MCID vs MDC: When a Clinically Important Change Is Still Noise

Two different questions wearing one number

Ask a researcher what the MCID means and the answer is usually some version of: the smallest change a patient would notice and care about. That is exactly right, and it is derived by anchoring a score change against a patient's own global rating of improvement, which is why a properly stated MCID is described as anchor-based. Jaeschke, Singer, and Guyatt formalised the concept in 1989, and every MCID published since answers one specific question: would this amount of change matter to the person who experienced it.¹

The MDC answers a different question entirely, and it is not a question about the patient at all. It asks whether an observed change is bigger than the noise the instrument generates every time it is administered, even to someone whose true condition has not changed. Two assessments of the same stable patient a week apart will not produce identical scores; some of that wobble is unavoidable measurement error, and the MDC is the threshold a change has to clear before it can be distinguished from that wobble with reasonable confidence. Beckerman and colleagues formalised this as the "smallest real difference" in 2001, and de Vet and colleagues later made the distinction from MCID explicit, precisely because researchers kept treating a real change and an important change as the same test.²^,³ The same authors returned to the point in 2010, arguing directly against using a distribution-based MDC as a stand-in for MCID: the two answer different questions, and treating "undetectable" as "unimportant" is a category error, not a rounding choice.⁶

Where the MDC number actually comes from

The calculation starts with test-retest reliability: administer the instrument twice to patients who have not clinically changed between assessments, and use the correlation between the two scores, usually an intraclass correlation coefficient, to estimate how consistent the instrument is. That reliability figure converts into a standard error of measurement:

SEM = SD × √(1 − ICC)

where SD is the spread of scores in the sample. The MDC at 95% confidence is then:

MDC₉₅ = 1.96 × SEM × √2

The extra √2 accounts for the fact that a change score involves two measurements, each carrying its own error. None of this touches a patient's opinion of whether the change mattered. It is a property of the instrument and the population it was tested in, derived from a different dataset and a different logic than the anchor-based MCID sitting next to it in the same methods section.

When the two numbers disagree

Because the two figures are calculated independently, nothing guarantees they land in a sensible order. A 2025 clinimetric analysis of 128 patients treated conservatively for lumbar spinal stenosis reported reliability and measurement error for three common low back pain outcome measures at three months, alongside whether each instrument's MCID actually cleared its own measurement error in patients who reported meaningful improvement.⁴

Table 1. Reliability, measurement error, and whether MCID cleared MDC in 128 patients with lumbar spinal stenosis.
Outcome measure	Test-retest reliability (ICC)	MDC₉₅	MCID exceeded measurement error?
Numeric Pain Rating Scale	0.55 (moderate)	2.3 points	Yes, in "much improved" patients
Oswestry Disability Index	0.86 (good)	8.5 points	Yes
Roland-Morris Disability Questionnaire	0.85 (good)	6.1 points	No, in either improvement category tested

The ODI and RMDQ had almost identical reliability in this cohort, 0.86 and 0.85, yet only the ODI's MCID survived contact with its own measurement error. A 2020 study of the 12-item WHODAS 2.0 among 1,988 patients with chronic musculoskeletal pain found the same pattern in sharper relief: the anchor-based MCID ranged from 3.1 to 4.7 points, while the distribution-based MDC was 8.6 points, almost double the upper end of the "important" threshold.⁵ Because the MDC exceeded the MCID by such a wide margin, the study's authors used the larger, measurement-error-based figure, rounded to nine points, as the practical cutoff for calling a change clinically significant, rather than the smaller anchor-based number that could not be trusted to represent a real change in the first place.

Both MCID and MDC are specific to the instrument, the population, and often the follow-up interval in which they were derived. A RMDQ figure from one back pain cohort does not transfer to a different population, and that applies equally to the reliability-based MDC as to the anchor-based MCID. Check that both numbers you are citing were derived in a population reasonably similar to yours before comparing one against the other.

Why this matters most in single-arm surgical outcome papers

Surgical literature runs on a specific sentence: "X% of patients achieved the MCID." It is almost always calculated from a single-arm, pre-post cohort with no comparator group, which means there is no P value protecting the claim the way there would be in a two-group trial. Each patient is simply counted as a responder if their score change clears the published MCID. If that MCID sits inside the instrument's own MDC for a comparable population, a share of the patients counted as responders may not have experienced a change distinguishable from the instrument's test-retest noise at all. The responder rate is then partly built from coin flips that happened to land on the favourable side.

What to put in your own methods and results section

Cite the MDC alongside the MCID whenever both have been established for your outcome instrument in a comparable population and follow-up window; a single sentence naming both numbers, and which one your observed change clears, closes this gap before a reviewer opens it. If only one of the two has been published for your instrument, say so explicitly rather than treating the number you have as sufficient on its own. And if you are reporting a percentage of patients who achieved the MCID, note whether that MCID is known to exceed the instrument's own measurement error in a population like yours; if it does not, or if that has never been checked, temper the claim accordingly rather than presenting the responder rate as settled.

