Conventional clinical study recruitment methods for rare diseases simply don’t work. Analyzing ICD-10 and prescription data to determine patient clusters, site selection and feasibility is impractical, as only 500 of 7,000 of these conditions have been formally classified, prevalence is typically less than 5 out of every 10,000 people, and 95% lack approved treatments. Programmatic advertising, social media outreach and search engine optimization is normally fruitless due to low disease awareness by both patient and health care provider alike. And while turning to patient registries, advocacy groups and Key Opinion Leaders (KOLs) may uncover an initial cluster of potential trial participants, an average of 81% will be deemed medically-ineligible. The fact is most medically-eligible patients will continue to be undiagnosed or misdiagnosed due to sporadic disease interception by a medical community that has difficulty making sense of symptomatology and test results.
With a 131 month average cycle time for a phase I-III rare disease investigation (68% longer than a common disease), costs exceeding $100 million for a phase III study, and six times the number of sites required for enrollment compared to those for non-orphan conditions, it’s no wonder that 90% of these studies are suspended or outright terminated. The impact is profound and necessitates a modern approach grounded in uncovering the ideal, medically-eligible patient population with the highest propensity for study enrollment and retention.
Discovering patients for rare disease clinical investigations while maintaining strict compliance with HIPAA guidelines can be streamlined through the use of machine learning and artificial intelligence in conjunction with a real world data universe, typically comprised of hundreds of millions of de-identified patient journeys over a decade or more, billions of social determinants of health indicators and first party sources such as research, laboratory and genetic data. The data pool is unified and tokenized to prevent any Personally Health Information (PHI) or Personally Identifiable Information (PII) from being compromised, while at the same time, providing the basis to detect patients that have a high likelihood of suffering from a rare disease.
The process begins with the definition of the ideal patient, from which evolutionary computation in the form of precision algorithms are developed to uncover, analyze and validate “look-alike” patients. Deciled clusters based on the propensity of a patient suffering from the affliction and likely to be medically-eligible for clinical study enrollment are generated and treating health care providers – whether primary care, specialty or inferred specialty physicians – are linked. This mapping empowers medical science liaisons (MSLs) or field operators to engage key treaters and provide disease interception as well as clinical study education that can be used by the physician to pinpoint the exact patient, confirm diagnosis and discuss trial enrollment. For sponsors initiating the recruitment phase of a study, protocol development, site selection and trial approach (centralized, decentralized or hybrid) can now be validated, while those in the midst of a challenging enrollment cycle can now confirm the need for new or additional sites, as well as protocol amendment based on real world data.
The outcomes are promising: Quicker approval of more efficacious treatments, the prospect of a higher quality of life for patients who have given up hope, and less burden on the medical community, as well as society as a whole. And for sponsors, markedly reduced study enrollment times, less patient recruitment costs and faster time to commercial impact.