Equine-Assisted Therapies for Children With Cerebral Palsy: A Meta-analysis

The protocol²⁹  of this systematic review and meta-analysis summarizes the prespecified methods in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA) guideline. When performing the review, we strictly adhered to this published protocol. In brief, results of randomized and nonrandomized controlled parallel-group or crossover studies were combined to evaluate the effect of equine-assisted therapies on motor capabilities as the primary outcome and QoL as a secondary outcome in children with CP.

Outcomes are summarized by mean difference (MD) or standardized mean difference (SMD) with 95% confidence intervals (CIs), P values, and estimates of the between-study variability τ² and I² to describe heterogeneity. More details of the methods used are provided in Supplemental Information.

A database search performed in May 2019 identified 1653 citations with potential for inclusion, and 2 additional studies^30,31  were identified from the reference lists. An updated search in February 2022 before the final analysis revealed 5 additional studies^{26,27,32–34}  that met our inclusion criteria. After title, abstract, and full-text screening, a total of 26 studies were selected to be included into the qualitative synthesis. Details of the selection process chosen to identify the randomized controlled trials to be included are depicted in a PRISMA flow diagram (Fig 1).

This meta-analysis reviews data from 26 studies on equine-related treatments conducted in 14 countries (Turkey, the United States, Egypt, Taiwan, Australia, Germany, Korea, Colombia, Spain, Poland, Japan, England, Thailand, and India) between 1998 and 2022. The study of Lerma-Castano,²³  which was excluded from the meta-analysis during bias analysis, is not contained herein or in the following sections, but details of this study are provided in the tables on bias assessment, study characteristics, and study data. The same applies to the studies of Lee et al¹⁵  because of the lack of a “standard of care” group. Also, the study of Park et al³³  was excluded from the meta-analysis because no other study reported on cardiac autonomic function. The studies partly differ considerably with regard to the number of patients, the duration of therapy, drop-out rates, severity of CP, and the tests used to describe outcome.

Overall, data from 755 patients were available for analysis. The male-to-female ratio could be calculated from 20 studies and was approximately 1.3:1. The lowest number of patients was reported in the studies of Benda et al¹⁷  and Silkwood-Sherer et al³⁴  (both n = 13), the highest in the study of Kwon et al¹¹  (n = 91). The duration of therapy varied between 1 day and 48 weeks.^17,18,35  So far reported, the severity of CP was mostly assessed by the GMFCS, less frequently by the modified Ashworth scale. Children with all types of CP were included, although most studies did not include the most severely affected children.

As mentioned above, we classified those horse-based studies as hippotherapy where the term hippotherapy was found in the title and/or text.^{8–12,17–19,22,23,25,30,33–35}  Fewer reported on the effects of therapeutic riding^{20,21,26,27,30,32}  or involved dynamic artificial horses.^14–16,31 

With the exception of 2 studies comparing living horses with artificial horses,^15,36  all remaining studies compared living horses or artificial horses with standard of care. In most studies, classic standard-of-care physiotherapy was applied to the treatment and to the control group.^{8–10,12,19–22,24,30,31,34}  In 3 studies, standard of care was discontinued in the treatment group or no data were provided,^{14,23,25–27,32,35}  whereas in the study of Kwon et al,¹¹  standard of care in the control group was supplemented by additional sports. In 2 of the 4 1-day studies, standard of care was applied by sitting on a barrel,^17,18  whereas sitting on a static artificial horse represented standard of care in the study of Quint et al.¹⁶ 

Two major types of outcomes were reported: outcomes focusing on motor abilities or QoL in general or those focusing on distinct motor or QoL subdomains.

Global motor function was assessed with the 66- and/or 88-item versions of the GMFM tool (n = 9).^26,37–39  This included 7 studies with living horses^{8,9,11,21,22,25,26,35}  and 2 studies with artificial horses, respectively.^14,31  If both GMFM measures were available, the GMFM-66 was preferentially included because “the GMFM-66 is a 66 item subset of the original 88 items identified through Rasch analysis to best describe the gross motor function of children with CP of varying abilities” (for details, see https://canchild.ca/en/resources/44-gross-motor-function-measure- gmfm).^37–40 

GMFM subtests can be used to describe distinct motor subdomains. Hereby, the GMFM-A subtest (lying) was reported in 4,^11,21,25,30  the GMFM-B (sitting) in 5,^{11,14,21,25,30}  the GMFM-C (crawling) in 4,^11,21,25,30  the GMFM-D (standing) in 6,^{9,11,21,22,25,30}  and the GMFM-E (walking) in 7 studies,^{9,11,21,22,25,30,35}  respectively.

