Does More Speech Therapy Lead to Better Results in Childhood Apraxia of Speech? A Systematic Review and Meta‐Analysis of the Available Evidence Authors: Patricia McCabe, Donna C Thomas, Jessica Xue Ming Choo, Rob Heard, Denise Ng Published: 2026 ABSTRACT ABSTRACTBackgroundIn recent years there has been a significant increase in the number and quality of childhood apraxia of speech (CAS) treatment research published. Much of this research has focussed on establishing treatment efficacy and identifying treatment components. A smaller subset of the literature has looked at treatment intensity, such as comparing therapy delivered twice per week with four times per week, however this literature has not been compared across interventions.AimsThis systematic review and meta‐analysis examines cumulative intervention intensity and its components (dose per session, dose frequency, intervention duration—number of weeks) and reported treatment effects to understand whether there is a relationship between dose and effect in specific treatments or overall.MethodsA systematic search and review was registered with Prospero (CRD420251107259). Twelve electronic databases were searched supplemented by handsearching references lists. Included studies were coded for CAS diagnosis confidence, methodological quality, dose characteristics, and treatment outcomes. Standard mean difference (SMD), an effect size metric equivalent to Cohen's d, was calculated for 25 included papers and weighted Pearson's correlations performed to identify relationships between dose characteristics and treatment outcomes.Main ContributionThis paper is the first to (1) create omnibus standard effect sizes across diverse interventions and (2) show the relationship between intensity and treatment outcome in children with CAS. The omnibus SMD across studies with three or more participants for treated items was 1.22 (confidence interval 0.76–1.67), a large effect, and for generalisation items was 0.66 (confidence interval 0.46–0.85) which is a moderate effect. These results show that CAS treatment works in well‐controlled studies when delivered at least twice per week for at least 45 min per session.The relationship between intensity variables and generalisation SMD was less clear with weeks of therapy being moderately correlated (r = 0.39) and none of the other variables showing any relationship with study SMD. These results should be interpreted carefully as there was significant missing data and limited range in some variables. Researchers need to include more standard intensity information in reporting their results.ConclusionsTreatment for CAS is effective and leads to moderate generalisation; more weeks of therapy leads to stronger gains than fewer weeks of therapy.WHAT THIS PAPER ADDS What is already known on this subject CAS is a neurological‐based motor speech disorder that originates in childhood for which there are a number of effective treatments. For this population, high intensity treatment is recommended—over 100 production trials per session, 2–5 sessions per week, across an extended period. In studies which directly compare intensity of variables, more therapy (e.g. 20 sessions) results in greater improvement than less therapy (e.g. 10 sessions). There are no previous meta‐analyses of intensity variables—dose, dose frequency, treatment duration, or cumulative intervention intensity in CAS What this study adds to the existing knowledge This study compiles data from 25 studies with a low risk of bias and high CAS diagnostic confidence in a meta‐analysis of the dose effects of CAS treatment. The meta‐analysis shows that CAS treatment in general works. Most studies had 2–4 sessions per week across 3–12 weeks. Within this narrow band of intensity, a significant correlation was shown between the number of weeks of therapy and outcome. There is no significant evidence for CAS therapy less frequently than twice per week at this time. Eight different treatments demonstrated efficacy for improving speech outcomes in children with CAS. What are the clinical implications of this study? Provided they are delivered at least twice per week for at least 12 sessions, with greater improvements when treatment extends over more weeks, selection of any of the included treatments should result in improved speech where the child being treated is similar to children in the research. The following treatments were shown to be effective for CAS: Integral stimulation treatments including Dynamic Temporal and Tactile Cueing, Kaufman‐Speech to Language Protocol, Motor Speech Treatment Protocol, Nuffield Dyspraxia Programme—3rd Edition, Rapid Syllable Transition Treatment, Speech Motor Chaining and Ultrasound Biofeedback alone and the combination of Speech Motor Chaining with Ultrasound Biofeedback.PRACTITIONER POINTS Therapy must be sufficiently frequent and sustained: Evidence shows that CAS treatment is effective when therapy is provided at least twice per week, over a minimum of 12 sessions, for 45 min per session with total weeks of therapy being the only intensity variable consistently associated with treatment outcomes. Dose matters, both within and across sessions: Clinicians should prioritise the number of production trials and ensure adequate cumulative intervention intensity. Sessions with low active practice time or insufficient trials are unlikely to produce meaningful change. Eight treatments are effective for CAS: Integral Stimulation including Dynamic Temporal and Tactile Cueing (DTTC), Kaufman Speech to Language Program (K‐SLP), Motor Speech Treatment Protocol (MSTP), Nuffield Dyspraxia Programme (3rd ed; NDP3), Rapid Syllable Transition Treatment (ReST), Speech Motor Chaining (SMC), Ultrasound Biofeedback, and SMC + Ultrasound Biofeedback combined. Clinicians should consider the client's speech needs and overall profile when selecting from the treatments that have demonstrated effectiveness.Keywords: apraxia, childhood apraxia of speech, dose, effect size, generalisation, intensity, speech sound disorder, treatment Introduction One of the most common recommendations in the treatment of childhood apraxia of speech (CAS) is that children should receive intensive therapy (see Maas 2024 for a review). When detail is provided about ‘intensive’ treatment, it often includes a recommendation for multiple sessions per week (e.g., American Speech‐Language‐Hearing Association [ASHA] 2007; Royal College of Speech and Language