Guides

Is that change real, or is it noise?

Two numbers decide whether a score has moved: one asks whether the change is bigger than the measurement error, the other whether it is big enough for the patient to care. They are not the same, and one is useless without the other.

Updated 9 September 2026 · Free to read

Two different numbers tell you whether a change on an outcome measure means anything, and they answer two different questions. The MDC asks whether the change is bigger than the measurement error. The MCID asks whether it is big enough for the patient to care.

What is the difference between MDC and MCID?

The minimal detectable change (MDC) is the smallest difference in score that is larger than the noise in the measure itself. Below it, you cannot tell a real change from the same patient being measured twice. The minimal clinically important difference (MCID) is the smallest change that someone — usually the patient — would recognise as an improvement worth having. The first is a property of the instrument. The second is a judgement about people.

You need both, and in that order. A change smaller than the MDC is not evidence of anything, however clinically appealing the number looks. A change larger than the MDC but smaller than the MCID is real and probably does not matter yet.

Where the numbers come from

The MDC is calculated from the measure’s test-retest reliability: the same patients, measured twice, close enough together that nothing has genuinely changed. Better reliability means a smaller MDC. It is often written MDC₉₅, meaning you can be 95% confident a change that size is not error.

The MCID is derived one of two ways, and the two disagree often enough that it is worth knowing which you are reading. Anchor-based methods ask somebody — the patient, the clinician, a family member — whether they felt better, and find the score change that separates yes from no. Distribution-based methods derive a threshold from the statistics of the sample instead, with no one asked whether anything improved.

MDC and MCID values on measures we have read a paper on

Every row below comes from a paper indexed on this site, and every value is the one that paper reports. This is deliberately not a general reference table: a table of thresholds with no source behind them is the thing this article is telling you not to trust.

Why do two studies give different MCIDs for the same measure?

Usually because they asked different people, or asked in a different way. The Extended Barthel paper is the clearest demonstration of this we have, because it did both at once.

The same paper also shows what happens when the second method is applied without judgement. Its ROC analysis produced area-under-curve values above 0.99, which looks like a near-perfect threshold. But specificity was 0.0% for the therapist raters and under 5% for the others — meaning almost every patient a rater judged “not improved” still scored above the cutoff the curve derived. The statistic was excellent and the threshold was useless. The authors recommend the anchor-based values for clinical use, and that is the right call.

What if the MCID is smaller than the MDC?

Then the measure cannot detect the smallest change that would matter, and you should stop using it to track that patient.

This is the trap the two numbers exist to catch, and it is not rare. It means the instrument’s own noise is larger than the improvement you are hoping to see, so any change small enough to be plausible is also small enough to be measurement error. The honest response is to change the measure, lengthen the interval between measurements, or accept that you are tracking something you cannot detect.

The shoulder dynamometry paper is a version of this. Internal rotation measured reliably; external rotation did not, in any position tested. A clinician recording both and comparing them week to week would be reading one real number and one random one, with nothing on the page to say which was which.

Does reaching the MCID mean the patient is satisfied?

No, and this is the most useful thing on this page.

That gap has a name. The Patient Acceptable Symptom State (PASS) is the score at which a patient says their current condition is acceptable — not that they have improved, but that they would be content to stay as they are. In that study it needed a 15.2-point change, nearly double the MCID, or an absolute score of 62 regardless of where they started.

MCID is a threshold for change. PASS is a threshold for state. A patient can clear the first every week and never reach the second, which is exactly the conversation that goes wrong when a discharge decision rests on “you’ve made meaningful progress”.

How should you use these numbers?

• Check the MDC first. If the change is smaller than it, nothing has been shown, and no amount of clinical reasoning about the number is going to fix that.
• Check which population the value came from. An MCID derived in acute stroke inpatients does not transfer to community-dwelling adults, and one derived in adolescent athletes does not transfer to anybody’s grandmother.
• Check who was asked. An anchor-based value carries a person inside it. If the paper does not say who, treat the number as provisional.
• Do not read MCID as a discharge criterion. It is the smallest change worth noticing, not the point at which someone is well.

Each paper on this site carries a certainty rating that tells you how much weight it can hold on its own; the guide on what the certainty rating means explains where it comes from. Every psychometric study above rates low — which is normal for this kind of work, and exactly why the population line in the table matters more than the number beside it.

Applied Evidence summarises peer-reviewed therapy research from the full paper. Browse the index.