There is an old adage in medicine: “When you hear hoofbeats, think horses, not zebras.” In other words, think of common illnesses when evaluating patients before looking to rare diseases. But of course, as anyone who works in rare diseases knows, sometimes the hoofbeats are zebras.
Systemic mastocytosis (SM) is a rare disease that can mask as any number of other more common illnesses due to its non-specific and heterogenous signs and symptoms that affect several systems in the body. 1,2 It also affects only about 32,000 people in the U.S.3
Currently, it takes an average of approximately six years between the onset of symptoms and a confirmed diagnosis of SM, meaning that patients may be undergoing years of debilitating and unpredictable symptoms – which may include skin rashes, gastrointestinal distress, and anaphylaxis – before being diagnosed.3
“We hear from some of these patients that they had been told it’s all in their head,” says Daniel Shaheen, Senior Medical Director at Blueprint Medicines. “How demoralizing is that, when you’re really sick, and you’re sick for a long time.”
As part of our effort to help doctors diagnose SM patients more accurately and earlier in their health journey, Blueprint Medicines has partnered with the University of Pennsylvania (Penn) to develop and publish an algorithm applying machine learning techniques that could be used to pinpoint potential SM cases using electronic medical records (EMRs). As part of the ongoing project, researchers at Penn are collecting de-identified data from the Penn Health EMR system from patients diagnosed with SM as a control group, as well as data regarding patients who have symptoms similar to SM patients but were confirmed not to have SM. Early findings from this research were presented at the American Academy of Allergy, Asthma and Immunology (AAAAI) Annual Meeting in 2023.
“Our objective is to effectively reduce time to diagnosis,” notes Shaheen. “If we can create connections from information presenting symptoms and data, we might create a clearer path for the healthcare providers to follow, potentially shortening the time to diagnosis.”
That’s where artificial intelligence (AI) comes in. There is so much data surrounding a patient and their journey; both structured and unstructured data – notes taken by the healthcare provider during each visit, each lab result, each referral – may be hard to decipher by one doctor one or two times per year. In totality, however, the volume of data creates the opportunity to identify patterns. All that data is fed to the algorithm to “teach” it to accurately predict when a patient should be tested for SM.
“Through natural language processing and machine learning, the algorithm can analyze all that information and pull out key phrases of patients that have actually received a diagnostic workup for SM,” says Shaheen. “It essentially creates a fingerprint or signature for patients that should be worked up for SM and, when it finds one, says these hoofbeats may actually be zebras.”
As soon as the algorithm demonstrates predictive capability within the Penn system’s cohort, it will be sent to other health care systems through the University of Michigan and the University of Utah, to be adjusted and tested against their electronic record systems. The long-term vision is ultimately an open-source algorithm that could be downloaded, incorporated into any institution’s EMR system and alerted for patients who may benefit from diagnostic testing for SM.
“As we know, health systems in the U.S. are extremely fragmented, and these systems are not dealing with the same sets of data,” Shaheen says. “So, if we can find an algorithm that can be scaled across diverse health systems, patients can hopefully be diagnosed more expeditiously.”
Shaheen says it’s reasonable that if the algorithm is ultimately successful, patients could be diagnosed and given appropriate care for months or even years earlier in their health journey. “I’m not suggesting that implementing this formula will make every single SM patient go to the door of their institution for diagnosis. It’s certainly going to take more work after this. We must, however, give these patients a greater voice and a better chance of being diagnosed.”
1Gilreath JA, Tchertanov L, Deininger MW. Novel approaches to treating advanced systemic mastocytosis. Clin Pharmacol. 2019;11:77-92. Published 2019 Jul 10.
2Cohen SS, Skovbo S, Vestergaard H, et al. Epidemiology of systemic mastocytosis in Denmark. Br J Haematol. 2014;166(4):521-528. doi:10.1111/bjh.12916
3Mesa, R.A., et al. (2022), Patient-reported outcomes among patients with systemic mastocytosis in routine clinical practice: Results of the TouchStone SM Patient Survey. Cancer, 128: 3691-3699.
4Herman, et al. (2023). Development of Scalable, Electronic Health Record (EHR)-Based Screening for Undiagnosed Systemic Mastocytosis: PREDICT-SM. Presented at AAAAI 2023 Annual Meeting.
More posts and webinars