Therapists [RCSLT] 2024). This recommendation rests on a number of foundation truths. Firstly, CAS is a motor speech disorder (ASHA 2007) and therefore treatment is believed to be most effective when informed by the principles of motor learning (PML) and neuroplasticity (Maas et al. 2008). The tenets of PML suggest that after an initial period of acquisition of an internal reference of correctness, treatment should be frequent, distributed, diverse, and high dose (Maas et al. 2008). As such, many treatments are developed with a focus on drill‐based interventions and high intensity practice within sessions (c.f., Dynamic Temporal and Tactile Cueing [DTTC, e.g, Strand 2020], Speech Motor Chaining [SMC, e.g. Preston et al. 2024], Rapid Syllable Transition [ReST, e.g., McCabe et al., 2020]). Secondly, existing treatment research appears to indicate that a greater overall amount of treatment leads to better outcomes, and more frequently delivered treatment is more effective than less frequently delivered treatment. For example, Namasivayam et al. (2015) showed that 20 sessions of Motor Speech Treatment Protocol (MSTP) was more efficacious than 10 sessions, and Thomas et al. (2024) showed that one session per week of ReST treatment was less efficacious than the same amount of treatment provided twice per week or four times per week. However, as we will show below, understanding treatment intensity characteristics is more nuanced than these assertions imply. Clinical reality frequently clashes with these recommendations for high frequency therapy (e.g., Gomez et al. 2019). In many settings across countries, children with CAS receive 1–2 sessions of therapy per week for limited weeks per year (e.g., Gomez et al. 2019; Gomez et al. 2022; Thomas, Gomez, et al. 2026; Wong et al. 2023). Clinicians report frustration that they cannot provide the recommended intensity, putting this limitation down to a range of factors including local policies, cost, and waiting lists (Gomez et al. 2019; Gomez et al. 2022; Wong et al. 2023). Although families may desire intensive CAS therapy for their children, they often find it difficult to access such therapy. Challenges include the availability of Speech and Language Therapists (SLTs) experienced with CAS (e.g., Dawoud et al. 2025; Highman et al. 2025; Thomas, Gomez, et al. 2026), calendar openings with these SLTs, and funding to support intensive therapy (Dawoud et al. 2025; Thomas, Gomez, et al. 2026). Families who have the financial means often supplement public services, such as those provided at a school or health centre, with private services (e.g., Thomas, Gomez, et al. 2026). Attending intensive therapy places significant burden on the whole family unit (Dawoud et al. 2025; Highman et al. 2025; Thomas et al. 2018), yet families report being willing to do “whatever it takes” to ensure that their child receives the most effective and efficient care (Thomas, Bricker‐Katz, et al. 2026). While the last 10+ years have seen the publication of a number of randomised control trials (RCTs) exploring various CAS treatments, there are few large scale, head‐to‐head studies which explicitly compare one treatment with another. The existing literature provides evidence that treatment, in general, is effective (Maas 2024); however, where two different interventions have been compared in an RCT, neither intervention has demonstrated greater efficacy over time (e.g., McCabe et al. 2023; Murray et al. 2015). It is possible that differences between treatments were not detected due to limitations in sample size or duration of treatment. If such differences exist, they may be uncovered during meta‐analysis due to the combination of results across studies. When describing components of overall intensity, it is helpful to consider the framework provided by Warren et al. (2007) that includes dose (i.e., the number of teaching episodes during a therapy session), dose frequency (i.e., the number of therapy sessions in a specific period, such as a week), and total intervention duration (i.e., the period of time over which therapy is provided). These components can be used to calculate an overall treatment intensity, known as cumulative intervention intensity. Cumulative Intervention Intensity Dose is the number of correctly delivered teaching episodes per session (Warren et al. 2007). In most motor speech treatment studies, dose is the number of speech production trials. For example, ReST treatment specifies the dose per session across the practice phase—100 practice trials of pseudowords per session (McCabe et al. 2020). Other CAS treatments report dose as the number of therapy minutes (e.g., Strand et al. 2006; van der Merwe and Steyn 2018). When the dose is not tightly controlled but well reported (e.g., Preston et al. 2024), there can be large variations in dose provided per session, across children in the same study. Dose frequency is the number of times per day or week that the treatment is provided. In CAS treatments, this is most frequently measured as the number of therapy sessions per week. A somewhat related concept is that of practice distribution, whereby therapy sessions can be clustered together or distributed across time. The distribution of therapy sessions can be measured using inter‐practice interval—the period of time between therapy sessions (Kaipa et al. 2023). The same dose frequency, for example twice per week, could have different inter‐practice intervals, depending on the spacing of those two sessions across the week. Inter‐practice intervals have been studied more in vocabulary acquisition than in motor speech learning (e.g., Kaipa et al. 2020; Kaipa et al. 2023) and the spacing of sessions is less frequently reported in CAS studies than in the language literature. Total intervention duration is the period of time over which the treatment is provided, measured in weeks or months (Warren et al. 2007). For example, the total intervention duration studied in the RCT of ultrasound biofeedback and SMC for children with CAS was 10 weeks (Preston et al. 2024). Typically, total intervention duration is reported for a treatment study rather than across the years in which a child receives CAS therapy, the exception being clinical case reports where a number of years of therapy may be reported (e.g., Baas et al. 2008). Cumulative intervention intensity is a term used to provide an overall indicator of the intensity of an intervention. It is calculated as