Visual Analogue Scale (VAS): Construction, Scoring and Minimal Important Change

A visual analogue scale (VAS) is a response format, not a questionnaire: a straight line of fixed length with a verbal anchor at each end and nothing in between. The respondent marks a point on the line; the score is the measured distance from the left anchor to that mark, conventionally in millimetres. That single mechanical fact — the score is a physical measurement of a drawn line — determines almost everything else that goes wrong with VAS in practice, from photocopied forms that quietly rescale the instrument to published minimal important change values borrowed across populations they were never derived in.

This page covers the construction decisions that make a VAS a valid instrument, the two conflations that most often invalidate a VAS analysis (VAS treated as a numeric rating scale, and MCID treated as a universal number), and the actual published minimal important change values with the conditions attached to each one. It assumes you are choosing, building or interpreting a scale for a real study, not looking for a definition.

What makes something a VAS

Four construction properties define the format. Break any of them and you have a different instrument with different measurement properties, and published VAS evidence no longer transfers to it.

1. A continuous line of stated length. Almost always 100 mm horizontal, printed or rendered to that exact length.
2. Exactly two verbal anchors, one at each end, describing the extremes of the construct.
3. No intermediate marks, tick marks, numbers or labels along the line. Kelly’s emergency-department validation study specifies a “100 mm, non-hatched VAS” for precisely this reason: intermediate marks create attraction points, and the response distribution clusters on them, which converts a continuous scale into a coarse categorical one without saying so.
4. Scoring by measured distance, not by the respondent selecting a value. The respondent never sees or reports a number.

Property 4 is the one that separates VAS from every other rating format, and it is also the one that makes VAS expensive: someone has to measure every line. That cost is the main practical reason numeric rating scales displaced VAS in routine clinical use, and the main reason electronic administration matters.

What a VAS is not

A VAS is not a Likert scale. A Likert item offers a small set of ordered, verbally labelled categories and is scored by which category was chosen; the argument about whether its integers may be averaged is genuinely unresolved and is covered on that page. A VAS produces a distance measurement on a continuous line and does not have that particular problem in the same form. It is also not a graphic rating scale with labelled intermediate points, and not a verbal rating scale (a short ordered list of words such as none / mild / moderate / severe).

The 100 mm convention, and why the length is not arbitrary

The 100 mm line is conventional rather than mandated, but the convention has an empirical basis. Seymour and colleagues (Pain 1985;21:177–185) administered VAS lines of 5, 10, 15 and 20 cm to 100 dental-pain patients. Scores across the lengths correlated highly, but the 10 cm and 15 cm scales had the smallest measurement error. Very short lines compress the available resolution; very long lines add error without adding information. 100 mm sits at the empirically supported end and has the arithmetic convenience that a millimetre reading is directly a 0–100 score.

Three consequences follow directly from “the score is a measured distance,” and each one is a real, recurring source of invalid data:

• Reproduction that rescales the line rescales the instrument. A form photocopied with any fit-to-page scaling, or a PDF printed with “shrink to fit” enabled, no longer has a 100 mm line. Every score collected on it is proportionally wrong, silently and consistently. Print a ruler check onto the master form and verify the length on the actual paper stock in use before a study starts.
• Digital rendering has the same problem in a different form. A line specified in CSS pixels or as a percentage of viewport width is a different physical length on a phone than on a tablet. Whether that matters is an empirical question about the specific implementation, not something to assume — see the digital-versus-paper section below.
• The measurement itself is a source of error. Reading marks by hand with a ruler introduces observer error on top of respondent variability. If two people score the same forms, that is an inter-rater reliability question and should be quantified, not assumed away.

Anchor wording is part of the instrument, not decoration

Because a VAS has only two labels, those two labels carry the entire definition of the construct being measured. Changing them changes what the scale measures, and therefore whether any published evidence about “the VAS” applies.

Seymour and colleagues also tested five different upper end-phrases (troublesome, miserable, intense, unbearable, worst pain imaginable) and concluded that “worst pain imaginable” was the best choice for comparing pain between groups. That phrase became the de facto standard for pain VAS, and it is what most published pain-VAS reference values assume.

The variability in practice is larger than most researchers realise. The European Palliative Care Research Collaborative’s systematic review (Hjermstad et al., Journal of Pain and Symptom Management 2011;41:1073–1093) found 24 different descriptors used to anchor the extremes across the 54 included studies of unidimensional pain scales. That is not a cosmetic difference; it is 24 slightly different constructs sharing a name.

