Motor learning is a set of internal processes, related to practice, leading to a relatively permanent change in the capability to perform a motor skill.1 To reflect its relative permanence, motor learning is commonly measured by a retention test, or by a comparison between pretest and retention test performance.1–5 While calculations using discrete pretraining and retention timepoints can represent motor learning well, they do not fully characterize how performance changes during the practice phase, hereafter termed practice.6,7Skill acquisition refers to performance changes occurring during practice. Typically, motor performance improves nonlinearly with practice, marked by rapid improvements early with eventual stabilization (ie, a performance plateau).3 These nonlinear properties (eg, rate of improvement and practice dose required to reach plateau) are lost in traditional retention test analyses.
One way to capture the nonlinear properties associated with learning is to model performance improvement using curvilinear functions (eg, exponential decay functions),8–10 which illustrate how motor skill acquisition progresses throughout practice. Nonlinear modeling is a valuable adjunct to test-retest comparisons because it typically incorporates all data points into the model, instead of providing summary statistics of specific timepoints. Examining the nonlinear characteristics of skill acquisition curves could provide critical insight into factors that facilitate or hinder learning. The amount, or dose, of practice needed to optimize motor learning is one such factor, and understanding it is essential to improving patient outcomes and guiding research inquiries.11 Wadden et al7 used nonlinear modeling to determine that the predicted optimal practice dose of a hand-tracking task was greater for participants poststroke than for healthy controls. The same study also found a significant relationship between skill acquisition rate and learning in healthy control participants, such that those who acquired the repeating sequence of a hand-tracking task faster during practice demonstrated better learning (ie, retention) of the task.
Idiopathic Parkinson disease (PD) causes motor and nonmotor deficits, which produce substantial disability.12 Dopamine-replacement medication (levodopa; l-dopa) forms the basis of PD medical management to reduce symptoms and improve mobility.13 However, even with l-dopa management, symptoms and motor function worsen, and falls increase with disease progression.13–16 In concert with medical management, physical therapy aims to improve balance and reduce falls through practicing anticipatory and reactive postural motor skills.17 Therefore, postural motor learning is particularly relevant to people with PD, as they seek to learn/relearn motor skills to remain safe and independent as their disease progresses. However, people with PD have impaired motor learning compared with those without neurologic conditions,18,19 which may be related to lower levels of endogenous dopamine and fewer dopaminergic receptors, both associated with the disease state.20 Dopamine-replacement therapy may overdose the basal ganglia and impair motor learning during early skill acquisition,20–22 but evidence regarding the effects of exogenous dopamine on postural motor learning is equivocal.23–32 If motor skill acquisition parameters and medication status reliably predict skill learning (ie, retention) in PD, this analysis method could allow therapists to individualize practice doses and levels of challenge, to individualize these parameters to the needs of the specific patient.
Motor learning has been investigated using a variety of experimental paradigms, including serial reaction time tasks (SRTTs).33–35 In an SRTT, participants are asked to map a visuospatial stimulus to a corresponding response key. Stimuli are presented in a series of either random or repeating sequences. The ability to improve responses to stimuli presented in random sequences is termed general learning, and represents learning of the general movement skill, whereas the ability to improve responses to a repeating sequence is termed repeated-sequence learning and represents learning of the general movement skill combined with learning of the underlying motor sequence (of which the participants are not explicitly aware due to blinding).35
We examined a postural motor task performed by individuals with PD. The purpose of this study was to characterize postural skill acquisition in people with PD, and to identify factors (such as acquisition rate and practice dose to plateau) that predict postural motor learning while controlling for participants' level of physical function.7 We hypothesized that skill acquisition rates, and whether or not a participant achieves plateau during practice, would significantly predict postural motor learning of a repeating sequence, over and above that predicted by participants' level of physical function. A secondary purpose of this study was to test whether l-dopa status during practice would explain additional variance in learning (beyond acquisition rate and level of physical function).
The current project was a secondary analysis of a randomized controlled trial (NCT02593812).36 The University of Utah Institutional Review Board approved the study, and participants provided written informed consent before enrolling. The study was conducted at the University of Utah and in participants' homes, depending on each participant's access to transportation. A physical therapist performed initial assessments, after which a computer-generated randomization schedule randomly assigned participants to practice either ON or OFF l-dopa. Within each group (ie, ON l-dopa and OFF l-dopa), stratified random assignment was used to ensure that the groups were balanced in regard to disease severity, based on Hoehn and Yahr (H&Y) disease stage (in which 0.0 = “asymptomatic” and 5.0 = “wheelchair bound or bedridden unless aided”). Participants within H&Y stages 2.0 or less were considered to have less severe PD, while participants with H&Y stages of more than 2.0 were considered to have more severe PD. Practice sessions were supervised by trained study staff. Delayed retention tests were performed by a physical therapist blinded to group assignment.