follows (Warren et al. 2007): Cumulative intervention intensity = dose x dose frequency x total intervention duration. As with total intervention duration, cumulative intervention intensity is typically reported across a research study rather than the entirety of a child's involvement with speech therapy, although the latter would probably be of more interest to parents and clinicians (Thomas, Gomez, et al. 2026). Cumulative intervention also can be measured in production trials (e.g., 1200 practice trials of pseudowords in ReST therapy; Murray et al. 2015) or in hours of therapy or minutes of over a treatment block (e.g., 24 hours of DTTC therapy; Iuzzini‐Seigel et al. 2024). Some studies report more than one block of treatment, which could be an important factor in considering any dose frequency measures. As discussed, intensive treatment has long been advocated for CAS (e.g., ASHA 2007; Strand et al. 2006). However, ‘intensity’ is not a unitary construct; rather, it is multidimensional. Within CAS, most of the research into intensity has sought to understand the efficacy of different dose frequencies within individual interventions (e.g., spacing of 12 sessions of ReST treatment, Thomas et al. 2024; distribution of 20 hours of ultrasound biofeedback, Preston et al. 2024; spacing of 24 h of DTTC, Iuzzini‐Seigel et al. 2025). However, it is important to note that such investigations are only meaningful if other factors such as cumulative intervention intensity or dose frequency are adequately controlled. When these confounds are controlled, it appears that dose frequencies greater than twice per week may have similar efficacy within CAS treatments. Using ReST as an example, there are study level indications that dose frequency may influence treatment effect, generalisation effect, and maintenance of treatment and generalisation effects. Analysis of individual studies indicates that dose frequencies of twice per week or more have similar treatment effects, but a dose frequency of four times per week may have superior performance in the maintenance phase of treatment (e.g., Murray et al. 2015 vs. Thomas et al. 2014). However, when Ng et al. (2022) analysed the dose effects in ReST therapy using an individual participant data meta‐analysis, the difference in dose frequencies was not significant. A single case experimental design paper published after the ReST individual meta‐analysis showed that a dose frequency of once per week has both poor treatment effect and generalisation (Thomas et al. 2024). Similar findings have been reported with dose frequencies of at least twice per week in DTTC treatment (Iuzzini‐Seigel et al. 2025). To date, there have not been sustained or substantial investigations of dose variables across research groups and treatments in CAS intervention. Although dose frequencies of one, two, three, and four sessions per week have been investigated for ReST, dose variables have not yet been carefully explored to similar depth in studies across other treatment approaches. For example, an attempt to compare dose frequency for MSTP was confounded because the children who had twice weekly therapy received twice as much therapy as the children who received weekly therapy (Namasivayam et al. 2015) across the same total intervention duration of 10 weeks. Similarly, an attempt to compare high dose frequency (two 30‐minute sessions per day for 2 weeks) with low dose frequency (one 60‐minute, session per week for 10 weeks) for a variation of the integral stimulation approach was confounded by having all children receive the high dose frequency prior to the low dose frequency treatment (Leonhartsberger et al. 2022). That said, there is some research which suggests that more intensive dose frequencies such as a camp model (Preston et al. 2016) or a ‘jump start’ model with a higher dose frequency at the start of a block than at the end of the block (Preston et al. 2024) may be of benefit to some children. From a naive reading of the literature, almost all treatments appear to work, almost all of the time, with few papers reporting interventions that work for less than half of participants at any given cumulative dose. It is therefore tempting to suggest that any given treatment with sufficient cumulative intervention intensity would be effective; however, this observation would be based on a false premise due to a range of biases. Primary among these is publication bias where unsuccessful studies are less likely to be published (e.g., Staples 2009) and selection bias where children in single case experimental design (SCED) studies are chosen for reasons other than just having CAS, such as convenience or likelihood of being suitable for that particular treatment. In conducting a systematic review, it is clear that such publication bias should be countered where possible. One of the biases evident in the CAS literature is the significant number of publications about a small range of interventions (see Maas 2024) with at least 20 DTTC/Integral Stimulation papers (Ng et al. 2026) and at least 10 ReST/Syllable Sequencing publications (see Ng et al. 2022; Thomas et al. 2024). Many other interventions have 1–3 SCED papers or case reports and only three treatments are known to have more than one RCT: ReST (McCabe et al. 2023; Murray et al. 2015), Nuffield Dyspraxia Program‐Third Edition (NDP3; McKechnie et al. 2020; Murray et al. 2015) and ultrasound biofeedback (McCabe et al. 2023; Preston et al. 2024) each of which have two published RCTs. Note: DTTC has at least two RCTs which are complete and have been shared in conferences but are not yet published (Iuzzini‐Seigel et al. 2025; McCabe et al. 2025). Aims and Research Questions The first aim of this paper is to examine the relationship between components of cumulative intervention intensity and effect sizes, based on the available literature on CAS interventions. As a secondary aim of the study, we aim to investigate whether there is a relationship between the treatment type and effect size, and, for this analysis, will group similar interventions together. For example, all integral stimulation (e.g., DTTC, Apraxia of Speech Systematic Integral Stimulation Treatment [ASSIST]; Maas et al. 2019) were grouped as DTTC/IS, and all nonword syllable sequencing studies were grouped (e.g., ReST, Treatment for Establishing Motor Program Organization [TEMPO]; Miller et al. 2021; Speech Motor Learning [SML]; van der Merwe and Steyn 2018) as