Apart from the GMFM-B, all further subtests were only performed in studies on living horses. Results of further tests also evaluating capabilities during sitting were available from the modified functional reach test,²⁰  from force plate testing,¹⁰  from SAS testing,^14,24  and from GMFM-B data reanalyzed by Rasch analysis.³⁶ 

Effects of equine-assisted treatments on abilities during sitting and standing were assessed by means of the PBS.^{11,15,22,26,31}  One additional method to study the abilities during walking is gait analysis.^22,26,34,35  Hereby, a limited set of gait analyses (cadence, step length, and/or walking speed) could be included.

Further tools focusing on spasticity were the modified Ashworth scale,^12,20,21  trunk asymmetry,^17,18  as well as Formetric analysis¹⁹  and the hip range of movement.^16,22,27 

As for QoL assessment, we only included data on caregivers’ assessment of their children’s QoL. In this context, 2 measures investigated QoL in general, for example, the KIDSCREEN^8,9  and the Pediatric Evaluation of Disability Inventory.²⁵  QoL subscores were provided by the CP-QoL-Child,⁸  the Pediatric Evaluation of Disability Inventory,²⁵  the Pediatric Quality of Life Inventory,³²  and the Child Health Questionnaire.^8,9  Hereby, 3 comparable QoL subscores were available from at least 2 studies: a physical subscore and a subscore on social wellbeing,^8,9,25,32  and an emotional subscore.^8,9,32 

Further descriptions of the individual study characteristics are provided in Table 1.

Risk of bias varied from low to high risk of bias. Bias in randomized studies was assessed with the Risk of Bias 2 tool (RoB 2) and was judged in all randomized controlled trials but 1 with “some concerns”, mainly because information on randomization and concealment was insufficient, the proportions of missing data were different across groups, and the assessor of the subjective outcomes was not blinded.

Bias in nonrandomized studies was assessed with the Risk of Bias in Nonrandomized Studies of Interventions tool. Four^20,21,25,27  out of 8 studies were judged with serious risk of bias mainly due to potential confounding. Three studies^24,30,31  were evaluated with a moderate risk of bias. Because of critical selection bias, the study of Lerma-Castano et al²³  was excluded from the meta-analysis.

We provided a summary of the bias assessment within studies in Tables 2 and 3.

The following different types of treatment were distinguished: equine-assisted therapy, treatment based on living and artificial horses; hippotherapy, true hippotherapy as described above; therapeutic riding, living equine-assisted treatment not fulfilling the criteria of hippotherapy; living equine-assisted therapy, comprising both hippotherapy and therapeutic riding; and artificial horse, nonliving equine-assisted therapy using a horse-riding simulator.

Global motor function was assessed by the GMFM total score, applying either the GMFM-66-tool or the GMFM-88 tool. We found evidence of improved motor function measured by GMFM total scores among children with CP and usual care compared with those with additional equine-assisted therapy (hippotherapy = 6; therapeutic riding = 2; and artificial horse = 2) (SMD, 0.24; 95% CI, 0.05 to 0.43; P = .01; τ² = 0.00; I² = 15%; Fig 2). In addition, a significant treatment effect could be shown for the subdomain E of the GMFM score (hippotherapy = 5; and therapeutic riding = 2) (SMD, 2.68; 95% CI, 1.02 to 4.34; P = .002; τ² = 0.0; I² = 0%; Fig 3). In no other subdomain of the GMFM score evidence for an improved motor function could be proven (Figs 4–7).

Motor function assessed while sitting was not only evaluated with GMFM-B but also with the modified functional reach test,²⁰  the SAS,²⁴  force plate testing,¹⁰  and GMFM-B reanalyzed by Rasch analysis.³⁶  Combined analysis of all these data also failed to demonstrate a significant treatment effect (SMD, 0.42; 95% CI, 0.00 to 0.84; P = .0524; τ² = 0.29; I² = 79%; Fig 6). Motor function as assessed by the PBS describes motor abilities during sitting and standing and was reported by 5 studies. Evidence of treatment effect could be identified on PBS testing associated with high statistical heterogeneity (SMD, 0.67; 95% CI, −0.12 to 1.23; P = .02; τ² = 0.29; I² = 85%; Fig 8).

