The effect of rehabilitation interventions on freezing of gait in people with Parkinson’s disease is unclear: a systematic review and meta-analyses

Abstract Purpose To summarize the effects of rehabilitation interventions to reduce freezing of gait (FOG) in people with Parkinson’s disease. Methods A systematic review with meta-analyses of randomized trials of rehabilitation interventions that reported a FOG outcome was conducted. Quality of included studies and certainty of FOG outcome were assessed using the PEDro scale and GRADE framework. Results Sixty-five studies were eligible, with 62 trialing physical therapy/exercise, and five trialing cognitive and/or behavioral therapies. All meta-analyses produced very low-certainty evidence. Physical therapy/exercise had a small effect on reducing FOG post-intervention compared to control (Hedges’ g= −0.26, 95% CI= −0.38 to −0.14, 95% prediction interval (PI)= −0.38 to −0.14). We are uncertain of the effects on FOG post-intervention when comparing: exercise with cueing to without cueing (Hedges’ g= −0.58, 95% CI= −0.86 to −0.29, 95% PI= −1.23 to 0.08); action observation training plus movement strategy practice to practice alone (Hedges’ g= −0.56, 95% CI= −1.16 to 0.05); and dance to multimodal exercises (Hedges’ g= −0.64, 95% CI= −1.53 to 0.25). Conclusions We are uncertain if physical therapy/exercise, cognitive or behavioral therapies, are effective at reducing FOG. Implications for rehabilitation FOG leads to impaired mobility and falls, but the effect of rehabilitation interventions (including physical therapy/exercise and cognitive/behavioral therapies) on FOG is small and uncertain. Until more robust evidence is generated, clinicians should assess FOG using both self-report and physical measures, as well as other related impairments such as cognition, anxiety, and fear of falling. Interventions for FOG should be personalized based on the individual’s triggers and form part of a broader exercise program addressing gait, balance, and falls prevention. Interventions should continue over the long term and be closely monitored and adjusted as individual circumstances change.


Introduction
Freezing of gait (FOG), defined as a brief, episodic, absence or marked reduction of forward progression of the feet despite the intention to walk, is a common phenomenon in people with Parkinson's disease (PwPD) [1]. Its incidence is about 50% and increases with disease duration and severity [2][3][4]. FOG is also known to impair mobility, increase risk of falls, and reduce healthrelated quality of life, resulting in significant health care costs [5][6][7][8].
Common interventions for FOG largely rely on pharmacological and surgical options [9]. Treatments are typically started when FOG emerges in the "off" state. However, as disease progresses, FOG emerges in both "on" and "off" states and may become less responsive to pharmacological and surgical interventions [10]. There is growing evidence for the use of rehabilitation interventions as an adjunct to medication to reduce FOG because FOG persists despite optimal medical intervention [10]. Rehabilitation interventions described in this review consist of non-pharmacological and non-surgical interventions "designed to optimize functioning and reduce disability in individuals with health conditions in interaction with their environment" [11]. Although the pathophysiological mechanisms underlying FOG remain unclear [12], there is emerging evidence to suggest rehabilitation interventions may be able to address underlying FOG pathology and are beneficial for FOG [13] and Parkinson's disease more globally [14,15]. Rehabilitation interventions for FOG may include cognitive, behavioral, and physical therapies as declining cognition, high levels of anxiety, and poor motor performance have established associations with FOG [12]. Several systematic reviews [16][17][18][19][20] have investigated the efficacy of rehabilitation interventions, with the most common being physical therapy [16,17,19,20]. Overall, physical therapy had a small effect on reducing FOG when compared to a control intervention. Specific physical therapy interventions that were suggested to reduce freezing were action observation training, treadmill training, and exercises consisting of cueing, movement strategies, and balance training [16,17]. However, effect sizes were small, and the clinical significance of these improvements remains unclear.
There are several limitations with the reviews published to date. First, three reviews were restricted to physical therapy only [16,19,20] and the one review of rehabilitation interventions more broadly searched within a limited timeframe (2013-2018) and did not report meta-analyses [18]. Second, the three reviews with meta-analyses [16,17,20] only reported effect sizes and 95% CI (i.e., the precision of the overall pooled effect), and did not report prediction intervals (PIs) (i.e., the dispersion or likely range of the true effect) so the true effect size in 95% of comparable studies were unknown [21]. Third, two reviews did not report the certainty of the evidence so results should be interpreted with caution [17,20].
To determine the most effective rehabilitation interventions to reduce FOG and gain insights into how such interventions should be delivered, a broader search of studies is required. Furthermore, PIs and certainty of the evidence should be reported to allow clinicians and researchers to interpret the pooled effect size and make decisions about treatment efficacy, selection, and application [21].
The objective of this review is to summarize the effects of rehabilitation interventions to reduce FOG in PwPD. Secondary aims include examining the effects of these interventions on gait, functional mobility, balance, balance-related confidence, activities of daily living, health-related quality of life, falls, and fear of falling.