Try it yourself

Get the reliability figure your MDC depends on

The MDC is not something you look up in isolation: it is built on the test-retest reliability of your instrument in your population. If your dataset holds two administrations on the same patients, both tools below give you the ICC, and the two formulas above turn it into an SEM and an MDC₉₅ you can compare against the published MCID.

In SPSS

Open the file with one row per patient and one column per administration (baseline score, repeat score).
Analyze → Scale → Reliability Analysis. Move both score columns into Items.
Click Statistics, tick Intraclass correlation coefficient, set Model to Two-Way Random and Type to Absolute Agreement, then click Continue and OK.
Read the single-measures ICC from the output. Get the SD from Analyze → Descriptive Statistics → Descriptives, then compute SEM and MDC₉₅ by hand: SPSS does not produce either one for you.

In StatsPlease

Upload the CSV.
Open the Analysis tab.
Set the two repeated administrations as the paired Outcome variables: the baseline score and the repeat score on the same patients.
Press the Before vs After preset button. There is no separate run step; pressing the preset is what runs the analysis, and the engine screens the paired data and routes to the right test itself.
Read the Results tab for the reliability and the per-administration means and SDs, written out in AMA by default (APA if you switch the reporting style), then feed the ICC and SD into the SEM and MDC₉₅ formulas above.

Compare: both paths return the identical ICC on identical data, because both are running the same computation rather than generating a plausible-looking number. Ask ChatGPT for the same figure and you get prose that sounds right and cannot be re-derived; the point of a reliability estimate is that you can re-run it in R or SPSS and get the same value back. The numbers here are computed, not generated.

Frequently asked questions

Can a change be clinically important but still not count as a real, measurable change?

Yes. A change can exceed the published minimal clinically important difference (MCID) while still falling inside the instrument's own measurement error, the minimal detectable change (MDC), meaning it can't be reliably distinguished from noise on repeated administration.

What is the difference between MCID and MDC?

MCID measures whether a change is large enough to matter to a patient, based on their own perception of benefit. MDC measures whether that same change is larger than the instrument's test-retest measurement error. They are derived from completely different data and don't have to agree.

Why does the MCID-versus-MDC relationship matter most in single-arm surgical papers?

Because responder-rate claims ("X% of patients achieved the MCID") based on a single-arm, pre-post design can't be trusted as reflecting real change if the MCID for that instrument sits inside its own MDC, since some of those "responders" may only be reflecting measurement noise.

You might also read

References

Jaeschke R, Singer J, Guyatt GH. Measurement of health status. Ascertaining the minimal clinically important difference. Controlled Clinical Trials. 1989;10(4):407–415. https://doi.org/10.1016/0197-2456(89)90005-6
Beckerman H, Roebroeck ME, Lankhorst GJ, Becher JG, Bezemer PD, Verbeek ALM. Smallest real difference, a link between reproducibility and responsiveness. Quality of Life Research. 2001;10(7):571–578. https://doi.org/10.1023/A:1013138911638
de Vet HC, Terwee CB, Ostelo RW, Beckerman H, Knol DL, Bouter LM. Minimal changes in health status questionnaires: distinction between minimally detectable change and minimally important change. Health and Quality of Life Outcomes. 2006;4:54. https://doi.org/10.1186/1477-7525-4-54
Young I, Dunning J, Escaloni J, Maselli F, Prall J, Mourad F, Hutting N, Fernández-de-las-Peñas C. Clinimetric analysis of the numeric pain rating scale, Oswestry disability index, and the Roland-Morris disability questionnaire in patients with lumbar spinal stenosis treated with conservative interventions. Journal of Clinical Neuroscience. 2025;143:111717. https://doi.org/10.1016/j.jocn.2025.111717
Katajapuu N, Heinonen A, Saltychev M. Minimal clinically important difference and minimal detectable change of the World Health Organization Disability Assessment Schedule 2.0 (WHODAS 2.0) amongst patients with chronic musculoskeletal pain. Clinical Rehabilitation. 2020;34(12):1506–1511. https://doi.org/10.1177/0269215520942573
de Vet HC, Terwee CB. The minimal detectable change should not replace the minimal important difference. Journal of Clinical Epidemiology. 2010;63(7):804–805. https://doi.org/10.1016/j.jclinepi.2009.12.015

Know whether the change is real before you call it important.

If your dataset includes repeated measurements on the same patients, StatsPlease computes the ICC and reports it in AMA format, the same reliability figure the SEM and MDC calculation both depend on, so you can see how much of an observed change is signal before you report it against a published MCID. Set the two administrations as your paired Outcome variables and press the Before vs After preset; pressing the preset is what runs the analysis, and every figure it returns is computed from your data, never generated.

Try StatsPlease free

MCID vs MDC: A Change Can Clear the Threshold and Still Be Measurement Noise