Practical rule: report both anchors verbatim in your methods section, along with the line length and orientation. A VAS described only as “a 0–100 VAS” is not a reproducible instrument, and a reader cannot judge whether a published minimal important change applies to it.

A VAS is not an NRS, and the scores are not interchangeable

This is the single most common error in applied use, and it is usually invisible because the two formats correlate so highly that nothing looks wrong.

A numeric rating scale (NRS) asks the respondent to state a whole number, typically 0–10, with the same kind of verbal anchors at the ends. It can be administered verbally, needs no measuring, and produces 11 discrete values. A VAS produces a continuous distance measurement, typically 0–100 mm. The temptation is to treat one as the other multiplied by ten. That transformation is arithmetically trivial and methodologically wrong.

What the comparison evidence actually shows

Hjermstad and colleagues reviewed 54 studies comparing NRS, VRS and VAS. Their findings, in the direction that matters here:

• NRS had better compliance in 15 of the 19 studies reporting compliance, and was the recommended tool in 11 studies on grounds of higher completion rates, responsiveness and ease of use. Twenty-nine studies expressed no preference.
• Overall, NRS and VAS scores corresponded — but with exceptions of systematically higher VAS scores. A systematic offset is exactly what a naive ×10 conversion cannot absorb.
• Eight different versions of the NRS were in use (NRS-6 through NRS-101). “NRS” is not one instrument either.

Correlation is not agreement — and the gap is the size of the MCID

Bahreini and colleagues (Journal of Emergency Medicine 2015;48:10–18) measured 150 adult emergency-department patients on VAS, NRS and a colour analogue scale at three time points. Spearman correlation between NRS and VAS was 0.94. On that number alone the two look substitutable, and the authors concluded the scales could be applied interchangeably for acute pain measurement at the group level.

Their Bland–Altman analysis, however, gave 95% limits of agreement between paired NRS and VAS of −2.0 to +2.6 on a 0–10 scale. Read that against the clinically important difference for an 11-point pain NRS — approximately 2 points, or about 30% (Farrar et al., Pain 2001;94:149–158). The disagreement between the two instruments spans the entire clinically important difference.

Both statements are true and they are not in conflict: agreement adequate for comparing group means is not agreement adequate for substituting one instrument for the other in an individual patient’s record, or for pooling scores collected on different formats into one change-score analysis. This is the same distinction as reliability-for-groups versus reliability-for-individuals, and it is a correlation-versus-agreement problem — see correlation coefficient for why a high correlation is compatible with a systematic offset.

The operational rules

1. Pick one format and use it for the whole study. Switching mid-study, or letting sites choose, introduces a between-instrument difference that will be indistinguishable from a treatment effect.
2. Do not convert VAS to NRS or back by rescaling. If you must combine data collected on both, model the instrument as a covariate and say so, rather than transforming.
3. Do not apply an NRS-derived MCID to VAS data, or vice versa, without saying explicitly that you are doing so and why. Farrar’s 2-point / 30% benchmark was derived on an 11-point NRS in chronic pain, from 2,724 subjects across 10 pregabalin trials. It is not a 20 mm VAS threshold just because 2 × 10 = 20.
4. Report which one you used, with its anchors, every time.

Digital versus paper VAS

Electronic administration removes the manual measurement step, eliminates out-of-range and illegible responses, timestamps entries, and is the reason VAS is viable again at scale. The equivalence question is whether an electronic VAS (e-VAS) measures the same thing as the paper version.

The broadest evidence is not VAS-specific: Gwaltney, Shields and Shiffman’s meta-analysis (Value in Health 2008;11:322–333) synthesised 65 studies directly comparing computer and paper administration of patient-reported outcome measures, with 46 unique studies covering 278 scales in the quantitative analysis. Across 233 direct comparisons, the average mean difference between modes was 0.2% of the scale range (about 0.02 points on a 10-point scale), and 93% of comparisons fell within ±5% of the scale range. Across 207 cross-mode correlations, the average weighted correlation was 0.90, with 94% at or above 0.75. Critically, in the four comparisons where both were available, the cross-mode paper-to-computer correlation (0.88) was essentially identical to the within-mode paper-to-paper test–retest correlation (0.91) — meaning the apparent mode difference was no larger than the instrument’s own retest noise.

