Diagnosing rare diseases with AI and machine learning

A rare disease is defined as a condition that affects fewer than 200,000 people, yet it’s estimated that 400 million people around the world are living with one-that’s greater than the population of the entire United States.

Rare diseases impact more people than cancer and HIV/AIDS combined, so why do 95% of rare diseases lack an FDA-approved treatment? 

Underlying causes of rare diseases are difficult to identify, and symptoms differ from person to person, making diagnosing the disease challenging and oftentimes too late in development. However, recent advances in technology-particularly in artificial intelligence and predictive modeling-are accelerating the accuracy and speed of rare disease diagnosis. 

Trending: ISMP: Top 10 medication errors and hazards in 2019

Imagine two scenarios, one where a physician is diagnosing a patient, and the other where a predictive algorithm trained to detect rare diseases assists a physician in diagnosing a patient. In the first scenario, the lone physician relies on their medical knowledge and years of experience treating patients. Keep in mind that a physician may only see one rare disease patient in their entire career, given the prevalence of rare diseases. 

Contrast this scenario with the one where the physician is gleaning insights from the ‘learnings’ of predictive algorithms that have been ‘trained’ on rare diseases. The predictive models leveraged hundreds of millions of healthcare data points and would be able to alert the physician to the similarities of the patients’ symptoms to a rare disease identified from among thousands of rare diseases. With the right predictive model and data resources, it’s possible to find patients hidden in healthcare databases and diagnose certain rare diseases in weeks compared to years.

As health technology becomes more powerful, we are beginning to leverage vast amounts of collected patient data to work with physicians to proactively identify undiagnosed patients, while respecting their privacy rights and shortening the time to diagnosis. This has been accomplished by applying scalable analytics such as AI and machine learning on data sources such as deidentified healthcare claims, genomics, and prescription data. These technologies are now mature enough to enable earlier diagnosis of these patients, enabling physicians to treat the right patients with the right therapeutics at the right time. 

Next Steps

The amount of healthcare data collected is growing exponentially and shows no sign of slowing down. A typical analyses of healthcare data can determine the incidence and prevalence of a disease and indicate the number of patients with that specific diagnosis; however, incidence and prevalence based on real-world data can now also inform how many patients are available to support drug development, execute a successful clinical trial, bring the drug to market, and determine whether an orphan designation can be sought for a therapy. 

Read More: Six EHR-related malpractice landmines to avoid

Access to orphan drugs requires both awareness and education. In addition, the data sources and machine learning and predictive modeling tools enable the discovery of undiagnosed rare disease patients. Discovering undiagnosed rare disease patients drives efforts to increase awareness to all stakeholders. The most effective technique to raise awareness is to connect insights from data with humanity by precisely targeting educational efforts that will have the greatest impact on getting undiagnosed rare disease patients connected to physicians who can prescribe the right therapy.

To bring the community together for Rare Disease Day, H4B Boston and HVH Precision Analytics are hosting an event Facing the future of rare together, which will feature innovative speakers, including biopharmaceutical group Ipsen and non-profits Global Genes and Beyond the Diagnosis.

Steve Costalas is the CEO of HVH Precision Analytics