How Do We Measure Recovery — An Introduction to Rehabilitation Outcome Measures

The conclusion of a rehabilitation paper is usually summarized in a single sentence: "Significant improvement was observed in the treatment group." This single sentence contains three different questions. What got better, how much better does it have to be to count as a true improvement, and does the patient actually feel that improvement? The sentence covers all three up together without answering any of them.

This article establishes the minimum foundation needed to read those three questions apart. How the Central Nervous System Recovers — A Primer on Neuroplasticity and How Peripheral Nerves Recover — An Introduction to Nerve Regeneration covered how recovery is possible. The central nervous system rewires its surviving circuits, and the peripheral nervous system actually grows severed nerve fibers back. The immediately preceding piece, Why Hold Several Rulers Against One Person — An Introduction to the ICF Framework, established the layers by which to divide and view that recovery. This article deals with the next question: how do we know whether recovery has occurred? Even if you know the principles of recovery, if you do not know the language of measurement, you will get lost again in front of a paper's results table.

Why We Need to Know — The Word "Improved" Means Three Different Things

A patient picks up a cup again three months after a stroke. What has improved can be read in three different directions.

• The impairment may have decreased. Impairment is injury to the body itself; if the quality of the motion of reaching out an arm has returned, the impairment has decreased.
• The capacity may have increased. Capacity is what the person can do, meaning the level they achieve when asked to do something in a condition with no help and no interference, like a laboratory.
• The performance may have increased. Performance is what the person actually does, meaning the level at which they actually use that arm in the home where they live right now.

The three do not move together. It is common for a patient to pick up a cup in the laboratory but still not use that arm at home. This gap—where someone is able but does not use it—is covered separately in "Able But Unused — An Introduction to Learned Non-Use and CIMT".

This is why rehabilitation papers include multiple layers of indicators together. Outcome indicators are tests, scales, and questionnaires used to record in numbers what a treatment has changed. Depending on which layer of the indicator showed improvement, the very same "significant improvement" takes on a completely different meaning.

This is where the first rule comes in. Before reading an outcome table, check first which layer that indicator measures.

The Big Picture — Metrics Live on Different Levels

These three branches can be translated into the language of the ICF established in the previous part. The ICF is a framework introduced in the primer Rehabilitation — Starting to Plan a Livable Home as the Fire Is Brought Under Control, which records a person's functioning using the layers of body functions and structures, activities, and participation, alongside the surrounding context. Impairment is a problem at the layer of body functions and structures. Capacity and performance are two scales attached to activity and participation, which separately record whether someone can accomplish the same task in an examination room and whether they actually do it where they live.

Which of these layers rehabilitation assessment tools cover, and to what extent, has been examined across various fields. Even in heart failure cardiac rehabilitation tools and hip fracture study outcome metrics, no single tool covered all layers (PMID 37806100, PMID 37221661). This means a single tool cannot capture a person's functioning in full.

Roughly sorting the metrics commonly used in stroke rehabilitation by layer yields the following.

A few names that will appear repeatedly in this article are introduced first. The FMA is an exam that scores the movement of paralyzed limbs item by item. The ARAT and Box & Block Test are exams that have the person perform tasks of picking up and moving objects, while the two walking tests in the table measure walking speed and walking endurance. The MAL is a scale that asks patients how often and how well they use their paralyzed arm in daily life. The mRS is a scale that rates how well people manage their daily lives without help from others after a stroke across several stages.

Making Core Concepts Easy

① An Indicator with the Same Name Is Not the Same Indicator

FMA is a recommended primary outcome measure for motor function in stroke rehabilitation, meaning the measure decided in advance to be looked at first when a trial determines effectiveness. However, FMA manuals varied, and those differences were undermining the reliability of assessments. As a result, it has been difficult to compare results across trials or pool data (PMID 41773341).

How fragmented it was remains in the numbers. There were 22 versions of the FMA. This is the number found across 36 studies by a systematic literature review, which is a method that thoroughly finds, gathers, and examines relevant studies using established rules (PMID 40533328).

To solve this problem, an international standard manual was created. 14 experienced users from 13 research institutions across 6 countries in 3 continents went through repeated consensus processes, confirming each section of the manual with an agreement of 79% or higher (PMID 41773341).

The manual also recommends evaluator training. It states that evaluators must have a solid foundation in neurorehabilitation and clinical assessment, and that first-time evaluators must undergo structured training that includes supervised practical sessions with patients (PMID 41773341).

Therefore, simply saying "measured with the FMA" does not allow results to be compared. Which version was used, and by an evaluator with what kind of training, determines whether the results can be compared.