ReST/SyllSeq for this section. In addressing the research aims, the primary research questions are as follows: Is there a relationship between cumulative intervention intensity and effect size on the primary outcome measure, and, if so, what is the nature of the relationship? Is there a relationship between cumulative intervention intensity and effect size on the primary outcome measure, and, if so, what is the nature of the relationship? Is there a relationship between dose frequency and effect size on the primary outcome measure, and, if so, what is the nature of the relationship? Is there a relationship between dose frequency and effect size on the primary outcome measure, and, if so, what is the nature of the relationship? The secondary question is: Does treatment type affect the results of questions (1) or (2)? Does treatment type affect the results of questions (1) or (2)? Materials and Methods The review protocol for this systematic review and meta‐analysis was registered with PROSPERO on 22 July 2025 (CRD420251107259) and can be accessed there. There was one change to the published protocol in the conduct of the review with the statistical method proposed in the protocol varied (as described below) due to missing data across included papers. The authors conducted this systematic review following the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) 2020 statement guidelines (Page et al. 2021). A university librarian was consulted for the search strategies used in this review. No ethical oversight was required by the auspicing university. The review was unfunded. Data and other materials not provided in the Supplementary Materials can be provided by the corresponding author. Eligibility Criteria Studies were eligible if they (1)were written in English, were written in English, (2)included children under the age of 18 with a primary diagnosis of Childhood Apraxia of Speech, and included children under the age of 18 with a primary diagnosis of Childhood Apraxia of Speech, and (3)reported quantitative outcomes from behavioural intervention/s aimed at improving the child's speech production accuracy, consistency, or intelligibility. Any behavioural intervention aimed at improving speech production accuracy or consistency or intelligibility in children with childhood apraxia of speech was included. CAS diagnosis certainty was dealt with in the quality evaluation stage. The searches were not limited by date of publication, study design, or service delivery modes. reported quantitative outcomes from behavioural intervention/s aimed at improving the child's speech production accuracy, consistency, or intelligibility. Any behavioural intervention aimed at improving speech production accuracy or consistency or intelligibility in children with childhood apraxia of speech was included. CAS diagnosis certainty was dealt with in the quality evaluation stage. The searches were not limited by date of publication, study design, or service delivery modes. Studies were excluded if they (1)included only children diagnosed with other childhood speech disorders as their primary speech disorder; children with acquired apraxia of speech; children with craniofacial anomalies. Studies relating to non‐speech dyspraxia (e.g., oculomotor apraxia, motor dyspraxia) or genetic studies without speech intervention were also excluded. included only children diagnosed with other childhood speech disorders as their primary speech disorder; children with acquired apraxia of speech; children with craniofacial anomalies. Studies relating to non‐speech dyspraxia (e.g., oculomotor apraxia, motor dyspraxia) or genetic studies without speech intervention were also excluded. (2)treated or measured non‐speech oral motor behaviours or other non‐speech goals as the primary outcome measure, or reported non‐numerical data (e.g., qualitative studies, narrative case reports). Drug treatments; surgical interventions; any type of cranial stimulation (e.g., direct current stimulation); other non‐behavioural interventions, interventions which focussed solely on non‐speech oral motor or oral stimulation were also excluded. treated or measured non‐speech oral motor behaviours or other non‐speech goals as the primary outcome measure, or reported non‐numerical data (e.g., qualitative studies, narrative case reports). Drug treatments; surgical interventions; any type of cranial stimulation (e.g., direct current stimulation); other non‐behavioural interventions, interventions which focussed solely on non‐speech oral motor or oral stimulation were also excluded. Information Sources A systematic literature search was conducted on the 4th of August 2025 across the following databases: CINAHL, Medline, PubMed, PsycINFO, Embase, Scopus, SpeechBITE, LLBA, ERIC, Google Scholar, Web of Science, and ProQuest. See Material S1 for search terms. No repeat searches were completed. The reference lists of systematic reviews obtained from the systematic literature search were also extracted and imported into EndNote for deduplication and addition to Covidence. The authors’ own libraries were hand searched for additional relevant references before being added to Covidence as appropriate. Search Terms The following search terms were used in combination with Boolean operators (such as *) to search within titles, abstracts, and text using keywords across the databases: (apraxi*, “childhood apraxia of speech”, dyspraxi*) AND (child*, adolescen*, boy, girl, toddler*, infant*, preschool*, pre‐school*, youth*) AND (therap*, interven*, treat*). A list of the detailed search terms for each database is provided in Material S1. A total of 8605 studies were identified, and references were exported to EndNote 21 (The Endnote Team 2013) where duplicates were removed. The resulting 3731 studies were uploaded to Covidence (Covidence 2025) for further screening. Selection Process Authors 1, 2, 3, and 5 contributed to the title and abstract screening, with agreement required from two authors independently reviewing each article. Any discrepancies were resolved by a third author. Following screening, full texts of the included references were retrieved and assessed for adherence to eligibility criteria. All screening authors contributed to the full text screening stage, with agreement required from two authors to independently review and choose to either include the article or provide a reason for exclusion. Any discrepancies were discussed between the