Between July 2016 and May 2017, people with PD were recruited from the Salt Lake City metropolitan area including a Movement Disorders Clinic, a PD wellness group, and PD support groups. Eligible participants were 50 to 80 years old, diagnosed with idiopathic PD by a neurologist, in H&Y stages 1.0 to 3.0, able to walk at least 10 m without physical assistance when OFF l-dopa, and on a stable medication regimen from 1 month prior to and throughout all study visits. Those with deep brain stimulation, not taking l-dopa, experiencing medication-resistant freezing of gait, functionally unresponsive to l-dopa, cognitively impaired (Montreal Cognitive Assessment score <18),37 or with any health conditions preventing safe performance of the assessments or motor task were excluded. Based on a priori power calculations for the original study aims,36 30 participants were sought for testing.
To ensure that our protocol was truly testing functional differences between the ON- versus OFF-medication states, volunteers were tested for functional l-dopa responsiveness. To do so, participants were tested twice during the initial assessment using the motor section of the Movement Disorders Society's revised version of the Unified Parkinson's Disease Rating Scale (MDS-UPDRS). First, the MDS-UPDRS motor section was scored while participants were OFF l-dopa following overnight withdrawal; it was then scored while participants were ON l-dopa (30-60 minutes after taking their usual dose). Participants whose MDS-UPDRS score while ON medication exceeded their score while OFF by more than the standard error of measurement were included in the study.38 All demographic and outcome measures were conducted ON l-dopa, the latter including self-selected and fast gait speeds,39 Four Square Step Test (4SST),40 and Mini Balance Evaluation Systems Test (mini-BEST).41
The motor task was a standing postural SRTT in which participants stood on an instrumented step mat,42 were presented with a stimulus on a computer screen, and stepped to the corresponding location on the mat as fast as safely able (see Figure 1A for an image of the motor task, and Supplemental Digital Content 2, available at: http://links.lww.com/JNPT/A251, for a video of the task). The step mat included 2 “home” pressure pads, consisting of 2 arrows at the back of the step mat. Four target pressure pads were also visible on the mat, consisting of 4 additional arrows (ie, “target arrows”), to the left, right, left front, and right fronts of the home pressure pads, respectively. An image of the step mat was visible on a computer screen in front of the participants, who were instructed to start by standing on the back arrows (ie, home position). When the stimulus appeared on the computer screen, participants stepped to the corresponding target arrow on the step mat using the foot ipsilateral to the location of the target (ie, the right foot stepped to targets to the right or right front of the participant, and vice versa for the left). The mat registered a correct step when approximately 66.35 N of force (ie, equivalent to a 6.77-kg stationary object) was applied to the appropriate target, at which point the stimulus dimmed and participants returned their step foot to home position. Subsequent stimuli only appeared after the mat registered that the participant had (1) reached the correct target and then (2) returned to home position. Response time was collected and defined as time from stimulus presentation to foot touching down on target, which includes reaction time plus movement time.1
Practice consisted of sequences of 12 steps that were either random or followed a repeating pattern. Random 12-step sequences were created by a computer randomization program, and participants encountered each random sequence only once during the entire study protocol. In contrast, the repeating 12-step sequence was repeated once during every practice trial. The 4 target arrows were equally represented within each 12-step sequence (whether random or repeating) such that each target was presented 3 times during every sequence.
Practice consisted of 6 blocks per day, over 3 consecutive days, followed by 2 delayed retention tests, performed 2 (retention 1) and 9 (retention 2) days after practice ended, respectively (Figure 1B). Every trial consisted of two 12-step sequences (1 random and 1 repeating) in random order. Every block consisted of 6 trials (144 steps) and participants completed 6 blocks per day for 3 days (108 total trials; 2592 total steps). Participants rested for 25 seconds (standing) between trials and 4 minutes (seated) between blocks, so each practice session lasted approximately 90 minutes. During seated breaks, participants were provided with verbal feedback43–45 of their median response time for the preceding block. Participants were blinded to the presence of the imbedded repeating sequence.
Pretest performance was defined as mean response time on the first block of practice (ie, trials 1-6 of day 1). Each retention test included a single block, as described earlier, lasting approximately 15 minutes. Retention 1 was defined as mean response time during the first retention test, while retention 2 was mean response time during the second retention test. Repeating and random sequence performance was separated for all calculations, and “learning” was defined as the difference between performance at pretest compared with retention test.
Participants randomly assigned to train OFF medication practiced the SRTT in the morning after overnight withdrawal from l-dopa, although they took all other antiparkinsonian medications at their typical times and dosages, including dopamine agonists.46 Participants randomly assigned to train ON l-dopa practiced approximately 1 hour after taking their morning l-dopa dose, to ensure optimal medication effects.
