By Alex Garner, Chief Product Officer, Raremark

The healthcare industry is increasingly focusing on niche patient populations. Around half of FDA approvals in the past two years were for rare or orphan drugs that serve fewer than 200,000 patients in total in the US and 1 in 2,000 patients in Europe. By 2024, orphan drug sales are expected to capture one-fifth of worldwide prescription sales.

However, finding these hard-to-reach patients is difficult and keeping them engaged over time even more so. Could machine learning platforms that deliver personalized experiences for patients and caregivers be part of the answer? Patient insight over time can help brands to understand niche patient populations, informing launch strategies, which in rare conditions can feel like launching in the dark or based on conversations with just a few people.

Chances are that at some point in the last few hours you’ll have used an application powered by machine learning in some form or other. Netflix, Facebook, Google and Siri all use machine learning to personalize how we experience their service. Machine learning is essentially feeding a computer lots of information for it to then find and act on patterns in the data. For example, Facebook’s machine learning algorithm analyzes how each user interacts with content on the platform and then, based on that, decides what content users should see next, making my Facebook feed look very different to yours.

Building a better road to diagnosis using machine learning

For healthcare, one benefit of machine learning lies in the ability to process enormous data sets and reliably find certain trends or insights that can improve and potentially disrupt the current levels of care patients are currently getting. For example, Microsoft is working on a way to automatically spot tumors from healthy tissue in radiological imagery. Other innovators are building prediction models to identify patients that could be at high risk of sepsis or heart failure and some are even developing facial recognition apps that help detect genetic disorders.  All of these are very much a work in progress, and we are only just scratching the surface of machine learning in healthcare. One aspect we can be sure of though is that machine learning relies heavily on big data sets – something not readily available in rare disease.

A huge challenge for patients with rare diseases is getting an accurate diagnosis. Patients typically may have waited eight years to get one, usually down to a lack of knowledge and awareness of their disease by healthcare providers. There are around 7,000 rare diseases with small globally dispersed populations, and detailed medical literature and research on each of these diseases is often scarce. 

There are some exciting developments happening in the rare disease space where innovators are using machine learning to try and improve diagnosis journeys.

Volv, a Swiss digital health and life sciences company, has made some great strides in this area. Their prediction model can diagnose patients with a rare disease with 97% accuracy using medical health records. Volv feeds information around symptoms, patient journeys, instances of misdiagnosis, clinical decision making and other clinical data points into its model to help it learn about a particular rare disease. Then they give it access to a huge dataset of anonymous medical records, which it analyzes and finds those patients at risk of a particular rare disease. The company recently shared a case study of its model in action and it found a whole new cohort of patients at risk of a rare disease, who did not have it as a diagnosis on their medical record. The anonymized patients found were being treated for other conditions. This enlightened approach could dramatically accelerate the diagnostic journey for many rare conditions.

Another area where machine learning can be used is in medical imaging. The award-winning breast cancer screening AI Mia, built by the British med tech company Kheiron, uses novel deep learning methods and radiology insights to find malignancies in mammograms. Kheiron was recently granted a UK government grant to help determine the best use of Mia to increase the automation of breast screening services. Boston-based biotech FDNA is also focusing on medical imagery to improve diagnosis. Their Face2Gene tool helps researchers analyze patient faces to determine whether they have a genetic disorder. In fact, it’s already being used in multiple studies investigating rare diseases, such as a 2020 study looking at Mucolipidosis type IV (ML-IV), a rare autosomal, recessive lysosomal storage disease, where researchers want to see whether patients with this disease share identifiable facial features not yet described in medical literature.

How is machine learning helping rare disease patients at the moment?

Alex Garner

A new wave of online patient platforms has emerged in the past decade, aimed at bringing patients together in one place to share experiences and learn from the wisdom of the crowd. Some of these platforms are researching  and discovering machine learning techniques to enhance the experience of users. Our platform Raremark is one of them. It’s the world’s largest patient experience network in rare disease. Our platform makes the right information available to patients at the right time. We understand that patients who have just been diagnosed have different needs and questions than someone who has been living with the condition for many years.

Raremark is continuing to research and develop new ways to match members to content and opportunities to share their lived experiences through a novel combination of machine learning techniques and behavioural science models. We believe that an effective matching algorithm for online health resources is to recognise that each member will have a different set of personal characteristics. These characteristics will determine how they confront the realities of living or caring for someone with a rare condition. Using these technologies to learn and automate when to recommend the right type of experiences to read or contribute on the platform help our members build a valuable knowledge base about their disease.

We have learnt the best way to find and engage with people affected by a rare disease is by firstly understanding their digital journeys and starting conversations on those channels first. Once a relationship has been established, we invite them to become a Raremark member, where we begin to build their trust by listening to and responding to their needs through our personalized content recommendation system. We can then go a step further and begin to study user behavior to gain some insight into areas like the motivations behind taking part in research and clinical studies or the reasons for treatment non-adherence for certain rare diseases. We keep our intentions clear and transparent – our members know that with their explicit consent we share certain member experiences and survey results with researchers and companies studying their disease to advance the field further.

Rare disease and the machine learning frontier

We still have a long way to go before the full potential of machine learning and AI are realized. It’s important not to overestimate the capabilities of machine learning and AI, we are still only touching the surface of its full potential. In rare disease, a challenge for all of these models is the small  data sets that come with small patient populations, as well as the format of rare disease research where insights are hidden in dense literature. Advances are already being made to find key information from reams of text.

Despite machine learning and artificial intelligence still being in its infancy, every day we’re seeing new and exciting innovations happening in health and with every new project or setback, we’re getting closer to making true artificial intelligence a reality. It’s an exciting road ahead.

About the author

Alex Garner, chief product officer, is responsible for the upkeep and future development of Raremark’s digital real-estate. He discovered a passion for building health-tech products from over five years of implementing and designing digital applications for the NHS. Along with a master’s degree in business management and innovation, Alex is a firm believer in the principles of user-centric design and constant learning