two authors to reach a consensus. If the chosen reason for exclusion was not unanimous, for example, when both reasons were applicable, the article was excluded with one of the two reasons chosen. If full text was not available either electronically or in the authors’ library, the corresponding author was contacted for it. At the full text screening stage, articles were excluded for the following reasons: (a) not in English, (b) adult population (c) not a treatment study (d) full text not available, (e) case report without data, (f) not CAS as primary diagnosis, (g) wrong outcomes (e.g., satisfaction, academic improvement, etc.), (h) wrong study design (e.g., qualitative, review, commentary, etc.), or (i) non‐speech intervention or not a behavioural intervention. Data Collection and Extraction The third and last authors extracted relevant data from all included papers into an Excel workbook. Extracted data included participant variables (e.g., age, gender, severity of speech sound disorder), treatment intensity information (e.g., frequency, duration, number of sessions, number of trials), and reported outcome measures of treated and/or generalisation probe data from the baseline, post‐treatment, and follow‐up timepoints if available. Since this review concerns the interaction between outcomes and intensity characteristics, treatment phase data was not extracted. Where available, data was extracted from published tables or hand‐measured from the published figures to re‐create numerical outcome data and rounded to the nearest whole number. Where data was not presented in the papers, study authors were contacted. Ten percent of the data extraction was repeated by another author blinded to the first extraction to ensure data reliability with a resulting percent exact match of 89.4% reliability on a point‐by‐point basis. All discrepancies were associated with hand measuring from figures and were within 5%. Quality Evaluation Certainty of CAS Diagnosis Definitions of CAS have changed over time and the amount of detail provided about diagnosis and the assumed features of CAS across papers varies substantially. Following the method in Murray et al. (2014), all papers included after full text screening were evaluated for certainty of CAS diagnosis. Table 1 shows the rating scale, adapted from Murray et al. (2014). TABLE 1Confidence rating scale for childhood apraxia of speech (CAS) diagnosis.LevelCharacteristics of CAS a ComorbidityAdapted from Murray et al. (2014)Mayo‐10 b DEMSS1(a)All three primary characteristics described: InconsistencyLengthened and disrupted coarticulatory transitions between sounds and syllablesInappropriate prosody At least 4 out of 10 across≤ 323 (significant evidence for CAS)CAS without another comorbid disorder (excluding expressive language delay).1(b)All three primary characteristics described: InconsistencyLengthened and disrupted coarticulatory transitions between sounds and syllablesInappropriate prosody At least 4 out of 10 features ≤ 323 (significant evidence for CAS) Clear cases of comorbid CAS where CAS was the primary diagnosis.2(a)Two of the three primary characteristics described: InconsistencyLengthened and disrupted coarticulatory transitions between sounds and syllablesInappropriate prosody 3 out of 10 features described, of which at least 2 have to include: 1, 4, 8, 10324–373 (some evidence for at least mild CAS)This includes the following: CAS without another comorbid disorder orClear cases of comorbid CAS, in which CAS was the primary diagnosis 2(b)Two of the three primary characteristics described: InconsistencyLengthened and disrupted coarticulatory transitions between sounds and syllablesInappropriate prosody 3 out of 10 features described, of which at least 2 have to be from this list: 1, 4, 8, 10324–373 (some evidence for at least mild CAS) Cases of CAS in which another comorbid disorder had the same severity (e.g., language delay, dysarthria). 3Only one of the three primary characteristics reported or incomplete/inadequate description of the primary characteristics of CAS was provided.3 out of 10 features described (any 3) 324–373 (some evidence for at least mild CAS) Unclear whether CAS was the primary diagnosis.4Diagnosis of CAS was reported or implied, but none of the three primary characteristics were described.CAS diagnosis was reported or implied but 0–2 of Mayo 10≥ 374Other comorbid disorders may be present. Notes: DEMSS = Dynamic Evaluation of Motor Speech Skill. a The most relevant column was used to rate the CAS diagnosis reported in the paper. If more than one was reported, priority was given to one of the two lists of features. b A list of 10 features of CAS by the Mayo Clinic (2025) system: (1) vowel distortions; (2) voicing errors; (3) distorted substitutions; (4) difficulty achieving initial articulatory configurations or transitionary movement gestures; (5) groping; (6) intrusive schwa; (7) increased difficulty with multisyllabic words; (8) syllable segregation; (9) slow speech rate and/or slow diadochokinetic rates; (10) and/or equal stress or lexical stress errors. Confidence rating scale for childhood apraxia of speech (CAS) diagnosis. Inconsistency Inconsistency Lengthened and disrupted coarticulatory transitions between sounds and syllables Lengthened and disrupted coarticulatory transitions between sounds and syllables Inappropriate prosody Inappropriate prosody Inconsistency Inconsistency Lengthened and disrupted coarticulatory transitions between sounds and syllables Lengthened and disrupted coarticulatory transitions between sounds and syllables Inappropriate prosody Inappropriate prosody At least 4 out of 10 features ≤ 323 (significant evidence for CAS) Inconsistency Inconsistency Lengthened and disrupted coarticulatory transitions between sounds and syllables Lengthened and disrupted coarticulatory transitions between sounds and syllables Inappropriate prosody Inappropriate prosody CAS without another comorbid disorder or CAS without another comorbid disorder or Clear cases of comorbid CAS, in which CAS was the primary diagnosis Clear cases of comorbid CAS, in which CAS was the primary diagnosis Inconsistency Inconsistency Lengthened and disrupted coarticulatory transitions between sounds and syllables Lengthened and disrupted coarticulatory transitions between sounds and syllables Inappropriate prosody Inappropriate prosody Cases of CAS in which another comorbid disorder had the same severity (e.g., language delay, dysarthria). 