When Does Recovery Happen — What Time Did, and What Treatment Did

In rehabilitation papers, "patients got better" mixes two things together. One part is played by treatment, and another part is played by time, regardless of treatment. This article covers how to separate the two.

The first two pieces in this series addressed different questions. How Do We Measure Recovery — An Introduction to Rehabilitation Outcome Measures separated what we use to measure, and How Deep Is the Recovery — Between Compensation and True Recovery separated how deep that improvement is. The same question has one more leg.

Did that improvement come from treatment, or did time do it?

The foundation was laid by the primer Neuroplasticity — A Brain That Builds New Roads Beside the Broken Ones. That piece covered that the speed of getting better after an injury is fastest at first and becomes more gradual over time, that the curve contains a portion that happens regardless of treatment, and that the brain responds more strongly to practice during the first few months. This article goes to the next depth. It is about why before-and-after comparisons without a control group are almost unreadable, and why our intuition fails twice on the question of "when and how much we should do."

A before-and-after comparison measures the same people before and after treatment and compares them. A control group is a comparison group that spent the same period without receiving that treatment. Let us define one term in advance. The primer used "recovery" in the narrow sense of the old way returning. Names like "spontaneous recovery" and "proportional recovery" appearing in this article broadly refer to improvement as a whole, so "recovery" in this article also has the broad meaning.

Why We Must Know — When the Background Is Large, the Signal Cannot Be Seen

There is spontaneous recovery after a stroke. Spontaneous recovery is the improvement that happens by itself as time passes after an injury, regardless of treatment. This is the definition given in the primer. This biological recovery process inside the body is not small in magnitude.

What reveals its magnitude most sharply is the debate surrounding the proportional recovery rule. The proportional recovery rule is the rule that a fixed percentage of lost function returns by itself. According to this rule, in an arm stroke, the degree of recovery can be accurately predicted just by looking at the initial severity of damage or damage to the corticospinal tract. This is regardless of how rehabilitation was performed. The corticospinal tract is the nerve bundle that sends movement instructions straight down from the motor area of the brain to the spinal cord.

One study notes its implications like this: current rehabilitation practices may play little or no role in brain recovery, and recovery is largely the result of natural biological recovery processes (PMID 29319405).

The same study asked whether this is a process conserved across species. In 593 laboratory rats (Sprague-Dawley rats), recovery was predicted using only the initial injury. Using hierarchical cluster analysis, a statistical method that groups and divides similar individuals, about 30% were classified as individuals fitting this rule. Those rats recovered 66% (95% CI 62–70%) of their initial impairment, and this range overlaps with the ranges of various human clinical trials (PMID 29319405). The CI in parentheses, meaning the confidence interval, is the range of values consistent with the data written alongside a single value. The authors concluded that proportional recovery is a cross-species phenomenon and can be used to search for the biological mechanisms contributing to recovery (PMID 29319405). If similar proportions appear in rats and humans, it means that an internal body recovery process may underlie that proportion.

What this fact establishes is that the background is large. The conclusion that rehabilitation is useless does not come from here. When the background is large, the effect of a treatment placed on top of it is not visible without a control group. Therefore, spontaneous recovery is the largest confounding factor in rehabilitation research. A confounding factor is something that steps in between the treatment and the outcome, making something done by another cause look like it was done by the treatment.

Whether the proportional recovery rule itself is correct is not covered in this article. This rule is currently being refuted from a methodological standpoint. It has been pointed out that mathematical coupling and compression to the ceiling inflate the estimates (PMID 33827246). Mathematical coupling is the correlation that naturally occurs between two numbers that share the same value, such as the initial score and the amount of change, and compression to the ceiling is a phenomenon where people closer to the highest point of a scale have no room left to rise, suppressing change. There is also a paper that reorganizes what the rule is and what it is not. This paper concludes that the pooled data generally match the rule at the population level (PMID 31524062). Statistical considerations when drawing conclusions about recovery are also being sorted out (PMID 33317423). Because it cannot be presented based on an ongoing debate, this debate is covered separately in "Does 70% of What Was Lost Come Back? — Stroke's "Proportional Recovery Rule" and the Illusion Statistics Built". In this article, it is used only as the background that spontaneous recovery is large.

