Diagnosing a rare genetic disease can be like putting a puzzle together. Doctors work to create a complete picture from many different pieces. They use lab and test results, bits of patient information, and limited knowledge about more than 7,000 rare diseases. The final picture is often unclear while pursuing a diagnosis—a process that can take time, money, and an emotional toll on patients and their families.
The good news is that efforts are being made to improve the time it takes to reach a rare disease diagnosis. Globally coordinated movements, such as Rare Disease Day (on February 28 this year), shine a light on the challenges posed by rare conditions including Gaucher disease. And leading-edge companies, such as Virginia-based ThinkGenetic, Inc., are working to shorten the diagnostic journey for people with rare genetic diseases. To meet their goal, they’re developing solutions using artificial intelligence (AI), the ability of computer systems to perform tasks that usually require human intelligence.
In 2017, ThinkGenetic, Inc. introduced SymptomMatcher, an AI tool that matches a patient’s self-reported symptoms with those of rare diseases. With their newest AI solution, ThinkGenetic, Inc. now hopes to identify patients with rare diseases who are currently undiagnosed. It’s called FindEHR, which stands for “Finding Individuals Needing Diagnosis through Electronic Health Records.” FindEHR leverages patient data and AI to screen patients for rare genetic diseases.
We spoke with two members of the FindEHR team: Jessica Dronen, MS, LCGC, a genetic disease research and information specialist, and Sarika Kondra, PhD, an applied researcher. Their collaborative efforts help build AI models that combine modern machine learning with genetic information. We asked them to explain how emerging technology is meeting the challenges facing rare disease diagnosis.
Here’s what you need to know about how AI tools like FindEHR are affecting people living with Gaucher disease and other rare genetic disorders.
How We Use AI in Healthcare
Healthcare systems collect a lot of data about patients, conditions, and treatments. They accumulate structured data, such as lab values, patient demographics, and other defined information. They also have unstructured data for each patient treated, such as clinical notes and reports written in free form by medical professionals.
All this data is protected by the Health Insurance Portability and Accountability Act of 1996 (HIPAA) and related rules. The Standards for Privacy of Individually Identifiable Health Information (HIPAA Privacy Rule) regulate how protected health information (PHI) is accessed and disclosed. The HIPAA Security Rule requires healthcare entities to safely store and protect PHI.
The amount of data healthcare systems gather is so large and varied in format, it’s hard for humans to make sense of it on their own. AI can comb through de-identified data, which doesn’t include any personally identifiable information (PII) and is HIPAA-compliant, to extract information and recognize patterns. The results inform and educate providers and other healthcare professionals.
Healthcare professionals use AI for help with:
- Diagnosing patients
- Defining disease characteristics
- Identifying potential treatments and which conditions will benefit from the treatments
- Predicting disease progression or response to treatment
- Understanding how a disease associates with other conditions
“Humans get tired, but machines do not,” Dr. Kondra says. “No matter how much information we input, machines still give us the results. Humans have limitations to how much they can handle and how much work that they can do.”
As advances in technology reveal new uses for AI, its role in the medical space continues to grow. Experts are also learning how to focus AI to meet specific challenges, like diagnosing rare genetic diseases.
Barriers in Rare Disease Diagnosis
The National Organization for Rare Disorders (NORD) surveyed over 1,000 people about living with a rare disease. They found that 28% of participants waited seven years or more for an accurate diagnosis. Approximately 38% of all those surveyed received a misdiagnosis during their journeys. In search of a diagnosis, patients endure endless testing and expenses, while missing out on years of possible treatment.
Dronen says the sheer number of rare genetic diseases makes diagnosis difficult. Doctors struggle to keep up with the latest research and details for each condition. Existing information about rare diseases also tends to come from details gathered about diagnosed patients. People with atypical cases often remain undiagnosed, so their signs and symptoms aren’t documented.
“With more testing available, we’re finding there are people who have atypical presentations of the disease,” Dronen says. “We’re also recognizing that there’s just a much larger phenotypic spectrum (range of observable traits or symptoms) to a lot of diseases, and they don’t necessarily behave the way that we thought they would.”
The same challenges in diagnosing rare diseases have also made it difficult to apply AI in this space. Each rare disease is only linked to a small number of patients whose disease tends to present in the same or similar ways. That makes it hard to collect the amount of data AI needs to predict diseases and outcomes with accuracy—especially for atypical cases.
Using AI to Help Diagnose Rare Genetic Diseases
Many types of AI used for general healthcare already help with diagnosing rare diseases. One example is a field of AI called computer vision, which enables computers to analyze digital images, videos, and other visual input. Computer vision applications help technicians understand X-ray images and identify anomalies otherwise missed.