324–373 (some evidence for at least mild CAS) Notes: DEMSS = Dynamic Evaluation of Motor Speech Skill. The most relevant column was used to rate the CAS diagnosis reported in the paper. If more than one was reported, priority was given to one of the two lists of features. A list of 10 features of CAS by the Mayo Clinic (2025) system: (1) vowel distortions; (2) voicing errors; (3) distorted substitutions; (4) difficulty achieving initial articulatory configurations or transitionary movement gestures; (5) groping; (6) intrusive schwa; (7) increased difficulty with multisyllabic words; (8) syllable segregation; (9) slow speech rate and/or slow diadochokinetic rates; (10) and/or equal stress or lexical stress errors. The CAS confidence rating was completed as follows. All authors completed training on the same three randomly selected papers to align diagnostic standards. All papers were independently rated by two authors, and no authors rated any papers for which they were authors. Any discrepancies were resolved by the two authors meeting and discussing the rating until a consensus rating was obtained using the following rules: ∘ If agreed on number but not on a or b, lowest was taken (e.g., 1a / 1b = 1b); If agreed on number but not on a or b, lowest was taken (e.g., 1a / 1b = 1b); ∘ If disagreed on level 3 or 4, the paper was excluded; If disagreed on level 3 or 4, the paper was excluded; ∘ If disagreed on number (e.g., 1 and 2, 1 and 3, 2 and 3, or 2 and 4), a 3rd rater determined the score (not an author on that paper). If disagreed on number (e.g., 1 and 2, 1 and 3, 2 and 3, or 2 and 4), a 3rd rater determined the score (not an author on that paper). Papers not excluded at the previous step received a rating on a scale of 1–4 and papers that did not receive a score of at least 2b were then eliminated. Studies were also assessed for the adequacy of their research designs for the study purpose. Studies with outcome measures which could not be converted to effect size measures, or had insufficient data to analyse, or had no post treatment data or, for SCEDs, had fewer than three measurements in the baseline or post treatment phases, were excluded. Despite a paper being included, individual participants were excluded if there was insufficient data for analysis for that child. Participant data were also excluded if they were reported in other included papers. Please see Material S2 for how each paper was excluded in detail. Risk of Bias Assessment The risk of bias in each included study was evaluated based on the type of study. Group studies and RCTs were evaluated using the 11‐item Physiotherapy Evidence Database—PsychBITE (PEDro‐P) scale adapted for NeuroBITE/speechBITE (Perdices et al. 2009). SCED studies were evaluated using the 15‐item Risk of Bias in N‐of‐1 Trials (RoBiNT) scale (Tate et al. 2013). PEDRO‐P gives a full score of 10, although due to the impossibility of blinding the clinician or the client in most SLT studies, a score of 8 is sometimes considered as a ceiling. The RoBiNT scale has a maximum possible score of 30, and again lack of blinding of clinician and client means a total of 28 could be considered the ceiling in SLT behavioural intervention studies. Both scales were developed for evaluation of allied health research and each has been used widely. It is noted that these risk of bias scores reflect methodological rigour which should reduce bias rather than the clinical effectiveness of interventions and that higher scores are desired but that qualitative descriptions of the risk of bias described by these tools are not statistically validated. Where available, the published PEDro‐P scores from speechBITE were used (https://speechbite.com/). Each of the remaining group studies were independently rated by any two of the first, third, and last authors of this paper, who had received training in the use of PEDro‐P, with no author rating their own paper. Item consensus was achieved through discussion of any items where authors initially disagreed. The scores for the included studies are reported as part of Material S2. SCED studies were evaluated using the RoBiNT scale. Authors 1, 2, 3 and 5 completed the RoBiNT training available from NeuroBITE prior to rating (https://training.neurorehab‐evidence.com/). Each study was independently rated by two authors, neither of whom was an author of the rated paper, with consensus achieved for items where ratings differed. Studies that received a score of 0 or 1 on Item 11 (Independent Variable) were excluded from the analysis as the information about intensity of the treatment required for this present investigation was missing. Effect Measures The effect size used for this meta‐analysis was a standardised mean difference (SMD), of the general form (Mean2 – Mean1)/sd, as with Cohen's d. In these calculations, Mean2 was the mean of the study's main outcome measure following the treatment administered in the study, which could have been obtained from treated or generalisation items. Where data from before‐to‐after treatment were available, Mean1 was the mean of the main outcome measure before treatment was administered. If no pre‐treatment data were available, Mean1 was the mean of the non‐treatment group. We preferred the pre‐to‐post comparison because it represented change associated with treatment, and in many cases where there was a comparison group design, dose was equated between groups and so was unrelated to any difference in treatment outcome between groups. The pooled standard deviation from all included time points formed the denominator for group effect size calculations. Calculation of this SMD is straightforward for group research studies, but not for single case experimental designs studies. The Beeson and Robey (2006) single case d 2 statistic has a superficially similar formula [(Mean2 – Mean1)/SD]. Their Mean1 and Mean2 are equivalent to the group design Mean1 and Mean2, but the SD is not. It estimates SD only from the data of a single person, rather than from data across observations of multiple participants. Within‐person variation is likely to be much less than between‐person variation, suggesting d 2 values are likely to be much larger than standard Cohen's d values (Beeson and Robey 2006; Shadish et al. 2014). SMDs for single case experimental designs can also be inflated by small numbers of observations, and autocorrelation in data series, and d 2 makes no adjustment for