Dr. Manouchehr Hessabi
← All writing
8 min readstroke outcomes · clinical trials · research methods

Why multinational stroke trials are harder than they look

The challenges of running stroke research across countries: different patients, care systems and outcome raters, and how trials keep results comparable.

By Manouchehr Hessabi, MD, MPH

Acute stroke trials need large numbers of patients, each treated within a narrow window of hours, so they recruit across many hospitals and often many countries. That breadth speeds enrollment and widens the range of people a result might apply to. It also brings in differences in who the patients are, what care surrounds the trial treatment, and how the outcome is scored, and a well-designed trial has to plan for all three.

Why are stroke trials run across many countries?

A trial comparing two stroke treatments is looking for a difference in outcomes that may be modest, so it needs thousands of participants to detect it with confidence. No single hospital, and often no single country, sees enough eligible patients quickly enough. Spreading enrollment across sites is the practical answer.

Trialists protect the fairness of the comparison with a simple tool. In stratified randomization, patients are randomized within each site or country, so every setting contributes patients to both arms. Whatever is unusual about one region is then present on both sides of the comparison there, rather than piling up in one arm. That keeps the head-to-head comparison fair. It does not make the regions alike, and the rest of this article is about why that matters.

How can the patients differ from one country to another?

A useful window comes from ENCHANTED, an international trial that compared a lower dose of the clot-dissolving drug alteplase with the standard dose. A secondary analysis compared participants enrolled in China with those enrolled elsewhere (Song and colleagues, Stroke and Vascular Neurology, 2017).

The two groups differed in observable ways:

  • Age and sex. Participants in China were younger on average (64.3 versus 68.2 years), and fewer were women (33.9 versus 40.8 percent).
  • Type of stroke. More participants in China had a final diagnosis of stroke caused by atheroma in a large artery (56.5 versus 26.4 percent), roughly twofold. Atheroma is the fatty build-up inside artery walls. At the same time, a blocked large artery seen directly on vessel imaging was recorded far less often in China (4.2 versus 24.3 percent), a reminder that how a characteristic is measured shapes what the comparison shows.
  • Timing. Participants in China were randomized later after symptom onset, at a median of 3.1 hours versus 2.4 hours elsewhere.

The authors suggested possible reasons for the later treatment, including delayed recognition of stroke symptoms, long transfer distances and traffic, the cost of alteplase, concern about bleeding, and the extra time that trial consent requires. They offered these as possibilities, not tested findings, and a careful reader should hold them the same way.

Song L, Wang X, Robinson T, et al. (2017). Characteristics, management and response to alteplase in China versus non-China participants of the ENCHANTED trial. Stroke and Vascular Neurology.DOI: 10.1136/svn-2017-000085

The methodological point is that a trial's headline result is an average over all of these populations. The standard way to check whether that average hides important variation is a heterogeneity test, a formal check of whether the treatment effect differs between subgroups more than chance would explain. In ENCHANTED, the authors reported no heterogeneity in the effect of low-dose versus standard-dose alteplase between the China and non-China groups on the main outcome of death or disability. The populations differed, but the comparison between doses held up across them.

Why does the care around the trial matter?

A trial tests its treatment on top of whatever care system it lands in. In the same ENCHANTED analysis, participants in China were less likely to receive endovascular clot retrieval, the catheter procedure that physically removes a clot, and less likely to be cared for in a dedicated stroke unit. The authors described them as having been "managed less intensively."

This distinction is worth making carefully. Randomization within a region keeps the comparison fair inside that region, because both arms receive the same background care. But background care can still change the absolute numbers: how many patients in either arm recover well, die, or are left with disability. A treatment tested where clot retrieval is routine and one tested where it is rare are being asked slightly different questions, even when the trial protocol is identical. The same caution applies when reading drug comparisons such as the tenecteplase trials.

Can different raters score the same patient differently?

Most acute stroke trials use the modified Rankin Scale (mRS) as their primary outcome, a 0 to 6 grade of overall disability in daily life. A companion article covers what the modified Rankin Scale measures. The issue here is narrower: the scale is assigned by a person, and different people can score the same patient differently.