Data sources and searches
A search of the CINAHL, Cochrane Library Central Register of Controlled Trials, EMBASE, MEDLINE, OTSeeker, PEDro, and Scopus electronic databases was conducted to 1 June 2021. Studies were included if they were randomized controlled or crossover trials (using data prior to crossover) with FOG reported as an outcome measure. The following search terms were used: Parkinson's disease, randomized, rehabilitation, physical therapy, occupational therapy, psychotherapy, cognitive therapy, and behavior therapy (Supplementary material 1 for full search strategy). There were no date or language restrictions.
Titles and abstracts were screened independently by two reviewers (LG and CC) to identify potential eligible studies. Full texts of these studies were assessed by two independent reviewers (LG and CC) to identify studies eligible for inclusion. Disagreements between reviewers over eligibility of studies were resolved through discussions with a third reviewer (NA). The reference lists of included trials were also reviewed to identify further eligible studies. This review was registered prospectively with PROSPERO (PROSPERO 2018 CRD42018116820).

Participants
Studies that recruited adults with idiopathic Parkinson's disease with FOG were included, even if they also included participants without FOG because most studies trialing rehabilitation interventions included participants with and without FOG. Studies that recruited adults with Parkinsonism or Parkinson's plus disorders were excluded.

Interventions
Studies were included if they reported rehabilitation interventions typically delivered by an allied health professional such as physical therapy, occupational therapy, psychotherapy, cognitive therapy, and behavioral therapy, compared with control (i.e., no intervention, usual care, or sham) or another rehabilitation intervention. For the purposes of this review, physical therapy was grouped with exercise (i.e., physical therapy/exercise) because of the significant overlap between these two types of interventions. Most physical therapy interventions for PwPD consist of exercise. However, exercise-based interventions may be prescribed by other instructors who are not physical therapists, e.g., dance instructors, yoga instructors, and exercise scientists/conditioning coaches. Only studies of interventions delivered for two weeks or more were analyzed in this paper to differentiate between studies examining immediate effects following a single intervention session and studies examining rehabilitation effects following a period of repetitive training.
Studies that examined pharmacological and surgical interventions such as deep brain stimulation, and interventions not usually prescribed by an allied health professional (e.g., repetitive transcranial magnetic stimulation, transcranial direct current stimulation) were excluded.

Data extraction and quality assessment
Data from the included studies were extracted by pairs of independent reviewers (LG, CC, NA, and JS Outcome data were extracted as follows: outcomes measured; time points; number of participants; medication state when outcomes were measured ("on" and/or "off"); and results at each time point. Where available, data were extracted as mean and standard deviation. If data were reported as median, range and/or interquartile range, the mean and standard deviations were calculated [51]. If standard errors or 95% confidence intervals were reported, standard deviations were calculated [52]. Data related to falls were extracted as rate ratio and risk ratio. If data were reported in a graph, it was extracted using the WebPlotDigitizer software [53]. Where follow up data were available, we extracted four to 16 weeks post intervention data. If there were multiple follow up data within this timeframe, the earlier timepoint was used.
The quality of each study was rated from 0 to 10 using the Physiotherapy Evidence Database (PEDro) Scale, with higher scores indicating higher quality [54]. PEDro scores were extracted from the PEDro website. Where a trial was not scored on the website, it was scored by two researchers independently (LG and JS), with any disagreements resolved with a third researcher (NA). All studies were included regardless of their PEDro scores.
The certainty of the FOG outcome was assessed using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) framework where the certainty is graded from very low to high [55]. Two researchers (LG and NA) scored the certainty of the evidence with any disagreements resolved by consensus with a third researcher (CC).

