To effectively diagnose and treat rare diseases, clinicians and researchers rely on genetic and diagnostic testing as well as phenotypic data to inform their decision-making.
For example, a patient’s commonly reported phenotypic data, which include quantified observable traits such as short stature, low-set ears and blood biochemistry, is generally insufficient to yield a definitive diagnosis. However, in combination with additional genetic data, phenotypic data offers the potential to unlock life-changing diagnoses for patients with rare cancers, pediatric diseases, inherited genetic syndromes and other conditions.
While the proliferation of testing and wealth of data continues to expand the knowledgebase for rare disease, this also creates challenges for clinicians and researchers seeking to pinpoint precise information to answer key questions. To reduce time-consuming manual searches and improve their findings, many organizations are turning to artificial intelligence technologies such as natural language processing (NLP). This enables organizations to more quickly and accurately assess and normalize far greater volumes of phenotypic and test data.
Understanding the challenges of rare disease diagnosis
A rare disease is defined as any disease, disorder, illness or condition that affects fewer than 200,000 people in the U.S., according to the Rare Action Network. An estimated 25 to 30 million Americans, or nearly 1 in 10, have at least one of approximately 7,000 identified rare diseases. There are more than 500 types of rare cancers, and all pediatric cancers are considered rare.
To initiate treatment as soon as possible for patients with rare disease, rapid diagnosis is essential – which is where NLP enters the picture. NLP automates the mining of complex, unstructured data so it can be transformed into curated, well-structured data that informs research and analysis. NLP enables the rapid reading, understanding and translation of the nuances contained in medical documentation, including free-form text in the notes and test result sections of electronic health records systems.
In cases of rare diseases, NLP can transform free text to Human Phenotype Ontology (HPO) terms to capture phenotypic data that is created when a patient is referred for testing. Clinicians can then use this to better understand the results of genetic tests based on any phenotype presentation. The presence of a genetic marker does not ensure that the disease itself will be present now or in the future.
NLP in action
Clinical and research institutions are adopting NLP to capture unstructured information from EHRs to improve diagnosis and identification. The following are two examples of NLP in action.
Identifying heart-disease patients
Researchers at a large health provider in California sought to understand the prevalence of aortic stenosis in their patient population. Because health systems often use procedure or billing codes that lack clinical nuance, it can be difficult for researchers to precisely identify complex clinical conditions among their patient population.
To overcome this challenge, researchers used NLP to comb through more than one million patient records and echocardiogram reports to identify certain abbreviations, words and phrases associated with severe aortic stenosis. In just a few minutes, the technology identified 54,000 patients with the condition, a feat that would have likely taken researchers years to accomplish using traditional methods of manual search.
The organization’s success pinpointing previously unidentified patients with complex diseases points to the promise for NLP to efficiently identify other conditions, empowering clinicians to more effectively manage the health of patients with rare disease.
Advancing personalized medicine research
Personalized medicine considers how a patient may respond to therapy based on their specific phenotypic and genetic characteristics. Clinicians and researchers seeking to advance personalized medicine, however, struggle to find the phenotypic information needed for personalization because details are often buried within EHRs in an unstructured format.
In their efforts to build a more comprehensive picture of individual patients with chronic disease, researchers one medical school leveraged NLP to extract unstructured information about patients with conditions such as Alzheimer’s disease, breast cancer, lung cancer, diabetes and obesity. The automated discovery of critical patient information has helped the organization advance its biomedical practices, such as identifying patterns in data and developing predictive analytics to determine patient outcomes.
Because rare diseases, by definition, are uncommon, patients who suffer from these complex and unusual conditions are often misdiagnosed and undertreated. Thanks to advances in AI-based technologies such as NLP, researchers and clinicians are now empowered to quickly gain new insights into rare diseases, offering hope for faster and more accurate diagnosis and personalized treatment.
Photo: Peach_iStock, Getty Images
