Capturing Critical Data Hidden in Health Records

A new industry-academia partnership hopes to leverage the latest language and data analysis software to transform the vast trove of unstructured healthcare data into actionable insights that could affect healthcare at large and on an individual patient level.

The Maryland-based analytics firm Inovalon this month announced a new collaboration with the University of Maryland’s Center for Health Information and Decision Systems (CHIDS) to bolster the company’s “ultra-high-speed” health data analysis platform. The goal is to use natural language processing, machine learning, and deep learning technologies to quickly and accurately analyze the unstructured data contained within medical documents and other clinical records.

Ritu Agarwal, MBA, MS, PhD, the founder and director of CHIDS, said the amount of health data available is massive, with estimates suggesting 2300 exabytes will be available by the year 2020. An exabyte is 10 to the 18th power worth of information units, or 1 billion gigabytes. However, Agarwal told Healthcare Analytics News™, all of that data comes in a wide range of formats and types.

“Think about images created by scans, notes made by the doctor during a consultation, experiences that a patient might share about their medication side effects in an online forum—the ability to harness the power of these diverse data in a timely fashion can contribute significantly to decision making quality,” said Agarwal, who is also senior associate dean for research and chair of information systems at the university’s Robert H. Smith School of Business.

“The insights generated through analysis of unstructured data when combined with traditional analytics definitely have the power to be transformative for discovery of new knowledge,” she added.

The hope is that the new capabilities developed through the partnership, along with Inovalon’s cache of EHR data, will force a major leap forward in healthcare analytics and decision systems.

Agarwal declined to describe the specific focus of the collaboration, which has not been made public. But a wide range of potential applications exist if she and her colleagues can find ways to better capture otherwise unreadable unstructured data, she said.

“One example is medication adherence, which is more likely caused by patient behavioral factors,” she said. “Traditionally, this type of information is not captured formally, but is embedded in the physician narratives. Being able to extract this type of information can help design intervention plans to improve patient decision-making in regard to medication adherence.”

Agarwal said such data can potentially be used to help physicians base treatment decisions on the outcomes of similar patients. This kind of patient-specific predictive modeling could bring about a long-awaited new era of healthcare efficiency.

“We could perhaps enable the vision of personalized medicine to become a reality,” she said. “[Artificial Intelligence] is also getting better at offering the ability to offer triage, decision support, and care management on demand for patients, helping serve as a catalyst for complementary digital therapies, and potentially relieving workforce bottlenecks.”

Aside from the technological implications, Agarwal said the deal shows how industry and academia can combine complementary skill sets to improve public health.

“At the Center for Health Information and Decision Systems, this is our preferred model for working with industry partners,” she said. “Such knowledge co-creation between industry and academia doesn’t happen frequently enough, in my opinion.”