either (Shadish et al. 2014). Shadish and colleagues (2014) suggest performing meta‐analysis on single case data with fewer than three study participants using studies with available raw data and converting the data from each study into a SMD statistic adjusting for short series length. Pustejovsky et al. (2024) provide an online calculator (https://jepusto.github.io/SingleCaseES/) which implements their formulae. In this meta‐analysis, the SingleCaseES site was used to calculate adjusted effect sizes for both treated words and generalisation words in SCED studies with fewer than three participants, where the data were available. Firstly, duplicate participants across studies were identified and removed from subsequent papers. For studies with more than one dose frequency, effect sizes were calculated for each dose frequency. For our calculation of effect sizes for SCEDs with fewer than three participants, we averaged data at each timepoint of measurement and calculated d 2 values using the SingleCaseES module (Pustejovsky et al. 2024) to produce d 2 statistics adjusted for short series length. These d 2 statistics are not fully equivalent to a group design d or g because the standard deviation still comes from observations of a single person. For SCEDs with three or more participants, we report an equivalent of between‐groups Cohen's d, calculated using the scdhlm module (Pustejovsky et al. 2023) and corresponding online calculator (https://jepusto.shinyapps.io/scdhlm). In contrast to SingleCaseES, which cannot provide an estimate of between‐person variance that would allow d 2 to be adjusted to a Cohen's d equivalent, scdhlm includes the between‐person variation estimate by having data from at least three participants, and is therefore able to do a full adjustment to between‐groups d, equivalent to the SMDs for between‐groups Cohen's d. Correlations Between Effect Sizes and Dose The main aim of the study was to examine the relationship between changes associated with treatment (effect size of a study) and the six available measures of intensity in the study: length of session; sessions per week; total number of sessions; number of weeks of therapy; total minutes of treatment, and total number of trials. Of these, length of session showed very little variation between studies so was not analysed, and total number of trials across studies had sufficient missing data to be unreliable and therefore was not analysed further. Studies varied in number of participants. The nature of treated items across studies included individual phonemes, nonwords, and real words which makes cross study comparison difficult, however, across studies, generalisation items were all real words. As argued elsewhere (e.g., McCabe et al. 2023), change in generalisation is measured with greatest face validity and therefore we report correlation of generalisation with dose variables here. We wanted to attach greater weight to studies with larger n, so used weighted Pearson correlations between generalisation effect sizes and dose, with number of participants as the weighting factor. Weighted correlation calculations used the wCorr library in R 4.4.1, with number of study participants as the weighting variable. All correlations used generalisation SMD as the outcome variable, and each dose measure in turn as a predictor. Weighted correlation accepts only one predictor variable at a time. To examine combined intensity variables, we used the prcomp command in R 4.4.1 to extract the first principal component of each of the 11 possible combinations of intensity variables. The first principal component is a single variable that summarises what the included measures have in common, more accurately than would an average. The correlations required effect size measures on the same scale across studies. As described above, the group research designs used a standard Cohen's d SMD. The single case studies with three or more participants had fully adjusted SMDs equivalent to a Cohen's d and so were included. The SCED papers with less than three participants were excluded from the correlations as a Cohen's d SMD could not be reliably calculated despite adjusting for short series length, because they include no estimate of between‐person variation. Synthesis Methods The meta‑analysis was conducted using the MAJOR module in Jamovi Version 2.7.6.0. A Restricted Maximum Likelihood (REML) random‐effects model was used to account for expected heterogeneity and to determine the pooled treatment effects across studies. Heterogeneity was evaluated using the I2 statistic (treated words I 2 = 72.4%; generalisation words I 2 = 48.9%). As expected, the I2 values were high due to methodological design differences between the RCTs, non‐RCT group study, and the SCED studies included in this review. Publication bias was not examined due to the mixed data set. Quality of Treatment Evidence Once data extraction was complete, risk of bias measures had been calculated, and omnibus effect sizes calculated, then a Grading of Recommendations Assessment, Development, and Evaluation (GRADE) score for Certainty of Evidence with the ROBINS‐I method (Schünemann et al. 2019) was independently allocated to each intervention type by two researchers. Differences were resolved by consensus. Interventions included in this rating had at least one RCT or at least two included SCED design studies. Results From the initial screening and review of 5002 papers, 84 papers mapped to the inclusion criteria were identified. The process to reach 84 is shown in Figure 1, a PRISMA flowchart (Page et al. 2021), and is detailed in the following section. Figure 2 shows the publication year distribution of the 84 papers initially included in the review. Preferred reporting items for systematic reviews and meta‐analyses (PRISMA) diagram for this review. Histogram of 84 initially included papers by year of publication. As described in the method, further screening was required to corroborate CAS diagnosis and rule out any studies not meeting risk of bias thresholds. Thus, the ultimate number of papers included in the review was 25. The number of papers eliminated at each stage is described below. Of the 84 papers extracted, 27, which had a Confidence in CAS diagnosis of 3 or 4, were excluded. The remaining 57 papers with ratings of 2b or better were reviewed for design adequacy. Of these, 31 papers were excluded for the following reasons: Outcome measure could not be