Neurodevelopmental Therapy for Cerebral Palsy: A Meta-analysis

Our objective was to determine the efficacy of NDT for any outcome in children and infants with CP and infants with a high risk of CP.

We hypothesized that (1) NDT is no more effective in improving outcomes in children and infants with CP than no intervention or passive approaches (hereby termed “control”), (2) NDT is less effective than activity-based approaches to improve motor function, (3) NDT is less effective than body function and structures-based approaches for body function and structures outcomes, (4) NDT is less effective than environment-based approaches on environment-based outcomes, (5) higher-dose NDT is no more effective than lower-dose NDT, and (6) NDT is not effective for improving motor function in infants with CP or risks for CP.

A systematic review and meta-analysis was conducted by using Cochrane methodology²⁶  and reported by using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.²⁷ 

Inclusion criteria using population, intervention, comparison, and outcome were (1) CP or high risk of CP based on diagnostic risk markers consistent with the era and >75% of participants <18 years, (2) intervention specified as Bobath, NDT, neurodevelopmental therapy, or “treatment,” (3) any comparison, (4) any outcome, (5) randomized controlled trial (RCT), and (6) published in English. No limits were placed on outcomes or date of publication. Exclusion criteria were (1) NDT versus NDT plus adjunct therapy (2) intervention not described as Bobath, NDT, neurodevelopmental therapy or treatment, (3) comparison of NDT at equal total dosage hours, but different frequencies, and (4) gray literature and conference abstracts. Search terms are presented in Supplemental Table 2.

Cumulative Index to Nursing & Allied Health Literature, Cochrane Library, Embase, and Medline databases were searched to March 10, 2021, with no lower date range. Studies were screened for eligibility by title and abstract by the first author and checked by 2 reviewers. Hand searching included the scanning of reference lists of all identified systematic reviews. Study characteristic extraction was divided between the authors and checked by the first author. Study quality was assessed at a domain level by using Cochrane Risk of Bias tool 2 (RoB-2)²⁸  by 2 independent raters (divided between authors) and discrepancies were resolved by a third independent rater.

Coding for synthesis and meta-analysis was decided by group consensus. The International Classification of Function, Disability and Health, Children and Youth Version¹⁸  was used to code comparison interventions into activity, body structures and function, and environmental-based groups. Additional comparators considered were NDT versus control, higher- versus lower-dose NDT, and NDT versus any intervention in infants. Active (child-generated and -controlled) and passive (therapist-generated and -controlled) components of NDT and comparator interventions were extracted. Motor function was defined as motor development and the acquisition of motor skills.²⁹ 

Inclusion criteria for meta-analysis included (1) 2 or more studies revealing similar outcomes, (2) mean and standard deviation outcome score and participant number could be extracted or calculated, (3) comparable outcome timeframe (long-term versus short-term), and (4) sufficient homogeneity of participants, interventions and outcomes to provide a meaningful summary.²⁶  Meta-analysis exclusion criteria were (1) no common outcome and for dose comparator only (2) both groups of NDT <30 hours dosage to exclude underdosed studies. This dosage threshold was based on evidence suggesting 30 to 40 hours of activity-based interventions are required to improve motor function.^20,30 

Meta-analyses were conducted by using Review Manager 5.4.1. Estimates of effect were assessed by using the standardized mean difference between comparison groups of posttest or change scores, with change scores preferentially used in meta-analyses. Effect size was considered small (0.2), moderate (0.5), large (0.8) and very large (1.3).³¹  Random effects were used because of the varied nature of outcome measures with 95% confidence intervals (CI) for certainty and I² for heterogeneity. Sensitivity analysis using fixed effects was also conducted. Clinically relevant subgroup analyses were conducted.

When mean and standard deviation were not available in text or from other systematic reviews, where possible, missing values were calculated by using REVMan using other variables. PlotDigitizer (version 2.6.9) was used to extract data from figures. If data were not retrievable using these methods, studies were excluded from the meta-analysis. If meta-analyses had considerable heterogeneity (I² >75%)²⁶  subanalyses were conducted to determine the heterogeneity source. Risk of publication bias was assessed by (1) visually inspecting funnel plots of each comparator meta-analysis for symmetry and (2) if >10 papers were included in a meta-analysis; Egger’s test was used to assess potential publication bias. Additionally, RoB-2 tool Domain 5 was used to assess reporting bias risk. The quality and strength of recommendations were further evaluated by using Grading of Recommendations Assessment, Development and Evaluation (GRADE).³²  A protocol was not prepared.

