CORE Symposium Recap: What Clin Ops Leaders Learned About Enrollment Forecasting

Enrollment forecast failures are not sudden. By the time they’re caught, the trial may be months behind, and the options to catch up are limited, expensive, or even nonexistent.

This challenge was the starting point for our CORE Symposium in Philadelphia, where clin ops leaders from nearly 30 pharma organizations met to rethink clinical trial forecasting.

The event opened with a panel discussion featuring leaders from Boehringer Ingelheim, Acadia Pharma, and Veristat. They looked at limitations of the more primitive forecasting methods teams still use, reframed the use of benchmark studies, and shared how to build a forecast that incorporates the right KPIs and risks so that teams can measure against these early enough to act on it.

Then we put the concepts to work. In a hands-on breakout session, attendees split into groups to stress-test sample forecasts, rank the impact of risks, and diagnose an enrollment plan that had gone off track. Here’s what we learned and what it says about the future of clinical trial enrollment forecasting.

The problem with how we've always done it

The panel first looked at the method that still drives most enrollment planning:

• Algebra 1 calculations based on patients per site per month (P/S/M)
• Excel spreadsheets
• Historical benchmarks

This method always creates dangerously naive forecasts, without taking into account the key inputs that actually drive enrollment.

For example, a spreadsheet model based on P/S/M always generates linear enrollment curves, when actual performance follows an S-shaped curve. It produces a tidy average, but that is always incorrect. It doesn’t account for site ramp-up, varying screen failure ratios, seasonal swings, or the fact that site and county performance is varied. During the actual enrollment, spreadsheets only show what happened after the fact, not what will happen next.  

Benchmarks have a different problem: they should be used to inform decisions, not dictate them. Unless the benchmark set is identical in every way to the study you are trying to forecast, those performance metrics will need adjustment.  

Multiple variables go into enrollment forecasting and this discipline deserves a best in class solution that enables rigorous and scalable outputs that hold up to senior leader scrutiny.

We asked the CORE audience: what is your current enrollment forecasting method? 

Enrollment curves vs. cycle time

A cycle time tells you how long it took to enroll a similar study. An enrollment curve shows what actually happens. These are related, but distinctly different. 

The panel walked through the operational realities that drive these differences: 

Staggered site startup. In a global program, sites activate at different times as they clear regulatory, translation, and contracting hurdles. Your first site going live says little about the rest, especially if the others may not ramp up for months. A P/S/M approach incorrectly assumes simultaneous site ready dates for all. If you're using this approach, your model is already broken. 

Protocol complexity. If your protocol is more complex than the study you’re using as a base, the forecast will be wrong. Complexity influences ability to enroll and screen failure ratios.  

Disease state and geography. Where you run a trial is shaped by where you plan to commercialize, the standard of care, and the patient pathway, which affects site selection, patient access, and yes, enrollment performance. 
 
Site relationships and patient characteristics. Visit and treatment patterns, patient profiles, and how well you know your sites all shape enrollment—especially in rare disease and studies with complex criteria. 

One idea that stood out was a site startup complexity score, which gives teams a way to plan for the fact that some sites will take longer to activate than others.  

Benchmarks are a compass, not a map

Benchmarks are a good indicator and a useful starting point. They’re especially helpful in understanding how complex your study is relative to the industry, and translating that complexity into expected impact. 

But benchmarks should never be taken at face value. First, no past study will ever repeat itself. Second, benchmark data is often older than teams realize. The context in which patients were recruited isn’t the same as now.  

As one panelist put it, a benchmark works like a compass: it can tell you north or south, but it can't give you turn-by-turn directions. 

We asked the CORE audience: what makes a good benchmark for a forecast?

What defines a quality forecast

The old models fall short. But even when you are utilizing modern methods, there are a few simple ways to contextualize a quality forecast.

A good forecast documents how every assumption was reached. You’ve got to be able to explain your assumptions. That transparency pays off later. If a benchmark screen failure ratio was 44%, why did we choose 56%? If Brazil was a poor enroller for an indication, why do we think it will be better now?  

A good forecast is reiterative: Fancy Gamma-Poisson models and Monte Carlo simulations are great early on. But once all the sites are selected, they are obsolete. You must build a final forecast based on site level assumptions.  

A good forecast is rigorously planned, not accurate. The “probability of success” of enrollment can be completely derailed if there is a delay with drug supply, regulators or contracting. Does that mean the forecast was incorrect or does life just happen? 

If your plan falls behind, what's the best way to catch up?

One of the strongest moments came from an analogy about good sleep habits. 

If you aim to get eight hours of sleep but you only get four, the intervention isn’t to “get more sleep.” The interventions happen before the sleep occurs—such as no screen time one hour before bed, no eating after 8 pm, or eliminating coffee after 4 pm. 

Enrollment works the same way. If you need 20 patients by the end of the month and you’ve only enrolled five, the solution isn’t “enroll more patients.” Without understanding the upstream root cause—slow site activation, higher than anticipated screen failure ratio, etc.—you will not pick the right interventions to accelerate enrollment. 

Putting it into practice

Exercise 1: The Ripple Effect 

With the panel setting the foundation, CORE attendees broke into teams and worked through a sample enrollment forecast. Their job was to weigh real setbacks, from slow site activation to protocol amendments to supply delays, and decide which ones would hit the trial the hardest downstream. 

This exercise made one thing clear: not all problems are created equal. Some delays are localized and can be fixed with operational changes. Others spread through the whole study. The groups often saw patient-side problems, like a jump in screen failures, as the hardest to recover from. Slow site startup in a country that holds a large share of the target population was a close second. 

As groups talked, they also added risks that rarely get a spotlight in the forecast: CRO involvement, diversity goals, vendor performance, patient density, lab supply. It all raised the question, “If you can’t model all of this up front, how will you measure it when it actually happens?” 

Exercise 2: Something Is Wrong 

The second exercise recreated a moment familiar to many clin ops leaders: enrollment is off track, but it’s not clear why. 

Groups started with the kind of high-level summary teams are typically handed when a study falls behind. 

At that level, it was easy to see that something had gone wrong, and at roughly what point in time. What wasn’t easy to see was why. Without the right data, the theories pointed in every direction. 

As the exercise went on, groups got more detailed data, first by country and then by site. With each new layer, the picture sharpened. Patterns emerged, the list of possible causes shrank, and the conversation shifted from guesswork to specific root causes and focused interventions. 

What it all tells us about trial forecasting

The panel explained why traditional forecasts break. The exercises made the room feel it firsthand. Together, they pointed to a few clear lessons for clinical trial forecasting.

• A single number isn’t a forecast. P/S/M, a spreadsheet, and a benchmark can orient you, but they can’t capture site ramp-up, protocol complexity, or shifting patient populations.

• The biggest risks are often the least controllable. Forecasts need to reflect patient, site, and medical realities, not just operational timeline goals.

• Forecasting must be iterative. Rigor, documented assumptions, and regular recalibration against real data are what keep a forecast reliable after first patient in.

• Aggregate data shows something is wrong, but not why. Comprehensive data, organized and visualized in the right way is what turns theories into real answers and fixes.

• Speed of insight is everything. The sooner a team can see a problem forming, the easier it is to act. Once enrollment falls behind, catch-up gets harder and riskier. 

Uncover enrollment problems before they compound

Want to see how ProofPilot gives sponsors real-time enrollment plan performance visibility? Explore our enrollment forecasting tools.