Newswise — Using one of the largest electronic health record datasets for ALS ever collected, a team of scientists and computational engineers led by Lawrence Livermore National Laboratory (LLNL) has identified several existing treatments that may be associated with improved survival for patients with amyotrophic lateral sclerosis (ALS).
Published in lancet digital healthThe study analyzed the health records of more than 11,000 U.S. military veterans diagnosed with ALS and treated within the Veterans Health Administration between 2009 and 2019. By combining causal inference techniques and machine learning (ML), researchers evaluated 162 drugs and identified those prescribed for other conditions that made a significant difference in survival. The study was conducted in collaboration with Stanford University School of Medicine, the Veterans Affairs Palo Alto Health System (VA Palo Alto), and the University of California, Los Angeles (UCLA).
The timing of the study was driven by a rare convergence of data access and funding, said LLNL Principal Investigator Priyadip Ray, a research scientist in LLNL’s Computational Engineering Department (CED). Since 2009, when ALS was officially recognized as a service-connected disease, the Veterans Administration has seen a rapid increase in the number of veterans being treated for ALS, creating more than a decade of detailed treatment data within a single health system. At the same time, new targeted funding programs are making it possible to advance ALS research on a scale that has historically been difficult for rare diseases.
The study was also motivated by recent setbacks in ALS drug development. Early enthusiasm for the drug Relibri, which was approved by the U.S. Food and Drug Administration in 2022, waned after large follow-up trials showed no efficacy and the drug was withdrawn from the market in 2024. The results highlight how difficult clinical trials for ALS are and helped spark interest in other discovery routes.
“We realized that our extensive experience with ALS treatment in the VA system could provide an alternative approach to identifying therapeutics for this disease,” Ray said.
Rather than relying on traditional machine learning approaches, the research team focused on causal inference. The researchers explained that this is a more demanding framework aimed at isolating potential treatment effects while accounting for bias, confounders, and uneven treatment patterns in real-world data.
“Our team has developed a series of methods that combine rigorous statistical methods with cutting-edge machine learning to isolate cause-and-effect relationships at the population level, even when data is not collected in a controlled manner,” said co-author Braden Soper, a data scientist at LLNL.
This analysis identified 27 drugs associated with statistically significant changes in mortality risk. Notably, multiple drugs within the same therapeutic class showed similar associations with improved survival, including statins, phosphodiesterase type 5 inhibitors, and alpha-adrenergic antagonists.
“What was striking scientifically was that in each of these groups there were multiple drugs with the same positive effect,” Ray said. “This gives us great confidence in the relationship between these drugs and slowing the progression of ALS.”
To investigate why these drugs influence disease progression, the research team used PathFX, a protein-protein interaction modeling tool developed by collaborators at UCLA and Stanford University. Network analysis suggested that several of the identified drugs clustered in common downstream protein pathways, pointing to common mechanisms and potential new molecular targets in ALS research. This research builds on LLNL’s extensive investments in causal modeling, ML, and computational tools for biomedical and national security applications.
Ray emphasized that while the study results do not prove clinical benefit, they provide a strong foundation for next steps, including deeper modeling that takes into account health factors that change over time and validation in independent datasets that include a more diverse civilian population.
“Sensitive medical data is difficult to share due to privacy and access restrictions, so we are committed to releasing our software pipeline as open source so researchers everywhere can apply these tools to their own datasets, diseases, and interventions,” Soper added.
Co-authors of the study include Andre Goncalves and Jose Cadena Pico of LLNL’s CED. Amy Grischke (formerly LLNL, currently at University of California, San Francisco, Innovation Ventures). Richard Reimer and Thomas Osborn of Stanford University School of Medicine and VA Palo Alto; Jennifer Wilson of UCLA. Paola Suarez of VA Palo Alto; and Kevin Grimes of Stanford University School of Medicine.