In total, 667 studies were screened and 35 studies met inclusion criteria with 2 studies^33,34  written about a single cohort (Fig 1). Studies included 1332 participants (n = 578 infants <2 years, n = 754 children ≥2 years). Of these, 21 studies (including 667 participants) met inclusion for meta-analysis (Fig 1).²⁷  Participants varied in terms of motor severity, motor type, and topography, reflecting known heterogeneity of children with CP.

Of 35 studies included, 11^4,35–44  were assessed as low risk of bias, 10^{14,33,34,45–51}  were assessed as having some concerns, and 14^52–65  were assessed as high risk of bias. Domain level and overall risk of bias for studies are presented in Fig 2 and Supplemental Figs 4 and 5.

Results of all studies are reported in Table 1. Of 35 studies, 6 of 35 (17%) favored NDT, 2 of 35 (6%) partially favored NDT, and 27 of 35 (77%) did not favor NDT (Table 1). Of 6 reports that favored NDT, 4 were assessed as high risk of bias. Active and passive components of interventions are described in Table 1. Motor function was the most common reported outcome (n = 28 studies)^{4,14,34–39,41–44,46,49–52,54–63,65}  followed by general development (n = 4),^34,50–52  reflex status (n = 4),^55,57–59  cognition (n = 3),^41,57,62  social development (n = 3),^41,50,60  mobility/walking capacity (n = 3),^14,39,40  quality of upper limb movement (n = 2),^35,44  muscle length (n = 2),^55,56  strength (n = 2),^46,47  tone/spasticity (n = 2),^58,59  mechanical efficiency (n = 2),^36,38  sitting function (n = 2),^48,61  and self-care (n = 2).^49,60 

We identified 9 studies (in 10 publications)^{33–35,45,52–57}  comprising 418 participants (Table 1) that compared NDT against a control. Controls consisted of no therapy (6 studies in 7 publications)^{33,34,45,53,55–57}  or passive movement approaches with no child self-generated movements (n = 3).^35,52,54  In total, 6 publications did not meet inclusion for meta-analysis for the following reasons: (1) data not extractable (n = 4),^53,55–57  (2) no common outcome (n = 1),⁴⁵  and (3) outcome timeframe not comparable (n = 1).³³  Four publications^34,35,52,54  (n = 138 participants), 2 of which had high risk of bias,^52,54  met inclusion criteria for meta-analysis for motor function outcome (Fig 2A). No difference was found between NDT and control for motor function with a pooled effect size of 0.13 (95% CI = −0.20 to 0.46), I² = 0%.

We identified 9 studies^{14,37–41,48,49,61}  comprising 307 participants (Table 1) for NDT versus activity-based approaches. In total, 8 studies^{14,37–41,48,49}  comprising 255 participants met inclusion for meta-analysis for motor function outcomes. A moderate pooled effect size of 0.76 (0.12 to 1.40), I² = 80% was observed, favoring activity-based approaches for motor function (Fig 2B). One outlier⁶¹  was excluded from meta-analysis because the intervention was primarily activity-based but used facilitation techniques. A meta-analysis with this outlier included is available in Supplemental Fig 4A. On subanalysis to account for heterogeneity, when studies with norm-referenced outcome measures^14,41,49  were excluded, both heterogeneity and pooled effect size decreased (0.42 [−0.08 to 0.93], I² = 36%) favoring activity-based interventions (Supplemental Fig 4B).

