timestamp that cannot be altered or erased.
“With limited data, you can only produce limited models that tend to have higher error rates. That is simply not acceptable in healthcare.”
Blockchain is viewed as an inherently secure technology due to the infeasibility of a bad actor altering every single copy of the shared ledger across all members of the community to change or erase the record of a transaction.
Healthcare stakeholders have identified a number of attractive use cases for this highly trustworthy data management structure, including governing claims transactions, monitoring pharmaceutical supply chains, and tracking professional credentials or certifications.
Houlding believes that blockchain is also ideally suited to becoming the fabric of trust that binds together disparate organizations and creating a secure environment for meaningful, collaborative health information exchange for developing artificial intelligence.
“The idea is to use blockchain as a mechanism to produce metadata about the datasets available at multiple organizations,” he said. “The metadata of the shared blockchain facilitates the discovery of information and the subsequent secure, peer-to-peer exchange of that data.”
This strategy allows blockchain to fill a role similar to that of a DICOM header: the blockchain creates a “pointer” to where the full data set is stored, allowing for discoverability, without requiring the entire data set to move each time someone wants to conduct a transaction.
As a result, organizations will be able to keep sensitive data, such as protected health information (PHI) and other personally identifiable information (PII) off the blockchain, reducing the risks of a breach or other improper usage.
“Our general guidance about what data to put on the blockchain is ‘minimal but sufficient,’” said Houlding.
“Keep the PII and PHI off the blockchain. That data should be stored in secure access, controlled enterprise systems. The blockchain is more for metadata: audit trails, cryptocurrency information, provenance information, and maybe some transactional data.”
A lightweight blockchain that only records metadata still has enormous value for artificial intelligence researchers, Houlding stressed, especially when it comes to tackling one of AI’s biggest pain points: unintentional bias in the data that can skew the results of a clinical decision support model or other algorithm.
“Our general guidance about what data to put on the blockchain is ‘minimal but sufficient.”
“The point of machine learning is that models change and adapt over time, so using an immutable tool like blockchain to monitor those changes can create an audit trail that allows developers to track that evolution and make adjustments accordingly,” he explained.
“Metadata stored on the blockchain allows us to track the provenance of data that goes into training an artificial intelligence model. If bias creeps into the model at some point, you can look back through the blockchain to see what data is causing the biasing and alter the model accordingly.”
In addition to preventing biasing during development and training, blockchain can help healthcare organizations validate the results of AI models as they mature.
Currently, validation is a relatively slow and laborious process. Clinical algorithms, for example, must typically be validated against data annotated by experts, which can in itself be fallible or hard to collect at scale.
Tools that purport to offer decision support must be thoroughly vetted for accuracy and safety before they can be integrated into the clinical environment, creating an incentive for researchers and developers to speed up the validation process.
“If the validation process is shared between organizations using blockchain, then participants could share those tasks and learn to trust a shared model faster,” said Houlding. “And blockchain offers the additional benefit of allowing for tokens or cryptocurrency to incent collaboration and sharing or the development of some new capability.”
Developing a financial ecosystem around model validation could be an important motivator for researchers, and may create a positive competitive atmosphere that accelerates artificial intelligence innovation even further.
CREATING A COLLABORATIVE BALANCE ACROSS PUBLIC AND PRIVATE COMMUNITIES
Blockchain comes in two main flavors: public chains that are open to any participant from anywhere, and private or permissioned blockchains which limit participation to pre-approved entities.
The major cryptocurrency platforms, such as Bitcoin and Ethereum, are fully public exchanges. There is no prerequisite for being allowed to engage in trading – just like any traditional monetary system, being in possession of currency automatically qualifies the individual to use it, and anyone is allowed to acquire currency.
Similar public blockchains have value for healthcare, says Dr. Jack Neil, a practicing anesthesiologist and Chief Technology Officer at MedStream and CEO of Zather, Inc, but a fully open community also brings challenges.
“Every participant needs to be able to see the entire ledger for most healthcare applications, which means a public blockchain is probably ideal from an architecture standpoint,” he told HealthITAnalytics.com.
“You don’t have to take someone’s word for it that they have modified or deleted something, because the entire chain and all its transactions are visible to members of the community. That is a significant change from the way we do transactions right now, and it’s a step in the right direction.”
“But a fully public blockchain is fully decentralized,” he continued. “If you are working with health data, you are going to need some sort of administrative function to govern the involvement of very sensitive data and make sure there is adherence to privacy rules. Any form of centralization is a hard pill for real blockchain evangelists to swallow, but it will likely be unavoidable for healthcare.”
