Can a Randomized Controlled Trial Also Be Pragmatic?

Can You Randomize Participants Without Creating an Artificial Research Environment?

You want to determine whether a new intervention works under ordinary conditions. Perhaps you are evaluating an AI-assisted learning platform in university courses, a clinical decision-support system in hospitals, or a professional development program delivered through existing institutional arrangements.

A conventional randomized trial may appear to require tightly standardized procedures, specially selected participants, and intensive monitoring. Yet those conditions may differ substantially from how the intervention would actually be used.

Does randomization necessarily make a study artificial? Can researchers preserve the advantages of random assignment while allowing interventions to operate within routine practice?

The distinction matters because a trial may produce a credible causal estimate without necessarily producing evidence that transfers well to the settings where decisions must be made.

Pragmatism and Randomization Are Compatible

How Pragmatic Randomized Trials Work

What Makes a Randomized Trial Pragmatic?

A pragmatic randomized trial is designed primarily to inform decisions about interventions under conditions resembling their intended use.

Rather than asking only whether an intervention can produce an effect under highly controlled circumstances, a pragmatic trial asks what happens when the intervention is offered or implemented within a relevant practice environment.

For example, a pragmatic educational trial might compare an AI-supported formative assessment system with existing instructional practice across ordinary university classes. Teachers would deliver the intervention using institutional resources, and student outcomes would be measured through assessments relevant to routine educational decisions.

Random assignment remains central to the experimental comparison. Participants, classrooms, institutions, or other eligible units are assigned through a random mechanism to the specified intervention conditions.

The pragmatic character arises from choices about eligibility, recruitment, implementation, follow-up, and analysis, not from abandoning randomization.

What Is the Difference Between Pragmatic and Explanatory Trials?

The distinction between explanatory and pragmatic trials is commonly associated with the work of Schwartz and Lellouch, who described different scientific attitudes toward therapeutic trials.

Explanatory trials generally emphasize testing whether an intervention produces an effect under conditions selected to clarify efficacy or a particular mechanism. Pragmatic trials emphasize evidence useful for choosing among interventions in practice.

These are orientations rather than mutually exclusive categories.

The table describes common tendencies, not fixed requirements. An explanatory trial can use intention-to-treat analysis, and a pragmatic trial may require standardized intervention components or additional safety monitoring.

Researchers should describe the actual design characteristics rather than treating pragmatic and explanatory as labels that automatically determine methodological quality.

How Does PRECIS-2 Help Researchers Design Pragmatic Trials?

PRECIS-2, developed by Loudon and colleagues, is a tool for helping researchers align trial design choices with the intended purpose of the study.

It evaluates nine domains on a continuum from very explanatory to very pragmatic.

Each domain is scored from 1, very explanatory, to 5, very pragmatic. The scores can be displayed using a PRECIS-2 wheel to show how the trial's design aligns with its purpose.

PRECIS-2 is primarily a design-support tool rather than a universal quality score. A higher score is not automatically better. A trial intended to answer a mechanistic question may appropriately include more explanatory features.

The value of the framework lies in making design choices explicit and identifying inconsistencies between the trial's purpose and its procedures.

Can Individual Randomization Be Used in a Pragmatic Trial?

Yes. Pragmatic trials may randomize individual participants when interventions can be allocated separately and the resulting comparison remains feasible within routine practice.

For example, patients attending participating clinics might be randomly offered one of two accepted care strategies, with outcomes collected through routine health records.

In education, students might be randomly assigned access to alternative digital learning tools when the tools can be delivered independently without substantial interference.

Individual randomization can be efficient, but it may be unsuitable when providers cannot maintain distinct intervention conditions or when participants influence one another's exposure.

The unit of randomization should therefore reflect the intervention mechanism and practical conditions, not simply the desire to make the trial more pragmatic.

When Is Cluster Randomization More Pragmatic?

Cluster randomization may be useful when the intervention is naturally implemented at the level of a classroom, clinic, institution, or community.

For example, a hospital-wide decision-support policy cannot necessarily be delivered independently to individual patients without changing how clinicians work. Randomizing hospitals or clinical units may better reflect implementation.

Similarly, a teaching intervention requiring all students in a classroom to participate in shared activities may be more naturally assigned by class.

However, cluster randomization introduces within-cluster correlation, potential recruitment bias, and other design considerations. It may require more participants or more independently randomized clusters than an individually randomized design.

Cluster randomization is not inherently pragmatic. A cluster trial conducted under highly artificial implementation conditions may still be predominantly explanatory.

Does Pragmatic Mean Researchers Should Avoid Standardization?

No. Pragmatic trials may allow flexibility in how interventions are delivered, but that does not mean intervention definitions can be vague.

Researchers still need to specify the essential components of the intervention and comparator so that treatment assignment has an interpretable meaning.

For example, an AI-assisted feedback intervention might permit instructors to decide when students need additional explanation while requiring all participating instructors to use the same core feedback system.

Such flexibility may reflect routine teaching practice. Yet researchers must document the intervention sufficiently to establish what was evaluated.

Without an adequate intervention description, readers may be unable to determine whether observed differences arose from the intended treatment, implementation variation, or other contextual factors.