All statistical analyses were conducted in R (v3.4.1; R Core Team, 2017) using packages for nonlinear curve-fitting and mixed-effects regression.47–49
Data Aggregation and Nonlinear Curve-Fitting
Visual inspection of the data revealed some outlying response time data points due to technical difficulties or participants being unaware that the trial had begun. Extreme outliers (greater than the mean ± 3 standard deviations for each individual) were excluded from analyses, resulting in 1.4% of data being removed and treated as missing. For each participant, mean response time was calculated for all repeating steps across all trials; the same was also calculated, separately, for all random steps across all trials.
Skill acquisition parameters were calculated for the repeating and random sequences, separately, for each participant. Individual exponential decay models (equation 1) were fitted to each participant's mean response time data (in seconds) as a function of trial number (trials 7-108 only, as trials 1-6 served as pretest),
where A is the estimated asymptote (ie, the final trial time that the exponential decay function approaches), C is the expected change in response time during practice (ie, the difference from the estimated upper limit to the estimated asymptote), R is the acquisition rate (ie, the decay constant), and x is the trial number (Figure 1C). These 3 parameters (A, C, and R) were estimated for each participant for random and repeating sequences, separately. Performance on each retention test was calculated as mean response time for the random and repeating sequences, separately.
Predicting Number of Trials Required to Reach Plateau
To determine whether motor learning (ie, retention) could be predicted based on the dichotomous variable of whether or not a participant reached a performance plateau during practice, we calculated the predicted number of practice trials required for each participant's performance to reach 1% over the predicted asymptote, A, consistent with calculations from Wadden et al,7
where Trialplateau is the expected trial number at which participants plateau in their performance (ie, reach 1% over trial time asymptote). Participants who were estimated to reach their Trialplateau after 108 trials or less (ie, within the number of practice trials completed) were categorized as “reached plateau,” while those whose Trialplateau was more than 108 trials were categorized as “did not reach plateau.” To determine whether reaching plateau affected learning, Trialplateau was included in a mixed-effects regression model, along with additional factors of the retention test and the retention test by plateau interaction. The factor of the retention test (retention 1 vs retention 2) was included to determine whether there was a significant difference in performance on the 2 retention tests.
Defining Physical Function Level
To control for varying function among participants, we included a measure of physical function in our analyses.7 In our data, several variables represented different aspects of physical function, but were highly correlated with each other. To reduce multicollinearity in our analyses, we conducted a principal component analysis of participants' (1) OFF-medication MDS-UPDRS scores, (2) mean self-selected gait speed, (3) mean fast gait speed, (4) mean 4SST time, and (5) mean mini-BEST score (Part 1 of Supplemental Digital Content 3, available at: http://links.lww.com/JNPT/A252, contains further principal component analysis details). Only 1 principal component was extracted, termed PC1. Both gait variables and mini-BEST scores positively loaded on this factor, while MDS-UPDRS scores and 4SST time negatively loaded on it. Therefore, PC1 was included in our analysis to represent physical function as a single variable (as PC1 captures all 5 variables); higher PC1 values represent better physical function.
The dependent measure was learning of the SRTT at both retention tests. Learning was defined as the difference between participants' response time on pretest compared with retention tests (pretest − retention test); thus, positive differences indicated larger improvements in response time. Learning of the blinded imbedded repeating sequence was termed repeated-sequence learning; learning of the general movement skill represented by the random sequences was termed general learning.
For both repeated-sequence and general learning, we tested mixed-effects regression models with a random effect of the participant (to control for retention test timepoint being a within-subject variable). The first model contained a fixed effect of the retention test (retention 1 vs retention 2) to determine whether there was a significant difference in performance on the 2 retention tests. The second model added a fixed effect of PC1, to control for level of physical function. The third model added a fixed effect of the acquisition rate parameter (R), to identify whether acquisition rate (ie, performance improvement rate) explained significant variation in learning when controlling for physical function. The fourth model added a fixed effect of medication status (ON vs OFF l-dopa). Model fit was assessed based on a 2-point change in the Akaike's Information Criterion (AIC), and P values for individual parameters were estimated using the Satterthwaite approximation.50,51
Twenty-seven volunteers were randomly assigned to practice the SRTT ON (n = 14) or OFF (n = 13) l-dopa (Table 1). All participants completed the entire study.
