This systematic review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines (15). The study protocol was developed a priori and registered with the International Prospective Register of Systematic Reviews (PROSPERO). The full protocol is publicly accessible on the PROSPERO website under registration number (CRD420251182969).

Studies were included if they met the following criteria: RCT design; participants were children (age ≤18 years) with diagnosed CP of any subtype; interventions comprised pharmacological agents administered for spasticity management, including BoNT-A (any formulation), oral baclofen, or other antispastic medications; comparators included placebo, no treatment, alternative interventions, or different dosing regimens; and outcomes included at least one measure of spasticity (MAS) or motor function (GMFM), and functional assessments. Studies were excluded if they were non-randomized designs, included adult populations exclusively, focused on surgical or non-pharmacological interventions alone, or lacked quantitative outcome data. The PICOS criteria for the study; (i) Population: children and adolescents (≤18 years) with CP; (ii) Intervention: pharmacological agents (BoNT-A, oral antispastics); (iii) Comparison: placebo, standard care, or active control; (iv) Outcomes: spasticity (MAS), motor function (GMFM), adverse events; (v) Study design: RCTs.

A comprehensive search was conducted across four major databases: PubMed/MEDLINE, Google Scholar, Web of Science, and Cochrane Library. The search encompassed all records from January 01, 2015, to November 01, 2025, with English language restrictions. The search strategy employed both controlled vocabulary terms (MeSH headings) and free-text keywords. For PubMed, the search string combined terms for CP, spasticity, pharmacological interventions, and study design: ((“cerebral palsy”[MeSH Terms] OR “cerebral palsy”[Title/Abstract]) AND (“spasticity”[MeSH Terms] OR “spasticity”[Title/Abstract]) AND (“botulinum toxins”[MeSH Terms] OR “botulinum toxin”[Title/Abstract] OR “baclofen”[MeSH Terms] OR “baclofen”[Title/Abstract] OR “antispastic agents”[Title/Abstract]) AND (“randomized controlled trial”[Publication Type] OR “randomized”[Title/Abstract])). Parallel searches with database-appropriate syntax were conducted in other databases. Reference lists of included studies and relevant systematic reviews were manually searched to identify additional eligible trials.

Search results were imported into Mendeley Reference Manager (Elsevier, London, United Kingdom), where duplicates were removed. Two reviewers independently screened titles and abstracts against eligibility criteria. Studies meeting initial screening criteria underwent full-text review by both reviewers independently. Disagreements were resolved through discussion, and the selection process was documented using a PRISMA flow diagram.

A standardized data extraction form was developed and piloted on three included studies. Two reviewers independently extracted data from each included study. Extracted information comprised study identifiers (authors, year, journal), participant characteristics (sample size, age, CP subtype, GMFCS levels), intervention details (agent, formulation, dosage, injection technique, adjunctive therapies), comparator specifications, outcome measures with assessment timepoints, quantitative results (means, standard deviations (SD), effect sizes, p-values), follow-up duration, and adverse events.

Risk of bias (RoB) was assessed using the Cochrane Risk of Bias tool version 2 for randomized trials (16). Two reviewers independently evaluated each study across five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of reported results. Each domain was judged as low risk, some concerns, or high risk according to signaling questions and algorithms specified in the tool. Overall risk of bias judgments were derived from domain-level assessments and disagreements were resolved through discussion.

Data synthesis employed narrative summary and, where appropriate, quantitative meta-analysis. Studies were grouped by intervention type, anatomical region (lower limb, upper limb), and outcome measure. To ensure methodological robustness, clinical homogeneity was assessed prior to pooling. Regarding drug type and formulation, different BoNT-A formulations (OnaBoNT-A, AboBoNT-A, and IncoBoNT-A) were included in the synthesis to evaluate the class effect. For spasticity outcomes, quantitative synthesis was restricted to the MAS to ensure comparability across trials; data from the MTS were not included in the pooled analysis. Regarding CP severity, studies including broad GMFCS levels (e.g., I–V) were eligible, and results were pooled using aggregate data provided by the trials, as stratified data by severity level were largely unavailable.

