Radial artery occlusion after angiography: checking a trial's sample size

A 2026 trial protocol plans 818 patients to test pressure-guided wrist compression against radial artery occlusion. We rework its sample size and power with fibub's sample size tool.

Many heart tests and stent procedures now go in through the radial artery at the wrist rather than the groin. Afterwards a band with an air balloon presses on the wrist to stop the bleeding. Press too hard or too long and the artery can clot shut: radial artery occlusion (RAO). It is usually silent, but it closes that artery for future procedures. A trial protocol published in PLOS ONE in 2026 plans to test a band with a pressure gauge and step-by-step release against the usual band. This article shows how the sample size tool lets anyone rework how many patients such a trial needs, and how the NNT calculator puts the hoped-for benefit in patient terms.

General education only, not medical advice.

The paper in brief

This is a protocol: the plan of a trial, published before the results. The trial is single-centre, randomised 1:1 and open-label (the doctors know which band is used), but the person who checks the artery with Doppler ultrasound at 24 hours does not. The main outcome is RAO at 24 hours.

The sample size rests on four choices: RAO of 4.3% with the usual band (taken from an earlier trial’s control arm), 1.0% hoped for with pressure monitoring, a two-sided 5% significance level and 80% power (an 80% chance of a clear result if the true rates really are 4.3% and 1.0%). The authors’ software gave 736 patients, and they plan 818 (409 per group) to allow for 10% dropping out.

Checked with the tools

Everything here is a recomputation from the protocol’s stated assumptions; there are no trial results yet.

Check 1: the 736. The Sample size tool, set to two proportions, 4.3% against 1.0%, gives 371 per group (exactly 370.69), 742 in all: six more than the protocol. The difference is the formula, not a mistake. The tool follows R’s power.prop.test, which uses the pooled rate under “no difference” for the significance part. The simpler unpooled formula, which many programs offer, gives 367.95, so 368 per group and exactly 736. Both are standard; they differ by less than 1%.

Check Protocol Our result
People per group before dropout 368 (736 in all) 371 with the tool’s pooled formula; 368 with the unpooled formula
After 10% dropout 409 per group, 818 in all 736 ÷ 0.9 = 817.8, so 818 (the tool: 413 per group, 826)
Power with 409 per group, if 4.3% against 1.0% 80% after dropout 83.7% if nobody drops out
Patients per occlusion avoided not stated NNT 30.3 at the planned rates

Check 2: dividing, not multiplying. The protocol inflates 736 by dividing by 0.9 (817.8, so 818), which is right: after 10% loss, 818 leaves 736. Multiplying by 1.1 instead would give 810 and leave too few. The tool’s Expected dropout (%) box divides in the same way.

Check 3: what if the benefit is smaller? A drop from 4.3% to 1.0% is a large effect: a relative reduction of about 77%. Figure 1 shows the trial’s power with 409 per group if the new band only brings RAO down to 1.5% or 2.0%: 67% and 47%. Figure 2 shows how fast the numbers grow: about 563 per group for 1.5% and 904 for 2.0% (before dropout). With small event rates, every bit of hoped-for benefit you give up costs many patients.

Power to detect fewer occlusions (4.3% in the control arm)Power to detect fewer occlusions (4.3% in the control arm)00.20.40.60.8102505007501e+3People per groupPower80% power409 plannednew arm 1.0%new arm 1.5%new arm 2.0%fibub.com
Figure 1. Power curves from the Sample size tool's two-proportion maths for the protocol's 4.3% control rate and three possible rates with the new method. The dashed line marks the 409 people per group the trial plans.
People per group needed as the expected benefit shrinksPeople per group needed as the expected benefit shrinks05001e+31.5e+32e+30.511.52Occlusion rate with the new method (%)People per groupprotocol: 1.0%fibub.com
Figure 2. People needed per group for 80% power at a two-sided 5% level, as the true occlusion rate with the new method rises from 0.4% to 2.4%. Worked out with the tool's maths; not trial data.

Check 4: in patient terms. If the rates are 4.3% and 1.0%, the risk difference is 3.3 points, so about 30 patients would need the pressure-monitored band for one fewer occluded artery at 24 hours. The NNT calculator gives this from any pair of counts, such as 43 of 1,000 against 10 of 1,000.

