* Total Points: 0
Back to Lessons Capstone: Audit the Power Statement 0 pts Module 6 · Lesson 5
Introduction

Capstone: Audit the Power Statement

One fictional manuscript, every trick in the module — you're the reviewer

Your Assignment

The (fictional) VELOCITY trial has landed in your journal club: a single-center RCT of drug-coated balloon (DCB) versus plain balloon angioplasty for femoropopliteal disease. Primary endpoint: 12-month primary patency. It concludes that DCB "provides no patency benefit."

Every number in this paper is arithmetically consistent — nothing is fabricated, nothing miscalculated. And the conclusion is still wrong. Your job is to find out why, using everything from Lessons 1–4: the delta audit, the calculator, attrition math, CI interpretation, and the post-hoc power fallacy.

This is the skill the module was building toward: reading a power paragraph the way you read an operative note — knowing what should be there, noticing what's missing, and never mistaking fluent prose for sound method.

Audit Step 1: The Power Paragraph

Here is VELOCITY's sample size section, verbatim. Read it like a reviewer — then we'll take it apart.

VELOCITY Trial — Methods (fictional)

Sample Size Calculation

Based on a retrospective series from our institution demonstrating 12-month primary patency of 55% with plain balloon angioplasty and 80% with DCB, we determined that 52 patients per arm would provide 80% power to detect this difference at a two-sided alpha of 0.05. Anticipating 15% loss to follow-up, we enrolled 124 patients (62 per arm) with 1:1 randomization.

Checks outThe arithmetic

Run 55% vs 80% through Lesson 3's calculator: ~52 per arm at 80% power. Correct. The attrition inflation is even done properly — dividing by 0.85 (52/0.85 = 61.2 → 62), not multiplying by 1.15. This is a competently executed calculation. Which is exactly why arithmetic-checking alone is never enough.

Red flagThe delta source

A 25-point absolute patency difference, taken from the authors' own retrospective series. Lesson 2's alarm bells should all be ringing: retrospective single-center comparisons are confounded by selection (who got DCB and why?), and effects that motivate trials are selected for being impressive. Published RCTs of this comparison report patency differences closer to 10–15 points. The trial is powered for an effect roughly twice what the literature supports.

Red flagThe missing MCID

Nowhere does the paper say what difference would matter. Would a 10-point patency improvement change practice? Most interventionalists would say yes — and detecting 65% vs 75% requires roughly 326 per arm, six times what VELOCITY enrolled. The study is a test of "is DCB spectacularly better?" quietly presented as a test of "is DCB better?"

Audit Step 2: Results vs. Design

Now the results and discussion — where design debts come due.

VELOCITY Trial — Results & Discussion (fictional)

Results

Follow-up was completed by 93 patients (46 plain balloon, 47 DCB; 25% attrition). Twelve-month primary patency was 68% with plain balloon versus 75% with DCB (difference 7 points, 95% CI −11 to +25; p = 0.45). Of nine prespecified secondary endpoints, freedom from target lesion revascularization favored DCB (p = 0.04).

Discussion

We found no significant difference in primary patency. A post-hoc power analysis using the observed 7-point difference yielded 11% power, suggesting our study may have been underpowered for small effects; nevertheless, our findings indicate that DCB provides no meaningful patency benefit, while significantly reducing reintervention.

CompoundingAttrition exceeded plan

They planned for 15% loss and suffered 25%, leaving ~46 per arm — below even the 52 the optimistic calculation demanded. Against the realistic 10-point MCID, the completed study had roughly 18% power. This trial was over before the first balloon inflated.

FallacyThe CI says "we can't tell" — the conclusion says "no benefit"

The CI runs from −11 to +25 points: compatible with meaningful harm, nothing, or two and a half times the MCID in benefit. Lesson 4's framework calls this an uninformative negative — yet the discussion converts it to "no meaningful patency benefit." Note also the observed 7-point difference is close to the 10-point MCID; framed honestly, this is a signal awaiting an adequately powered trial, not a null result.

FallacyPost-hoc power, by the book

"Observed power of 11%" computed from the observed difference — the exact circular maneuver from Lesson 4. It restates p = 0.45 and adds nothing. The legitimate version would have been power against a prespecified MCID, disclosed as a limitation.

MultiplicityThe secondary endpoint rescue

One significant result among nine secondaries (a ~37% chance of at least one false positive even if nothing works), in an underpowered trial where any "hit" is inflated by the winner's curse — promoted to the discussion's closing claim. The abstract will say "DCB reduces reintervention." Journal club should not.

The Power Audit Checklist

Seven questions, in order. They take two minutes and catch nearly everything.

⚡ The StatUp Power Audit

  1. Where did delta come from? Clinical importance (MCID)? Or a pilot, a retrospective series, or the feasible n in disguise?
  2. Is delta plausible? Compare it to published effects in the field. A delta twice the literature is a rigged calculation.
  3. Are the assumptions realistic? Control event rate and SD — check them against real data, not hope.
  4. Does the arithmetic check out? Thirty seconds with a calculator. Necessary, never sufficient.
  5. Did reality match the plan? Attrition, event counts, allocation — what n actually reached analysis?
  6. Was the analyzed endpoint the powered endpoint? Watch for composites, switches, and secondary-outcome promotions.
  7. Do the conclusions respect the CI? A negative claim requires a CI that excludes the MCID. Post-hoc "observed power" counts for nothing.

Applied to VELOCITY: items 4 and (half of) 5 pass; items 1, 2, 6, and 7 fail outright. A paper can be arithmetically flawless and still be a machine for laundering "we couldn't tell" into "there's no benefit." The checklist is how you catch it in real time, in journal club, with the authors' fluent prose working against you.

Exercise: The Full Audit

Eight questions on VELOCITY — and on the audit skills you'll reuse forever.

Question 1 of 8

Module Complete!

0
Total Points Earned
Exercise (0/8 correct) +0 pts
Lesson Completed +100 pts

Capstone: Audit the Power Statement

Module 6 - Sample Size and Power complete

What You Can Now Do

  • Explain power as the four-way trade-off between n, effect size, alpha, and variability — and why negative ≠ equivalent.
  • Choose a defensible delta: the MCID, not the pilot estimate, not the feasible-n reverse-engineer, not Cohen's folklore.
  • Run the numbers yourself for binary and continuous outcomes, including dropout inflation and the big exchange rates (halve delta ≈ 4× n).
  • Handle a fixed sample honestly: feasibility check first, estimation framing, CIs against the MCID — and never post-hoc observed power.
  • Audit any power paragraph in two minutes with the seven-question checklist.

Next time you're handed a "negative" study in journal club — or asked "how many patients do we need?" on rounds — you have the whole toolkit. Pair this module with Module 3 (Understanding Your Output) for interpretation and Module 1 (Reading a Paper) for the rest of the audit.