Truly decentralized public blockchains also make it impossible to change or update code, Neil added.
“There’s no ability to perform customer service,” he said. “Once you write the code and send it out into the world, that’s it – it’s final. If you haven’t foreseen every single possibility that could come up before you hit ‘send,’ you’re stuck with what you’ve got.”
“There may be some situations where that’s acceptable, but it seems unlikely that anyone in healthcare would be comfortable without the ability to make adjustments if necessary.”
“Any form of centralization is a hard pill for real blockchain evangelists to swallow, but it will likely be unavoidable for healthcare.”
Houlding agrees that the Platonic ideal of a completely decentralized blockchain may not be appropriate for the majority of healthcare use cases in the present landscape.
“The need for governance doesn’t go away just because you start using blockchain,” he said. “Code doesn’t write itself. It gets written by people, and those people need to consult with the entire consortium to make sure the code meets the requirements and parameters for the situation at hand.”
“Blockchain may turn the role of the intermediary into a governance role, which is good for democratizing data access and reducing bottlenecks, but it doesn’t eliminate the need for someone to create consensus across disparate groups.”
Consensus is the key to how blockchain operates, and it is exceedingly difficult to generate efficiently on a truly open, public blockchain, says Emily Vaughn, Director of Blockchain Product Development at Change Healthcare.
In a public model where anyone can join, all participants remain anonymous to one another to preserve their privacy, which sounds as if it should be highly valuable to healthcare organizations. But anonymity can swing both ways.
In such an environment, “it is very difficult to understand the intent of the person who is going against what everyone else is saying or acting in a way that raises suspicions,” Vaughn said earlier in 2018. “That means that the consensus model has to be designed to account for anonymous individuals who may be malicious.”
“Bitcoin and Ethereum have to deal with that by creating much more complex consensus mechanisms for making sure everyone is in agreement. That requires a lot more processing power, which can slow down the network and limit the volume that can move through the system.”
In private or permissioned communities, on the other hand, all of the participants know each other and understand that they all share a common purpose within the community, creating a backdrop of trust.
“That is why the vast majority of healthcare organizations piloting blockchain right now are using private or permissioned blockchains,” pointed out Houlding.
“With a private, permissioned blockchain, healthcare organizations can be certain that they are only inviting highly trusted participants that are already known to them.”
Healthcare organizations may be particularly inclined to embrace private communities due to the fact that blockchains acting as “pointers” to artificial intelligence training data are, in fact, connected to the systems where those full datasets are stored, he added.
“Blockchain is not a stand-alone thing,” said Houlding. “It can augment enterprise systems, such as the electronic health record or a payer platform. The blockchain component may be acting in a decentralized way, but it will still be connected to those enterprise systems.”
“The EHR or the claims management application acts as the user interface for blockchain. Blockchain is simply the behind-the-scenes engine that will make the systems we’re used to seeing more secure and interoperable.”
“In a private community, providers don’t have to worry about some random person in a random country accessing the public community and getting their hands on something they shouldn’t.”
Instead, if a member of a permissioned community acts in a way contrary to the best interests of the consortium, they can be easily identified and removed from the group – a threat which could have larger financial implications for any organization that gains a bad reputation for managing privacy and security.
“In a private community, providers don’t have to worry about some random person…getting their hands on something they shouldn’t.”
“Not even blockchain is zero risk in terms of security,” Houlding cautioned. “There is no such thing, especially when there are humans involved. But blockchain does help to mitigate the chance of a negative event, and gives more visibility into what happened when a negative event occurs. I believe the benefits will significantly outweigh the risks when using this technology.”
Vaughn believes that permissioned communities are the natural place for healthcare organizations to start, but she envisions a shift to more open and public blockchains as the technology matures.
“In order to get healthcare stakeholders comfortable with the idea of blockchain, especially more open blockchains, we have to prove that it’s viable and prove that the value will be greater than what we have at present,” she said.
“I do believe that we will have permissioned blockchain networks that will create enterprise value within the next few years, and we will certainly keep the goal of more open blockchains in mind as we build on the technology and really start to find out what it can do.”
CREATING THE RIGHT CONDITIONS FOR ADOPTION
Healthcare organizations are, understandably, reticent to embrace any new technology that might have an impact on the way they protect and manage their data, observed Neil, which may slow the adoption of blockchain across the industry.
“Healthcare isn’t known as being an early adopter of pretty much anything,” he said.