Table 1. -
Baseline Demographic Data by Groupa
||OFFb (n = 13)
||ONc (n = 14)
||All (n = 27)
|Time since diagnosis, y
|OFF MDS-UPDRS motor (0-132)e
|OFF H&Y: 2, 3
||13 (100%), 0 (0%)
||12 (86%), 2 (14%)
||25 (93%), 2 (7%)
|ON MDS-UPDRS motor (0-132)
|ON H&Y: 2, 3
||13 (100%), 0 (0%)
||14 (100%), 0 (0%)
||27 (100%), 0 (0%)
|Self-selected gait speed, m/sd,e
|Fast gait speed, m/sd,e
|4SST time, se
Abbreviations: 4SST, Four Square Step Test; H&Y, Hoehn and Yahr stage; LED, levodopa-equivalent dose; MDS-UPDRS, Movement Disorders Society sponsored version of the Unified Parkinson's Disease Rating Scale; mini-BEST, Mini Balance Evaluations Systems Test; MoCA, Montreal Cognitive Assessment.
aData presented as mean (standard deviation) or n (%).
bParticipants who practiced the serial reaction time task in the morning following overnight withdrawal from l-dopa.
cParticipants who practiced the serial reaction time task approximately 1 hour after taking their morning dose of l-dopa.
dHigher score indicates better performance.
eAll of the variables incorporated into PC1, which in turn broadly represents physical function in a single variable (higher PC1 values represent better physical function).
For the exponential curves, the model failed to converge for 1 participant's random-sequence curve, and another participant's acquisition rate parameter was an extreme outlier, resulting in both participants' exclusion from random sequence analyses. Furthermore, exponential curves showed performance worsening during practice for 4 participants' repeating trials (ON l-dopa = 2, OFF l-dopa = 2) and 6 participants' random trials (ON l-dopa = 4, OFF l-dopa = 2). These participants were excluded from the respective analyses to be consistent with Wadden et al7 and to keep the interpretation of the acquisition rate parameter (R) consistent across participants, as the interpretation of R depends on the sign of C (Part 2 of Supplemental Digital Content 3, available at: http://links.lww.com/JNPT/A252, explains these exclusions, and how to interpret R). After these exclusions, there were 23 participants whose data remained for subsequent analyses of the repeating sequence, and 19 whose data remained for random sequence analyses.
Relationships Between Skill Acquisition Parameters and Learning
Acquisition Rate and Repeated-Sequence Learning
Overall, participants demonstrated significant repeated-sequence learning, defined as the difference between participants' response time on the repeating sequence at pretests compared with retention tests (P < 0.001), with no difference in learning between retention 1 and retention 2. There was a significant negative effect of PC1, such that there was less learning from pretest to retention in less impaired participants. There was also a significant negative effect of the acquisition rate parameter, such that participants with faster rates of improvement during practice learned less (Figures 2A and 2B). There was no difference in learning based on medication status. The best-fitting model was model 3, including fixed effects of PC1 and acquisition rate (Part 3 of Supplemental Digital Content 3, available at: http://links.lww.com/JNPT/A252, contains detailed model fits). For completeness, Table 2 shows all parameters from model 4, which included all main effects.
Table 2. -
Fixed Effects of Learning Data for Repeating and Random Sequencesa
|Retention test timepointc
|Retention test timepoint
aAll variables were mean-centered to reduce potential collinearity and simplify the interpretation of the main effects.
bStatistically significant using an alpha level of .05.
cRetention test 1 or retention test 2.
dExtracted via principal components analysis, PC1 broadly represents physical function in a single variable, in which higher PC1 values equate to better physical function.
Rate of improvement (ie, decay rate), as calculated via equation 1
fWhether the participant practiced the serial reaction time task while ON- or OFF-levodopa medication.
Acquisition Rate and General Learning
Overall, participants demonstrated significant general learning, defined as the difference between participants' response time on the random sequences at pretest compared with retention tests (P < 0.001), with no difference in learning between retention 1 and retention 2. Unlike the repeating trials, there was no significant effect of acquisition rate. However, similar to repeated-sequence findings, there was a significant negative effect of PC1, such that there was less learning from pretest to retention in less impaired participants. There was also a significant effect of medication status, with participants who practiced ON l-dopa learning more. These effects are shown in Figures 2C and 2D. The best-fitting model was model 4 (Table 2), including fixed effects of PC1, acquisition rate, and medication status (Part 3 of Supplemental Digital Content 3, available at: http://links.lww.com/JNPT/A252, contains detailed model fits).
Reaching Plateau and Learning
Most participants included in the analysis reached their estimated plateau on the repeating sequence (16 reached Trialplateau) and random sequences (12 reached Trialplateau) during practice. For both repeated-sequence and general learning, there was no effect of having reached plateau (repeated-sequence P = 0.59, general P = 0.41), nor was there a retention test by plateau interaction (repeated-sequence P = 0.94, general P = 0.20; Table 3).