Data synthesis and analysis
Separate analyses were performed for studies comparing an intervention to control and studies comparing two interventions. If outcomes were reported by three or more comparable studies, data were pooled and reported as standardized mean differences (Hedges' g ± 95% CI). This reflects the precision of the pooled effect, i.e., how precisely the mean effect size has been estimated using effect sizes from studies in the meta-analyses [21]. Where data were pooled, the post intervention standard deviation was used to standardize the results. Random-effects models were used, with the heterogeneity reported as the ratio of true to observed variance (I 2 ). The dispersion of the pooled effect was reported as 95% PIs. This reflects how the effect size varies across studies, i.e., the true effect in 95% of comparable studies if the entire population could hypothetically be enrolled in the study [21]. The PI reports values on the same scale as the pooled effect size. If heterogeneity was unlikely to be important (i.e., I 2 ¼0%), the 95% PI was not shown as all studies shared a common effect size.
A sensitivity analysis was conducted for FOG to explore the effect of removing any studies responsible for high heterogeneity. To estimate mean difference of an outcome, the standardized mean difference was multiplied by the standard deviation of the study with the largest number of participants [52]. For studies comparing three groups (e.g., two interventions and one control), the sample size of the control group was equally divided so that participants were not counted twice for each pairwise comparison. A mixed effects model was used to compare the effect size in different subgroups. If there were 10 or more studies included in a meta-analysis, funnel plots were used to evaluate any publication bias.
Subgroup analyses were performed to explore the effect of type of exercise, presence of FOG in participants at baseline (yes/ no), and dose of intervention (min/week, dichotomized at the median) on FOG outcomes. We were unable to conduct preplanned subgroup analyses to explore the effect of disease severity, adherence, and cognition because we were unable to categorize these factors: there was a wide range of disease severity within most studies, adherence was overall high, and people with substantially impaired cognition were excluded. We conducted an additional post hoc subgroup analysis based on the type of interventionist (i.e., physical therapist versus another type of exercise instructor) to determine if there was any evidence of a differential effect between physical therapy and other forms of exercise.
The meta-analyses were completed using the software Comprehensive Meta-Analysis version 3 (Biostat 2013, Englewood, NJ) and 95% PI were calculated using the software Comprehensive Meta-Analysis Prediction Interval (Biostat 2021, Englewood, NJ).

Flow of studies through the review
Database searches identified 7734 potential records. After duplicates were removed, 4764 records were screened for eligibility. Of these records, full text of 535 reports were retrieved of which 65 studies met the inclusion criteria ( Figure 1).

Studies characteristics
Of

Interventions
Despite the broad eligibility criteria intended to capture a wide range of rehabilitation interventions, studies eligible for inclusion were from two major categories: physical therapy/exercise or cognitive and/or behavioral therapy. Sixty-two studies trialed physical therapy/exercise interventions, including seven dance [76,80,83,84,107,108,120], three yoga [59, 115,119], and two music therapy [98,103]. Two-thirds of the physical therapy/exercise interventions were delivered by a physical therapist, including two interventions that delivered physical therapy as a part of a multidisciplinary rehabilitation program. The remaining physical therapy/exercise interventions were delivered by a dance instructor, yoga instructor, exercise scientist/conditioning coach, physiatrist, or unspecified therapist/personnel. Five studies trialed cognitive and/or behavioral therapy typically delivered by a psychologist or neuropsychologist [81,88,97,121,126].
Studies in the physical therapy/exercise category were subgrouped according to the primary nature of the intervention of interest: multimodal exercises (multiple exercise types); cueing; dance; treadmill; action observation training plus physical practice of movement strategies to overcome FOG; physical practice of movement strategies to overcome FOG with or without sham observation training; strength; aquatic; gait training; yoga; balance; music therapy; Nordic walking; step training; Alexanderbased exercises; proprioception; Qigong; and trunk exercises. Subgroups for studies in the cognitive and/or behavioral therapy category were: cognitive training; Cognitive Behavioral Therapy; and Acceptance and Commitment Therapy (Supplementary material 4).
The dosage of the interventions was highly variable, with the frequency ranging from 10 to 240 min per day, 1-7 days per week, over 2-52 weeks.