But diagnosing rare genetic diseases is not a straightforward process. Doctors must consider factors that include a patient’s family history, medical history, and genetic reports. That information often exists in written reports or as part of the clinical notes collected at appointments. In other words, it’s found within unstructured data.
The role of natural language processing
Natural language processing (NLP) is a branch of AI used to extract information from unstructured text data. It can scan freeform text and pull data about specific symptoms, gene sequencing, or medical history. That information helps physicians screen patients for genetic conditions and diagnose rare diseases.
“To diagnose a rare disease, we have to dig deep into the unstructured data that is clinical notes, and we have to extract that information,” Dr. Kondra says. “This is where NLP plays a major role.”
FindEHR: NLP in action
FindEHR is a screening tool that uses NLP, along with other AI techniques, to comb electronic health records (EHR). Healthcare systems engage FindEHR to flag at-risk patients for a specific underlying genetic condition. FindEHR extracts information from both structured and unstructured EHR data using only de-identified and relevant pieces of the EHR (not the entire health record).
If FindEHR flags a patient, a genetic counselor reviews the information and passes it along to the healthcare facility. ThinkGenetic, Inc. also provides doctors with any education, next steps, and resources they need.
Benefits of AI within the rare disease space
AI techniques, such as NLP, offer many benefits when used to diagnose rare diseases:
- Patients: AI minimizes the time, money, and effort it takes to get a diagnosis. An earlier diagnosis provides an opportunity for earlier treatment, which typically improves outcomes. AI can identify approved therapies and clinical trials related to the rare disease.
- Providers: Tools like FindEHR give doctors the help they need to diagnose rare diseases. AI predictions are usually precise and reduce the element of human error in the diagnostic process.
- Payers: A timely and accurate diagnosis saves payers from the costs of unnecessary tests and misdiagnosis. When patients receive treatment earlier, it can lessen the number of health concerns that develop later.
Other Emerging AI Applications for Rare Disease
FindEHR is one of many AI applications in development for rare diseases. Experts are designing AI technology to help predict, diagnose, and identify treatment for rare diseases. Some up and coming AI applications for the rare disease space include:
- Fabric GEM by Fabric Genomics, which uses AI to analyze possible gene variants and quickly identify the root cause of the disease so that patients receive the right treatment sooner.
- Face2Gene by FNDA, an AI-powered smartphone app that uses computer vision to analyze facial features and identify rare genetic disorders.
- Healnet by Healx, a platform that uses AI drug discovery to predict which known drugs or drug combinations have the highest chance of success in treating rare diseases.
How AI Affects People With Gaucher Disease
The power of AI is especially exciting for people living with Gaucher disease and other lysosomal storage disorders. People with typical or severe symptoms of Gaucher disease can use a standard blood test for diagnosis. But those who are asymptomatic or present with mild or atypical signs often go undiagnosed. FindEHR and other AI applications may change that.
“The exciting thing about FindEHR is hopefully finding those atypical patients sooner. They often have a longer wait for diagnosis, and they’re wasting time that they could be getting helpful treatment,” Dronen says.
Diagnosing Gaucher disease when it’s mild or atypical allows patients to connect with a Gaucher specialist and obtain proper disease management earlier. If deemed appropriate, the person can start enzyme replacement therapy (ERT) or substrate reduction therapy (SRT) before the disease progresses. “When a condition’s treatment includes enzyme replacement or substrate reduction therapy, patients have better health outcomes over their lives if they start treatment earlier.”
Join the National Gaucher Foundation in Celebrating Rare Disease Day
NGF and ThinkGenetic, Inc. are advocacy partners, working together to provide support and information about rare diseases like Gaucher disease. You can help raise awareness and celebrate our community by getting involved with Rare Disease Day, which takes place annually on the last day in February. Make a gift or learn more about Rare Disease Day.
- Orphanet Journal of Rare Diseases: The Use of Machine Learning in Rare Diseases – A Scoping Review — https://ojrd.biomedcentral.com/articles/10.1186/s13023-020-01424-6
- National Gaucher Foundation: Gaucher and Other Rare Disease Diagnosis — https://www.gaucherdisease.org/blog/gaucher-rare-genetic-disease-diagnosis/
- National Organization for Rare Disorders: Barriers to Rare Disease Diagnosis, Care, and Treatment in the US – A Thirty-Year Comparative Analysis — https://rarediseases.org/wp-content/uploads/2020/11/NRD-2088-Barriers-30-Yr-Survey-Report_FNL-2.pdf
- ThinkGenetic (video): FindEHR by ThinkGenetic – Supporting Healthcare Providers in Diagnosing Rare Genetic Conditions — https://www.youtube.com/watch?v=JseuFKOIxSw