Subanalysis of 2 studies^37,39  (n = 38 participants) that tested upper limb interventions and were analyzed for upper limb motor outcomes revealed activity-based approaches improved upper limb motor function over NDT with a large, pooled effect of 0.83 (0.16 to 1.50) I² = 0% (Supplemental Fig 4C). However, there was no difference of effect for lower limb interventions for lower limb motor function outcomes from 4 studies^14,38–40  (n = 165 participants, outlier⁶¹  excluded), with a pooled effect of 0.68 (−0.18 to 1.54), I² = 82% (Supplemental Fig 4D). No change in results were seen when norm-referenced outcome measures^14,39  were excluded accounting for heterogeneity in lower limb motor function (0.03 [−0.39 to 0.46], I² = 0%) (Supplemental Fig 4E). All studies were rated low risk of bias or some concerns, except for the outlier⁶¹  which was assessed as high risk of bias.

We identified 7 studies^{36,46,47,58–60,65}  with 175 participants (Table 1) that tested NDT versus body function and structures-based approaches. Only 1 study⁶⁰  did not meet inclusion criteria for any meta-analyses in this comparator because data were not extractable. Motor function was the most common outcome reported. In total, 5 studies^{36,46,58,59,65}  met inclusion for a motor function outcome meta-analysis, with a moderate pooled effect size of 0.77 (0.19 to 1.35), I² = 61% favoring body function and structures approaches (Fig 2C). Other outcomes were muscle strength, muscle tone, and primitive reflex status. In total, 2 studies were combined for strength outcomes^46,47  with no effect found (0.02 [−0.56 to 0.61]) (Supplemental Fig 4F). In addition, n = 2 studies,^58,59  which tested the neurofacilitation of developmental reaction (NFDR) approach, were combined and analyzed for effect on muscle tone and primitive reflex status. NFDR was favored for improving muscle tone outcomes (0.97 [0.43 to 1.51]) (Supplemental Fig 4G); however, no effect was found for primitive reflex status (0.35 [−0.16 to 0.86]) (Supplemental Fig 4H). Notably, both these studies^58,59  were assessed as high risk of bias.

We identified 2 studies^62,63  comprising 47 participants who were all infants that compared NDT to environmental-based approaches. One study⁶³  partially favored NDT over a nonspecific handling intervention and 1 study⁶²  favored a parent coaching model over NDT. Comparison interventions of studies, parent coaching, and passive handling were not similar enough for meaningful meta-analysis. Additionally, both studies had a high risk of bias.

We identified 7 studies^{4,42–44,50,51,64}  with 385 participants that compared 2 doses of NDT. In total, 3 studies^4,44,51  met inclusion for meta-analysis for motor function outcomes. Average low dose was 35 (standard deviation = 3) hours versus 90 (36) hours high dose. Reasons for exclusion from meta-analysis included (1) data not extractable (n = 2 studies)^43,50  and (2) dose <30 hours threshold (n = 2).^42,64  No difference between higher and lower dose NDT on motor function was found with an effect size of 0.32 (−0.11 to 0.75), I² = 30% (Fig 2D). Interpretation of all meta-analyses did not change using a fixed effects model.

Across all included studies, we identified 11^{33,34,41,50,51,57,58,60,62–64}  publications that enrolled infant participants <2 years, including 2 articles^,34  written about the same cohort. These infant studies comprised 578 infants, of which 259 (44.8%) had confirmed CP. Studies were spread across all comparators including NDT versus control (n = 3),^33,34,57  NDT versus activity-based approaches (n = 1),⁴¹  NDT versus body function and structures-based approaches (n = 2),^58,60  NDT versus environment-based approaches (n = 2),^62,63  and NDT lower versus higher dose (n = 3).^50,51,64  In total, 8 studies^{33,34,41,51,58,62–64}  had extractable standardized mean difference data; however, no 2 studies were sufficiently homogenous in terms of participants, interventions, outcomes, or outcome measures to conduct a meaningful meta-analysis. Of 11 infant studies, 2 of 11 (18%) favored NDT, 1 of 11 (9%) partially favored NDT, and 7 of 11 (64%) did not favor NDT (Table 1). When studies with some concerns or a high risk of bias were excluded, only 1 study remained⁴¹  that revealed infant stimulation, and an activity-based approach improved motor skills over NDT with a large effect size (2.62 [1.83 to 3.41]) (Fig 2B).