The Big Picture — So "When" Becomes the Issue

With a broad background, to see if an intervention adds something, we have to find the window where the intervention works best. This is where the sensitive period hypothesis comes in. A sensitive period is a short window of time after a brain injury when the brain responds unusually strongly to training. We met this term in the primer, and it is also called a critical period.

Animal literature points in this direction. Neuroplasticity is the ability of the brain to reorganize itself by creating new neural connections. The mammalian brain has a strong potential for recovery after an acute injury through this neuroplasticity mechanism, and that mechanism can be nudged by training methods and neurotransmitter manipulation (PMID 25972803).

The crucial factor is timing. In rodents, there is persuasive evidence for timing effects in rehabilitation. Motor training given at a specific time after a stroke can be more effective than the exact same training given earlier or later (PMID 25972803). This means that even with the same amount of training, when you deliver it can change how well it works. If we can find similar sensitive periods in humans, we can allocate our current clinical resources much more efficiently.

The same paper adds a caveat. Most of these findings have not been verified or confirmed in rehabilitation settings, because it is hard to move conceptually simple experiments from the lab into meaningful, rigorous clinical trials in humans (PMID 25972803).

CPASS (Critical Period After Stroke Study) is a Phase II randomized controlled trial designed to ask whether that window exists in humans as well. This phase is a setting to gauge a treatment's effects for the first time, and a randomized controlled trial is a test that divides groups like a draw of lots to compare them. CPASS added the same amount of arm practice at different time points to see if the effects changed depending on the timing (PMID 25972803). Those results were covered by "Neuroplasticity — A Brain That Builds New Roads Beside the Broken Ones" in the primer.

What preclinical research — meaning animal experiments done before using a treatment in humans — shows about neuroplasticity and what it still cannot show is summarized in a separate review (PMID 31390969).

Key Concepts — Dose Has Two Axes

After the timing, the next thing to ask about is the amount. In rehabilitation, dose (dose) is how much you practice. The primer introduced that dose is commonly measured by practice time or repetition count, and that how often and how hard it was done is also looked at together.

In this article, we must split the two axes within that. How often you do it (frequency) and how long you do it at once (duration per session) are different variables, and as seen below, the directions of their effects can be opposite. If they are bundled into a single "total treatment time", these two axes get mixed up.

How It Actually Manifests — "Earlier and More" Missed the Mark Twice

The first trial is AVERT (A Very Early Rehabilitation Trial). It compared usual care against very early and frequent out-of-bed mobilization in patients admitted to a stroke unit within 24 hours of stroke onset. Mobilization is the act of getting patients out of bed to sit, stand, and walk. A prespecified dose-response analysis of this trial enrolled 2,104 people, with 2,083 (99.0%) followed up at 3 months (PMID 26888985). A dose-response analysis is an analysis that looks at how outcomes change when the amount received varies.

The results diverged between frequency and duration. First, as the number of out-of-bed sessions per day—in other words, the frequency—increased, the odds of a favorable outcome consistently rose. This was true for both efficacy outcomes and safety outcomes. The odds ratio (OR) was 1.13 (95% CI 1.09–1.18, p<0.001), adjusted for the time to first mobilization and the total volume of mobilization (PMID 26888985). The odds represent the probability of a favorable outcome occurring divided by the probability of it not occurring, and the odds ratio is how many times greater those odds are. When it is greater than one, it leans toward a favorable outcome; when it is smaller, it leans the other way.

Conversely, as the daily duration of mobilization (minutes per day) increased, the odds of a favorable outcome decreased. The odds ratio was 0.94 (95% CI 0.91–0.97, p<0.001) (PMID 26888985). In the exact same patients, frequency and duration moved in opposite directions alongside the outcome.

In a CART analysis, which splits variables like a tree to find which ones drive the outcome, session frequency was the third most important variable after age and initial stroke severity (PMID 26888985). The authors concluded that short, frequent mobilization following acute stroke is associated with better outcomes (PMID 26888985).

