August 2026
M	T	W	T	F	S	S
	1	2
3	4	5	6	7	8	9
10	11	12	13	14	15	16
17	18	19	20	21	22	23
24	25	26	27	28	29	30
31

Mark Hensley: The Fragility Index – How Robust Are Clinical Trial Results?

Mark Hensley/LinkedIn

Aug 20, 2026, 10:38

Mark Hensley: The Fragility Index – How Robust Are Clinical Trial Results?

Mark Hensley, Founder and Managing Director at Hensley Biostats, shared a post on LinkedIn:

“One of my clients sent me a competitor’s Phase 3 result last month and asked whether the data was strong enough to anchor his own trial design around.

Three hundred patients. p is equal to 0.04.

The kind of number that shows up in a Series C proposal.

I ran a quick calculation.

How many patients in this trial would need to have had a different outcome for that p-value to cross 0.05?

The answer was two. Two patients, out of three hundred, were the entire distance between significant and not significant.

That number is called the Fragility Index.

It measures how many patient outcomes would need to be different to flip a trial’s statistical significance.

And a systematic review of 399 RCTs from the NEJM, Lancet, JAMA, and BMJ found that a quarter of them had a Fragility Index of 3 or less.

The median across major medical journals was only 8.

In critical care, the median dropped to 2, and over 40% of trials had an index of 0 or 1.

These are the studies that change clinical practice. And a meaningful share of them are a couple of patients away from saying nothing at all.

How does fragility interact with dropout?

Well, in over half of published trials, the number of patients lost to follow-up exceeds the Fragility Index.

If the index is 2 and 5 patients left the study early, you’re not looking at a proven result.

You’re looking at the assumption that every missing patient would have had a good outcome.

A rough way to gauge fragility before running the formal math: check how close the p-value sits to 0.05, then compare the dropout count to the number of events separating the two arms.

If the dropouts approach or exceed that margin, the published conclusion is leaning on data nobody collected.

I told that founder his competitor’s result was probably real.

But the evidence behind it was thinner than the paper suggested.

Two patients thinner.

When you’re building a clinical program on someone else’s published result, it helps to know how wide that ledge is before you step onto it.”

Stay updated with Hemostasis Today.