Table 3. -
Participants Who Reached Their Estimated Plateau in Response Time Compared With Those Who Did Nota
||Learning (Pretest - Retention Test)
Data presented as mean (standard deviation). Number of trials needed to reach plateau (ie, Trialplateau
) was estimated based on equation 2
“Not reached” refers to participants who did not reach their estimated plateau during practice, while “reached” refers to participants who did reach their estimated plateau. Note that the reported “n” is different for the repeating and random sequences due to exclusions from the exponential curve-fitting.
Using nonlinear mathematical models to estimate individual skill acquisition parameters for each participant, we investigated repeated-sequence and general learning of a postural SRTT in people with PD, with learning defined as the difference between performance at pretest compared with retention tests. Overall, participants demonstrated significant repeated-sequence and general learning; of note, this learning was persistent, as there were no differences in retention test performance when tested 2 (retention 1) and 9 (retention 2) days after practice ended. As hypothesized, skill acquisition rates significantly predicted repeated-sequence learning over and above that predicted by participants' level of physical function.7 Interestingly, achieving plateau during the prescribed dose of practice did not predict learning. Contrary to the dopamine-overdose hypothesis,20l-dopa status did not negatively affect learning; specifically, l-dopa showed no effect on repeated-sequence learning and significantly predicted general learning (over and above acquisition rate and level of physical function), with those who practiced ON l-dopa learning (ie, retaining) more. Collectively, these findings emphasize that individuals at various levels of physical function who have PD can learn a postural motor skill. Importantly, individual skill acquisition characteristics can inform how patient-specific factors may influence postural motor learning and balance rehabilitation.
Nonlinear Analyses Reveal Individual Learning Differences
While most participants acquired and retained both the general skill and the repeating sequence within the provided practice dose, individual curve characteristics were consistently influenced by the level of physical function. The finding that more impaired participants demonstrated more learning partially agrees with findings from a study of reactive stepping in people with PD,52 but is also likely due to the strong correlation between pretest performance and level of physical function, suggesting that higher functioning participants experienced a floor effect on this task. In addition, a minority of participants demonstrated an unexpected lack of skill acquisition. Such variability is certainly present in clinical situations and should be considered when prescribing balance rehabilitation components, such as types of activities and practice doses.
Skill Acquisition Characteristics
Skill Acquisition Rate
Previous studies of postural motor learning in PD have not investigated skill acquisition rate. Interestingly, participants in our study who improved more quickly during practice demonstrated less repeated-sequence learning of the task. Although this finding is contrary to what Wadden et al7 found among participants poststroke, it is consistent with robust findings of a performance-learning distinction, observed when methods of motor practice that result in better performance during practice result in worse long-term learning (eg, blocked vs random practice).53–55 This finding suggests that, in a clinical setting, slow performance improvement during initial practice should not be considered a negative predictor of learning. In addition, rapid improvement during practice may reflect an inappropriate level of challenge, and should suggest that therapists progress task difficulty to optimize motor learning.56
Research examining motor learning in PD is fairly consistent in reporting that PD slows learning, but is relatively silent regarding practice dose effects on learning.57 While participants in our study completed 108 trials (2592 steps) of the SRTT, this practice dose was preset and did not differ depending on an individual's specific needs or characteristics. While the dose of practice provided was sufficient for most participants to achieve performance plateau, there was no significant relationship between reaching performance plateau during practice and how much general or repeated-sequence learning occurred. To our knowledge, this is the first examination of this construct in people with PD; however, prior studies of various populations suggest that continuing to practice after reaching performance plateau may improve motor learning.58–60 Our nonsignificant finding may be due to unequal numbers of participants achieving plateau compared to not. A more controlled method for determining the role that reaching a performance plateau plays in motor learning would be to prospectively identify when plateau has been reached in real time, and to randomly assign participants to either practice until reaching performance plateau, or continue practicing even after reaching plateau.
Since l-dopa is the most common treatment to improve motor symptoms associated with PD, it is important to consider potential adverse effects on the other roles of the basal ganglia, such as motor learning. Antithetical to the dopamine-overdose hypothesis,20 our results indicate that l-dopa did not negatively influence postural motor learning, regardless of what was practiced (ie, repeating or random sequences). Results of previous studies examining how exogenous dopamine affects motor learning appear to differ based on the task, effector, and type of learning. For upper extremity reaching/pointing tasks, l-dopa has resulted in positive32 and negative26 effects on explicit motor sequence learning, trivial effects on implicit motor learning of an SRTT,27,28 and negative effects on feedback-based learning.31 Exogenous dopamine has a complex relationship with postural control and motor learning as evidenced by studies indicating that l-dopa may negatively affect performance of reactive postural control and gait,24,25 while not impairing learning of gait,30 in-place postural control,19,23 or reactive postural control.29 In our study, it appears that the reduced burden of motor deficits while ON l-dopa outweighed any potential negative effect on learning. Based on these results, individuals with PD should be encouraged to participate in postural motor training while ON medication.