Motor function as assessed by gait analysis was reported by up to 4 studies on hippotherapy applying 3 different subtests (eg, step length, walking speed, and cadence. Hippotherapy compared with standard of care resulted in significantly higher step length (MD, 10.66; 95% CI, 2.73 to 18.59; P = .008; τ² = 28.21; I² = 84%; Fig 9) but not in improved walking speed (SMD, 2.34; 95% CI, −0.11 to 4.80; P = .06; τ² = 4.67; I² = 95%; Fig 10) or cadence (MD, 2.22; 95% CI, −25.31 to 29.75; P = .87; τ² = 332.39; I² = 84%; Fig 11).

No evidence of improved motor function measured by different spasticity parameters was identified (hippotherapy = 5; therapeutic riding = 3; and artificial horse = 1) (SMD, −0.44; 95% CI, −0.90 to −0.02; P = .06; τ² = 0.31; I² = 64%; Fig 12). However, the I² (of 64%) points to a considerable heterogeneity between studies. Subgroups with respect to different instruments assessing spasticity reduced heterogeneity to 32% between studies using Ashworth, 19% using hip range of motion, and 59% using electromyography asymmetry scores (Fig 12).

QoL assessment was performed in 4 studies,^8,9,25,32  whereas only data on global QoL as well as on physical, social, and emotional QoL subscores could be included. QoL data from Silkwood-Sherer and McGibbon’s study³⁴  were not included, as they were derived from children and parents. Our analysis could not demonstrate evidence for the improvement for any of the QoL scores (global QoL: SMD, −0.11; 95% CI, −0.15 to 0.37; P = .40, τ² = 0.0; I² = 0%; Fig 13; physical QoL: SMD, 0.08; 95% CI, −0.26 to 0.42; P = .63; τ² = 0.06; I² = 43%; Fig 14; social QoL: SMD, −0.02; 95% CI, −0.26 to 0.22; P = .87, τ² = 0.0; I² = 0%; Fig 15; emotional QoL: SMD, −1.58; 95% CI, −1.36 to 4.51; P = .29; τ² = 6.49; I² = 97%; Fig 16).

As we could not pool more than 10 studies, we did not create a funnel plot to explore possible small study biases for the primary outcome. A registered study protocol in a publicly accessible database was available for the studies of Herrero et al^14,41  and Ahn et al.³²  Whereas the protocol included several outcomes, only data on the GMFM, SAS testing, or PedsQL were reported in the final study. Mutho et al¹³  provided a protocol as a supplement²⁹ . For this study, no deviation between publication and protocol was noted.

A comparison of GMFM total measurements between types of equine-assisted therapy revealed a nonsignificant difference between hippotherapy, therapeutic riding, and artificial horse (P = .07; Fig 17) as well as a nonsignificant difference between therapy with living and therapy with artificial horses (P = .83; Fig 18).

Comparison of motor function between fewer or at least 8 weeks’ duration of therapy also did not show a significant difference with respect to GMFM total score (P = .80; Fig 19).

Limiting the analyses to studies at low risk or some concerns of bias provided evidence for improved motor function as assessed by the GMFM total (SMD, 0.25; 95% CI, −0.04 to 0.46; P = .02; τ² = 0.02; I² = 18%; Fig 20) and by the GMFM subdomain E (hippotherapy = 4; therapeutic riding = 1) (SMD, 2.51; 95% CI, −0.68 to 4.34; P = .007; τ²=0.00; I² = 0%; Fig 21).

A large number of reviews have already addressed the impact of equine-assisted treatments on children with CP, including 6 meta-analyses.^{2,4,5,7,28,42}  This, however, is the first meta-analysis that is based on a prepublished protocol, includes studies on living and artificial horses, and focuses on both motor function and QoL.

Riding on a horse involves a large number of body muscles, which confounds its use for treatment of motor disturbances. The therapeutic use of animal horses in general is called therapeutic riding, whereas the term hippotherapy implies guidance by a trained therapeutic person, such as a physiotherapist or occupational therapist, experienced in this treatment.⁴³  For all animal-based studies included in the present meta-analysis, the involvement of trained therapists was reported, indicating a high standard of treatment. Only those 14 studies, however, that contained the term hippotherapy in the manuscript were finally classified as hippotherapy studies.

