What does an AI business sell when it helps run a clinical trial?

An idea for making clinical trials more efficient with AI becomes a service that a pharmaceutical company can buy only when it can be used in an actual trial. Prediction accuracy is not enough: the service also has to cover statistical analysis, site operations, patient safety and the evidence presented to regulators.

SURPASS, a programme for which the US Advanced Research Projects Agency for Health (ARPA-H) opened a funding call on September 30, will put public money into that practical demonstration. It aims to incorporate predictive models, analysis of incoming trial data and operational automation into an infrastructure that can evaluate multiple drugs over time. The goal is to reduce the need to build a new trial operation for every drug and create a system that later entrants can also use. SURPASS programme ↗

For an AI company, this is certainly an opportunity to improve its product and establish a track record it can show customers. But the programme is not simply a chance to have a software demonstration evaluated. Participants must work with clinical, statistical and regulatory specialists to run real trials.

The solicitation says ARPA-H will consider fully funding trial costs for the first three treatment arms, with support declining thereafter. This suggests an intention to fund the first users publicly, then use that experience to move towards a system in which subsequent pharmaceutical participants pay for access. Solicitation document, public copy ↗

Rights to the drug and rights to the AI are separate questions

One question is whether an AI company supporting a trial can also gain rights to the drug being evaluated.

We found no provision in the solicitation that automatically allocates drug development rights or sales revenue to participating technology providers. ARPA-H's general guidance also states that intellectual property arrangements are determined through agreement negotiations. ARPA-H submission FAQs ↗

Existing drug patents and commercial rights, rights to software or models improved during the trial, and permission to use patient data are distinct matters. Working on a trial does not automatically permit a company to use its data to train models for other customers. SURPASS requires arrangements governing data access, confidentiality, publication and model management. Solicitation document, public copy ↗

The most immediate business model is to be paid for trial design, analysis and operations. Moving to a model in which the technology provider shares development costs and failure risk in return for rights or revenue upon success would require a separate agreement with the pharmaceutical company.

This distinction also connects with our earlier article on research AI and collaborative drug discovery. Providing software and owning rights to a drug developed with it may be connected technically, but they are not the same contractual arrangement. Our article on research AI and joint drug discovery ↗

Can a publicly funded demonstration lead to the next contract?

SURPASS envisages making some outputs available for use in other development programmes. For example, at least one set of documentation that allows a model to be referenced in regulatory submissions must be made public following the specified review. Additional documentation may remain private to protect intellectual property. This does not mean that all patient data or models will be publicly released. Solicitation document, public copy ↗

The companies that emerge from this programme will not necessarily be AI businesses that own drugs. A company may also build a viable business by getting trials started sooner, obtaining results that support decisions from limited patient numbers, and providing regulators with an adequate account of how those results were produced.

In that case, customers are buying a trial operation they can rely on more readily than one they would have to assemble themselves—not the AI label. The programme may bring publicity. Its real commercial value, however, will depend on whether the methods and experience gained in the initial demonstration become a service that the next customer is willing to pay for.