Equipping Frontline Providers with Tools for Diagnosis and Referral
Objective: Equip frontline providers with the knowledge and tools to quickly and effectively identify patients who may have a rare disease and take appropriate action.
Half of all rare diseases begin in childhood. Therefore, frontline providers (i.e. pediatricians, primary care physicians) are generally the initial point of contact for parents of patients with abnormal clinical findings. These physicians play a critical role as both the health care provider and the medical gatekeeper. But when it comes to a child who could have a rare disease, these physicians may lack the awareness to consider a rare disease and, if alerted, may not be equipped with the knowledge, tools, and time required to make the best decision about what to do next.
A frontline provider may never encounter a patient with a specific rare disease during his/her medical career. To date, more than 6,000 rare diseases have been described and new diseases are regularly reported in medical literature. While an individual disease might be labeled as “rare,” the number of people suffering from a rare disease is estimated at 350 million. In Europe, more than 30 million people are suffering, roughly 6-8% of the EU population. Therefore, there is a very high likelihood that physicians will see many patients with a rare disease in their routine practice, including some whose symptoms have been attributed erroneously to a common condition. Considering rare diseases as part of the differential diagnosis pathway is essential. However, even if a physician suspects that a patient has a rare disease, pre-diagnosis is difficult because many rare diseases do not have standard pathways to an accurate diagnosis.
Without the knowledge and tools for following up on a suspected rare disease or condition, frontline providers may delay making a referral to a specialist – extending the diagnostic odyssey.
Frontline providers might not feel confident that they are seeing something out of the ordinary or may not know what kind of specialist the patient should see because the patient has a cluster of seemingly unrelated symptoms or does not resemble the “textbook case” of someone with the same disease.
For all these reasons, patients may be referred for evaluation to a variety of specialists – sometimes across different centers, hospitals, or regions of the country and with little communication between institutions. The result is a frustrating process of referrals from one specialist to the next, delaying diagnosis and increasing the risk of misdiagnosis and potential mistreatment.
Artificial intelligence, such as machine learning and deep neural network technology, at global scale has the potential to overcome the primary barrier to early and efficient recognition of a patient with a rare disease.
How it would work
The Global Commission proposes a technology that first “learns” from medical literature based on identified symptoms for each rare disease using natural language processing (NLP) techniques. Then, the technology scans patient health records, highlighting symptoms and matching those symptoms to diseases. It ranks the likelihood that the “matched diseases” are the proper diagnoses for the specific patient. Patient information stored in the cloud would be aggregated globally from all rare disease patients, providing the high volume of data needed to train the technology to detect rare diseases.
Over time, the technology continues to learn and becomes more accurate in extracting symptoms from health records and matching those symptoms to the appropriate disease. Physicians continuously train the technology by inputting a range of data on their undiagnosed patients, culled from medical records and patient reports. The technology then applies learnings from previous analysis to narrow down possible diagnoses. Rather than providing a definitive diagnosis, this technology would identify a category of likely diseases, giving physicians an evidence-based jumpstart in making the right diagnosis. Ultimately, this technology could be applied preemptively, constantly scanning or “mining” electronic health records to flag patients who may have a rare disease in real time, thereby further reducing the diagnostic journey.
Why this is promising
The benefit of using machine learning algorithms and global, cloud-based data to increase the accuracy of diagnosis is being proven across many health conditions, including cancer, heart diseases, and Parkinson’s disease. Its globally applicability is demonstrated in diabetes care in the U.S. and India. Furthermore, it is unrealistic to expect that one physician could identify symptoms for thousands of rare diseases to accurately diagnose every rare disease patient. These types of tools, and other innovative uses of technology advancements, are needed to help front-line providers navigate the complexity of a rare disease diagnosis with speed and precision.
A key challenge to implementing this technology globally is the interoperability of the machine learning technology with existing electronic health systems and physician workflows. The technology should be integrated seamlessly with existing systems, platforms, and workflows to encourage physicians to adopt it as standard practice and ensure that no additional training or system changes are required.
A technology enabled solution can solicit guidance from geneticists and other experts to inform decision-making about who should receive diagnostic testing.
The field of genetics and genomics is evolving rapidly—holding great promise for diagnosing rare diseases since 80% are genetic in origin. At the same time, the cost of genetic testing, or genome-wide sequencing (GWS), has gone down dramatically and will continue to fall. Nevertheless, there is limited access to GWS, in part because of the assumption that it is costly to the health system and the patient, even though it may lead to savings if patients are diagnosed and treated earlier. Also, many specialty centers have a long waiting list to receive GWS because of the lack of experts qualified to analyze the findings.
How it would work
This platform allows frontline providers to provide patient information – based on medical records and other relevant information, including family history – to a panel of experts via an online input system. These experts then work with the physician to make informed decisions about whether genetic testing is warranted and, if so, which test to order. Ideally – keeping in mind variance by healthcare system and availability of genetic counselors – the primary care physician or pediatrician who requested the consult will be able to order the test. Equipping physicians with this information will enable them to order the appropriate genetic test based on each patient’s set of symptoms, reducing unnecessary testing and better pinpointing the right test. The results are faster and more accurate diagnosis and savings to the health care system.
Why this is promising
An easily accessible tool that frontline providers and insurers trust to provide the best available guidance on testing would reduce costs and speed up diagnosis time for patients.
With the current rate of advancements in technology and diagnostics, a traditional triage solution may become obsolete over the next 5-10 years. As prices for GWS decrease, more is known about how to interpret genetic data and more experts are trained to analyze complex genomic data, it is likely that GWS will become the standard of care for many people in many countries. Mirroring this demand for testing is the increase in demand for genetic counselors and the need for optimizing their time and resources so they can best serve patients. Furthermore, understanding of sequencing data continues to evolve and improve (i.e. variants of currently unknown significance). This knowledge would need to be incorporated on an ongoing basis. In the interim, evidence of efficacy will be required to support reimbursement for this approach.