The primary question of this meta-analysis was related to motor outcome, which we at first addressed by pooling studies on living and artificial horses, which used the GMFM total tool. We preferentially included data on the GMFM-66 tool, as this was specifically developed for patients with CP,^37–40  and found a significant effect, suggesting that equine-assisted therapy does positively affect global motor function. Astonishingly, this effect was chiefly based on studies using the GMFM-88 tool, although subgroup analyses on three studies that reported on GMFM-66 and GMFM-88 data did not reveal a significant difference between both measures.^11,22,25  Moreover, this effect of different horse-based treatments on the GMFM total score could still be observed when restricting the analysis to studies with no or low risk of bias (Fig 20).

Previous meta-analyses have already focused on the impact of horse-related treatment on global gross motor function. Tseng et al⁴  did not find improved total GMFM-66 scores or GMFM-88 scores when pooling studies with hippotherapy and therapeutic riding, respectively.^{4,18,21,22,44,45}  No controls had been provided by 2 of the included studies.^44,45  Wang et al⁷  included studies on artificial and living horses. They found no significant effects when studying GMFM-66 total or GMFM-88 total data alone but a significant effect in favor of the intervention when pooling GMFM-66 with GMFM-88 total data. Stergiou et al²  and Charry-Sanchez et al³  analyzed the same 4 studies on children with CP and did not find a significant effect of hippotherapy on motor function as assessed by the GMFM-66 total measure. Santos et al²⁸  also failed to detect a treatment effect of hippotherapy when analyzing 3 different studies.

During further subgroup analyses (Figs 17–19), we addressed differences in GMFM total scores when comparing living with artificial horses, therapeutic riding with hippotherapy, as well as longer with shorter treatment duration. Hereby, no significant differences between these groups could be detected. Kwon et al¹¹  compared children of different Gross Motor Function Classification System (GMFCS) levels reporting on significantly increased GMFM-66 for children with levels II, III, and IV and significantly increased GMFM-88 scores for children with all GMFCS levels, suggesting no major impact of the GMFCS on the effect of equine-assisted therapy. In our previous study,⁹  we did not find a relationship between the number of treatments and the GMFM total measure.

We further addressed the impact of horse-related treatment on distinct motor subdomains mostly assessed by the GMFM tool focusing on lying (dimension A), sitting (dimension B), crawling (dimension C), standing (dimension D), and walking (dimension E). These data were complemented by data on subdomains assessed by further instruments such as gait analysis, SAS, or PBS.

As for dimensions A, C, and D, and for motor function during sitting, we did not find significant treatment effects. A significant treatment effect, however, was detected for GMFM dimension E based on 5 studies with hippotherapy and 2 studies with therapeutic riding. This effect was still visible when confining the analysis to 5 studies with low and moderate risk of bias indicating a strong effect (hippotherapy, n = 4; and therapeutic riding, n = 1). Analyzing results from studies with gait analysis as outcome parameter confirmed this positive effect on gait, revealing significant effects of animal-based treatments on step length (Fig 9). The 2 studies analyzed for step length showed significant heterogeneity, so this effect should be taken with caution. Compared with the study of Mutoh et al,³⁵  the study of Kwon et al²²  recorded a decrease of cadence, a stronger increase in step length, and a similar increase in walking speed. In the study of Kwon et al,²²  children were younger and less severely affected (GMFCS I/II vs II/III), which may contribute to this heterogeneity. As sufficient data were not available, GMFCS as a possible source of heterogeneity could not be evaluated. Studies on motor function during sitting showed a comparably high degree of heterogeneity, possibly because the different tests applied did not measure identical sitting capabilities. Moreover, the study of Temcharoensuk et al³⁶  was the only one involving artificial horses and a reanalysis of GMFM data by Rasch analysis.

Spasticity is one further major problem in children with CP. Combined analysis of 9 studies on different spasticity parameters did not reveal a significant reduction of spasticity during equine-assisted therapy. This contrasts with the recent meta-analysis of Hyun et al,⁴²  who, however, also included case series and repeated measure designs.