On RoB-2 only 5 of 35 publications^{52,53,57,62,63}  had high risk and 5 of 35^{14,45,48,49,60}  had some concerns of reporting bias (assessed on Domain 5 of RoB-2), meaning that, overall, there was a low risk of reporting bias among the included studies. A funnel plot of each meta-analysis was visually inspected, with no asymmetry detected for any comparator, indicating a low risk for reporting bias.

Overall, based on GRADE methodology, the body of evidence was rated as moderate-quality. Recommendations for the use of NDT for children and infants with CP are (1) strong recommendation for the use of activity-based approaches in preference to NDT for improving motor function (with decisions informed by evidence certainty, benefits versus lost opportunity, family and clinician preferences, cost-effectiveness) and (2) strong recommendation against the use of NDT at any dose for improving motor function (with decisions informed by evidence certainty, benefit versus lost opportunity, cost-effectiveness, parent preference, feasibility, and family burden). Evidence and reasoning for the strength of recommendations are detailed in the Supplemental Material 1. The effect size and comparative effectiveness of each meta-analysis are depicted in a bubble chart traffic map using GRADE³²  recommendations and the Evidence Alert System²²  (Fig 3).

Our primary objective was to determine the efficacy of NDT on any outcome in children and infants with CP or high risk for CP. We found 35 RCTs evaluating NDT efficacy, which is a larger body of evidence than for most topics in CP rehabilitation. Despite the moderate quality of the evidence, the claim that insufficient evidence exists to judge the efficacy of NDT can be seriously challenged. We found no difference in effect between NDT and control or between different doses of NDT. In addition, activity-based and body function and structures-based approaches improved motor function over NDT with a moderate effect. Findings reveal top-down interventions are superior for improving motor function compared with bottom-up interventions.

We have made strong recommendations for the use of activity-based interventions in preference to NDT and against the use of NDT at any dose to improve motor outcomes in infants and children with CP. The certainty of evidence (moderate quality) in combination with the effect size of interventions, the value people place on effective motor interventions, costs, impact on health equity, acceptability, and feasibility of NDT versus alternative effective interventions using the GRADE methodological approach³²  were all considered.

It is worth discussing the exclusion of the outlier study⁶¹  from the NDT versus activity-based meta-analysis. In this study, the NDT intervention consisted of facilitation handling in combination with a range of activity-based elements, including trunk activation, reaching during standing or sitting activities. The comparison intervention consisted primarily of passive stretching with some active tasks therefore, not truly an activity-based approach that could be pooled with other studies. Results from the outlier study⁶¹  suggest a child-active approach is the effective element in neurorehabilitation.

Overall body functions interventions improved motor function over NDT, but the effect size was influenced by the studies with a high risk of bias, meaning this result should be interpreted with caution. Studies that compared NDT with other body function and structures-based approaches for muscle strength^46,47  used strength measures taken from single muscle groups; therefore results are unlikely to have functional implications. The 2 studies that compared NDT with body function and structures-based approaches for muscle tone outcomes^58,59  tested NDT versus the NFDR approach. Both studies had a high risk of bias based on invalid use of the Modified Ashworth Scale;⁶⁶  therefore the result that NFDR is superior to NDT for improving muscle tone should be interpreted with caution. The same 2 studies^58,59  revealed no difference between NDT and NFDR on primitive reflex status. This is not surprising given it is scientifically unlikely that distal manual interventions will impact central nervous system reflexes.

We only found 2 studies^62,63  that tested NDT versus environmental approaches. The low methodological quality of these studies meant no meaningful results could be extrapolated in this review.^23,62,63  However, another systematic review revealed environmental enrichment improves motor function in infants with CP with a small effect size (0.39 [0.05 to 0.72]).⁶⁷  Other effective interventions (ie, constraint-induced movement therapy)⁶⁸  also use enriched environments to induce repetition, practice, and learning. Because home environments are typically more enriching, we recommend practice in a child’s real-life environment.⁶⁹ 

A previous systematic review from 2001¹⁹  found NDT is not more effective in a higher than lower dose and our review support these results. One study,⁴  with a low risk of bias, is highly cited when advocating the effectiveness of NDT for improving motor function. However, on close examination, the CIs for the effect size (0.25 [−0.42 to 0.93]) cross the line of no effect suggesting no genuine statistical difference between doses.