One caveat comes with reading this result. Although this analysis was prespecified, it grouped patients by the actual dose they received, regardless of the group they were randomized to. The authors also classify this as Class III evidence, meaning a lower grade of evidence than comparisons between randomized assignments. This means the association emerged outside the protection of random assignment. While age and severity were accounted for, this analysis cannot rule out other differences between patients who received a lot and those who received little. Therefore, this result does not establish which frequency and duration are optimal.

What this result points toward is "frequent and short" rather than "more." Lumping frequency and duration together into a single "more" causes one to read the exact same paper backward.

The second trial is VECTORS. Constraint-Induced Movement Therapy (CIMT) is a bundle that combines restraining the less-impaired arm, intensive training with the more-impaired arm, and behavioral techniques to transfer that use into daily life. It is one of the most well-developed approaches for motor recovery of the arm (PMID 19458319).

VECTORS is a single-center, single-blind Phase 2 trial comparing three groups during acute inpatient rehabilitation: usual arm therapy, dose-matched CIMT, and high-dose CIMT (PMID 19458319). Single-blinded is a design where either the participants or the assessors do not know who is in which group.

Fifty-two participants were randomized at an average of 9.65 ± 4.5 days post-stroke. The mean NIHSS was 5.3, and the mean baseline ARAT total score was 22.5. The NIHSS is a score that rates the severity of stroke symptoms, where higher means more severe, and the ARAT is a test that measures the ability to grasp and move objects with the arm and hand, where higher means better. The primary outcome was the total ARAT score of the paralyzed side at 90 days post-stroke (PMID 19458319).

As expected, all groups improved over time. Had only a single group been observed, all of this improvement would have been read as a treatment effect. However, a significant interaction appeared between time and group (F=3.1, p<0.01). An interaction means that the shape of change over time differed depending on the group. The group that drove the difference was the high-dose group, which improved significantly less at 90 days (PMID 19458319).

There was no significant difference at 90 days between the dose-matched CIMT group and the control group. MRIs of some participants showed no evidence of activity-dependent lesion enlargement—in other words, no evidence that the lesions grew larger because they were used heavily (PMID 19458319).

The authors' conclusion is as follows. CIMT was just as effective as an equivalent amount of traditional therapy, but not better. The authors go a step further. The high-dose CIMT showed even less motor improvement at 90 days, suggesting an inverse dose-response relationship where more dose led to worse outcomes. Therefore, they write that motor intervention trials must control dosage, and especially in the early stage of stroke, one cannot assume that a higher dose is more beneficial (PMID 19458319). However, this conclusion comes from a small single-center trial.

The two trials tested different interventions but point in the same direction. Increasing intensity in the acute phase does not automatically translate to a benefit. These same two trials also yielded other findings. In AVERT, short and frequent mobilization was associated with better outcomes, and in VECTORS, dose-matched CIMT performed as well as usual care. The conclusion that moving early in itself is harmful does not stem from these two trials. However, because this analysis in AVERT controlled for the time to first mobilization, it does not answer when it is best to start moving for the first time.

Common Misconceptions

There is a widespread belief that "recovery ends in 3 to 6 months." This notion has been widely accepted based on the idea that there is a "critical window" of recovery within that timeframe, and it has been used as the rationale for generally ending physical therapy during the subsequent chronic phase (PMID 31141442). That evidence is being shaken.

One piece of evidence comes from individual patient data of 219 patients with mild or moderate hemiparesis in the arm. Hemiparesis is a condition where the limbs on one side of the body lose strength or cannot move. In this data, body functions and structures could improve even in the late chronic phase (PMID 31141442). Body functions and structures refer to the layer that looks at whether the body's parts are doing their jobs, among the ICF layers introduced in the primer "Rehabilitation — Starting to Plan a Livable Home as the Fire Is Brought Under Control". In a bootstrapping analysis, which tests how much results fluctuate by repeatedly resampling the data, the slope showing increased sensitivity to treatment continued past 12 months after onset. The authors write that rehabilitation clinical guidelines should be revised to reflect this temporal structure (PMID 31141442).