Results from comparable meta-analyses on motor subdomains in the literature are contradictory. As for the GMFM-E domain, Tseng et al⁴  analyzed 2 studies with hippotherapy and 2 studies with therapeutic riding.^21,22,44,45  They did not find significant effects, neither when studying both groups separately nor when pooling the 4 studies. As for stride length, they analyzed 2 studies,^22,46  also not finding any improvement. As for the asymmetry score, they found reduced asymmetry after hippotherapy.^17,18  Little et al⁶  analyzed 3 studies with hippotherapy and 3 with therapeutic riding, among others focusing on GMFM domains E and D. They did not find significant effects for dimension D or dimension E. Wang et al⁷  pooled data from studies with artificial and living horses. They did not observe a significant effect with regard to dimension B and dimension D, whereas a significant effect in favor of the intervention was observed for dimension E. Dominguez et al⁴⁷  performed a meta-analysis on artificial horses, not detecting a significant effect on GMFM dimension B. Zadnikar et al⁵  performed a meta-analysis on 11 studies applying hippotherapy, therapeutic riding, or artificial horse therapy, which measured outcome by a variety of different tests. In their pooled analysis, they reported a significant effect on postural control and balance. Santos et al²⁸  did not find significant effects when analyzing 3 studies for GMFM dimension E, 3 studies for GMFM dimension D, and 2 studies for cadence, stride length, and walking speed, respectively.

The interaction between living horses and human beings may also affect QoL. Focusing on this hypothesis, we did not detect a significant effect regarding global QoL, the physical, social, or emotional subdomains, respectively. Several reasons may account for this observation: The low number of studies available and the heterogeneity of QoL domains assessed in different studies made it difficult to combine studies for meta-analytic procedures. Equine-assisted treatments might also affect QoL domains other than those included in this analysis, such as family cohesion.⁸  Marked heterogeneity was also observed with regard to emotional QoL outcome, which may be related to the fact that the Kidscreen-27 tool used by Deutz et al⁹  was developed to measure general health-related QoL, whereas the CP-QoL tool used by Davis et al⁸  focuses on QoL in children with CP.⁴² 

Major limitations of this meta-analysis are the heterogeneity of tools used in different studies and the low number of studies addressing QoL issues, factors that limit the number of studies available for distinct analyses. Apart from the GMFM total measure and the GMFM subdomains, the capabilities assessed by various measures differed markedly, which could explain the heterogeneity found in some analyses.

We found strong evidence for a positive effect of equine-assisted therapy on GMFM total, particularly on motor capabilities during walking. This effect may in part be based on improved balance and can be effectively monitored by applying the GMFM total, the GMFM-E, and the PBS tools, respectively. The effect of equine-assisted treatments on the GMFM total score and on the GMFM dimension E was strong and still present when excluding studies with high risk of bias. The treatment standard in the studies on therapeutic riding that were included in the analysis of GMFM total (n = 2) and GMFM dimension E (n = 1) were high and comparable to hippotherapy standards. This recommends living horse-assisted therapy, particularly hippotherapy, as a therapeutic tool for children with CP who are about to learn walking or already have achieved the ability to walk. The high professional therapeutic level described in most studies also recommends providing comparably high treatment standards in clinical settings. Whereas the overall treatment duration ranged from 1 day to 1 year, most children received weekly sessions for a period of 2 to 3 months, so first therapeutic effects should be observed within this timeframe. Further studies on the effects of equine-assisted therapy on motor function should always include the GMFM tool to facilitate data comparison, although additional methods, for example, focusing on spasticity, will be necessary to broaden the view on different therapeutic effects. As for QoL assessment, a broad approach may be necessary to detect more QoL subdomains that are positively influenced by equine-assisted therapies.

We thank Dr Dorothée Debuse (Faculty of Health and life Sciences, Northumbria University, Newcastle Upon Tyne, UK) and Mrs Sanna Mattila-Rautiainen (President of HETI, the Federation of Horses for Education and Therapy International) for advising us with the selection of studies to be included.

CI

confidence interval

CP

cerbral palsy

GMFCS

Gross Motor Function Classification System

GMFM

gross motor function measure

PBS

Pediatric Balance Scale

QoL

quality of life

RoB 2

Risk of Bias 2 tool

SAS

Sitting Assessment Scale

SMD

standard mean difference