② Depending on Who Measures, the Value Changes — Reliability

Reliability is the degree to which the same value comes out when the same target is measured again or measured by a different person. When reliability is low, it is impossible to know whether a difference in scores is due to a difference in the patients or a difference in the measurers.

mRS has been adopted as the primary outcome in most recent acute stroke clinical trials. This scale was designed to be used together by clinical personnel from various professions, but the values assigned do not match evenly from person to person. That misalignment can lead to errors in clinical care and research (PMID 38546172).

There is a comparison examining how much this agreement changes when the judgment procedure is changed. In the group that used an automated flowchart (iRankin) guiding the judgment through step-by-step questions, the weighted kappa between nurses was 0.76, while in the group rated by the conventional method, it was 0.30. The weighted kappa is a value indicating how much the grades assigned by two evaluators agree beyond the share that would match by chance, with a larger value indicating better agreement. This was a comparison that randomly divided 59 personnel from 6 comprehensive stroke centers into two groups (PMID 38546172).

However, the confidence intervals of the two values overlap with each other. A confidence interval is written alongside the value produced by a study, showing the range of values that do not contradict this data ("How Much Does a Number Counted Once Wobble? — The Two Numbers in the Parentheses, the Confidence Interval"). Because the two ranges overlap, the magnitude of the difference is still wide open. The authors also note that additional research is needed to generalize this (PMID 38546172). Still, this comparison shows that agreement can change by this much simply by changing the judgment procedure without altering the scale itself.

Measured values are always mixed with error. The next concept comes from this error.

③ Change That Is Not by Chance — Minimum Detectable Change (MDC)

Even when the same patient is measured twice with the same tool, the values differ slightly. Standard error of measurement (SEM) is the range by which a value wobbles each time it is measured again, and minimum detectable change (MDC) is the smallest amount of change that can be said to have gone beyond that wobble. A change smaller than the MDC cannot be distinguished as whether it actually improved or is just a wobble produced by measuring again.

The MAL 4/5 for patients with severe upper extremity paralysis is a good example. This is an MAL version made separately for people with moderate-to-severe or severe arm paralysis (PMID 38265026). This version kept its values well when measured again. The degree to which values come out the same when the same person is measured again over time is called test-retest reliability and is written as a value called ICC; the larger it is, the more stable. The ICC was high at 0.84 for amount of use (AOU) and 0.90 for quality of use (QOU). The measurement error and MDC were 0.3 and 0.8 points for AOU, and 0.2 and 0.5 points for QOU (PMID 38265026).

How to read this is as follows. Let us say that the AOU of a patient measured with this version (MAL 4/5 for patients with moderate-to-severe upper extremity paralysis) went up by 0.5 points. That improvement falls short of the MDC of 0.8 points, so it cannot be distinguished from measurement error. If the version and target group change, the MDC to compare against also changes.

④ The Patient's Felt Change — The Minimal Clinically Important Difference (MCID)

A change that clears the margin of error and a change large enough for a patient to feel that their life has actually shifted are two different thresholds. The minimal clinically important difference (MCID) is the smallest amount of change that leaves a patient feeling meaningfully improved.

An MCID does not emerge from looking at scores alone; it requires an anchor. An anchor is an external benchmark that asks separately from the score, "Do you feel you've gotten better?" After dividing people into those who answered yes and those who did not using this benchmark, the score change that best separates the two groups is set as the MCID (PMID 37742304, PMID 39145516).

In people with chronic stroke and moderate-to-severe arm paralysis, the MCID for the MAL-14 was 0.89 points for amount of use (AOU) and 0.77 points for quality of movement (QOM). A subscale of the Stroke Impact Scale (SIS) was used as the anchor (PMID 37742304).

For the 6-minute walk distance (6MWD) in acute stroke, the value diverged depending on whose judgment was used for the anchor. This distance, measured by how far someone walks in a set time, looks at walking endurance. When the anchor was the change rated by the patients themselves, the MCID was 78.7 to 100.0 meters; when the anchor was the change rated by the therapists, it was 95.2 to 99.5 meters (PMID 39145516).

The reliability of the two anchors also differed. The patient anchor successfully sorted out those who had improved (discriminative power AUC > 0.8). Discriminative power (AUC) is a value indicating how well a criterion separates those who have improved from those who have not, with higher numbers indicating better separation. The therapist anchor met only 2 out of 5 core criteria for anchor reliability, thus failing to pass the reliability threshold. The authors noted that therapists may have failed to accurately capture patients' fatigue or subjective symptoms, and proposed the patient-anchored threshold of 78.7 m as the final MCID (PMID 39145516).

The MCID varies depending on the target population (acute/chronic), severity, and which anchor was used. Therefore, when a paper cites an MCID, one must look together at which group that value came from.

⑤ Ceiling and Floor — The Zones Indicators Cannot See

Every scale has a point beyond which it cannot go higher (the ceiling) and a point below which it cannot go lower (the floor). The ceiling effect is a phenomenon where many people bunch up near the highest point of a scale, so that even if they get better, their score does not go up. The floor effect is the opposite phenomenon, where people bunch up near the lowest point and changes do not show in the score.