VAS-specific studies point the same way. Maarj and colleagues (JMIR Pediatrics and Parenting 2022;5:e41930) compared an e-VAS mobile application against paper VAS in 43 children and adolescents with hypermobility spectrum disorder, and reported an ICC of 0.87 (95% CI 0.78–0.93) with a Bland–Altman mean difference of 0.19 (SD 0.95) and limits of agreement of −1.67 to 2.04.

What that evidence does and does not license. It supports the general claim that a well-implemented electronic PRO is equivalent to its paper form. It does not certify your implementation. Before treating an e-VAS as equivalent to paper, check:

• Does the rendered line have a defined, device-independent length, or does it stretch with the viewport? A scale whose physical length varies by device is a different instrument on every device.
• Does the interface display a number as the respondent drags the slider? If it does, you have built an NRS with extra steps — the respondent is now selecting a value, not marking a position, and the defining property of a VAS is gone.
• Is there a default starting position? A slider pre-positioned at the midpoint anchors responses toward it. A line with no mark until the respondent makes one does not.
• What is the touch resolution? A 100 mm line captured to integer millimetres has 101 possible values; a fat-finger touch target may reliably resolve far fewer.
• Is the scale of the recorded value the same? Storing to one decimal place when paper was read to the nearest millimetre changes the granularity of the change score.

Scoring, and the analysis question

Score by measuring from the left-hand (zero) anchor to the respondent’s mark, to the nearest millimetre, giving a 0–100 value. Fix the following in the protocol before data collection, not after:

• Marks outside the line: pre-specify whether these are scored at the nearest endpoint or treated as missing. Do not decide once you have seen the data.
• Multiple marks or a range: pre-specify the rule (commonly the midpoint of the range, or missing).
• Vertical versus horizontal: both are used; they are not automatically equivalent and should not be mixed within a study.
• Who measures, and blinding: the person measuring the lines should be blind to allocation and to the prior score for that respondent.

On levels of measurement: a VAS yields a continuous variable and is conventionally analysed as interval-level, which is a stronger assumption than it is usually stated to be — it assumes a 10 mm difference means the same amount of change at 15 mm as at 85 mm. In practice the distribution is very often floor- or ceiling-constrained (a post-treatment pain VAS piles up near zero), so check the distribution and consider a rank-based test rather than assuming normality by default. See descriptive statistics for how to report the shape honestly, and the Wilcoxon signed-rank test for the paired non-parametric alternative on change scores.

The minimal important change: a conditional range, not a number

Ask what “the MCID of the VAS” is and you will be given a single figure, usually somewhere between 10 mm and 20 mm. There is no such number. The minimal clinically important difference is a property of an instrument in a population, at a baseline severity, derived by a stated anchor method — not a property of the scale. The concept originates with Jaeschke, Singer and Guyatt (Controlled Clinical Trials 1989;10:407–415) as the smallest change patients perceive as beneficial; the definition is inherently anchored to a patient judgement, and different judgements give different numbers.

Here are the values that are actually in the literature, each with what it was derived from. Read the conditions column, not the number column.

Value	Population and setting	Derivation	Source
13 mm (95% CI 10–17)	Acute pain from trauma, urban ED, 48 patients / 80 usable contrasts	Mean change on 100 mm VAS when the patient reported “a little less” or “a little more” pain at 20-minute intervals	Todd et al., Ann Emerg Med 1996;27:485–489
12 mm (95% CI 9–15)	Adult ED patients, 156 enrolled / 88 usable comparisons	Same anchor design; also reported by severity band: mild 11 mm (4–18), moderate 14 mm (10–18), severe 10 mm (6–14)	Kelly, Emerg Med J 2001;18:205–207
16 mm (95% CI 13–18)	Acute abdominal pain, two urban EDs	Same anchor design, 30-minute intervals; VAS retest ICC 0.99 at 1 minute	Gallagher et al., Am J Emerg Med 2002;20:287–290
−19.9 mm (−40.8%) knee OA −15.3 mm (−32.0%) hip OA	Knee and hip osteoarthritis outpatients, 4-week cohort, MCII estimated in 814 patients	Patient-opinion anchor (5-point response-to-treatment rating); reported as minimal clinically important improvement	Tubach et al., Ann Rheum Dis 2005;64:29–33
Median 23 mm (IQR 12–39) or 34% relative (IQR 22–45)	Chronic pain, systematic review of 66 studies, 31,254 patients	Mean-change approach; threshold approach gave median 20 mm (IQR 15–30) and 32% (IQR 15–41). Heterogeneity I² = 99% (absolute), 96% (relative)	Olsen et al., J Clin Epidemiol 2018;101:87–106

The spread is not measurement noise. It is the answer: values derived in acute emergency-department pain cluster around 12–16 mm, values derived in chronic pain and osteoarthritis cluster around 15–24 mm and 30–40% relative change, and the chronic-pain literature is heterogeneous to the point where a pooled point estimate is not meaningful (I² = 99%).