Quality of included studies
The overall mean PEDro score of the included studies was 6.1 (Supplementary material 5).

Certainty of the pooled FOG outcomes (GRADE)
The overall certainty of the pooled FOG outcomes was very low, with ratings downgraded for serious or very serious risk of bias due to lack of concealed allocation, no assessor blinding, inadequate follow up, and/or no intention-to-treat. Ratings were also downgraded due to indirectness, as most studies recruited both people with and without FOG; and imprecision, as sample sizes were small and CI and/or PI were wide (Supplementary material 6). Funnel plot analyses did not show evidence of publication bias (Supplementary material 7).

Effect of physical therapy/exercise interventions versus control post intervention
Data were available from 21/24 included studies with 23 comparisons (n ¼ 1129) (Table 1) We are uncertain whether physical therapy/exercise reduces FOG as GRADE indicated the certainty of the evidence is very low. The pooled mean estimate showed a small to medium effect in favor of physical therapy/exercise interventions to reduce FOG (Hedges' g¼ À 0.39, 95% CI¼ À 0.60 to À 0.18), with substantial heterogeneity in the estimates (I 2 ¼64%, p < 0.001). However, the 95% PI indicated the true effect size in 95% of comparable studies lies between À 1.25 and 0.47, indicating physical therapy/exercise may have no benefit on FOG (Figure 2(a), see also Supplementary material 8, Figure 8a for the detailed plot).
Wr� oblewska et al. [122] (n ¼ 40) had a larger effect size than the other studies and was responsible for the substantial heterogeneity. When a sensitivity analysis was completed by removing this study, the pooled result continued to show a small effect in favor of physical therapy/exercise to reduce FOG (Hedges' g¼ À 0.26, 95% CI¼ À 0.38 to À 0.14, I 2 ¼0%), with 95% of comparable studies also sharing this effect (Figure 2(b), see also Supplementary material 8, Figure 8b for the detailed plot). This pooled estimate represented a mean difference of approximately À 1.1 points on the NFOGQ. Nonetheless, we remain uncertain whether physical therapy/exercise reduces FOG (GRADE very low certainty evidence).
Due to its significant contribution to heterogeneity and small effect on the result, Wr� oblewska et al. [122] was excluded from further analyses.
Three subgroup analyses were planned. Results indicated there were no effects of exercise type (Q ¼ 1.74, df ¼ 10, p ¼ 1.00) (Supplementary material 8, Figure 8c) Figure 8d). Given that studies in the meta-analyses included participants with and without FOG, a subgroup analysis was completed to determine if there were any differential effects on FOG between studies that included participants with FOG only and studies that included participants with and without FOG. Overall, there was also no difference between studies that recruited participants with FOG only (n ¼ 83) [90, 127,128] and studies that recruited people with and   Figure 8e). However, there were limited data from the FOG only subgroup. An additional subgroup analysis was completed post hoc to determine if there was a differential effect on FOG between studies that trialed physical therapy (i.e., intervention delivered by a physical therapist) and studies that trialed exercise (i.e., intervention delivered by another instructor); no differential effect was seen (Q ¼ 0.06, df ¼ 1, p ¼ 0.80).

Effect of physical therapy/exercise interventions versus control post intervention at follow up
Five studies contributed follow up data (n ¼ 267) [64,85,103,109,110] between 4 and 12 weeks post intervention. We are uncertain whether physical therapy/exercise reduces FOG in the longer term (Hedges' g¼ À 0.23, 95% CI¼ À 0.48 to 0.02, I 2 ¼0%, GRADE very low certainty evidence) (Supplementary material 8, Figure 8f).

