- Investing in big data analytics is a problematic proposition for many healthcare organizations, which often face the perennial issue of purchasing technologies that become outdated before the implementation process is even finished.
New systems, tools, applications, and interfaces are revealed on a weekly basis, as innovative developers clamber for a piece of a rapidly growing market – and each novel offering seems to bring its own crop of inscrutable buzzwords.
“Cognitive computing,” “graph databases,” and “semantic computing” are some of the terms that have exploded in popularity over the past few years, as tech giants like IBM pour billions of dollars into cutting-edge analytics technologies that leverage the latest in computer science.
But what do these terms really mean, and why are they so important for healthcare? In this installment of our advanced data analytics series, HealthITAnalytics.com explains the basics of semantic computing and how this approach to healthcare data might soon revolutionize the industry.
What is semantic computing, anyway?
Semantic computing relies on the notion that computers can be “taught” to approach concepts and problems in a similar way to humans. By linking together certain natural language concepts instead of just solving mathematical equations, computers can make inferences about data sets that might not be hard-coded into the system from the start.
Semantic analytics requires curators and data scientists to write algorithms that group carefully developed categories of data elements, such as patient names, diagnoses, locations, or economic statuses, into possible relationships, instead of creating a new equation involving specific data elements each time there’s a new question to ask.
For example, a provider may have a list of patients in a traditional relational database that includes Joan Smith and Maria Smith. That is very handy when Joan and Maria come in for their yearly physicals, but not so meaningful when Joan wants to talk about her family history of breast cancer or heart disease.
That’s because the sentence “Maria is Joan’s sister” requires a semantic understanding of key terms. “Sister” implies a relationship between three data elements: “Joan,” “Maria,” and the hidden, shared link that is “Joan and Maria’s mother.”
A human can understand this concept in a split second, and may not even think about the logical leaps it takes to make the connection. In order for a computer to come to the same conclusion, however, it would need to be explicitly told that “Joan Smith” and “Maria Smith” are siblings, that both Joan and Maria are female (and therefore sisters instead of brothers), and that they share the same biological mother.
While it is possible to infer that two patients with the last name “Smith” might be related, there is no guarantee that this is the case. Without a stronger connection between the two entries, providers run into patient matching risks that could have serious consequences for patient care.
And when more data elements are added to the problem, the semantic gymnastics get even more complex.
“Who is Joan Smith’s cousin?” is an easy question to answer with a semantic approach. A cousin is an individual’s parent’s sibling’s child. In this case, tracing that relationship from Joan’s perspective leads straight to George, even if he doesn’t share anything else in common with her.
But with a standard relational database that only includes a simple list of Joan, Maria, George, Caroline, Joe, and Betty, there’s no way to know that the Smith family and the Jones family are even biologically related, let alone that Joan Smith is referring to George Jones when she tells her physician that her cousin has diabetes.
Why is this approach important for healthcare?
Population health management and clinical decision support are all about relationships, too. The relationships between a patient and her environment, or a medication and an allergy, require computers to understand how one category of data elements impacts another to produce a result.
As providers become increasingly responsible for everything that happens to a particular patient across multiple care settings, having a good grasp of the full scope of these relationships is becoming critical for reimbursement and better outcomes.
Many healthcare organizations have been working quickly to integrate multiple sources of patient information to perform big data analytics that will allow them to assign risk scores to patients or flag missed opportunities for preventative care.
But while traditional relational databases can help uncover important insights, semantic computing can help providers go above and beyond the basics of population health management or responsive, point-of-care clinical decision support.
The benefits for patient management are clear, as HealthITAnalytics.com explained in the first installment of this series. With a minimal number of data sets linked semantically, providers can unlock a deeper, more comprehensive understanding of what challenges prevent patients from achieving optimal health.
When it comes to clinical decision support for precision medicine or personalized care, cognitive computing and semantic analytics can also change the way clinicians approach patient care.
If a computer can mimic the human ability to make inferences across seemingly unrelated data sets, yet can also store millions of times more data without memory lapses, it should be able to out-perform even the most brilliant clinical mind.
Developers haven’t yet created a true artificial intelligence that can replace a discerning, experienced physician just yet, but the technology holds great promise for patient care.
Cognitive computing engines can make connections between genes, drugs, treatments, and previous experiments that may never occur to a human clinician, and may be able to accelerate the development of therapies for cancer, neurodegenerative diseases, and other conditions targeted by the Precision Medicine Initiative.
Who is leveraging this type of technology?
Self-aware, all-knowing supercomputers may still be in the realm of science fiction, but they may not stay there for long. Semantic computing already underpins some of the most common consumer technologies, such as the digital personal assistants available on many smartphone platforms, and cognitive computing functionalities are being put to good use in healthcare organizations every day.
Semantic analytics can help providers save time and money due to the flexible nature of the underlying architecture, said Parsa Mirhaji, MD, PhD, an informaticist and researcher at Montefiore Medical Center. Montefiore’s data lake relies on semantic technology to tackle emerging questions in population health.
“We don't have the time and resources to build silos and specialized systems for specific needs,” Mirhaji said. “Relational databases require a very fine structure that you have to plan out before you can use it - you have to frame your problems in a very specific way.”
While relational databases can offer some powerful functionalities for healthcare providers, their static and rigid structure requires data scientists to build each database for a specific, limited purpose. Mirhaji compares the process of developing traditional big data analytics tools to building a city, where everything must be pre-planned in order to function together.
“The costs of changing your mind or your requirements are huge,” he pointed out. “And that's why you end up with these data silos. You end up with different architectures for different problems, because you have to box the problem before you begin.”
While IBM Watson may be the most famous cognitive computing option at the moment, organizations like Montefiore and Partners Healthcare have been diligently developing their own semantic computing offerings for both in-house use and commercial applications.
Other companies such as Microsoft, Epic Systems, and Dell are developing the core competencies to integrate semantic computing into their offerings for healthcare and other industries.
If semantic computing is the future, why invest in traditional big data analytics anymore?
With cognitive computing looming on the horizon, it may seem fruitless to spend money on current iterations of population health management or clinical decision support technologies. But for many healthcare organizations, holding off on development until semantic tools become commonplace simply isn’t an option.
Accountable care and value-based reimbursement contracts that require robust population health management capabilities aren’t waiting for the market to deliver cheap and mature cognitive computing technologies.
Providers looking to make the leap into leveraging their big data for these new payment arrangements may not have the luxury of taking the next three to five years to design sophisticated next-generation analytics system from scratch. They must harness the data they have right now with the tools currently available to them.
That may not seem like a great reason to invest in a large-scale big data initiative, but the truth is that for many healthcare organizations, a good relational database can help accomplish many of the tasks necessary at this point in the health system’s evolution.
Cognitive computing and semantic databases will certainly open up additional options for advanced patient care, but for organizations still investigating how to identify their diabetic patients or predict medication non-adherence, the implementation of traditional tools can make an enormous difference.
For healthcare providers, it is not a question of choosing either semantic analytics or traditional analytics, but a matter of adopting the right tools at the right time to do produce the highest possible quality of care.