Another piece of evidence is a meta-analysis comparing physical therapy administered at least 6 months after onset—that is, in the chronic phase—against placebo or no treatment. A meta-analysis is an analysis that pools the results of multiple trials together, and this one included 15 RCTs and 700 people. The pooled effect was significant, with an effect size of 0.29 (95% CI 0.14–0.45). An effect size is a value obtained by dividing the difference between two groups by the spread of the results, in order to place results measured in different units on the same scale.

When walking is divided into short and long distances, the effect was particularly distinct. There were weighted mean differences of +0.05 m/s (95% CI 0.008–0.088) for walking speed and +20 m (95% CI 3.6–36.0) for walking distance. A weighted mean difference is the average difference where different weights are assigned based on the size of each trial. Improvements in activities of daily living were also greater in the treatment group, but they were not statistically significant (PMID 20826872). The authors concluded that various physical therapies improve functional outcomes even when applied late after stroke, and that these results cast doubt on the concept of a plateau in functional recovery (PMID 20826872).

"There is a sensitive period" and "It is useless afterward" are not the same thing. The animal evidence in the previous section states that the effect is larger at a specific time point, not that the effect is zero at any other time.

Proposals to rethink this period classification itself have also already appeared in the literature. They point out the paradox that while randomized trial evidence for rehabilitation started in the chronic phase has steadily accumulated, expectations and resources placed on chronic-phase rehabilitation remain low (PMID 24529594, PMID 23192711).

To Go Deeper

When reading a paper, the first clue is whether there is a control group. Without a control group, that improvement is entirely mixed up with natural recovery. If there is a control group, the next clue is how many days after onset the random allocation was done. Trials that allocated patients at an average of 9.65 days, like VECTORS, and chronic-stage trials answer different questions.

Dosage must be read by separating frequency from volume. AVERT shows that these two can be associated with outcomes in opposite directions. In papers reporting only "total treatment time," this distinction is erased.

If this article covered "when," the remaining questions are "what and how during that time." Motor learning principles such as repetition, task specificity, feedback, and practice schedules are covered in "Why Therapy Is Designed That Way — An Introduction to Motor Learning Principles," the relationship between dose and effect in "How Much Is Enough — An Introduction to Dose-Response in Rehabilitation," and learned non-use and CIMT in "Able But Unused — An Introduction to Learned Non-Use and CIMT."

Because methodological debates are ongoing, the proportional recovery rule is examined separately in "Does 70% of What Was Lost Come Back? — Stroke's "Proportional Recovery Rule" and the Illusion Statistics Built" (PMID 33827246, PMID 31524062, PMID 33317423).

Source Papers

• Does Stroke Rehabilitation Really Matter? Part A: Proportional Stroke Recovery in the Rat. PMID 29319405 (2018)
• Critical periods after stroke study: translating animal stroke recovery experiments into a clinical trial. PMID 25972803 (2015)
• Prespecified dose-response analysis for A Very Early Rehabilitation Trial (AVERT). PMID 26888985 (2016)
• Very Early Constraint-Induced Movement during Stroke Rehabilitation (VECTORS): A single-center RCT. PMID 19458319 (2009)
• A critical time window for recovery extends beyond one-year post-stroke. PMID 31141442 (2019)
• Efficacy of physiotherapy interventions late after stroke: a meta-analysis. PMID 20826872 (2011)
• Inflated Estimates of Proportional Recovery From Stroke: The Dangers of Mathematical Coupling and Compression to Ceiling. PMID 33827246 (2021)
• What the Proportional Recovery Rule Is (and Is Not): Methodological and Statistical Considerations. PMID 31524062 (2019)
• Statistical Considerations for Drawing Conclusions About Recovery. PMID 33317423 (2021)
• Preclinical Studies of Neuroplasticity Following Experimental Brain Injury. PMID 31390969 (2019)
• Rethinking the continuum of stroke rehabilitation. PMID 24529594 (2014)
• Time to rethink long-term rehabilitation management of stroke patients. PMID 23192711 (2012)