“We’re just now adopting cloud – we probably won’t get to blockchain for a while. A lot of chief information officers are very wary of anything that has the word ‘blockchain’ in it, because they know it’s still early in its evolution and they are not willing to be the first to experiment and potentially expose their organizations to a problem.”
“That’s completely understandable – it’s very similar to where we are with some artificial intelligence applications, too,” Neil continued. “It’s an uphill battle to get some organizations to develop confidence either of those things. Providers are very results-oriented and they want to see proof that they – and their patients – can benefit from that kind of investment.”
The proof is not yet sufficiently forthcoming for a large number of CIOs, a Gartner survey from May of 2018 revealed.
Across multiple industries, only one percent of CIOs said that they have adopted any kind of blockchain technology, and just 8 percent had any near-term plans to do so.
Participants identified the lack of available development talent as a major contributor to their reluctance to dive into the world of blockchain. Changing the culture of the IT department was another significant challenge.
Adoption was highest in the financial services and insurance industries, which doesn’t surprise Houlding.
“Blockchain has found a home in the financial services, because blockchain is really good at transaction-based use cases,” he said. “There are a number of very similar opportunities in the healthcare industry where B2B networks already exist – and that is where blockchain is slowly but steadily gaining traction.”
At Change Healthcare, blockchain is performing that supplementary role by operating in parallel with the company’s existing claims management technology, said Vaughn.
“It provides an additional window into the status of a claim based on data captured on the blockchain, which can be very helpful for providers waiting for reimbursements to clear,” she explained.
With $2 trillion moving through Change Healthcare’s claims processing tools each year, the added visibility creates value by helping providers plan their revenue cycles more effectively.
Blockchain also has the potential to streamline existing networks that are simply too nebulous and complex to be effective in their current state, said Houlding.
“In the case of provider identity management, for example, there is a lot of waste that is easy to identify but harder to eliminate using the traditional tools we have at our disposal,” he said.
Multiple industry efforts are underway that aim to use provider identity management as a high-value prove point for blockchain.
Currently, obtaining proof of licensure or certification typically requires multiple phone calls, faxes, and paper-based verifications, all of which have to be repeated for each site of care where an individual is seeking authorization to practice.
“There is a lot of waste that is easy to identify but harder to eliminate.”
In March of 2018, the group released a provider identity tool that uses blockchain to securely exchange data related to a practitioner’s credentials to practice at certain levels or locations.
The system doesn’t replace the need to produce the set of documents required by each state medical board or other associations, says Hashed Health CEO John Bass, nor does it constitute a new or separate credentialing process.
Instead, “an individual can put a certificate of attestation of licensure on the blockchain and other states are able to see that on the network,” Bass explained to HealthITAnalytics.com in October of 2017. “That offers a faster, better, cheaper way to ensure that everyone remains up-to-date without all the back-and-forth of traditional communication.”
“It’s a good starting place because the information involved isn’t overly sensitive. It isn’t a politically charged issue, and there aren’t a lot of existing competitive interests. Everyone is searching for a better solution to a shared problem, which makes it a perfect place to innovate with low barriers to entry for those who want to try new strategies.”
Hashed Health isn’t the only group that shares the opinion. In April of 2018, five heavyweight stakeholders - Humana, MultiPlan, Optum, Quest Diagnostics and UnitedHealthcare – announced a plan to tackle changes to provider demographic data in a similar effort to drive waste out of the industry.
“Health is hard, but great breakthroughs may come from industry players collaborating around emerging, innovative technologies to make life easier for doctors and patients,” said Busy Burr, Vice President and Head of Healthcare Innovation and Trend at Humana.
“We think industry leaders can work together to eliminate technology barriers in the health care system and promote better health outcomes.”
Houlding believes that these early projects addressing transaction-based use cases will provide the substrata of trust necessary to bring blockchain and artificial intelligence closer together.
“These prove points are a little bit easier to get our hands around, and they will act as a necessary stepping stone along the evolutionary curve to discovering the intersection of innovation and value,” he said.
“The blockchain use cases that achieve satisfactory ROI will spread, and then we’ll start to see rapid growth in terms of layering on additional functionalities in more areas of the industry. I firmly believe that data sharing to fuel artificial intelligence will be one of the functionalities that we will see sooner rather than later.”
“As the industry becomes more familiar and comfortable with blockchain, it will support the flow of data across disparate systems, and we will be able to use blockchain as a foundation for AI, machine learning, and big data analytics. We will see a very rapid acceleration of innovation, and blockchain will be at the core of those breakthroughs.”