The baseline-severity finding, and its apparent contradiction

Two well-conducted studies appear to disagree, and the disagreement is informative rather than a reason to distrust either.

Kelly (2001) pre-defined mild (≤30 mm), moderate (31–69 mm) and severe (≥70 mm) bands and found no statistical difference in the minimum clinically significant difference between them — 11, 14 and 10 mm respectively. Olsen and colleagues (2018), across 66 chronic-pain studies, found the opposite: absolute MCID was strongly associated with baseline pain, which explained approximately two-thirds of the variation between studies, with the operational definition of “minimum pain relief” and the clinical condition contributing further. Farrar (2001) reported the same pattern on the NRS: the percentage change corresponding to a clinically important difference was consistent regardless of baseline, while higher baseline scores required larger raw changes.

The reconciliation: Kelly’s study was a single acute ED population over a two-hour window, where the range of baseline severity and the range of achievable change are both narrow. Olsen’s finding is a between-study association across chronic conditions with far wider variation in baseline pain, and a between-study association is not the same estimand as a within-population subgroup comparison. The practical consequence is the same either way, and it is the one worth acting on: at high baseline severity, prefer a relative (percentage) criterion; a fixed millimetre threshold is easier to clear from 90 mm than from 30 mm.

Choosing an MCID for your study

Pre-specify it in the protocol or statistical analysis plan, before unblinding, and justify it against these five questions:

1. Same condition and care setting? An acute-trauma ED value does not transfer to chronic low back pain, and vice versa. This is the largest single source of error.
2. Same baseline severity? If your cohort starts substantially higher or lower than the source population, use a relative criterion or a value derived at a comparable baseline.
3. Same derivation method? Mean-change and threshold (ROC-optimised) approaches give different numbers from the same data — in Olsen’s review, medians of 23 mm and 20 mm respectively. Anchor-based values, which incorporate a patient judgement of what is important, are preferred over distribution-based values, which do not.
4. Improvement or deterioration? They are not symmetric. Tubach’s values are explicitly for improvement (MCII).
5. Absolute or relative? State which, and do not mix them across endpoints in the same report.

Then report the value, its citation, and its derivation population in your methods — not just the number. A responder analysis defined against an unattributed threshold is not interpretable, and it is the kind of thing that gets flagged in review.

MCID and MDC are different questions — check both

The minimal important change tells you how much change matters. It says nothing about how much change your instrument can actually detect above its own measurement error. That is the minimal detectable change, and the two are routinely conflated.

The decision rule, the SEM-based calculation, the √2 in the formula and the four-cell interpretation are covered in full on minimal detectable change (MDC) — read it alongside this page rather than treating the MCID values above as sufficient. The short version of why it matters here: if your VAS implementation’s MDC exceeds the MCID you have chosen, an individual patient’s change score cannot distinguish a clinically important improvement from measurement noise, and no threshold you pick will fix that. Note also that the measurement error parameter, not the reliability coefficient, is the more transferable quantity between populations — a point established in de Vet et al. (Health Qual Life Outcomes 2006;4:54) and set out on that page.

A closely related distinction, at the level of the trial rather than the instrument, is covered in statistical significance vs. clinical significance: a 4 mm mean difference on a 100 mm VAS can be highly statistically significant in a large trial and still fall well below every MCID in the table above.

Reporting checklist

A VAS is fully specified in a methods section only when all of these are present:

• Line length in millimetres and orientation (horizontal or vertical).
• Both anchor phrases, verbatim.
• Whether the line carried any intermediate marks (it should not).
• Administration mode (paper or electronic), and for electronic, the device class and whether a numeric value was displayed to the respondent.
• Recall period — “pain right now”, “worst pain in the last 24 hours” and “average pain this week” are three different measurements.
• Scoring resolution and the rules for out-of-range or ambiguous marks.
• Who measured the marks, and whether they were blinded.
• The MCID or responder threshold used, with its citation and derivation population.
• Whether the threshold was pre-specified, and where.

