Getty Images

Supporting Providers with Innovative Data Extraction, Augmented Intelligence

Unstructured data is a major barrier to quality care, but data extraction and augmented intelligence can give providers the information they need to make the right decisions.

For healthcare providers, the key to delivering the best possible patient care is having a clear, complete picture of each patient's health history. That complete picture is made up of many data points from providers across the care continuum, but obtaining access to each of those unique data points often poses a problem for providers, given that so many different pieces of data are housed in disparate systems.

"In healthcare, there are so many systems where data is housed—the EHR, lab systems, pharmacy systems, payer systems, and even personal health records," says Christopher Larkin, Chief Technology Officer at Concord Technologies. "The question becomes: Who has the complete picture of either patient, the site of care (for example, an emergency room), or the intake process? There is a huge opportunity within health data exchange to integrate all these data sources to form a complete picture of the patient's information."

Each day, massive volumes of protected health data move between healthcare organizations to try and establish the complete picture needed for providers to deliver high-quality, appropriate care. But even with data flowing between health organizations, interpreting the patient picture they paint is often still a problem.

The reason that these disparate pieces of data are problematic is due to the way that it's structured – or more accurately, not structured.

Unstructured data is one of three main sources of critical patient information, the others being structured and hybrid data. But unstructured data is by the far the hardest for providers to process and implement at the point of care, according to Simran Bagga, Program Director - AI and Machine Learning at Concord Technologies.

Structured data is, of course, the easiest to process, as it lives in a clearly defined tabular format; unstructured data, by contrast, is the most difficult to process, as is it is not organized in any kind of pre-defined manner. Semi-structured data is then a hybrid of the two types: it doesn't rely on a tabular structure, but it does contain tags or other markers to denote fields and record hierarchies within the data.

As an example of the latter, Bagga points to clinical notes, a treasure trove of information with some grammatical structure. "In their notes, physicians don't speak in machine terms: 'semi-colon, Simran Bagga, Date of Birth, semi-colon,' But this is still a goldmine of data that has some degree of shape and form that is part of the patient's overall picture of health and must be included in that picture."

Because the majority of data that passes between healthcare facilities is not structured, critical information can be difficult for providers to interpret at the point of care. At best, this difficulty of interpretation can dramatically slow clinicians down as they try to rapidly put together the right care plan; at worst, it can leave care teams to make decisions based on limited views of a patient's health. Making sense of unstructured data is possible. However, providers typically must endure the manual process of combing through largely unstructured data sources to find nuggets of information critical to quality patient care. And like any manual process, it's prone to error.

But thanks to advancements in intelligent data extraction, forms of artificial intelligence can efficiently parse and highlight key data elements for providers to interpret more quickly and clearly. Therefore, automation can reduce the burden on providers by preparing the data for interpretation and analysis, allowing clinicians to practice at the top of their license.

However, healthcare professionals have some reservations about what automation means to clinical decision-making.

"Within healthcare, people are generally uncomfortable being in a situation where machines are making decisions about them or for them," Larkin says. "For this reason, augmented intelligence is so important in healthcare because of its ability to assist a human with administrative or clinical work. Augmented intelligence works by using algorithms and data to present information that a human can then interpret and act on."

By leveraging AI as a support system, though, healthcare organizations can rely on augmented intelligence to surface important information that enables skilled healthcare professionals to gain a comprehensive view of a patient's health and make timely decisions to improve outcomes from there.

"A holistic perspective is needed to understand the patient's care or the population's care," explains Bagga, who also sees augmented intelligence as a crucial support system for the implementation and use of AI.

"Healthcare workers take all aspects of patient care seriously, and they're aware of how different each patient's situation is. This variability can make healthcare providers understandably wary of automation," she says. "They do want to sit in the driver's seat, they do want the most complete data possible, they do want AI-driven insights, but they don't want the machine deciding for them. That's the difference between artificial intelligence and augmented intelligence."

So what is augmented intelligence, and where does intelligent data extraction fit into the mix?

"When we talk about augmentation of the intelligence," Bagga continues, "we're basically saying we're going to use AI methodologies to learn from and derive insights from data and present them to the decision-makers to use their clinical expertise to agree or disagree with the information presented."

The provider then finds herself in a unique position: receiving unified insight instead of disparate data points present across documents. Care providers shouldn't have to waste time hunting manually through inbound documents for the details they need.

"We want to give that person insight data from all of our AI methods so that individuals can make the best decisions for the patient, the best decisions for the site, the best decisions for the healthcare entity. That human intelligence is crucial to achieving the full potential of augmented intelligence," Larkin adds.

Particularly important to the development of augmented intelligence are both data labeling and annotation, training signals to the system that show where vital pieces of information live in a document and what form they generally take.

"Extraction of intelligence from faxed forms in healthcare is fueling innovation in health data exchange," Bagga emphasizes. "A key aspect of that technology is a term called 'data annotation' or 'data labeling.' Eighty percent of the work is data preparation. We are preparing annotations and label data from these documents to provide the correct output, or what we call 'the golden data,'          back to our users."

Augmented intelligence solutions can speed up the data handling process so that healthcare providers receive immediate access to patient data that will inform the right decisions at the right time. Taking them a step further, these solutions can lead to significant advances in interoperability by automating the input of data into an EHR from numerous systems, bringing all the pieces that complete the picture of a patient's health and well-being. By reducing the friction around health data exchange, providers and patients can have meaningful interactions that pave the way for quality and health improvements.

________________________________________

Concord is leading the development of new Artificial Intelligence technologies to extract data from documents and ease the burden of managing fax communications. With our 97% customer retention rate and delivery reliability that is unparalleled in the industry, Concord is committed to being the partner you can trust with your data.

Dig Deeper on Artificial intelligence in healthcare