The appropriate balance depends on whether flexibility is part of the intervention strategy or an uncontrolled deviation from it.

How Does the Control Condition Affect Pragmatism?

Pragmatic trials frequently compare a new intervention with existing practice because the decision of interest concerns whether the new approach should be adopted.

For example, a university might compare an AI-assisted advising service with its existing advising arrangements rather than with an artificial inactive activity.

However, usual care can vary across sites. One institution may provide extensive support, while another offers limited services.

This variation may be relevant to the real-world decision, but it also affects interpretation. Researchers should describe the comparator and consider whether treatment effects differ across settings.

The choice among placebo, usual-care, active, and no-treatment controls should follow the intended causal contrast and ethical requirements.

What Outcomes Are Most Appropriate for Pragmatic Trials?

Pragmatic trials generally prioritize outcomes that matter to patients, learners, practitioners, institutions, or policymakers making the relevant decision.

In healthcare, these may include clinical events, functional outcomes, quality of life, or service utilization. In education, they might include learning achievement, course completion, retention, or meaningful measures of student experience.

The outcome should reflect the intended benefit rather than merely being easy to measure.

For instance, an AI learning platform may generate extensive engagement analytics, but the number of clicks or prompts submitted may not establish whether students learn more effectively.

Pragmatic trials may use routinely collected data to reduce participant burden and study costs. However, routine data are not automatically valid or complete for research purposes.

Researchers should evaluate measurement quality, missingness, differences across sites, and whether the data capture the intended outcome consistently.

Does Pragmatic Randomization Establish Causation?

Random assignment can support causal inference by reducing systematic baseline differences between treatment groups in expectation.

Nevertheless, pragmatic trials remain vulnerable to implementation failures, missing outcomes, nonadherence, contamination, and other complications.

Randomization does not guarantee that every participant receives the assigned intervention or that outcome measurement is unbiased.

The causal interpretation depends on the estimand and whether the design and analysis appropriately address events occurring after randomization.

A pragmatic trial often focuses on the effect of assignment to an intervention strategy, which may include real-world variation in uptake and adherence.

That effect can be highly relevant to policy decisions, but it is not necessarily the same as the effect that would occur if every participant adhered perfectly.

Why Is Intention-to-Treat Analysis Often Relevant?

Intention-to-treat analysis generally preserves participants in their randomized groups for the comparison, regardless of whether they fully receive or adhere to the assigned intervention.

This approach often aligns with pragmatic questions about the consequences of offering or implementing a treatment strategy.

For example, a university deciding whether to introduce an AI tutoring system may need to know what happens when access is offered to students, including those who use it infrequently.

However, intention-to-treat analysis does not automatically resolve missing outcome data or every post-randomization complication. The target estimand, missing-data assumptions, and sensitivity analyses must still be specified.

Per-protocol or adherence-related analyses may address additional questions, but naïvely comparing participants according to observed adherence can introduce confounding because adherence is not generally randomized.

What Are the Main Limitations of Pragmatic Randomized Trials?

Pragmatic trials may encounter substantial variation in implementation, heterogeneous participant populations, and differences in routine practice across settings.

These features can improve relevance to practice, but they may also increase outcome variability or complicate the interpretation of treatment mechanisms.

Contamination may occur when providers or participants share intervention components across assigned conditions. Recruitment and consent procedures may also affect who participates.

Some pragmatic trials rely on routinely collected records, which can contain incomplete or inconsistently defined measurements.

Furthermore, a trial conducted in ordinary practice is not automatically generalizable to every other setting. Applicability still depends on the characteristics of the participating population, institutions, intervention resources, and healthcare or educational systems.

The pragmatic label should therefore describe the trial's purpose and design characteristics rather than serve as a substitute for evaluating validity.

Evaluating an AI-Assisted Learning System in Ordinary University Classes

This design is pragmatic in several respects because recruitment, intervention delivery, and follow-up are embedded in ordinary university operations.

It remains randomized because course sections are assigned through a random allocation procedure. The analysis must account for the cluster structure, and the relatively small number of randomized sections may limit statistical precision.

The findings would inform whether introducing the AI system under similar institutional conditions improves outcomes. They would not necessarily establish the maximum benefit achievable under perfect adherence or intensive research supervision.

The distinction is valuable because the university's actual decision concerns implementation, not merely whether the software can produce favorable results under ideal conditions.

Common Misconceptions About Pragmatic Randomized Trials

How to Decide Whether a Pragmatic RCT Fits Your Research Question

Begin by clarifying whether the primary objective concerns efficacy under controlled conditions or effectiveness when the intervention is implemented through routine practice.

Then examine whether eligibility criteria, recruitment, delivery, follow-up, and analysis are consistent with that objective.

Use PRECIS-2 during protocol development to examine whether each design domain aligns with the intended purpose.

The broader issue of balancing internal validity with real-world relevance remains important even when the trial is explicitly pragmatic.

Finally, report which aspects of the design were pragmatic, which were more explanatory, and why those choices were appropriate.

Before Designing a Pragmatic Randomized Trial

Questions About Pragmatic Randomized Controlled Trials

Randomization and Real-World Relevance Can Coexist

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.