For rare diseases and oncology, where data is scarce and patients are hard to find, the margin for error is razor-thin. We believe achieving success with AI requires a strategic commitment to three core pillars:

1. The Right Data Foundation with Integrated Patient Data™: Finding quality data in Oncology & Rare Diseases has always been a significant challenge. The depth and accuracy of data dictate the success or failure of any AI-based solutions. Integrated Patient Data™ curated via Patient-Mediated Multimodal Data Curation is the answer. 

• • • Integrated Patient Data™: is not just EMRs or claims. It must be a comprehensive 360° view, aggregating medical records, genomics, device data (wearables), social determinants of health (SDOH), and patient-reported outcomes (PROs).
    • The Tokenization Trap: Many turn to tokenization to link syndicated, de-identified datasets. However, this probabilistic approach is highly limited in rare diseases, often resulting in incomplete data cubes.
    • Patient Consent is the Key: We advocate for patient-mediated data curation. By engaging directly with patients (often through partners like advocacy groups and panel companies) and obtaining their consent, we can ethically collect accurate, real-time, and linkable longitudinal records, overcoming data scarcity. For serious diseases, patient consent rates are often high—sometimes over 90%—as patients are highly motivated to contribute to research for themselves and others.
    • The Unstructured Barrier: Between 80−95% of all healthcare data is unstructured—locked within physician notes, SOAP summaries, and discharge reports. Without access to this data, any AI model is inherently blind to crucial details like precursor symptoms, rationale for treatment switching, and precise diagnosis details.

2. The Context Challenge: Once the data challenge is solved, the next critical challenge is building the right clinical context into any AI-based solution. A few approaches are recommended. 

• • • Medically Smart AI™: To extract meaningful value, AI must be built on domain-specific data—meaning, a model for non-small cell lung cancer must be trained primarily on non-small cell lung cancer data, not just general oncology data.
    • Human in the Loop: Furthermore, successful models require meticulous expert annotation and validation by human clinicians, ensuring that the AI has the necessary context to avoid generating inaccurate or misleading information.
    • Multi-Stage Training: Training disease-specific models requires a careful, systematic approach, which starts with continual pre-training, and continues to curriculum-based fine-tuning and Reinforcement Learning from Human Feedback. 

3. Patients as Partners: Last but not least, none of what is described above is possible without partnering with rare disease and oncology patients who share their data and experiences. Approaches to access the patients include setting up close ties with a variety of organizations, including advocacy groups, patient panel companies, NGOs, patient groups, etc. 

These organizations bring the patients to the table, a key role that is needed at all different phases of the project. 

The AI Advantage – Time, Cost, and Evidence:  

When the right data foundation is in place, the strategic impact of AI is immediate and quantifiable.

• Workload Reduction: AI agents are poised to free up 25−40% of employee capacity in key commercial and medical functions.
• Financial Lift: This translates directly to an estimated 3−7% revenue growth and 6−8% cost efficiency for pharmaceutical companies.
• Rapid Real-World Evidence (RWE): During our webinar demonstration, we showed how AI agents can unify thousands of longitudinal records to answer critical Real-World Evidence and Market Research questions—such as the most frequent lines of therapy or concomitant medications for a complex indication like DLBCL—in minutes, not months.

The era of merely experimenting with AI is over. For rare disease and oncology leaders, the focus must shift to building AI solutions on a foundation of integrated, patient-mediated data and medically smart AI solutions to achieve the velocity and evidence needed to succeed in these complex therapeutic areas.

References:

1. Reimagining life science enterprises with genetic AI-McKinsey & Company, Sept 2025 https://www.mckinsey.com/industries/life-sciences/our-insights/reimagining-life-science-enterprises-with-agentic-ai  
2. Advancing Principled Pharmacoepidemiologic Research to Support Regulatory and Healthcare Decision Making: The Era of Real-World Evidence: Clinical Pharmacology & Therapeutics, Dec 2024 https://pubmed.ncbi.nlm.nih.gov/39807817/
3. https://clinakos.com/