Effect of one rehabilitation intervention versus another rehabilitation intervention post intervention
Comparable studies comparing the effect of one rehabilitation intervention to another, and reporting FOGQ/NFOGQ data, were pooled as follows: six studies (n  (Table 2). We are uncertain if exercise with cueing reduces FOG compared to exercise without cueing. While there was a medium effect size in favor of cueing (Hedges' g¼ À 0.58, 95% CI¼ À 0.86 to À 0.29, I 2 ¼28%), the effect in 95% of comparable studies included no benefit (95% PI¼ À 1.23 to 0.08) (Supplementary material 8, Figure 8g) and the certainty of the evidence was very low. We are also uncertain if action observation training plus physical practice reduces FOG compared to physical practice studies comparing physical therapy or exercise interventions to control. The standardized mean differences and 95% confidence intervals of 20 studies fall closely around the vertical line representing no effect, with results from one study from Wr� oblewska et al. [122] strongly favoring the intervention group. The pooled standardized mean difference shows a small effect in favor of intervention; however, the prediction interval crosses zero. (b). Forest plot showing the effect on freezing of gait from 20 studies comparing physical therapy or exercise interventions to control after removing the study by Wr� oblewska et al. [122] which was responsible for substantial heterogeneity as part of a sensitivity analysis. Removing this study did not substantially alter the result. alone (Hedges' g¼ À 0.56, 95% CI¼ À 1.16 to 0.05, I 2 ¼62%, 95% PI¼ À 3.01 to 1.89, GRADE very low certainty evidence) (Supplementary material 8, Figure 8h) and if dance reduces FOG compared to multimodal exercises (Hedges' g¼ À 0.64, 95% CI¼ À 1.53 to 0.25, I 2 ¼74%, 95% PI¼ À 10.98 to 9.70, GRADE very low certainty evidence) (Supplementary material 8, Figure 8i).
Data from the remaining studies comparing one rehabilitation intervention to another (including physical therapy/exercise, Cognitive Behavioral Therapy, and Acceptance and Commitment Therapy) could not be pooled (Supplementary material 3) as the types of intervention prescribed and their effects on freezing outcomes were highly variable, giving no clear indication of any intervention being more effective than another. Their results are summarized in Supplementary material 9, Figure 9a.

Effect of cognitive training versus control post intervention
There were insufficient comparable outcomes from three studies of cognitive training [88,97,121] compared to control, to pool in a meta-analysis (Supplementary material 2) and results are summarized in Supplementary material 9, Figure 9b. Computerized cognitive training showed a large reduction in "on" state FOG (measured using percentage of time spent frozen) in a subset of participants where freezing was triggered during physical testing at baseline [121]. The other two studies did not show an effect on FOG [88,97].

Effect of physical therapy/exercise interventions versus control post intervention
The pooled mean estimates of the effects, including PIs, are presented for the secondary outcomes in Table 3, with the detailed forest plots presented in Supplementary material 8, Figure 8l to 8v. There was a small effect of physical therapy/exercise on: increasing comfortable walking speed (Hedges' g ¼ 0.23, 95% CI ¼ 0.04 to 0.43, I 2 ¼0%); increasing stride length (Hedges' g ¼ 0.30, 95% CI ¼ 0.04 to 0.55, I 2 ¼0%); reducing dual task TUG (Hedges' g¼ À 0.26, 95% CI¼ À 0.51 to À 0.02, I 2 ¼0%); and reducing fear of falling (Hedges' g¼ À 0.25, 95% CI¼ À 0.37 to À 0.13, I 2 ¼0%). These effect sizes represented an increase in comfortable walking speed by 0.05 m/s; increase in stride length by 0.07 m; a reduction of 2.1 s on dual task TUG; and reduction of 2.9 points on the FES-I.
There were small to medium effect sizes in favor of physical therapy/exercise on improving single task TUG, balance, and activities of daily living; however, PIs indicated the true effects in 95% of comparable studies included no benefit. Physical therapy/exercise did not improve fast walking speed, health-related quality of life, or falls. There were insufficient data to pool balance-related confidence measures.

Effect of one rehabilitation intervention versus another rehabilitation intervention post intervention
There were medium to large effect sizes in favor of exercise with cueing versus without cueing, on single task TUG, balance, and activities of daily living, but PIs indicated the effects in 95% of comparable studies included no benefit (Supplementary material 8, Figure 8w to 8y). Action observation training did not improve balance compared to physical practice alone (Supplementary material 8, Figure 8z).

