Read Trials on the Pragmatic-Explanatory Continuum

Key points#

• Identify the decision the trial was designed to inform before calling it pragmatic or explanatory.
• Use a continuum across design domains rather than sorting whole studies into two boxes.
• Pragmatic design aims to estimate effects under conditions resembling intended routine use; explanatory design isolates an effect under more controlled conditions.
• Pragmatism does not excuse bias, vague intervention descriptions, incomplete outcomes, or weak analysis.
• Interpret the result together with implementation, adherence, usual care, missing data, and variation across sites.

Two intentions, many design choices#

An explanatory trial asks whether an intervention can produce an effect under conditions chosen to clarify efficacy or mechanism, while a pragmatic trial asks how a strategy performs under conditions close to those in which a decision will be made. These are intentions, not mutually exclusive study types.

The distinction matters because design changes the effect being estimated. Intensive training, selected participants, frequent follow-up, and tightly standardized delivery may help isolate a biological or behavioral effect; broader eligibility, ordinary staff, routine follow-up, and flexible delivery may estimate the result of offering the intervention through a real service.

Neither estimate is universally superior. A controlled efficacy question can be the right early question. A policy or comparative-care decision may need evidence that includes ordinary variation in delivery and uptake. Trouble begins when the design answers one question and the conclusion is written as if it answered the other.

Use the nine PRECIS-2 domains#

PRECIS-2 was developed to help trial teams align design choices with purpose. Each domain is rated from very explanatory to very pragmatic relative to the intended usual-care setting.

Eligibility#

Narrow criteria can reduce clinical heterogeneity or protect participants during early evaluation. Broad criteria can better represent people who would be offered the intervention. Read the exclusions, screening log, and enrolled population rather than accepting “real world” in the title.

Recruitment#

Special campaigns, repeated invitations, or incentives can produce a study population different from people encountered through ordinary care. More pragmatic recruitment occurs through pathways close to intended use. The method also affects consent, selection, and who never enters the denominator.

Setting#

A trial conducted in specialized centers may answer a question for those centers. Multiple community and health-system settings may improve applicability, but only if their characteristics and contribution are reported. Site count alone does not establish diversity.

Organization#

Extra staff, equipment, expertise, and protected time can support reliable delivery while making replication difficult. A pragmatic design uses resources close to those expected after the study. Cost and workforce assumptions belong in the interpretation.

Flexibility of delivery#

An explanatory protocol may control timing, provider behavior, and co-interventions to test a defined treatment. A pragmatic protocol may allow ordinary variation. Too much flexibility can leave you unsure what was actually offered, so treatment components and site adaptations still need documentation.

Flexibility of adherence#

Trials differ in how strongly they monitor and promote participant adherence. Extra reminders and visits can estimate benefit under sustained use. Routine levels of support can estimate the consequence of offering the intervention through normal care. Both adherence and support should be reported, because they help explain the observed effect.

Follow-up#

Research-only visits may improve outcome completeness but alter participant behavior and workload. Routinely collected data reduce additional contact while introducing coding, capture, and missing-data problems. “From the electronic record” is a data source, not proof of outcome validity.

Primary outcome#

Explanatory trials may use a sensitive physiological or surrogate measure connected to mechanism. Pragmatic trials often prioritize outcomes meaningful to patients, clinicians, or systems. The outcome must still be valid, prespecified, and measured comparably across groups.

Primary analysis#

An intention-to-treat analysis estimates the effect of assignment and preserves the protection of randomization. That often aligns with a pragmatic question about offering a strategy. Per-protocol or adherence-adjusted analyses can address different questions but introduce assumptions and potential bias. The estimand should be explicit.

A trial can sit at different points in these nine domains. A broad, multicenter study may still use intensive research follow-up. A tightly controlled intervention may measure a patient-important outcome. The wheel is a design conversation, not a total score of quality.

Pragmatic does not mean uncontrolled#

Randomization remains valuable because ordinary-care data contain confounding. Allocation concealment, appropriate masking where feasible, prespecified outcomes, complete follow-up, transparent analysis, and accurate reporting still matter.

Some pragmatic trials randomize clinics, hospitals, or care teams rather than individuals to avoid contamination or to study a system-level intervention. Cluster randomization changes the unit of analysis, sample-size calculation, baseline-balance problem, and consent context. The report should state how clustering was handled and how many clusters contributed to each group.

Open-label delivery may be necessary when staff must know which pathway they use. That knowledge can alter co-interventions, documentation, and outcome assessment. Independent or objective outcome measurement may reduce some risks, but every claimed safeguard should be checked rather than assumed.

Routine data can cover large populations at lower research burden, and they can also miss care received elsewhere, lag behind events, change definitions during the study, or encode billing behavior instead of the intended clinical outcome. Outcome algorithms need validation in the studied data environment.

Understand what nonadherence means#

Suppose an intervention has a strong effect among people who receive it exactly as planned, but many people assigned to it do not complete it under ordinary service conditions, though an explanatory analysis may clarify the effect under sustained delivery. A pragmatic intention-to-treat result may be smaller because it includes the realities of uptake.

That smaller estimate is not automatically a failed trial. It may be the correct answer to “What happens if this health system offers the strategy?” It does not answer “What is the effect among every person who would fully receive it?”

The interpretation needs a process description:

• What proportion received the assigned intervention?
• How much delivery varied across staff and sites?
• Was the comparator contaminated by similar care?
• Which co-interventions differed?
• Did implementation mature during the trial?
• Were reasons for nonuse measured?

Avoid treating post-randomization adherence groups as if randomization still balanced them. People who adhere often differ from those who do not. Causal analyses beyond intention-to-treat require explicit assumptions and sensitivity checks.

Define usual care instead of naming it#

“Usual care” varies by site, clinician, time, access, and policy. It may change during a long trial. A pragmatic comparator is informative only when you can see what the comparison group actually received.

Report baseline practice, permitted alternatives, co-interventions, major policy changes, and site variation. If the intervention adds a new process to variable usual care, estimate both the assignment effect and relevant implementation measures, because a pooled average across sites can hide benefit in one delivery context and little difference in another.

Applicability is similarly local. Broad eligibility helps, but you still have to compare study patients, settings, resources, and background care with the decision you have. A pragmatic design supports transfer when those conditions align; it does not erase contextual differences.

Read the article in a fixed order#

When you open the paper, work through these in order:

1. What decision and estimand were specified?
2. Where does each PRECIS-2 domain sit relative to intended routine care?
3. Were randomization and other bias protections sound?
4. What intervention and comparator were actually delivered?
5. How complete and valid were the outcomes?
6. What did adherence, contamination, and site variation contribute?
7. Are benefits, harms, burden, and resource use reported?
8. Which populations and settings remain outside the evidence?

The CONSORT pragmatic extension exists because you need these details to judge applicability. The label in a title is not enough.

Match the evidence to the decision#

Read explanatory evidence for what it can establish under controlled conditions. Read pragmatic evidence for the effect of a strategy under the conditions actually represented. When both exist, differences can reveal how delivery, population, adherence, or background care shapes outcomes.

The useful question is not which trial philosophy wins. It is whether the design, conduct, analysis, and conclusion all point to the same decision. The guide to keeping a systematic review current explains how changing practice can also alter the relevance of a body of trial evidence over time.