For spasticity outcomes, MAS changes from baseline were extracted, and for motor function, GMFM scores or changes were recorded. When studies reported multiple timepoints, the primary endpoint specified by study authors was used; if unspecified, the assessment closest to the expected peak treatment effect (12 weeks) was selected. Meta-analysis was planned for outcomes reported by two or more studies with sufficient homogeneity. Given anticipated heterogeneity in dosing, techniques, and adjunctive interventions, random-effects models were specified a priori. Statistical heterogeneity was quantified using the I² statistic. All analyses were conducted using R statistical software version 4.3.2 with the meta package. Effect sizes are reported as mean differences for continuous outcomes measured on identical scales (MAS and GMFM) or standardized mean differences when different instruments assessed the same construct. Statistical significance was set at p < 0.05 and Confidence intervals (CI) are reported at 95% level.

The database searches identified 175 records. After removal of 28 duplicates, 147 unique records underwent title and abstract screening. Of these, 43 studies proceeded to full-text review. Following detailed assessment, 18 RCT met all inclusion criteria and were included in the qualitative synthesis. The primary reasons for exclusion at full-text review were non-randomized study designs (n = 12), non-pharmacological interventions exclusively (n = 8), adult populations (n = 3), and absence of relevant outcome measures (n = 2). The study selection process is illustrated in Figure 1.

Although the search strategy included terms for a broad range of pharmacological interventions (e.g., oral baclofen, diazepam, tizanidine), no randomized controlled trials for these agents met the inclusion criteria. Consequently, all 18 included studies exclusively investigated BoNT-A. The six included RCTs in the meta-analysis are summarized in Table 1, encompassed multicenter phase-3 dose-comparison studies and placebo-controlled efficacy trials across upper and lower limbs in children and adolescents with CP (GMFCS I–V; ~2–17 years). Two large, double-blind, placebo-controlled lower-limb trials—Delgado et al. (17) (AboBoNT-A) and Dimitrova et al. (18) (OnaBoNT-A)—randomized participants 1:1:1 to two active doses versus placebo, with standardized physiotherapy (PT) mandated or permitted across arms, and prespecified early efficacy windows around 4–6 weeks [Dimitrova et al. (18) averaged weeks 4 and 6]. One assessor-blinded two-arm study [de Souza et al. (19)] compared BoNT-A plus a structured PT program with the same PT alone at ~30 days and ~3 months. Three double-blind active-dose RCTs without placebo completed the set: Heinen et al. (20) (IncoBoNT-A 4/12/16 U kg⁻¹; two cycles; lower limb) and two upper-limb trials—Delgado et al. (21) (AboBoNT-A with a tailored home program) and Dabrowski et al. (22) (IncoBoNT-A 2/6/8 U kg⁻¹)—each using a double-blind main period to compare licensed dose tiers.

Across trials, injections targeted clinically relevant patterns (e.g., triceps surae for dynamic equinus; upper-limb flexor/pronator patterns), with total/body-region dose caps per protocol and image-guided or localization-assisted techniques. Concomitant therapy was standardized in the placebo-controlled studies and balanced across arms in the dose-comparison designs; in de Souza et al. (19) both groups received identical structured PT to isolate the incremental effect of BoNT-A. Outcomes aligned with contemporary pediatric practice: spasticity was captured primarily with the MAS at ~4–6 weeks, often as change-from-baseline and, in pivotal trials, reported as model-based least-squares estimates; dynamic muscle response was assessed with the MTS where available. Functional status was measured with GMFM-66 (20) and GMFM-88 domain scores (19) at ~1–3 months, alongside goal-level metrics (e.g., GAS) in selected trials. For meta-analysis, multi-arm studies were combined within-trial to preserve independence; MAS effects entered as mean differences or via generic inverse-variance when only adjusted effects were reported, and GMFM outcomes were pooled as standardized mean differences to accommodate instrument differences.

To address the heterogeneity in pharmacological protocols, we extracted detailed dosing information from all included RCTs. Table 2 summarizes the drug formulations, injection sites, and specific dosage ranges (units/kg) utilized in the primary studies. The analysis includes three distinct formulations of BoNT-A (AboBoNT-A, OnaBoNT-A, and IncoBoNT-A) targeted primarily at the triceps surae for lower limb spasticity and specific flexor patterns for upper limb interventions. Dosing strategies varied significantly by formulation, necessitating a comparison against standard manufacturer guidelines.