Discussion
This systematic review with meta-analyses is the most comprehensive review of rehabilitation interventions for FOG to date and provides evidence that the effect of rehabilitation on FOG is unclear. We  . Overall, we found uncertainty about whether rehabilitation interventions reduce FOG post intervention when compared to control (i.e., no intervention, usual care, or sham), or another intervention. This contrasts with previous reviews which have interpreted their results more favorably.
Our effect size in favor of physical therapy/exercise compared to control (Hedges' g¼ À 0.26, 95% CI¼ À 0.38 to À 0.14, I 2 ¼0%) is consistent to previous reviews [16,17]. However, we interpret this result very cautiously as the certainty of the evidence is very low. This effect size only represents a change of 1.1 points on the NFOGQ. This reduction is unlikely to be meaningful as it is well within the measurement error for the NFOGQ, which has a minimal detectable change of 10 points on a 28-point scale [130]. Our subgroup analyses of physical therapy/exercise showed no differential effects of exercise types and intervention time on reducing FOG. Overall, the small effect of physical therapy/exercise interventions on reducing FOG and the lack of differential effects from the subgroup analyses suggest high-efficacy FOG interventions and relationships between FOG and intervention dosage are not yet identified.
Of the 40 studies which compared two intervention types, we were only able to pool studies trialing exercise with cueing versus without cueing (n ¼ 6), action observation training plus physical practice versus physical practice alone (n ¼ 4), and dance versus multimodal exercises (n ¼ 3). The only comparison that showed a pooled effect in favor of the intervention of interest was exercise with cueing versus without cueing. However, the true effect in 95% of comparable studies included no effect. At follow up, the effect of cueing is promising but the certainty of the evidence is very low. The lack of clear evidence may be due to the wide range of cueing strategies used, and limited evidence of cueing type and frequency being tailored to the individual.
In contrast to a previous review [16], the pooled effect for action observation training plus physical practice of movement strategies compared to physical practice alone did not favor either intervention. We believe our meta-analysis to be more robust as FOG at baseline was considered when calculating effect Table 3. Standardized mean difference, 95% confidence interval (CI), heterogeneity (I 2 ), and 95% prediction interval (PI) of effect of physical therapy/exercise interventions compared to control on secondary outcomes post intervention. sizes. Nonetheless, further research into the use of action observation training is warranted as it has potential to address both motor and non-motor aspects of FOG and has been shown to increase activation of frontoparietal mirror neurons that facilitate motor learning and movement [131]. The pooled effect for dance compared to multimodal exercises also did not favor either intervention. To our knowledge, this is the first meta-analysis comparing these two types of interventions, with previous analyses comparing dance to control also showing no effect [132][133][134].
The overall small effect sizes with wide PIs may be a result of intervention effects being diluted by participants who did not have FOG at baseline. Although Gilat et al. [17] provided preliminary data to support this hypothesis, this proposal could not be confirmed since the effect of presence of FOG at baseline was limited by a small number of participants in this subgroup. Furthermore, physical therapy/exercise interventions in this review primarily addressed motor aspects of freezing. Although the pathophysiology underlying FOG is complex and remains poorly understood, freezing is also known to be associated with non-motor impairments such as anxiety, cognitive decline, and sensory-perceptual disorders, as well as environmental factors [135][136][137][138].
Our review is the only review of the effect of rehabilitation interventions on FOG to include secondary gait, functional mobility, balance, balance-related confidence, activities of daily living, health-related quality of life, falls, and fear of falling outcomes. Physical therapy/exercise interventions eligible for this review had a small effect in favor of reducing fear of falling post intervention compared to control. Given that people with FOG are at higher risk of falls, it is possible that people with FOG are less fearful and anxious once they have developed strategies to manage their freezing. The pooled mean estimate in our meta-analysis represented a change of À 2.9 points on the FES-I (range 16-64) but the clinical significance of this reduction remains unclear as the minimal detectable change for the FES-I has not been reported in PwPD. There were also small effects in favor of physical therapy/ exercise interventions to improve comfortable walking speed, stride length, and dual task TUG. This is unsurprising as the exercise types included in the meta-analyses included gait, step and balance training, cueing, dual task training, and lower limb strengthening, which are likely to improve these outcomes. Although there was no evidence of an effect on other secondary outcomes and balance-related confidence measures were not able to be pooled, other reviews examining the effect of physical therapy/exercise on gait and balance with broader selection criteria report improvements in gait, balance [139], and falls [140].
With growing evidence of the contribution of non-motor impairments to FOG [135,141], a small number of studies examining the effect of cognitive and/or behavioral therapy on FOG were eligible for this review. At present, there is insufficient evidence to determine if cognitive and/or behavioral therapy are effective in reducing FOG.
FOG is a major concern for PwPD, a significant contributor to reduced mobility and falls, and must be addressed. With the lack of clear evidence about effective interventions, it is recommended that rehabilitation is comprehensive and personalized. This includes assessment of FOG using both self-report and physical measures, accounting for cognition, anxiety, and fear of falling. Although untested to date, the Characterizing Freezing of Gait Questionnaire [142] may be a useful tool to identify FOG triggers and possible strategies to manage FOG. A proposed framework to assist physical therapists in selecting appropriate FOG interventions suggests exercises relevant to FOG such as gait, balance and dual task training may be prescribed for individuals with occasional FOG, and exercises specifically aimed at the immediate alleviation of FOG such as cueing and movement strategies should be prescribed for individuals with frequent FOG [17]. Based on the currently available evidence, use of individually tailored cueing as part of an exercise program addressing gait, balance, and falls prevention should be considered and the effects of interventions closely monitored [143,144]. Given that this review and others have shown little to no maintenance of effects at follow up, interventions for FOG should be implemented and sustained for the long term.
Future research trials should explore rehabilitation interventions that address both motor and non-motor aspects of FOG using a cognitive and/or behavioral approach in addition to motor rehabilitation, to maximize the potential for reducing FOG [126]. It is likely that personalized rehabilitation for FOG, where interventions are individually tailored to the PwPD's specific FOG profile, and are compared to inactive control interventions, will optimize results [17]. In addition, trials should only include participants with FOG, or report subgroup data separately for people with and without FOG. Trials should also consider using physical FOG measures in addition to self-report measures, as the use of FOGQ and NFOGQ alone may not be adequate in capturing FOG severity. Although the FOGQ and NFOGQ were reported to be reliable and valid [22,23], outcomes from the NFOGQ did not correlate to duration or frequency of FOG episodes from video analyses of FOG-provoking physical tests [145]. Furthermore, the NFOGQ may not be suitably responsive to detect changes in FOG over time as a recent study reported the minimal detectable change for the NFOGQ to be 10 points on a 28-point scale [130]. Rehabilitation interventions specifically targeting PwPD with more severe disease and/or cognitive impairment also need to be developed and tested.

