Biomedical Industry · global
BMS Partners with Pathos and Tempus to Improve Drug Development Using Neurological Disease Data
A five-year data collaboration targeting 500,000 patient records will support research into Alzheimer’s disease progression and trial design; a separate cancer initiative focuses on referrals and community engagement to improve patients’ access to research.
Developing new drugs requires both understanding how diseases evolve and giving suitable patients opportunities to participate in trials. Two collaborations recently announced by Bristol Myers Squibb (BMS) address these two challenges: one uses artificial intelligence to analyze neurological disease data, while the other improves pathways to participation in cancer clinical trials. Both represent investments in research and development infrastructure, whose effectiveness remains to be assessed through actual research and implementation results.
Pathos AI announced on October 23 that it will combine patient data from Tempus, its own Foundry platform, and BMS’s neuroscience research and development experience to study Alzheimer’s disease and other neurological diseases. Specific uses include analyzing disease progression, generating hypotheses about disease biology, and helping identify patient populations that may benefit and design clinical trials. BMS has obtained exclusive rights to use Foundry in neuroscience for three years, along with nonexclusive use of the platform across its broader research and development portfolio.
The scale of the data and payment terms are clearer in documents Tempus filed with the U.S. Securities and Exchange Commission: the collaboration aims to deliver 100,000 de-identified patient records annually over five years, totaling 500,000 records. Pathos will pay up to $55 million, with data licensing payments dependent on the number of records actually delivered. This is therefore an arrangement to build a dataset year by year and should not be taken to mean that the complete dataset is already available.
The same filing also disclosed a separate oncology data and services agreement under which BMS must pay Tempus fixed payments totaling $137.5 million through the end of 2031. This is a separate arrangement from the neurological disease data transaction. Because Tempus and Pathos have overlapping senior executives and equity ownership, the neurological disease agreements were reviewed and approved by a special committee of disinterested Tempus directors.
However, there remains a gap between obtaining large volumes of patient records and demonstrating that AI can improve research and development. Pathos said Foundry has already been used in its oncology drug development decisions; this announcement provided no quantitative validation for neurological disease applications and presented no comparative results showing improved trial success rates. Associations identified in existing patient records still require checks on data completeness and patient representativeness, as well as subsequent research to establish whether their predictions are reliable.
Whether patients can enter research is the focus of the other collaboration. The Institute for Healthcare Improvement (IHI) and BMS announced on October 22 that they will launch the three-year Cancer Collective Action Collaborative, which is expected to bring together 24 teams, with a particular focus on lung and breast cancer. The work includes improving referral processes, strengthening community engagement, and helping populations disproportionately affected by cancer access trials; applications to participate close on November 30.
This cancer initiative and the neurological disease AI collaboration are independent programs, and the announcements did not state that they would share data or link recruitment efforts. Yet the problems they address echo one another: models need data that adequately reflect differences among patients, and research needs systems that enable diverse patients to participate. Whether the former can produce more reliable research and development judgments, and whether the latter can improve actual participation, both require evidence of results beyond collaboration announcements.