Though it is an example from orthopedics, the structure is the same. Significant ceiling effects were reported in questionnaire indexes answered by patients after knee and hip replacement surgery, which limited the ability to capture changes over time (PMID 40170702, PMID 41840038). This means there may be patients who got better than what appears in the score (PMID 41840038).

The reason this is especially problematic in rehabilitation is the breadth of the patient population. When people ranging from those who can barely move to those whose function has almost fully returned are measured with a single scale, the scale easily misses changes at one end or the other. That is why separate versions dedicated to severe cases are made. The MAL 4/5 cited earlier is precisely that case, a version created for patients with moderate to severe or greater upper limb paralysis (PMID 38265026).

How It Actually Appears

What to Measure — International Consensus

If researchers pick different measures, data from different studies cannot be combined. For this reason, international consensus has been moving forward on what to measure. The Third Stroke Recovery and Rehabilitation Roundtable (SRRR) brought together 13 global experts to recommend a core set of measures for the domains of balance and mobility (PMID 37837351).

What was recommended is as follows:

• Lower extremity motor function — Fugl-Meyer Assessment lower extremity subscale
• Seated balance — Trunk Impairment Scale (TIS)
• Standing balance — Mini-BESTest and Berg Balance Scale (BBS)
• Walking independence — Functional Ambulation Categories (FAC, 0–5)
• Walking speed — 10-Meter Walk Test
• Walking endurance — 6-Minute Walk Test
• Complex walking — Dynamic Gait Index (DGI)

It was recommended that if the FAC, which grades walking independence, is under 3 points, standing balance should be measured with both the Mini-BESTest and the BBS, and the three walking tests (10MWT, 6MWT, DGI) should be marked as "Not Applicable" (PMID 37837351). This means that people who have difficulty walking alone are not forced to perform walking tests, and standing balance is looked at more closely instead. This recommendation aims to increase comparability between studies and make it possible to combine data (PMID 37837351). Consensus efforts of the same kind are expanding to other domains, such as post-stroke fatigue (PMID 37424273).

What Is Measured Versus What Is Actually Used — Placing Clinical Indicators and Sensors Side by Side

Recent research places wearable sensors next to clinical indicators. This is to check whether the ability measured in the test room and the degree of actual use in daily life measure the same thing. To give the answer first, past the early stage, the two generally moved together and diverged in their power to capture the changes felt by the patients (PMID 39881332).

This is the result of a longitudinal cohort that repeatedly measured the same people at 3 days, 10 days, 28 days, 90 days, and 365 days after their first stroke. A longitudinal cohort is a study that follows the same people over time and measures them repeatedly. As clinical indicators, FMA, ARAT, Box & Block, and the 14-item MAL were measured, and five sensors were worn on the torso, both wrists, and both ankles for three days. Thirteen arm use indicators were calculated using only functional movements excluding the walking intervals. This is because the indicators would be biased if arm swinging while walking or arm movements unrelated to daily tasks got mixed in (PMID 39881332).

At all time points after 28 days, clinical indicators and the use indicators of the paralyzed arm showed moderate to high positive correlations ($r_s$ 0.57 to 0.88). The correlation coefficient ($r_s$) is the degree to which two values rise and fall together, and the larger it is, the more they move in the same direction. What diverged was responsiveness. Responsiveness is the power of an indicator to move following actual changes. This study measured responsiveness by how well the changes in the indicators matched the changes felt by the patients themselves. That correspondence was clearly weaker for sensor indicators, at 0.28 to 0.43, whereas clinical indicators were 0.60 to 0.73 (PMID 39881332). This means that while sensors reflect how much one actually uses [the arm], their power to track the improvement felt by the patient was weaker than that of clinical indicators.

In the MAL study for 4/5 severe patients, a high correlation also appeared between the amount of use answered by the patients and the value from an accelerometer worn on the wrist, namely a small sensor that records movement (AOU $r$ = 0.67, QOU $r$ = 0.76) (PMID 38265026). This means that the amount of use reported by the patient and the amount of use measured by the machine overlap considerably.

Translating Clinical Importance into Other Units

The concept of the MCID extends beyond clinical scores. In the realm of acute reperfusion therapy, which reopens blocked blood vessels, imaging markers are converted into a patient's probability of an outcome. Infarction volume is the volume of brain tissue that has died because its blood supply was cut off.

