HHS’s SURPASS program could fundamentally reshape drug development by replacing rigid Phase I–III trials with adaptive, “phaseless” studies that use real-time data analysis, automation, shared controls, and predictive modeling. The upside is compelling: shorter development timelines, lower costs, faster failure of ineffective drugs, and potentially fewer patients needed for control groups. But speed carries risks. Biology can’t always be accelerated, long-term safety still requires time, adaptive analyses can introduce statistical complexity, and AI-driven decisions raise questions about bias and accountability. SURPASS’s real test isn’t whether it makes clinical trials faster—it’s whether it can deliver trustworthy evidence faster without lowering the scientific bar.
On September 30th, ARPA-H introduced SURPASS—Simulation-augmented, Real-time Platform Adaptive Seamless Trials. It is meant to help us shorten the time it takes to develop a new drug or biologic. According to ARPA-H, developing a new medicine typically takes 10+ years and costs anywhere from $1 to $2 billion, with a failure rate of about 90%. So SURPASS hopes to prove that we can develop new drugs and biologics in less than 4 years.
The pharma industry should take notice.
SURPASS hopes to leverage AI to shorten the development timeline. But it is questioning why we even conduct drug development in Phases. With advances in technology, we should be able to learn continuously throughout development.
ARPA-H believes that we can run perpetual adaptive platform trials. They incorporate predictive algorithms, common infrastructure, control groups, and real-time data analysis. Their three pillars include a “phaseless” platform design engine. That would incorporate digital twins and other technologies. An inference engine that can learn from data that continues to accrue. And an agentic layer that can perform various functions from starting up a trial to cleaning data and creating datasets.
The FDA has also been trying to implement adaptive and complex innovative trial designs. Their current program for complex innovative trial designs includes adaptive, Bayesian trials that use simulations. They have also stated in guidance that these trial designs should yield evidence that the product is safe and effective.
I think this new initiative has many positives.
First, we spend too much time in between phases. You have to create protocols, open sites, recruit patients, create databases, etc. When moving into the next phase, you have to repeat the process. With platform trials, you can open sites once, use common controls, and reuse the same infrastructure across multiple trials. The FDA has said this can help expedite development by allowing you to share control arms, clinical sites, trial oversight committees, and data-monitoring infrastructure.
SURPASS can also help us fail drugs quicker. Which can be just as important as getting them approved quicker. If data continues to accrue and you know that a certain dose or treatment isn’t working. Why keep spending money and enrolling patients when your development plan says you’ll review the data in a few months? Adaptive designs can help you know when a drug isn’t working. This includes dropping treatment arms, changing the ratio of patients to each arm, and stopping your trial for futility or efficacy.
Patients are another reason that I believe in this concept. If we can use shared control arms, we may not need as many patients on placebo or standard of care. Since we can have a control group for all of our investigational drugs. Maybe we can also predict better doses and trial designs.
SURPASS also isn’t the only thing ARPA-H is trying to improve.
They have 3 other projects that I think will also help us get drugs to patients quicker. STACK will use AI to open sites faster and help sites that are not used to doing research enroll patients. COMMONS will help create national consent efforts and a privacy-friendly data infrastructure that meets regulatory requirements. CINCH will help patients collect their real-world data, manage their health, and find trials that they may be eligible for.
Together, these pieces show that SURPASS is more than a way to shorten clinical trials.
It’s trying to help us create a new operating system for drug development.
However, I do have some concerns about SURPASS. Drugs take so long to develop because it takes time to figure out if something is safe. We have computers that can crunch data in the blink of an eye. But they can’t tell you whether a patient will have an adverse event after 6 months when your trial lasts only 6 weeks.
Some side effects take time to manifest. You need to have patients exposed to your drug long enough to know if something is dangerous. Diseases can take time to worsen, and you may need more patients to identify rare adverse events. So having a “phaseless” trial may help remove some of the artificial time that we waste. But we can never remove time from biology.
One concern that I have is with the term “real-time analysis” of data.
As you look at your data in real time, you continuously learn about your drug. This differs from taking a single look at your data at the end of your study. If you continue to adapt your trial, you need statistics that support your design. The FDA has issued guidance on how to properly conduct adaptive trials and make sure you aren’t drawing false conclusions. Make sure you get accurate estimates of your treatment effect and pre-specify your design.
We don’t want to learn continuously and fish for answers that we may not want to hear.
Using digital twins or computational controls can be beneficial as well. But they’re only as good as the data you put into them. If you use historical data to build your model, does that truly reflect what is happening today?
My main concern is that we can be more efficient but less certain.
SURPASS could help us run smaller trials with fewer patients. But if we remove patients from our conventional trial design, we need to prove we are getting the same results with these methods. The FDA has mentioned that you may need to simulate these types of trials to show you have adequate power and Type I error rates.
We can use automation to clean data and create datasets. Who is responsible if something goes wrong? Is it the sponsor? CRO? Programmer? Site?
There are already many vendors that we work with when developing drugs. By removing some of that time, we can get our products to market faster. But we may also create another black box between the patient and the FDA.
I also think the industry will push back.
Shortening your trial doesn’t always mean you have a better drug.
If we can get our drugs to market in less than 4 years, we can put more of our R&D budget toward more products. We will spend less money developing our drugs and have more time left on our patent when we go to market. We may also spend less money overall by having smaller trials.
But maybe we’ll spend that money developing more products we may not need.
I think the question is not how fast we can get our drugs to market.
It’s how quickly we can get reliable data on whether our product will benefit our patients and what the risks are.
So if we can partner with the FDA to create methods that they trust. we can use these concepts to develop more drugs that help our patients. The FDA has been incorporating these concepts into its guidance. They recently updated their draft guidance on master protocols in 2026.
I believe that SURPASS can change how we develop drugs in the future. We need to find ways to improve our current trial design. We don’t need to recreate our infrastructure for each phase. We can also reduce the number of patients we need in a control group. We shouldn’t wait months to analyze data we already have, or use people to do tasks computers can do.
But we need to make sure that we are not removing time that we need to have in our trials to know that our drug is safe and effective.
If we can do that I think SURPASS will change the way we develop drugs. If not we are just making our trials faster. Which isn’t what patients need. They need data that they can trust sooner.