Agreement between raters is usually summarized with kappa, a statistic that measures agreement beyond what chance alone would produce. A kappa of 1 means perfect agreement; 0 means no better than chance. A 2024 systematic review and meta-analysis pooled 46 studies with 8,608 participants (Tvrda and colleagues, European Stroke Journal, 2024). Its key findings:

  • Across 29 studies, overall inter-rater reliability was kappa 0.65 (95 percent confidence interval 0.58 to 0.71).
  • For the standard mRS, used without added structure, kappa was 0.55 (0.46 to 0.64) across 13 studies.
  • For the Rankin Focussed Assessment, a structured interview approach, kappa was 0.94 in 2 studies.
  • Reliability was kappa 0.56 in studies before rater training and 0.69 in studies after it.

A multinational trial adds a further layer that these numbers do not directly measure. The mRS depends on a conversation about a patient's daily life, and that conversation happens in different languages, with different expectations about work, family roles and independence. It is reasonable to expect this to add variation between sites, which is part of why trialists standardize the assessment as tightly as they can. No pooled figure for that specific effect was identified here, so it is best treated as a reason for care rather than a measured quantity.

What do trialists do to keep results comparable?

Several standard design tools address the problems above:

  • Rater training and certification. Assessors are trained on the scale before they score trial patients, sometimes with a formal certification step.
  • Structured interviews. Fixed questions in a fixed order narrow the room for individual judgment, consistent with the higher agreement for structured approaches in the 2024 review.
  • Blinded outcome assessment. The person scoring the outcome does not know which treatment the patient received, so expectations cannot tilt the score toward one arm.
  • Central review of recorded assessments. Video-recorded interviews can be scored by a central panel. In the 2024 review, video review agreed well with face-to-face assessment (weighted kappa 0.87 across 3 studies), and the authors noted that central video review may be more reliable than a telephone interview.
  • Stratified randomization and central imaging review. Randomizing within sites, and reading scans in a central core laboratory, keep regional differences from becoming imbalances between arms.

The review's own conclusion is a fair summary of the field: single interventions do not completely remove reliability concerns. The tools work best in combination.

What does this mean for who a result applies to?

Generalizability is the question of whether a trial's result applies to people outside it. It depends on knowing who was enrolled, and that information is not always reported. A review of 12 acute ischemic stroke trials conducted in the United States and Canada between 2010 and 2020, with 7,955 participants, found that only 5 of the 12 publications reported race and ethnicity. Where it could be assessed, Asian participants made up 2 percent and Hispanic participants 11 percent of enrollment. The authors concluded that this underreporting impedes interpretation of how far the trials generalize (Acton, Abbasi and Kasner, Journal of the American Heart Association, 2022).

Multinational enrollment is one way to broaden who a trial represents, but only if the trial reports who it enrolled and checks whether its effect held across them. A reader weighing a stroke trial can ask four questions:

  1. Where were patients enrolled, and who were they?
  2. What care did they receive alongside the trial treatment?
  3. How was the outcome scored, and by whom?
  4. Was the effect consistent across regions, and was that formally tested?

Frequently asked questions

Why not run a stroke trial in a single country?

Mainly speed and size. A single country may not enroll enough eligible patients within the treatment window in a reasonable time, and a trial confined to one care system says less about how a treatment performs elsewhere.

Is the modified Rankin Scale reliable?

Reasonably, with caveats. The 2024 meta-analysis found moderate agreement for the standard form and substantially higher agreement for structured approaches and after rater training.

Does a regional difference mean a treatment works differently there?

Not by itself. Differences in patients and care are expected. The question is whether the treatment effect differs, which is what a formal heterogeneity test examines, as ENCHANTED reported for its dose comparison.

For readers who want to go further, the stroke outcomes research program and the peer-reviewed publications are the natural next stops.

About the author. Dr. Manouchehr Hessabi is a physician-epidemiologist and Senior Research Scientist at the BERD core of UTHealth Houston's Center for Clinical and Translational Sciences. See his peer-reviewed publications or research programs.