Common errors

• Quoting “the” MCID. Any single figure presented without its population, baseline severity and derivation method is not a usable threshold.
• Converting between VAS and NRS by multiplying by ten. High correlation is not agreement; the limits of agreement span the clinically important difference.
• Adding tick marks or numbers to “help” respondents. This changes the instrument and voids the published evidence base for it.
• Photocopying or PDF-printing the form with scaling on. Silent, systematic, and undetectable after the fact.
• Treating a slider with a live numeric readout as a VAS. It is an NRS with more resolution.
• Using a distribution-based value (0.5 SD, 1 SEM) and calling it an MCID. Those estimate detectability, not importance.
• Applying an improvement threshold symmetrically to deterioration. They are separately estimated quantities.
• Comparing VAS means across studies with different anchors. With 24 anchor descriptors in circulation, the scales are not the same measurement.

Frequently asked questions

What is a visual analogue scale?

A response format consisting of a fixed-length line — conventionally 100 mm — with a verbal anchor at each end and no marks in between. The respondent places a single mark on the line, and the score is the measured distance in millimetres from the left anchor.

Is a VAS the same as a numeric rating scale?

No. An NRS asks the respondent to state a whole number (usually 0–10); a VAS records a mark on a continuous line and is scored by measurement. They correlate very highly — around 0.94 in emergency-department comparisons — but 95% limits of agreement of roughly −2.0 to +2.6 on a 0–10 scale mean the disagreement is as large as the clinically important difference. Do not substitute one for the other in an individual record, and do not convert between them by rescaling.

How long should a VAS line be?

100 mm is the convention. The empirical basis is Seymour et al. (1985), who found the smallest measurement error at 10 cm and 15 cm among lines of 5, 10, 15 and 20 cm. Whatever length you use, state it, and verify it survives printing or screen rendering unchanged.

What is the MCID for a VAS?

There is no single value. Published anchor-based estimates range from about 12–16 mm in acute emergency-department pain (Todd 1996; Kelly 2001; Gallagher 2002) to about 15–20 mm in hip and knee osteoarthritis (Tubach 2005) and a median of 23 mm, IQR 12–39, across 66 chronic-pain studies (Olsen 2018). Heterogeneity in the chronic-pain literature reaches I² = 99%, so a pooled point estimate is not meaningful. Choose a value derived in a comparable condition, setting and baseline severity, and cite it.

Does the MCID depend on how severe the pain was to begin with?

Across studies, yes: Olsen et al. found baseline pain explained roughly two-thirds of the between-study variation in absolute MCID. Within a single acute ED population, Kelly found no significant difference across mild, moderate and severe bands. The practical rule that satisfies both findings is to use a relative (percentage) criterion when baseline severity varies widely.

Can I use an electronic VAS instead of paper?

Generally yes, but verify your implementation rather than relying on the general literature. Across 46 studies and 278 scales, electronic and paper PRO administration differed by an average of 0.2% of the scale range with a mean cross-mode correlation of 0.90 (Gwaltney et al., 2008), and a VAS-specific paediatric comparison reported ICC 0.87 (Maarj et al., 2022). Check that your rendered line has a device-independent length, shows no numeric readout, has no default slider position, and records at the same resolution as the paper version.

Is VAS data interval or ordinal?

A VAS yields a continuous variable and is conventionally analysed as interval-level, but that assumes a 10 mm difference means the same amount of change everywhere on the line. VAS distributions are frequently floor- or ceiling-constrained, so check the distribution before defaulting to parametric methods.

How is a VAS different from a Likert scale?

A Likert item presents a small number of labelled ordered categories and is scored by which one is chosen; a VAS presents an unmarked continuous line and is scored by measured position. The long-running argument about averaging Likert responses arises from that categorical structure and does not apply in the same form to VAS — see the Likert scale guide for that debate.

What is the difference between MCID, MIC and MCII?

They are largely the same idea under different names. MCID (minimal clinically important difference) is the original term from Jaeschke et al. (1989); COSMIN prefers MIC (minimal important change); MCII (minimal clinically important improvement) is used in the osteoarthritis literature specifically to flag that the value applies to improvement, not deterioration. What matters more than the label is the derivation method and the population.