Limitations and Future Directions
While this study is the first to examine individual postural motor learning trajectories in participants with PD in an attempt to understand factors that facilitate or impede learning, it should be interpreted with caution. The participants were not withdrawn from all antiparkinsonian medications, as they continued to take dopamine agonists throughout the study, which may limit the findings of the OFF-state. In addition, considering that the association between posture and impairments differs in those with versus without freezing of gait, it would be valuable for future studies to assess the relationship between freezing of gait and postural skill acquisition.61 Finally, the analysis methodology employed required a full set of a predetermined number of practice trials to examine skill acquisition curves and determine curve characteristics. Interestingly, we encountered unforeseen challenges to using nonlinear models to describe these data, despite this method of retrospective individual curve-fitting being used successfully in other populations, tasks, and effectors.7,8 A number of participants were excluded for several reasons, including the model failing to converge or participant performance worsening overall, which was unrelated to any measured variable but could be due to an unmeasured confounding variable, such as boredom, “fatigue,” or motivation.
Participants who worsened overall were excluded because interpreting the acquisition rate parameter depended on the sign of the change. Although this variability in skill acquisition patterns is not ideal, it strengthens our position that clinicians need more individualized methods for analyzing patients' performance to make data-driven decisions about practice conditions and dose. Future research should examine alternative analysis methods that can individually characterize skill acquisition efficiently during balance rehabilitation. Such analyses would allow clinicians to recognize and accommodate interindividual variation in acquisition characteristics and prescribe individualized postural motor learning programs.
Following a multiday bout of practice, postural motor learning occurs in individuals with PD, even among those more severely impaired. Slow improvers learned more than faster improvers. Examining individual skill acquisition characteristics (eg, rate) and other patient-specific factors (eg, level of physical function and medication status) may influence how postural motor learning is delivered during balance rehabilitation.
The authors wish to thank Amy Ballard, Jacqueline Hill, Kirsten Gorski, Jane Savier-Steiger, Alicia Dibble, Shelby Dibble, Orin Ryan, Dylan Wile, Anna Lundgren, and Marianne Wilson for their assistance with data collection.
1. Schmidt RA, Lee TD. Motor Control and Learning: A Behavioral Emphasis. 5th ed. Champaign, IL: Human Kinetics; 2011.
2. Boyd L, Winstein C. Explicit information interferes with implicit motor learning
of both continuous and discrete movement tasks after stroke. J Neurol Phys Ther. 2006;30(2):46–57; discussion 48-49.
3. Whitall J. Stroke rehabilitation research: time to answer more specific questions? Neurorehabil Neural Repair. 2004;18(1):3–811.
4. Liu Y, Cao C, Yan JH. Functional aging impairs the role of feedback in motor learning
. Geriatr Gerontol Int. 2013;13(4):849–859.
5. Al-Sharman A, Siengsukon CF. Time rather than sleep appears to enhance off-line learning and transfer of learning of an implicit continuous task. Nat Sci Sleep. 2014;6:27–36.
6. Schaefer SY, Duff K. Rapid responsiveness to practice predicts longer-term retention of upper extremity motor skill in non-demented older adults. Front Aging Neurosci. 2015;7:214.
7. Wadden KP, Asis K, Mang CS, et al. Predicting motor sequence learning in individuals with chronic stroke. Neurorehabil Neural Repair. 2017;31(1):95–104.
8. Lang CE, Bastian AJ. Cerebellar subjects show impaired adaptation of anticipatory EMG during catching. J Neurophysiol. 1999;82(5):2108–2119.
9. Deuschl G, Toro C, Zeffiro T, Massaquoi S, Hallett M. Adaptation motor learning
of arm movements in patients with cerebellar disease. J Neurol Neurosurg Psychiatry. 1996;60(5):515–519.
10. Schaefer SY, Dibble LE, Duff K. Efficacy and feasibility of functional upper extremity task-specific training for older adults with and without cognitive impairment. Neurorehabil Neural Repair. 2015;29(7):636–644.
11. Lohse KR, Lang CE, Boyd LA. Is more better? Using metadata to explore dose-response relationships in stroke rehabilitation. Stroke. 2014;45(7):2053–2058.
12. Jankovic J. Parkinson's disease: clinical features and diagnosis. J Neurol Neurosurg Psychiatry. 2008;79(4):368–376.
13. Kim SD, Allen NE, Canning CG, Fung VSC. Postural instability in patients with Parkinson's disease. CNS Drugs. 2013;27(2):97–112.