Strengths and limitations
This review adds to existing knowledge from previous reviews by extending its scope beyond physical therapy and reporting PIs. However, this review also has its limitations. Many studies that did not have an explicit aim to reduce FOG were included in this review provided FOG was reported as an outcome. In addition, there were many studies excluded from meta-analyses due to the unavailability of FOG data or insufficient comparable interventions or outcomes from the wide range of interventions and measures used to target and measure FOG. It was also not possible to discriminate the effects of interventions on FOG during "on" or "off" states as the only FOG outcome that was able to be pooled were from the FOGQ and NFOGQ, which were designed to assess FOG over an entire day and in the past week during both "on" and "off" states.

Conclusions
Using highly robust methodology, our results suggest the effects of rehabilitation interventions are less favorable compared to conclusions from previously published reviews. We are uncertain whether rehabilitation interventions reduce FOG. While there was a small, pooled effect in favor of physical therapy/exercise compared to control, the certainty of this evidence is very low, and its effect is unlikely to be clinically meaningful. Additionally, we are uncertain if exercise with cueing, action observation training and dance are more effective at reducing FOG compared to exercise without cueing, physical practice of movement strategies alone, or multimodal exercises, respectively. The small number of studies focusing on cognitive and/or behavioral therapy are inconclusive. The development and testing of rehabilitation interventions that are individually tailored to the PwPD's specific FOG profile, and measures of FOG that are responsive and easily implemented are urgently needed to advance knowledge in this area.