Data from patients in the AcT trial who received thrombolytic therapy—a treatment in which a drug that dissolves the clot blocking a blood vessel is injected intravenously ("What a Stroke Actually Is — Blocked, Burst, and the Clock")—serve as an example. The adjusted odds ratio for a favorable functional outcome (mRS 0–1) per 1 mL increase in 24-hour final infarct volume (FIV) was 0.97. An adjusted odds ratio compares the likelihood of an outcome occurring after adjusting for other conditions, and this value means that as the infarct volume grows, the likelihood of achieving a favorable outcome decreases slightly. The differences in infarct volume corresponding to a 1% and 5% increase in the probability of a favorable outcome were a median of 1.5 mL and 7.3 mL, respectively (PMID 41128272).

The authors suggest that this value can be used in designing early-stage trials that use infarct volume as a primary outcome (PMID 41128272). It is a ruler that translates a difference of a few milliliters on an image into the probability of the outcome a patient will obtain.

Common Misconceptions

"If the p-value is small, the finding is clinically important." The p-value is the probability that, assuming there is no difference at all, you would get a result as extreme as the current one or more so ("What Does "Statistically Significant" Actually Say? — The p-Value, Explained with a Coin Toss"). Therefore, a small p-value only means the result is hard to write off as a fluke; it does not tell you how large the change actually is. The languages for judging that size are MDC and MCID.

"The MCID is a fixed number intrinsic to that scale." As we saw in the previous two examples, the MCID varies depending on the target group, severity, and anchor. When a paper states that a result "crossed the MCID," you have to check which specific group that MCID came from.

"Since they used the FMA, the results must be comparable." There were 22 versions of the FMA (PMID 40533328). The international standard manual came out precisely because of that problem (PMID 41773341).

"Differences between raters are minor." In the mRS, the weighted kappa for nurses in the standardized procedure group was 0.76, while the routine method group was 0.30 (the confidence intervals overlap) (PMID 38546172).

"The score didn't go up, so the patient didn't get better." It might be a ceiling effect (PMID 40170702). Different scales also differ in their power to capture the change patients actually feel. In the same cohort, sensor-based arm use metrics showed weaker correspondence than clinical metrics (PMID 39881332).

To Go Deeper

Result tables are read in this order: which layer the indicator measures, which version was measured by whom, whether the amount of change exceeded the MDC, whether it exceeded the MCID, and in which group that MCID was derived.

The next installment, How Deep Is the Recovery — Between Compensation and True Recovery, is the partner to this piece. When a score goes up, it is the question of whether that means the arm has returned or that a different way of doing things has been learned. The meanings of those two terms are exactly as established in the primer, Neuroplasticity — A Brain That Builds New Roads Beside the Broken Ones. Recovery is the return of movement or function in the same way as before the injury, and compensation is accomplishing the same task by using another body part or a different method instead of the lost way.

When choosing an indicator, it is safe to start from international consensus recommendations such as SRRR (PMID 37837351).

Source Papers

• Standardized International Manual of the Fugl-Meyer Assessment of Motor Function After Stroke. PMID 41773341 (2026)
• Fugl-Meyer Assessment for upper extremity in stroke: A psychometric systematic review. PMID 40533328 (2026)
• Standardized measurement of balance and mobility post-stroke: Consensus-based core recommendations from the third Stroke Recovery and Rehabilitation Roundtable. PMID 37837351 (2024)
• A roadmap for research in post-stroke fatigue: Consensus-based core recommendations from the third Stroke Recovery and Rehabilitation Roundtable. PMID 37424273 (2024)
• An automated flowchart for the Modified Rankin Scale assessment: A multicenter inter-rater agreement analysis. PMID 38546172 (2024)
• Estimating the minimal clinically important difference of upper extremity outcome measures in chronic stroke patients with moderate to severe impairment. PMID 37742304 (2024)
• Estimation of minimal clinically important difference for 6-minute walking distance in patients with acute stroke using anchor-based methods and credibility instruments. PMID 39145516 (2024)
• Grade-4/5 motor activity log-Brazil for post-stroke individuals with a severely impaired upper limb: a validity, reliability and measurement error study. PMID 38265026 (2024)
• Construct validity and responsiveness of clinical upper limb measures and sensor-based arm use within the first year after stroke: a longitudinal cohort study. PMID 39881332 (2025)
• Association of Minimal Clinically Important Difference With Infarct Volume in Clinical Poststroke Outcomes. PMID 41128272 (2025)
• The ability of the Knee Osteoarthritis Outcome Score to detect changes over time is limited in patients with patellar instability due to substantial ceiling effect. PMID 40170702 (2025)
• Floor and ceiling effects of the international hip outcome tool-12 in patients undergoing hip preservation surgery: A national registry study. PMID 41840038 (2026)
• Which Components of The International Classification of Functioning, Disability and Health (ICF) are Covered by Cardiac Rehabilitation Assessment Tools among Individuals with Heart Failure? PMID 37806100 (2024)
• Providing evidence for content validity of the most frequently used hip specific recovery outcome measures in hip fracture studies: an International Classification of Functioning approach. PMID 37221661 (2024)