14. Eggers C, Pedrosa DJ, Kahraman D, et al. Parkinson subtypes progress differently in clinical course and imaging pattern. PLoS One. 2012;7(10):e46813.
15. Heinzel S, Bernhard FP, Roeben B, et al. Progression markers of motor deficits in Parkinson's disease: a biannual 4-year prospective study. Mov Disord. 2017;32(8):1254–1256.
16. Hely MA, Reid WG, Adena MA, Halliday GM, Morris JG. The Sydney multicenter study of Parkinson's disease: the inevitability of dementia at 20 years. Mov Disord. 2008;23(6):837–844.
17. Paul SS, Dibble LE, Peterson DS. Motor learning
in people with Parkinson's disease: implications for fall prevention across the disease spectrum. Gait Posture. 2018;61:311–319.
18. Felix K, Gain K, Paiva E, Whitney K, Jenkins ME, Spaulding SJ. Upper extremity motor learning
among individuals with Parkinson's disease: a meta-analysis evaluating movement time in simple tasks. Parkinsons Dis. 2012;2012:589152.
19. Hayes HA, Hunsaker N, Schaefer SY, et al. Does dopamine replacement medication affect postural sequence learning in Parkinson disease? Motor Control. 2015;19(4):325–340.
20. Vaillancourt DE, Schonfeld D, Kwak Y, Bohnen NI, Seidler R. Dopamine overdose hypothesis: evidence and clinical implications. Mov Disord. 2013;28(14):1920–1929.
21. Dan X, King BR, Doyon J, Chan P. Motor sequence learning and consolidation in unilateral de novo patients with Parkinson's disease. PLoS One. 2015;10(7):e0134291.
22. Kwak Y, Müller ML, Bohnen NI, Dayalu P, Seidler RD. Effect of dopaminergic medications on the time course of explicit motor sequence learning in Parkinson's disease. J Neurophysiol. 2010;103(2):942–949.
23. Anderson ED, Horak FB, Lasarev MR, Nutt JG. Performance of a motor task learned on levodopa deteriorates when subsequently practiced off. Mov Disord. 2014;29(1):54–60.
24. Curtze C, Nutt JG, Carlson-Kuhta P, Mancini M, Horak FB. Levodopa is a double-edged sword for balance and gait in people with Parkinson's disease. Mov Disord. 2015;30(10):1361–1370.
25. de Kam D, Nonnekes J, Oude Nijhuis LB, Geurts AC, Bloem BR, Weerdesteyn V. Dopaminergic medication does not improve stepping responses following backward and forward balance perturbations in patients with Parkinson's disease. J Neurol. 2014;261(12):2330–2337.
26. Feigin A, Ghilardi MF, Carbon M, et al. Effects of levodopa on motor sequence learning in Parkinson's disease. Neurology. 2003;60(11):1744–1749.
27. Ghilardi MF, Feigin AS, Battaglia F, et al. L-Dopa infusion does not improve explicit sequence learning in Parkinson's disease. Parkinsonism Relat Disord. 2007;13(3):146–151.
28. Muslimovic D, Post B, Speelman JD, Schmand B. Motor procedural learning in Parkinson's disease. Brain. 2007;130(pt 11):2887–2897.
29. Peterson DS, Horak FB. The effect of levodopa on improvements in protective stepping in people with Parkinson's disease. Neurorehabil Neural Repair. 2016;30(10):931–940.
30. Roemmich RT, Hack N, Akbar U, Hass CJ. Effects of dopaminergic therapy on locomotor adaptation and adaptive learning in persons with Parkinson's disease. Behav Brain Res. 2014;268:31–39.
31. Shohamy D, Myers CE, Geghman KD, Sage J, Gluck MA. l
-dopa impairs learning, but spares generalization, in Parkinson's disease. Neuropsychologia. 2006;44(5):774–784.
32. Tremblay PL, Bedard MA, Langlois D, Blanchet PJ, Lemay M, Parent M. Movement chunking during sequence learning is a dopamine-dependant process: a study conducted in Parkinson's disease. Exp Brain Res. 2010;205(3):375–385.
33. Nissen MJ, Bullemer P. Attentional requirements of learning: evidence from performance measures. Cognitive Psychol. 1987;19(1):1–32.
34. Robertson EM. The serial reaction time task: implicit motor skill learning? J Neurosci. 2007;27(38):10073–10075.
35. Schwarb H, Schumacher EH. Generalized lessons about sequence learning from the study of the serial reaction time task. Adv Cogn Psychol. 2012;8(2):165–178.