Studies including infants <2 years revealed a range of outcomes and interventions across all comparators. Most infant studies did not favor NDT. Palmer and colleagues’ publication⁴¹  was the only infant study assessed as low risk of bias to reveal a clear effect that infant stimulation, an activity-based approach, improved motor outcomes over NDT. The Palmer study reflects an emerging trend in CP early interventions in which early, activity-based, top-down approaches are more effective than bottom-up approaches in improving motor function.^23,70,71  In addition, findings are consistent with results in older children with CP outlined in this systematic review. Intervention efficacy for infants is a well-known gap in the CP literature. Until recently, early detection of CP was not routine, leading to a lack of infant intervention trials. Now, with guidelines for accurately identifying infants with CP,²⁴  more studies testing interventions in infants with CP are underway.

Historically, it was argued that the evidence quality was too poor and evidence volume too small to adequately test NDT. The quality of studies varied in this review; 11 of 35 studies had a low risk of bias, and the overall body of evidence was rated moderate-quality by using GRADE.³²  However, we did find 35 RCTs with 1332 participants with outcomes principally pointing in the same direction. In addition, results in this review mirror recommendations for children⁷²  and adults⁷³  after stroke, which, although a different population, have similar mechanisms of action for neurorehabilitation.

In this review, ineffective interventions for motor function had the following common features: facilitation, handling and positioning to normalize or minimize atypical motor behavior (movement), tone and reflexes/reactions, passive stretching, sensory support, and vestibular input. In contrast, effective interventions had the following common features: learning through active self-generated movement, progressive, repetitive, and varied specific task practice, performance feedback, real-life meaningful goals, and child problem-solving, all of which are based on neuroplasticity principles.^74,75 

Most studies were published before Consolidated Standards of Reporting Trials-compliant RCT⁷⁶  reporting. The NDT versus control comparator included studies primarily from the 1970s and 1980s. It is no longer ethical to provide no therapy or passive comparison interventions. Studies were screened for “Bobath,” “NDT,” and “Neurodevelopmental therapy” or “treatment” search terms in titles and abstracts. It is likely studies that used NDT but described it as “physiotherapy,” “occupational therapy,” or “traditional therapy” in the full text may have been missed. However, given the volume of literature found and the use of hand searching, this is unlikely to have influenced our findings. Heterogeneity of the NDT versus activity-based approaches meta-analysis was considerable but accounted for by diverse outcome measures on subanalysis. Heterogeneity of infant studies is a limitation of the evidence; however, with likely effective interventions in this population,^71,77  a strong recommendation against the use of NDT in infants was made. Despite some studies with small sample sizes and dated methods being a limitation, it is still striking that most studies did not favor NDT. Lastly, excluding studies with <30 hours dose for the dosage comparator could be a limitation because there are some effective interventions with <30 hours dosage.

Results from this review support the deimplementation of NDT in clinical practice. Deimplementation requires evidence-based, multifaceted, and context-specific strategies,⁷⁸  which include policy changes, restructuring funding, and financing training for effective alternatives to NDT.⁷⁹  Multiple alternatives for NDT exist;²³  for example, task-specific training,⁸⁰  goal-directed training,⁸⁰  treadmill training,⁸¹  constraint-induced movement therapy,⁸²  action observation,³⁰  and bimanual therapy.^82,84–86   Clinical education in these effective alternatives is required because Bobath or NDT training is considered the best rehabilitation training in some countries.

We found activity-based interventions are more effective than NDT and no difference between higher and lower dose NDT, nor between NDT and control for improving motor function in CP. We give strong recommendations for the use of activity-based interventions in preference to NDT and against the use of NDT at any dose to improve motor function in CP. Deimplementation of NDT in clinical practice is warranted and will provide the best outcomes for children with CP.

We acknowledge Callum McEwan for his contribution to data extraction.

CI

confidence interval

CP

cerebral palsy

GRADE

Grading of Recommendations Assessment, Development and Evaluation

NDT

neurodevelopmental therapy

NFDR

neurofacilitation of developmental reaction

RCT

randomized controlled trial

RoB-2

Cochrane Risk of Bias-2 tool