As a new, first-time mom, I sometimes find myself diagnosing my daughter with rare and terrifying diseases. My baby wasn’t even a day old when I became convinced that she might have Hirschsprung Disease — until the hospital nurse who changed her diaper assured me her intestines were indeed doing their job.
As a physician, I know I should be more logical. After all, I do remember my medical school professors drilling into us that when we hear hoofbeats behind us, we should think of horses, not zebras. In other words, when making a diagnosis, clinicians should first consider a more commonplace explanation (the “horse”) versus the rare and more exotic disease (the “zebra”).
Yet sometimes, patients do have rare diseases.
The advocacy group Rare Action Network defines a rare disease as any disease, disorder, illness, or condition that affects fewer than 200,000 people in the U.S. An estimated 25 to 30 million Americans — nearly one in 10 — have at least one of approximately 7,000 identified rare diseases.
Rare disease diagnosis has challenges because often the relevant “clues” about a condition are not obvious until all other diagnoses have been ruled out. The clues may be in the patient’s chart but can be difficult to find if they are buried in the free text section of a chart note, in an old lab report, or as an unstructured comment in the patient portal. To uncover those clues, a clinician must manually comb through different sections of the EHR – which can be a time-consuming, error-prone, and often a non-feasible task. In addition, a deeper understanding can be gained from a review of published papers on the disease – but capturing the key information from literature databases is also a slow and painstaking task.
When critical information about a rare disease is overlooked, diagnosis can be delayed, and patient outcomes can be compromised. If clinicians are armed with better tools to augment their search efforts, however, we have the potential to advance the diagnosis and treatment of rare disease.
Leveraging augmented intelligence technologies to enhance diagnosis and treatment
To help clinicians diagnose rare disease more quickly and accurately, many healthcare organizations are embracing technology solutions to aid in the process. For example, natural language processing (NLP) tools can create augmented intelligence workflows that facilitate the rapid search of unstructured clinical data from multiple data sources, such as EHRs, patient portals, health information exchanges, or other systems.
Instead of manually searching through multiple records or databases, NLP software can look through both structured (discrete fields) and unstructured (free text) data and extract relevant details. This provides clinicians with a more accurate, 360-degree view of each individual patient so they can make a more precise diagnosis sooner, which is often critical when treating patients with a rare disease.
At the University of Iowa Stead Family Children’s Hospital, for example, researchers and clinicians wanted to minimize burdensome manual searches when identifying rare diseases. The organization successfully deployed NLP to deliver significantly faster and more accurate information that helped identify clinical phenotypes in infants based on clinical records. Using this method, the organization curated phenotypes 200X faster using NLP, compared to manual methods.
Leveraging the right data, NLP-driven analytics can similarly deliver a clearer diagnosis picture for whole-disease populations, allowing for the creation or enhancement of rare disease registries, the identification of cohorts for clinical trials, or finding insight on common signs/symptoms or comorbidities. When de-anonymized, data can be utilized by biopharmaceutical companies for insights in the development of new therapies.
Improving understanding and treatment of genetic disease
Consider, for example, how the biopharmaceutical company Shire used NLP tools to systematically examine gene-disease associations to assess potential disease severity for patients with Hunter Syndrome. Shire had innovated a therapeutic intervention that could greatly enhance the life of patients with a severe form of Hunter syndrome that presents at 2-4 years of age and progressively worsens, usually leading to premature death in the second decade of life. Before initiating a clinical trial, however, Shire wanted a means to identify patients with the greatest potential to benefit, in part because the new intervention was invasive and unpleasant for young children.
Shire research specialists used NLP text mining to map and compare Hunter syndrome patient genotypes and phenotypes, which would enable genetic screening for patients who could benefit from the intervention. The researchers created a suite of NLP queries to search unstructured scientific abstracts (published in PubMed) to identify and classify every patient with Hunter Syndrome or related symptoms, and also associated gene variants and mutations, This quickly yielded high value, relevant results. Researchers were then able to fine-tune the results to identify the location of specific mutations within a gene and associate specific gene mutations with specific phenotypes.
Shire’s use of NLP text mining produced excellent results that matched or even bettered results from other available genetic databases of reported genotypes, confirming the ability of NLP technologies to advance the understanding and treatment of rare disease.
Other pharmaceutical companies have used NLP in similar fashion to search large volumes of published literature for the nuggets of information on rare disease patients and find the associations with genes and gene variants. For example, Agios developed a virtual portfolio for orphan disease by using NLP to systematically map the space around inborn errors of metabolism and link diseases to targets. Stuart Murray, Research Fellow/Director Informatics, Agios Pharmaceuticals, said “Our rare genetic program was founded on an understanding of the global space of rare genetic diseases which we mined extensively with NLP to identify candidate diseases and candidate target genes.”
With augmented intelligence tools and NLP, clinicians and biopharmaceutical researchers are better equipped to work together to efficiently identify real zebras and advance the treatment and outcomes for individuals with rare disease.
Linguamatics is a proud sponsor of the annual Findacure essay competition, which raises rare disease awareness and highlights patient challenges among the doctors and researchers of tomorrow. More information about the competition is available here.