How we made the graphs

The charts and numbers come from scripts/blog-figures/radial-artery-occlusion-hemostasis-trial-sample-size.mjs in fibub’s public code. It calls powerProp and solveN from src/lib/sci/power.ts, the maths behind the sample size tool, and compare from src/lib/sci/nnt.ts, the maths behind the NNT calculator. The unpooled formula is written out in the script for comparison only. Every input (4.3%, 1.0%, 5%, 80%, 10% dropout, 409 per group) is from the protocol; the alternative rates of 1.5% and 2.0% in the figures are our own what-ifs.

How we test the maths

On every change to fibub, tests/sci.mjs and tests/sci-tools/sample-size.mjs check the power maths against R:

  • 4.3% against 1.0% at 80% power must give 370.689 per group, R’s power.prop.test result, and the unpooled formula must give 367.946 (the protocol’s 368). This check was added with this article.
  • 50 per group, 50% against 75%, must give power 0.74017, as power.prop.test does; and the earlier check of 76.7 per group for 90% power.
  • For means, 20 per group with an effect size of 1 must give power 0.86895 (R’s power.t.test), and 64 per group with 0.5 must give 0.80146.

Check it yourself

On the sample size tool:

  1. Set Comparing to Two proportions (percentages).
  2. Type 4.3 in Group 1 (%) and 1 in Group 2 (%).
  3. Leave Significance level α on 0.05 and Power on 0.8.
  4. The result reads 371 per group, with Without dropout 371 per group (exactly 370.69) and Total 742.
  5. Type 10 in Expected dropout (%): it reads 413 per group and Total 826.
  6. Change Group 2 (%) to 2: the result jumps to 905 per group before dropout. The power curves under it show the same story as Figure 1.

On the NNT calculator, type 10 and 1000 for the first group and 43 and 1000 for the second: the big result reads NNTB 30.3.

What researchers can use it for

  • Recheck a protocol or grant before review: type in its stated rates, level and power and see whether the n matches, as here.
  • Explain small differences between programs (pooled against unpooled variance) rather than suspecting an error.
  • Stress-test the assumptions with the power curves: how much power is left if the benefit is smaller than hoped?
  • Plan the dropout allowance correctly by dividing, with the Expected dropout (%) box.
  • Report the benefit in patients with the NNT calculator, and later pool similar trials in the meta-analysis tool.

The fine print

  • The 1.0% target is optimistic; the trial will show whether it holds. Our what-if figures are not predictions.
  • Neither formula here adds a continuity correction; with it, 4.3% against 1.0% would need about 430 per group. Programs and protocols differ on this.
  • RAO at 24 hours is a short-term outcome; some arteries reopen later, which the planned 30-day check may show.
  • A single-centre, open-label trial can be affected by the team’s experience with the new band; the protocol plans a learning-curve analysis for that.

Sources

  • Zhang, X., Zou, L., Zhang, D., Yao, B., Chen, J., Wei, T. et al. (2026), Balloon pressure monitoring for radial artery hemostasis after transradial coronary procedures: protocol for a randomized controlled trial, PLOS ONE 21(6): e0350563, doi:10.1371/journal.pone.0350563. Licence CC BY 4.0, stated on the article page, checked 2026-11-16. Values used here, from the sample size section: occlusion at 24 hours assumed 4.3% with usual care (from the PROPHET-II control arm) and 1.0% or less with pressure monitoring; two-sided α 0.05, power 80%, 1:1 allocation, PASS 15.0.5; 736 patients before and 818 (409 per group) after allowing 10% dropout; registration ChiCTR2400088258.
  • R Core Team, power.prop.test (stats package): the pooled-variance normal approximation the Sample size tool follows.
  • Fleiss, J. L., Levin, B. and Paik, M. C. (2003), Statistical Methods for Rates and Proportions, 3rd ed., Wiley (sample size formulas for two proportions).

If you are an author or publisher and want a credit or a passage changed or removed, tell us and we will act quickly.

Tools in this article

Read next