The MAS forest plot, synthesizing data from trials utilizing AboBoNT-A (17, 21), OnaBoNT-A (18), and IncoBoNT-A (20, 22), demonstrates a consistent early reduction in spasticity favoring BoNT-A across trials, with all point estimates lying on the benefit side. Between-study heterogeneity is moderate, largely introduced when upper-limb dose-comparison trials and the PT-controlled study are pooled with the placebo RCTs. Subgroup inspection shows the most homogeneous and precise effects in the placebo-controlled lower-limb subset, whereas adding upper-limb dose-tier contrasts broadens variance without changing the direction of effect. Study weights are appropriately driven by the larger, multicenter RCTs; leave-one-out analyses do not materially shift the pooled estimate, and fixed- versus random-effects models yield concordant conclusions. With fewer than 10 contrasts, formal small-study bias tests are underpowered; visual inspection shows no dominant outlier and only mild funnel asymmetry, supporting a robust signal of short-term MAS improvement at 4–6 weeks.

The GMFM forest plot, synthesizing GMFM-66 (dose-comparison) and GMFM-88 (two-arm) data at ~1–3 months, indicates a small functional benefit of BoNT-A. The pooled standardized effect is positive, though heterogeneity is higher than in MAS analyses, reflecting instrument differences (66 vs. 88), scoring metrics (change vs. post), and co-intervention intensity. Sensitivity analyses restricting to matched windows or removing either study preserve the direction of effect, with magnitude shifting within a narrow small-effect range; prediction intervals encompass small benefit and, at worst, minimal change—consistent with the expectation that functional gains depend on concurrent, goal-directed therapy. However, it is important to note that this synthesis is derived from only two trials, which limits statistical power. Consequently, these results should be interpreted with caution as a preliminary signal of efficacy rather than a definitive confirmation of functional benefit. Overall, the forest plots support a coherent pattern: BoNT-A yields a short-term reduction in tone that translates into early but modest functional improvement, with the most precise estimates arising from placebo-controlled lower-limb trials and increased dispersion when heterogeneous designs are pooled (Figure 2).

Risk of bias was judged largely acceptable across the evidence base, with two placebo-controlled lower-limb RCTs (17, 18), rated low risk in all key domains (adequate random sequence generation and allocation concealment, double-blinding of participants/personnel, blinded outcome assessment, minimal and balanced attrition, and prespecified outcomes). The assessor-blinded two-arm trial (19) was rated high risk overall due to unavoidable lack of blinding of participants and therapists (performance bias) and insufficient detail on allocation concealment, although attrition and outcome reporting were acceptable. The active dose-comparison trials (20–22) were low risk within their designs (central randomization, double-blinded dosing, blinded assessors, low and balanced attrition), but they lacked a placebo/usual-care comparator and were therefore not considered in the main placebo-controlled efficacy pool. Sensitivity analyses restricting the MAS synthesis to the low-risk placebo-controlled trials yielded nearly identical effect estimates with reduced heterogeneity, indicating that overall conclusions were robust to exclusion of higher-risk or non-placebo designs.

Across trials directly evaluating BoNT-A in children with CP, early reductions in lower-limb spasticity (measured by MAS) and supportive functional signals are consistently described, while durability varies and depends on rehabilitation context. In children with equinus who had previously received toxin, AboBoNT-A again improved clinical measures of ankle spasticity and was well tolerated, reinforcing benefit in a pre-exposed population (22). A pragmatic randomized trial that paired BoNT-A with a structured PT program reported superior short-term functional capacity (GMFM-88 domains) and lower post-treatment MAS scores versus PT alone, suggesting that chemodenervation plus targeted therapy can translate tone reduction into early motor gains (19). Dose-ranging work with IncoBoNT-A in a large phase-3 pediatric cohort further supports efficacy across licensed dosing bands with an acceptable safety profile, albeit without a usual-care placebo comparator (20).

Evidence on repeat exposure and programmatic delivery complements these findings. A secondary analysis focused on repeated injections for equinus found that clinical improvements can be reproduced across cycles in appropriately selected candidates (23). Safety-oriented randomized work with IncoBoNT-A injections to the gastrocnemius in children confirmed a favorable adverse-event profile, dominated by mild, transient events typical of local chemodenervation (e.g., injection site pain) (24). Similarly, the large placebo-controlled trial of OnaBoNT-A reported that adverse events were generally mild and self-limiting, with no significant increase in serious adverse events compared to placebo. Development and implementation papers around upper-limb programs emphasize that standardized, goal-centered home therapy packages are integral to maximizing functional impact after injections, operationalizing the “window of opportunity” following tone reduction (25, 26). Together, these studies support a model in which BoNT-A reduces spasticity reliably in the short term and delivers functional benefit most consistently when embedded within structured, goal-directed rehabilitation.