Related CASRAI guides

• Minimal detectable change (MDC) — the SEM-based calculation and the MDC-vs-MCID decision rule.
• Psychometrics — how unobservable constructs are measured at all.
• Reliability in research and criterion validity — the measurement properties a new scale must demonstrate.
• Clinical outcome assessment validation and the clinical outcome assessment (COA) term — the regulatory framing when a VAS is a trial endpoint.
• Questionnaire design and survey question types — choosing a response format in the first place.
• Response bias — the rating-scale biases a VAS does and does not avoid.
• Research methods pillar — the full cluster.

Sources

• Seymour RA, Simpson JM, Charlton JE, Phillips ME. An evaluation of length and end-phrase of visual analogue scales in dental pain. Pain 1985;21(2):177–185. DOI 10.1016/0304-3959(85)90287-8. PMID 3982841.
• Todd KH, Funk KG, Funk JP, Bonacci R. Clinical significance of reported changes in pain severity. Annals of Emergency Medicine 1996;27(4):485–489. DOI 10.1016/s0196-0644(96)70238-x. PMID 8604867.
• Kelly AM. The minimum clinically significant difference in visual analogue scale pain score does not differ with severity of pain. Emergency Medicine Journal 2001;18(3):205–207. DOI 10.1136/emj.18.3.205. PMID 11354213; PMC1725574.
• Farrar JT, Young JP Jr, LaMoreaux L, Werth JL, Poole MR. Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain 2001;94(2):149–158. DOI 10.1016/S0304-3959(01)00349-9. PMID 11690728.
• Gallagher EJ, Bijur PE, Latimer C, Silver W. Reliability and validity of a visual analog scale for acute abdominal pain in the ED. American Journal of Emergency Medicine 2002;20(4):287–290. DOI 10.1053/ajem.2002.33778. PMID 12098173.
• Tubach F, Ravaud P, Baron G, et al. Evaluation of clinically relevant changes in patient reported outcomes in knee and hip osteoarthritis: the minimal clinically important improvement. Annals of the Rheumatic Diseases 2005;64(1):29–33. DOI 10.1136/ard.2004.022905. PMID 15208174.
• Gwaltney CJ, Shields AL, Shiffman S. Equivalence of electronic and paper-and-pencil administration of patient-reported outcome measures: a meta-analytic review. Value in Health 2008;11(2):322–333. DOI 10.1111/j.1524-4733.2007.00231.x. PMID 18380645.
• Dworkin RH, Turk DC, Wyrwich KW, et al. Interpreting the clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT recommendations. The Journal of Pain 2008;9(2):105–121. DOI 10.1016/j.jpain.2007.09.005. PMID 18055266.
• Hjermstad MJ, Fayers PM, Haugen DF, et al. Studies comparing Numerical Rating Scales, Verbal Rating Scales, and Visual Analogue Scales for assessment of pain intensity in adults: a systematic literature review. Journal of Pain and Symptom Management 2011;41(6):1073–1093. DOI 10.1016/j.jpainsymman.2010.08.016. PMID 21621130.
• Bahreini M, Jalili M, Moradi-Lakeh M. A comparison of three self-report pain scales in adults with acute pain. The Journal of Emergency Medicine 2015;48(1):10–18. DOI 10.1016/j.jemermed.2014.07.039. PMID 25271179.
• Olsen MF, Bjerre E, Hansen MD, Tendal B, Hilden J, Hróbjartsson A. Minimum clinically important differences in chronic pain vary considerably by baseline pain and methodological factors: systematic review of empirical studies. Journal of Clinical Epidemiology 2018;101:87–106. DOI 10.1016/j.jclinepi.2018.05.007. PMID 29793007.
• Maarj M, Pacey V, Tofts L, Clapham M, Gironés Garcia X, Coda A. Validation of an electronic visual analog scale app for pain evaluation in children and adolescents with symptomatic hypermobility: cross-sectional study. JMIR Pediatrics and Parenting 2022;5(4):e41930. DOI 10.2196/41930. PMID 36287606; PMC9647467.
• Jaeschke R, Singer J, Guyatt GH. Measurement of health status. Ascertaining the minimal clinically important difference. Controlled Clinical Trials 1989;10(4):407–415. DOI 10.1016/0197-2456(89)90005-6.
• de Vet HCW, Terwee CB, Ostelo RWJG, 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. DOI 10.1186/1477-7525-4-54. PMID 16925807; PMC1560110.