Despite the medical advances of the modern age, more than 300 million people around the world still suffer from diseases for which there are no cures.
Many cancers remain a death sentence. Alzheimer’s disease and Parkinson’s disease wreak havoc on millions of people. And more than 30 million Americans continue to suffer from rare diseases that have no effective treatments.
The lack of treatments is not for lack of trying on the part of researchers. Developing a drug—from initial laboratory tests through the requisite clinical trials to prove safety and effectiveness—takes an enormous amount of both time and money. The National Institutes of Health estimates that developing a new drug costs between $173 million and $2.6 billion, and takes an average of 12 years. Only 12% of potential drugs are ever approved by the Food and Drug Administration (FDA).
But what if there were a way to speed things up, to shave time off the development of new therapeutic treatments and new medical devices and to expand therapeutic uses for drugs already on the market, without endangering the patients who will ultimately take them?
There may well be, and researchers at UC Berkeley School of Public Health are in the vanguard of the international effort to bring it to fruition. The Berkeley home for this groundbreaking program is the Center for Targeted Machine Learning and Causal Inference.
Speeding up drug development
The research mission can be traced back to December, 2016, when the U.S. Congress passed the 21st Century Cures Act of 2016. The law was designed to speed up development of drug treatments and other medical products, getting them to patients faster.
The law spurred the FDA to create a framework to evaluate the potential use of what has been called “real-world evidence”— clinical data derived from electronic health records, medical claims, and information from digital health technologies or disease registries.
Proponents of real-world evidence hoped it might be used to support approval of new uses for drugs that were already approved or to help satisfy the FDA’s requirements for post-market studies. But critics believed reliance on real-world evidence was too risky – that it would be a mistake to replace clinical trials – long the gold standard of drug approval – and would lead to unsafe drugs and devices. They argued that it would be very difficult to decide what kind of real-world evidence should be acceptable as a basis for regulatory decisions; and that the FDA wouldn’t know how to weigh such data.
That’s where CTML comes in. The center’s expertise in causal inference—a type of scientific inquiry that uses formal mathematical frameworks to move from studying statistical associations to studying cause-and-effect relationships—is particularly suited to this thorny problem. Targeted learning is a subfield of statistics, which brings together causal inference, machine learning, and statistical theory to help answer scientifically impactful questions with statistical confidence.
The Center’s mission is to harness both cutting-edge causal inference and targeted learning toward robust discoveries and informed decision-making to improve public health.
“Clinical trials, particularly large Phase 2 and 3 trials, are one of the biggest reasons it can take more than a decade and over a billion dollars to bring a new drug to patients. CTML is working with the FDA and pharmaceutical partners to help change that,” said Dr. Michael C. Lu, Dean of UC Berkeley School of Public Health. “By combining the power of machine learning with the rigor of modern biostatistics, our researchers are developing new methodologies—such as using real-world health data to simulate placebo trials—that could make clinical trials faster, more efficient, and more affordable while maintaining the highest standards of scientific rigor.”
“This is exactly why public research universities matter,” Lu added. “CTML exemplifies how the intellectual firepower of Berkeley can help solve some of the world’s most pressing problems.”
CTML is co-directed by Dr. Maya L. Petersen, professor of biostatistics and epidemiology, and co-director of the UCSF UCB Joint Program in Computational Precision Health; Dr. Mark van der Laan, professor of Biostatistics and Statistics; and Dr. Alan Hubbard, professor of Biostatistics.
Van der Laan, the leading force behind the development of the field of targeted learning, has worked with the FDA since 2010, when the agency asked him and Dr. Susan Gruber, then his doctoral student, to conduct a demonstration project on the use of targeted learning to adapt real-world evidence to drug safety studies. Over the years both Van der Laan and Petersen have led workshops for FDA personnel, to explain how the agency could change their methods to speed drug study and approval.
In 2020, with a $3.2 million research gift from Novo Nordisk, the Center launched the Joint Initiative for Causal Inference, known as JICI. The center’s expertise in causal inference—a type of scientific inquiry that uses formal mathematical frameworks to move from studying statistical associations to studying cause-and-effect relationships—is particularly suited to this thorny problem.
The initiative has since grown to include research collaboration with Copenhagen University, Oxford University, Harvard University, and University College London, among other institutions. JICI develops ways to optimally and robustly draw causal conclusions from observational and clinical trial data sets using state-of-the-art advances in machine learning through the targeted learning framework.
JICI researchers create statistical techniques to replace drug trial randomization and allow for causal conclusions to support drug development programs, by reducing the cost of required randomized clinical trials, by integrating clinical trial data with observational data, and by producing valid analysis of randomized trials. They are also developing techniques that let researchers speed up the screening of candidate drugs.
The success of the JICI initiative led to a new UCBPH partnership with Gilead Sciences, on a similar project investigating the best uses of real-world evidence. Gilead has contributed $1.5 million to that effort, which began in 2024, over three years.
To learn more about their work, UC Berkeley School of Public Health talked to Petersen—who is also co-director of the UCSF-UC Berkeley Joint Program in Computational Precision Health—and Van der Laan, about their work.