Comparators and adjuncts highlight where BoNT-A sits among broader spasticity strategies. A randomized cross-over study comparing radial extracorporeal shock-wave therapy with BoNT-A for lower-extremity spasticity suggested that non-injectable modalities may produce clinically relevant changes in tone for some children, underscoring the importance of individualized planning and shared decision-making (27). Casting studies conducted as adjuncts after BoNT-A for equinus—or independently for upper-limb deformity—demonstrated additional mechanical benefits attributable to prolonged stretch and positioning, indicating that orthotic or casting strategies can complement chemodenervation in selected goals (28). Beyond injectables, trials of functional electrical stimulation cycling and technology-supported arm rehabilitation illustrate alternative or additive rehabilitation pathways to drive practice intensity and participation, even if they do not target neural over-activity directly (29, 30).

Longer-term and measurement-focused studies provide context for interpretation. A national cohort study exploring hip and spine outcomes after BoNT-A highlights uncertainty around structural sequelae and the need for longer follow-up beyond typical 12-week efficacy windows (31). Method papers applying shear-wave elastography offer objective, quantitative insights into muscle stiffness change after injections, potentially refining outcome measurement alongside clinical scales such as MAS and GMFM (32). Protocol publications for walking-focused RCTs emphasize the importance of prespecified functional endpoints and standardized therapy doses to improve comparability across trials (33). Finally, observational and regional reports underscore variability in practice patterns and the promise of data-driven selection of children most likely to benefit from neurotoxin treatment (34, 35). Collectively, the corpus indicates that BoNT-A is a useful, generally safe tool for short-term tone reduction; meaningful functional gains are most likely when injections are planned within a comprehensive program that includes goal-based therapy, adjunct casting/orthoses where appropriate, and outcome measures sensitive to change.

This review integrates high-quality randomized evidence across complementary designs and outcomes, strengthening confidence in the conclusions. Placebo-controlled lower-limb RCTs provided precise efficacy estimates for short-term spasticity reduction (17, 18), while dose-comparison trials broadened external validity by testing licensed dose ranges in larger, multicenter cohorts (20, 22). The inclusion of a pragmatic two-arm trial embedding BoNT-A within a structured PT package directly linked tone reduction to functional change under real-world therapy conditions (19). Synthesizing both impairment (MAS) and activity-level outcomes (GMFM-66/88) offered a fuller view of clinical impact, and parallel reports on goal-directed home programs and goal attainment enriched interpretation of how rehabilitation dosing and goal setting translate injections into patient-relevant gains (25, 26).

This study has several limitations. First, although our search encompassed all pharmacological interventions, only BoNT-A trials met inclusion criteria; consequently, findings are specific to this agent and cannot be extrapolated to other treatments. Second, generalizability is constrained by clinical heterogeneity (e.g., dosing, GMFCS levels) and a reliance on short-term outcomes (typically 4–6 weeks), with limited data extending to 3 months (19). Methodologically, the synthesis was challenged by the reporting of least-squares means rather than raw data and a scarcity of functional outcomes, restricting the GMFM meta-analysis to only two trials. This results in low statistical power, rendering evidence for functional gain preliminary. Finally, potential publication bias, the predominance of high-income study settings, and a search strategy requiring the specific term “randomized” may have reduced sensitivity for eligible trials lacking this metadata (27, 28, 49).

Future trials should prespecify harmonized functional endpoints and reporting conventions—e.g., GMFM-66 change scored with consistent software, standardized Goal Attainment Scaling procedures—and provide arm-level means and SDs at key time points to enable conventional meta-analysis alongside adjusted estimates (33). Crucially, trial designs ought to quantify rehabilitation “dose” and systematically evaluate adjuncts to define combinations that best convert short-term tone reduction into sustained activity gains (25, 26). This evolution in therapy design extends beyond traditional mechanical aids (e.g., casting) to the transformative potential of artificial intelligence (AI) and the metaverse. As recent reviews highlight, these immersive environments offer novel avenues to complement the “therapeutic window” of BoNT-A by providing high-intensity, gamified motor learning that overcomes traditional barriers to adherence and accessibility (50). Longer follow-up across repeated injection cycles is needed to characterize durability, cumulative effects, and longer-term structural and participation outcomes, informed by signals from observational cohorts (31). Work to identify predictors of response (e.g., phenotype, GMFCS level, selective motor control), incorporate objective muscle-property measures (e.g., shear-wave elastography) alongside clinical scales, and address implementation barriers (access to guidance technology, therapy capacity) will improve targeting and real-world impact (32). Comparative effectiveness and economic evaluations across pharmacologic and non-pharmacologic strategies remain priorities to guide resource-sensitive care pathways.