Applications of AI in Rare Disease

Every year more companies enter the rare disease space due to heavy competition in the larger disease area and a faster path to market due to smaller trial sizes needed and greater precision medicine treatments of the conditions in smaller targeted populations. Rare disease is classified as being a rare disease when it affects fewer than in 1 in 2000. Although over 7000 rare diseases have been identified, for the majority of these, no treatment is available. Many of these are from childhood and persist throughout the life of the patient (whether that life is short, or long).

The trend of growing interest in this space and advances in technology and science to enable understanding of these rare diseases allows more targeted therapies to be developed. This focus on rare diseases is supported by the FDA creating accelerated approval pathways, as well as greater ability for patients to create awareness through enhanced communication channels. In addition, larger disease areas are now being broken up into smaller more targeted spaces due to genetic information and biomarkers.

Innovations in science, technology and the use of big data and AI analytics are enabling great leaps forward with rare disease. Areas that AI is assisting pharma already are:
Discovery and pre-clinical applications of AI in rare disease
Development of more targeted therapies through greater volume of and accuracy of data processing and identification of genetic markers

AI can be used to support more effective use of existing data to target therapies for rare diseases with faster speed than ever before. The amount of data is impossible for humans to process effectively. For example, the average researcher reads around 270 articles a year when the actual volume of articles in scientific journals is around 50 million per year and in cancer alone is around 260,000 per year. AI can scan the entire 50 million in minutes and identify learnings that are pertinent to the area of discovery. This means we can connect the dots far more rapidly than ever before. Huge amounts of pre-clinical and clinical data are sitting in pharma R & D organizations that could be analyzed very quickly and effectively to identify new compounds and how to better target specific conditions.

Faster clinical trials
Although currently most companies are focusing on discovery with AI, there are other uses that vastly reduce costs to pharma. These include identifying these patients for clinical trials, automating the data collection in the trials through the use of sensors and wearables, and even automating the compilation of the Clinical Study Reports that are 80-90% complete in less than an hour. This reduces the time for completing each CSR by 4-6 weeks and can reduce the internal resource utilization costs by around 250 hours. A drug earning $9 billion a year, would be earning around $24 m per day so if 6 weeks of time saved to get to market sooner, that would be equivalent to around an extra $1b in revenue. In addition, once a drug is launched, AI can continuously churn through the data and identify any new patients that enter the system.

Eularis Case Study: Faster Clinical Trial Patient Recruitment

 
THE CLIENT
    •    Had a PIII clinical trial underway in a specific type of cancer but could not find more than 153 patients to participate which was wasting time (and money)
    •    Turned to Eularis to see if AI could identify and recruit more patients

THE SOLUTION
    •    Use big data (EHR and claims) and AI to:
    ◦    Understand which patients were appropriate for the trial
    ◦    Who their physicians were
    ◦    Created a visualization of all potential patient locations and could pull up doctor ID and other data on physician to allow contacting them for trial participation for their patient
    ◦    Allowed selection of clinical trial sites based on where patients were located

THE OUTCOME
    •    Within 10 minutes following completion of the system, a further 6,472 eligible patients were identified
Faster approval and reimbursement using AI analysis of dossier submission
From a business perspective, each day a drug is delayed from reaching the market, the company loses millions. Pharmaceutical Executive report that a delay in launch can cost a company an average of $15 million per drug, per day. This varies by country but the stakes are high and increasing speed to reimbursement is critical for companies seeking to maximize their returns. From a human, ethical perspective, with every day of delay patients in dire need of effective new treatments are denied access to treatments that have been thoroughly reviewed and ultimately approved by the regulators to provide substantially improved clinical outcomes over existing treatments. Using AI (a combination of NLP and ML), we can assess the factors affecting time-to-reimbursement, and if the factors identified are able to be modified, then it is possible to describe the path to optimize the process to gain faster reimbursement.

Patient diagnosis for identification of which physicians the reps should see

Clearly one aspect that has the most potential to accelerate diagnosis and patient identification in rare disease is the use of big data and AI analytics. In the past delays in diagnosis due to the rarity of the conditions, often leading to exacerbated severity, meant many patients were never diagnosed, or the condition was so late stage by the time they were, it did not help them. The average time taken to diagnose a rare disease without technology is 7.6 years according to Shire Pharmaceuticals, after countless tests and physician visits. This causes a lot of unnecessary costs to the healthcare system and much suffering for the patient.



Now, using AI, we can identify all of these patients within the data sets (EHR and claims data) within minutes after the initial time spent data wrangling and creating the algorithms. Then every time a new patient enters the healthcare system, that patient is immediately identified using the algorithms.



In fact, the potential of this approach, and the success of the work in this space means that the trend in our AI project requests has moved from HCP marketing to identifying patients in both classic rare diseases, or sub populations of specific cancer types. After the initial time data wrangling and creating algorithms to identify the patients, it takes a matter of minutes to identify all rare disease patients in the data sets (typically EHR and claims data). This means patients are found faster, and treated quickly, and pharma do not waste vast sums of money educating potential physicians who are likely to never see a patient in their lifetime.