Here’s how life sciences organizations can develop safe and compliant AI-based solutions that improve care and protect patient privacy.

In any industry, quality data is crucial to building accurate, reliable and validated solutions. With upcoming AI regulations in life sciences seeking to ensure that solutions are robust and secure, quality becomes all the more important.

Fulfilling these regulatory requirements will require high levels of data protection and data cleaning, while ensuring patient privacy. These responsibilities call for a best-in-class data management strategy for developing artificial intelligence (AI) solutions.

Through our work with clients, we’ve identified three key tenets for crafting a data management strategy:

1. Data gathering and storage. Define procedures to ensure data quality and avoid data compromise, gather verified information from trusted sources, and anonymize personal data appropriately and in real time.
   
   
2. Data modeling and training. Clearly define and understand data modeling and training requirements, as this leads to robust and accurate data models that improve and evolve over time. When building data models, it is important to expose them to a diverse set of inputs to reduce data bias and make the models more adaptable to changing parameters over time.
   
   
3. Solution testing. Define the appropriate data requirements, key metrics and documentation procedures to clearly understand the solution’s performance and ensure accuracy, explainability and reliability. To demonstrate the safety and accuracy of the algorithm, organizations need to submit auditable data while documenting proofs of validation at varied levels of competency requirements.  

Applying the data management tenets

To understand how each of these tenets applies to life sciences AI, consider an AI algorithm-based solution to assist dermatologists in diagnosing skin cancer. This solution helps determine whether a mole is cancerous or benign, considering various factors such as symmetry, color and size, to identify a pattern or key characteristics.

• Tenet 1: Data gathering and storage. For the AI solution described above, it would be imperative to collect all images from trusted institutions such as the International Skin Imaging Collaboration (ISIC) or Interactive Atlas of Dermoscopy. Meanwhile, to protect patient privacy, images and personal data should be anonymized appropriately before use, using techniques like data masking, data swapping and pseudonymization.
  
  
• Tenet 2: Data modeling and training. Organizations would need to build data models by feeding images of all skin colors and types from different age groups, geographic regions, races and genders. The images should include those of healthy skin and confirmed melanoma lesions, and with varied image qualities such as those taken under optimal or suboptimal lighting conditions.
  
  
• Tenet 3: Solution testing. In our AI example, solution testing would translate to validating the solution’s capability by comparing verifiable data of its performance with certified dermatologists’ diagnoses and subsequently documenting proof of accuracy and competence for different forms of cancer.

Growing AI prevalence calls for more data discipline

As the use of AI becomes more prevalent in life sciences, it is imperative that pharma companies implement secure information management and exchange systems that not only protect privacy but also expose the data model to diverse set of inputs and characteristics, in order to develop scalable and accurate AI solutions.

Lastly, pharma companies should demonstrate and submit extensive proofs of safety and conditions under which the solution is tested before releasing for commercial use.

These three tenets are designed with the intent to help pharma companies access the right data at the right time to build safe and compliant AI-based solutions.

For more details on this three-step approach, read our white paper “AI Regulation Is Coming to Life Sciences: Three Steps to Take Now.”

Mini Nair and Shirali Desai — both members of Cognizant’s Life Sciences Consulting Practice — also contributed to this blog post.