Collaboration in the Time of COVID-19: Legal Considerations for Successful AI and Healthcare Partnerships

The last decade has seen an explosive growth in healthcare solutions using Artificial Intelligence (“AI”). Medical imaging diagnostics, drug discovery and repurposing, new target identification, manufacturing efficiencies, clinical trial planning and enrollment, personal medicine, evaluation of drug adherence and dosage, are just a few examples which have exploited the powerful predictive results of AI.

This surge in innovation in the healthcare/AI technologies is reflected in investment dollars as well as rising numbers of patent filings. The potential impact of such technologies on all our lives is significant, and current hopes lie in such technologies playing a part in developing new diagnostic methods, treatments and vaccines for COVID-19.

Who Are the Players in Healthcare/AI Technologies? 

Current players include traditional healthcare companies diversifying

Current players include traditional healthcare companies diversifying towards data science solutions, as well as technology giants such as Google and IBM investing in healthcare solutions. For any given project, expertise in both AI/ data sciences as well as the healthcare discipline (biological sciences, biochemistry, drug development, etc.) is needed. To the extent that this broad range of expertise cannot be found in-house, external collaboration is needed, between companies in traditionally very different spaces, speaking different business and technology languages, which can have its own set of challenges.

 data science solutions, as well as technology giants such as Google and IBM investing in healthcare solutions. For any given project, expertise in both AI/ data sciences as well as the healthcare discipline (biological sciences, biochemistry, drug development, etc.) is needed. To the extent that this broad range of expertise cannot be found in-house, external collaboration is needed, between companies in traditionally very different spaces, speaking different business and technology languages, which can have its own set of challenges. 

So, how can the different parties ensure a smooth collaboration? How best to go about structuring, negotiating and implementing collaboration agreements for developing a vaccine?

The key is for the different players to understand each other’s perspectives, objectives and contributions, and for these to then be defined using common terminology. Let’s take vaccine development as an example.

What Are the Perspectives of the Vaccine Component?

Vaccines comprise antigens that will trigger an immune response in the patient when the patient is exposed to the virus without making the patient sick. The antigen can be any part of the virus or a weakened version of the virus. A typical virus comprises genetic material (DNA or RNA) encapsulated by layers of protein with hundreds of thousands of subcomponents. In developing a new vaccine, a suitable target subcomponent of the protein of the virus must be identified from all the possibilities which is where data analytics can play a part. The candidate vaccine with the target will then undergo preclinical and clinical testing. AI can also play a part in analysing the clinical data. 

What Are the Perspectives of the AI Component? 

AI is an umbrella term describing the programming of machines to mimic human tasks. AI includes a number of subsets such as Machine Learning (ML), which can build models (algorithms) for predictive uses. Different subsets of AI have very different data dependencies, human input requirements, execution times, and interpretability. ML itself can be subdivided into supervised learning, unsupervised learning and reinforcement learning. Broadly, training data is used for creating and fine-tuning a model. The model is then used on production data to provide an AI output. In the case of vaccine development, the AI output is the antigen which has the desired immune response properties without side effects. 

Common Terminology 

The collaboration agreement should therefore identify these five key elements:

• the data analytics solution (training process and trained model)
• the training data (e.g. historical clinical data)
• the production data (e.g. clinical data to identify prediction)
• the AI output (e.g. prediction)
• the AI evolution. (the “trained” model)

For each component, the collaborating parties must agree on the following elements:

• who provides the component
• who will use the component
• how the component will be used
• who owns the component. 

Specific Considerations

How you deal with key elements of such a collaboration agreement will greatly depend on the specifics of the AI methodology, the data, and the partners’ roles in the collaboration. 

For example, is the technology partner using proprietary AI training methods, or will they apply off the shelf methods to the training data? Who will handle data privacy matters?

Is the proposed AI method compatible with stringent regulatory requirements? Deep learning has the potential for huge technological value but explain ability is difficult. For this reason, machine learning methods may sometimes be preferred. 

What is the source of the training data and the production data? Who owns it? Is data cleaning required? Who will own the cleaned data? Data ownership must be covered in the agreement itself as Canadian and US copyright laws cannot be relied on for data protection. Europe on the other hand has sui generis Database rights.

One of the challenges in negotiating a collaboration agreement in the field of AI is the diversity of background of the different players involved. 

These collaborations often include a large pharmaceutical or medical device company and a technology start-up, which evolve at different paces and have opposite appetites for risks. By performing appropriate due diligence and being accompanied by seasoned experts, it is possible to avoid pitfalls and reap the full benefits of the collaborations.

BCF’s team can help optimise the value of your innovations and partnerships. Should you have any questions regarding this article, written in collaboration with Andréanne Auger, Ilya Kalnish and Julien Lacheré, or about collaboration agreements in general, please do not hesitate to contact us.