RCT Recruitment Prediction: Better Statistics, Better Trials

RCT recruitment prediction is an incredibly important part of clinical trial design, especially as sponsors face persistent uncertainty over whether studies will recruit enough participants, quickly enough, and the statistical approach used to forecast recruitment can materially affect how realistic those expectations are. 

Image

Recruitment Forecasting During Trial Design

Recruitment is often treated as an operational issue that can be fixed once a trial is underway. That approach can be expensive.

The latest evidence shows that 37% of RCTs fail to meet their recruitment target, highlighting the need to assess recruitment feasibility before sites open and patients begin screening. 

Recruitment forecasts influence much more than a study timeline. They can affect:

• Site selection and activation strategy
• Trial duration and resource planning
• Budget assumptions
• Patient and investigator burden
• The feasibility of the proposed protocol

The Clinical Trials Transformation Initiative (CTTI) has previously argued for a broader recruitment planning approach covering protocol design, feasibility and site selection, and communication.

Recruitment should be designed into the trial, rather than treated as a problem to solve after launch.

What Is the Problem with Simple Patient Recruitment Estimates?

Historically, recruitment projections have often relied on straightforward calculations. A sponsor might estimate an average number of participants per site and multiply that figure across the expected recruitment period.

The appeal is obvious because the calculation is simple, transparent, and easy to communicate.

The problem is that patient recruitment is rarely perfectly predictable.

Image

A 2019 systematic review identified 13 statistical models for predicting recruitment at the design stage. Most use stochastic approaches, which attempt to account for the randomness inherent in patient arrival.

The models differ in how they handle time, recruitment rates, centre-specific performance, and site initiation. 

This creates a difficult balance.

More sophisticated models can represent recruitment more realistically. However, they also require more assumptions and parameters. That can make them harder for trial teams to implement and validate.

The result is a familiar pharma problem: A model can become statistically sophisticated without becoming operationally useful.

View interview Is AI Quietly Creating a GDPR Crisis in Clinical Research?

Is AI Quietly Creating a GDPR Crisis in Clinical Research?

Generative AI is transforming clinical trials, but a failure to distinguish between pseudonymised and anonymised data is creating costly regulatory gaps.

Image

Why Poisson Models Could Improve RCT Recruitment Prediction

New 2026 research compares deterministic, Poisson, and Bayesian approaches using data from six previously conducted RCTs. 

The analysis found that Poisson methods were more appropriate for recruitment prediction at the design stage than the alternatives tested.

Their advantage is that they treat recruitment as a process involving random events rather than assuming that participants will arrive according to a fixed schedule.

For a single-centre trial, the researchers recommend a homogeneous Poisson process, where the recruitment rate is assumed to remain broadly constant.

For multicentre studies, a non-homogeneous Poisson process can be more appropriate because recruitment rates can change over time.

This matters because a multicentre trial rarely starts with every site recruiting simultaneously. Sites activate at different times, recruitment capacity varies, and patient flow can change during the study.

Earlier research has demonstrated how non-homogeneous Poisson models can incorporate changing recruitment intensity, staggered centre openings, seasonal effects, and planned interruptions.

The 2026 research therefore builds on an established statistical idea while asking a more practical question: Which method is useful when sponsors are still making decisions at the trial design stage?

Image

What Bayesian Recruitment Models Still Offer in Trial Design

Bayesian approaches remain relevant.

The RECRUIT-IT project was established to investigate recruitment patterns across a large international sample of RCTs and assess the performance of an existing Bayesian recruitment model.

Its proposed dataset targeted around 300 completed or discontinued trials, with the aim of developing practical tools for planning and monitoring recruitment. 

Bayesian approaches can incorporate prior information and update predictions as recruitment data accumulate.

That can be valuable once a trial is underway.

However, the recent 2026 research subsequently found that Bayesian approaches using informative priors produce wider prediction intervals than the Poisson approaches evaluated.

The authors also identified difficulties around parameter selection and the availability of suitable software. 

This highlights an important distinction between statistical flexibility and practical usability.

The best model is not necessarily the most complex one, but is the one that reflects the recruitment problem sufficiently well while giving trial teams information they can act on.

From Recruitment Forecasts to Recruitment Intelligence

The bigger opportunity is to move recruitment prediction further upstream.

Pharma teams increasingly have access to historical trial data, site performance information, electronic health records, epidemiological datasets, and other sources that could improve feasibility assessment.

Pharmatica’s analysis of EHR-supported patient recruitment illustrates how digital systems are already being explored to identify potential participants and improve trial matching. 

Similarly, advances in deep learning in clinical trials could eventually support more sophisticated modelling of recruitment, site performance, and protocol feasibility.

But better data does not automatically produce better predictions.

Sponsors need to understand the assumptions behind the model, the quality of the historical data, and the uncertainty around the forecast.

A recruitment forecast should be treated as a range of plausible outcomes, not a promise.

RCTs should also report the methods and parameters used to generate recruitment predictions. Greater transparency would make forecasts easier to evaluate, reproduce, and improve. 

What Better Recruitment Prediction Changes

Recruitment prediction sits at the intersection of statistics, clinical operations, and trial strategy.

Relatively straightforward stochastic approaches can provide more useful uncertainty estimates than deterministic calculations, while more complex models may require information that is unavailable when sponsors are making early design decisions. 

The priority should therefore be fit-for-purpose forecasting, not simply asking “How many patients will we recruit?” but, “How confident are we in that estimate, what assumptions support it, and what should we do if recruitment follows a different trajectory?”

At Pharmatica, we examine the systems, strategies, and technologies shaping Clinical Development, connecting evidence and innovation with the decisions that determine how efficiently therapies move towards patients.

Pharmatica: Insight. Connection. Impact.

Image