Healthcare data is often stored and processed in silos with only a fraction of the information used to profile the risks of patients. So what if there was a tool to change that?


“Imagine a scenario where two individuals are applying for enrolment in a health plan,” says Anand Shroff, Founder and former Chief Development Officer of Health Fidelity, a California-based healthcare tech startup. “The first individual is 75-years-old, with diabetes and chronic obstructive pulmonary disease (COPD). While the second individual is young and healthy. Under a value-based care model where payments are determined by patient outcomes, the health plan is likely to enroll the latter and avoid a chronically ill person totally.

“This is where risk adjustment reimbursement methodologies kick-in, to remove the perverse incentives that could potentially steer health plans away from patients with chronic conditions.” Yet, in reality, individuals’ characteristics are not so distinctly defined as the above example. Often we are only using a fraction of the patient-specific clinical knowledge and information that exists to determine if they are most at-risk or would benefit from certain interventions. Determined to create a change, Shroff founded Health Fidelity in 2011, with the aim to leverage natural language processing (NLP) and inference platform to analyze unstructured data for clinical and financial insights.

Health Fidelity’s NLP engine, known as Lumanent Insights, was developed in close collaboration with Dr. Carol Friedman, a professor at the Department of Biomedical Informatics of Columbia University. Over the years, Dr. Friedman continued to help in the advancement and refining of the engine’s accuracy through a supervised machine learning feedback loop to ensure a blend of clinical specificity and general linguistics applicability contributes to its robustness.

The company has since developed a suite of Lamanent products targeting different healthcare workflows. “We built the engine to benefit from, and amplify, human expertise and to recognize general medical language patterns,” says Dr. Friedman. “As a result, it understands clinical language and how it’s used, even as physician notes and health care facilities change, without having to be retrained.”

“The focus of health IT innovation like Lamanent, is for people to get better care from other people by delivering insights based on all available information — clinical results, claims data, patient-specific diagnostic data, etc — at key decision points along the patient lifecycle,” adds Steve Whitehurst, CEO of Health Fidelity. “Technologies such as natural language processing are removing the barriers to accessing, analyzing, interpreting and utilizing this information, freeing it to be delivered at the precise moment when it can inform and improve decisions by those who need it.”

Shroff notes the benefit of using NLP is twofold. “NLP engine brings precision; the ability to detect clinical findings accurately,” he explains. “For example, if a patient record states that the patient suffers from diabetic neuropathy, the NLP engine will not surface diabetes but diabetic neuropathy.” An NLP engine is also able to recall as many clinical findings as possible. “An NLP engine will not only recall 25% of the clinical findings in a medical record but everything,” Shroff adds. “Moreover, some of the more competitive NLP engines, like Lamanent, have extensive lexicons that include commonly used acronyms, shorthand, and other jargon. They also include grammar and disambiguation modules that allow them to successfully navigate different writing styles.”

Shroff believes any health institution could begin their NLP journey if they fulfilled three criteria. The first is the availability of data and the infrastructure required to process NLP. “Organizations get data from multiple data sources and many different EHR systems,” Shroff says. “To be able to organize that data and meaningfully manage them through NLP is a big hurdle.” The second is organizational alignment around priority use cases and expected outcomes. “Unless there’s agreement on which use cases deliver the most value to the organization, it’s challenging to find long-term funding for the use of NLP,” Shroff continues. The third is incorporating NLP-generated results into existing workflows.

In the long run, Health Fidelity aims to empower NLP with voice technology. “Voice technology is already being used widely by physicians to document patient encounters and there are a host of companies trying to bring voice-powered virtual assistants into the clinical setting,” says Whitehurst. “Natural language interfaces, especially voice, are slowly restoring a more organic, familiar means of interacting with technology. As it proliferates and relevant clinical data is more effectively captured and channelled to the point of care, patients and clinicians will be able to re-engage in a more humanistic way.

“Treatment plans will then be far more specific to a patient’s unique circumstances and background. This is Health Fidelity’s ultimate mission, but it should also be the mission of health IT, collectively, even as we acknowledge that the deployment of new technologies has so far stood in opposition to that experience.”