36. Paul SS, Schaefer SY, Olivier GN, Walter CS, Lohse KR, Dibble LE. Dopamine replacement medication does not influence implicit learning of a stepping task in people with Parkinson's disease. Neurorehabil Neural Repair. Published online November 9, 2018. https://doi.org/10.1177/1545968318809922
37. Hoops S, Nazem S, Siderowf AD, et al. Validity of the MoCA and MMSE in the detection of MCI and dementia in Parkinson disease. Neurology. 2009;73(21):1738–1745.
38. Martinez-Martin P, Rodriguez-Blazquez C, Alvarez-Sanchez M, et al. Expanded and independent validation of the Movement Disorder Society-Unified Parkinson's Disease Rating Scale (MDS-UPDRS). J Neurol. 2013;260(1):228–236.
39. Steffen T, Seney M. Test-retest reliability and minimal detectable change on balance and ambulation tests, the 36-item short-form health survey, and the unified Parkinson disease rating scale in people with parkinsonism. Phys Ther. 2008;88(6):733–746.
40. Langford Z. The Four Square Step Test. J Physiother. 2015;61(3):162.
41. Franchignoni F, Horak F, Godi M, Nardone A, Giordano A. Using psychometric techniques to improve the balance evaluation systems test: the mini-BESTest. J Rehabil Med. 2010;42(4):323–331.
42. Schoene D, Lord SR, Verhoef P, Smith ST. A novel video game-based device for measuring stepping performance and fall risk in older people. Arch Phys Med Rehabil. 2011;92(6):947–953.
43. Chiviacowsky S, Campos T, Domingues MR. Reduced frequency of knowledge of results enhances learning in persons with Parkinson's disease. Front Psychol. 2010;1:226.
44. Dobkin BH, Plummer-D'Amato P, Elashoff R, Lee J. International randomized clinical trial, stroke inpatient rehabilitation with reinforcement of walking speed (SIRROWS), improves outcomes. Neurorehabil Neural Repair. 2010;24(3):235–242.
45. Sidaway B, Ahn S, Boldeau P, Griffin S, Noyes B, Pelletier K. A comparison of manual guidance and knowledge of results in the learning of a weight-bearing skill. J Neurol Phys Ther. 2008;32(1):32–38.
46. Reichmann H, Emre M. Optimizing levodopa therapy to treat wearing-off symptoms in Parkinson's disease: focus on levodopa/carbidopa/entacapone. Expert Rev Neurother. 2012;12(2):119–131.
47. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67(1):1–48.
48. Wickham H. Ggplot2: Elegant Graphics for Data Analysis. New York, NY: Springer-Verlag; 2009.
49. Ritz C, Baty F, Streibig JC, Gerhard D. dose-response analysis using R. PLoS One. 2015;10(12).
50. Kuznetsova A, Brockhoff PB, Christensen RHB. lmerTest package: tests in linear mixed effects models. J Stat Softw. 2017;82(13):26.
51. Long JD. Longitudinal Data Analysis for the Behavioral Sciences Using R. Thousand Oaks, CA: Sage; 2012.
52. Peterson DS, Dijkstra BW, Horak FB. Postural motor learning
in people with Parkinson's disease. J Neurol. 2016;263(8):1518–1529.
53. Park H, Schweighofer N. Nonlinear mixed-effects model reveals a distinction between learning and performance in intensive reach training post-stroke. J Neuroeng Rehabil. 2017;14(1):21.
54. Kantak SS, Winstein CJ. Learning-performance distinction and memory processes for motor skills: a focused review and perspective. Behav Brain Res. 2012;228(1):219–231.
55. Kim S, Oh Y, Schweighofer N. Between-trial forgetting due to interference and time in motor adaptation. PLoS One. 2015;10(11):e0142963.
56. Guadagnoli MA, Lee TD. Challenge point: a framework for conceptualizing the effects of various practice conditions in motor learning
. J Mot Behav. 2004;36(2):212–224.
57. Hayes HA, Hunsaker N, Dibble LE. Implicit motor sequence learning in individuals with Parkinson disease: a meta-analysis. J Parkinsons Dis. 2015;5(3):549–560.
58. Driskell JE, Willis RP, Copper C. Effect of overlearning on retention. J App Psychol. 1992;77(5):615–622.
59. Jones MB. Nonimposed Overpractice and Skill Retention. ARI Research Note 86-55. Alexandria, VA: US Army Research Institute for the Behavioral and Social Sciences; 1986:28.
60. Chasey WC, Knowles CJ. Effects of overlearning on retention and relearning of gross-motor skill by mentally retarded males. Percept Motor Skills. 1973;36(2):503–509.
61. Forsyth AL, Paul SS, Allen NE, Sherrington C, Fung VSC, Canning CG. Flexed truncal posture in Parkinson disease. J Neurol Phys Ther. 2017. doi:10.1097/NPT.0000000000000171