converted to a suitable effect size: n = 1. Outcome measure could not be converted to a suitable effect size: n = 1. Insufficient data to calculate an effect size: n = 2. Insufficient data to calculate an effect size: n = 2. No unique participants identified: n = 2. No unique participants identified: n = 2. No post‐treatment data: n = 7. No post‐treatment data: n = 7. SCEDs with fewer than three measurements in the baseline or post‐treatment phase: n = 19. SCEDs with fewer than three measurements in the baseline or post‐treatment phase: n = 19. Therefore, 26 papers were retained for Risk of Bias evaluation. The five group studies were evaluated for bias using the PEDro‐P scale (Perdices et al. 2009) with scores ranging from 4–8 (M = 6.4, SD = 1.8) which can be interpreted as a moderate to low risk of bias arising from issues with internal or external validity or statistical power (Perdices et al. 2009). The 21 single case experimental design studies were evaluated using the RoBiNT (Tate et al. 2013) scale with scores from 16–24 (M = 19.7, SD = 2.1). These scores can be seen as having moderate to low risk of bias (Tate et al. 2013). One paper (Martin et al. 2016) was excluded at this point due to insufficient data on the independent variable, which included treatment intensity, leaving 25 papers for inclusion in the meta‐analysis. Description of 25 Included Studies Details of the 25 included papers are included in Table 2. The papers consisted of 5 group and 20 SCED studies. The group studies comprised 4 RCT and 1 non‐RCT designs. In some studies, more than one treatment was evaluated. These studies have been broken into their intervention arms and each arm is counted individually. For example, from the Murray et al. (2015) RCT, ReST and NDP are each counted as separate arms. The distribution of arms across interventions in the group studies is therefore: ReST (n = 2), NDP3 (n = 2), ultrasound biofeedback (n = 1), SMC + ultrasound biofeedback (n = 2), SMC (n = 2), MSTP (n = 2). Across the 5 group studies there were 135 unique participants (range = 14–48, M = 27.0, SD = 14.3). TABLE 2Characteristics of included papers.Dosage characteristicsIncluded articleDesignIntervention approach/es by study arms a Reported effect sizeIncluded unique participants (n) Number of sessionsDose (min/ session)Dose frequency (session/ week)Total intervention duration (weeks)Number of treatment phase/sCumulative intervention intensity (min)Reported or estimated cumulative number g of production trials (aka dose) Mean (SD) GROUP STUDIES McCabe et al. 2023 RCTReST r = 0.862 h 12602617201198 (5) # Ultrasound biofeedback712602617202222 (822) # McKechnie et al. 2020 RCTNDP3 i not reported1412604317201874 (488) # Murray et al. 2015 RCTReST d = 1.311312604317201204 (11) # NDP3 d = 2.161312604317201216 (14) # Preston et al. 2024 RCTSMC Distributed versus intensive η 2 = .108 Use of ultrasound η 2 = .052 16206021011200 4960 * , c 4760 * , d SMC + Ultrasound biofeedback82060210112004600 * SMC172060Titrated b 511200 6720 * , c 4240 * , d SMC + Ultrasound biofeedback72060Titrated b 5112008760 * Namasivayam et al. 2015 Non‐RCT MSTP e (low intensity) d = 0.731210451101450Not reported(high intensity) d = 1.85; d = 2.632120452101900Not reported SCED STUDIES Condoluci 2020 SCEDDTTC/IS d1 = −0.17–3.1821660242960Not reportedGomez et al. 2018 SCEDK‐SLPvarious212604317201308 # Gomez et al. 2024 SCEDK‐SLPvarious412604317201176 (276) # Grigos et al. 2024 SCEDDTTC/IS d = .2674545361810Not reportedIuzzini‐Seigel et al. 2024 SCEDDTTC/IS d1 = 1.63–13.902 j 24603811440Not reportedKovacs 2017 SCEDDTTC/IS Distributed d1 = 1.05–1.28; Massed d1 = 2.39–2.65 21650242800Not reportedLim et al. 2019 SCEDDTTC/ISC‐IRD 39%–53%, T‐IRD = 100%110015105215004550 * C‐IRD = 48%–49%; T‐IRD = 100%150155527501175 * Lim et al. 2020 SCEDDTTC/ISIRD = 12%–57%4801510421200Not reportedMaas and Farinella 2012 SCEDDTTC/IS d 2 = 0.62–4.04424503421200Not reportedMaas et al. 2012 SCEDDTTC/IS High frequency d 2 = 1.32; Low frequency d 2 = 0.83–2.55 41660242960Not reportedMaas et al. 2019 SCEDDTTC/IS Massed d 1 = 1.47–3.50; Distributed d1 = 0.61–3.46 61650242800Not reportedNamasivayam et al. 2024 SCEDK‐SLPvarious5121504311800Not calculable, minimum 90 per goal per sessionOlivares 2020 SCEDVideo self‐modellingPND = 62.5%–100%348403821920Not reportedPreston et al. 2016 SCEDUltrasound biofeedback d 2; various328602721680Not reportedPreston et al. 2017 SCEDUltrasound biofeedback d 2; various614602718404115 * Scarcella et al. 2021 SCEDReST d 2; various21260341720Target 1200 * Thomas et al. 2014 SCEDReST d 2; various41260261720Target 1200 * Thomas et al. 2016 SCEDReST d 2; various512604317201188 (112) # Thomas et al. 2018 SCEDReST d 2; various51260 d 431720Target 1200 * Thomas et al. 2024 SCEDReST d 2; various512601121720Target 1200 * Notes. min = minutes. Cumulative intervention intensity (minutes) was calculated as Dose (minutes per session) x Dose frequency (sessions per week) x Total intervention duration (total number of weeks) x number of treatment phases.IS/DTTC = Integral Stimulation/Dynamic Temporal and Tactile Cueing; K‐SLP = Kaufman Speech to Language Protocol; MSTP = Motor Speech Treatment Protocol; NDP3 = Nuffield Dyspraxia Programme, Third Edition; ReST = Rapid Syllable Transition Treatment; SMC = Speech Motor Chaining. r = Pearson's r; d = Cohen's d; d2 = Robey & Beeson's d2 ; C‐IRD = composite Improvement Rate Difference (IRD); TC‐IRD = total IRD (C‐IRD) and total IRD (TC‐IRD = t). a Some group studies made comparison between two intervention approaches. In such cases, both approaches were listed separately. b Titrated frequency of 10 times (twice a day for five days) in the first week, three times a week for two weeks, and two times a week for two weeks giving a total of 20 sessions. c Mean and SD of participants who received SMC via face‐to‐face intervention. d Mean and SD of participants who received SMC via telepractice intervention. e Studies included two groups using the same intervention with different dosage characteristics. f Initial dose reported was number of trials. Dose information in minutes obtained via personal correspondence with the author. g Where an actual mean total trials was provided in the paper it is reported (#), where it was not reported the estimated mean was calculated by multiplying mean session trials by number of sessions or the target number of total productions is reported (*). h Five of the children in the ReST arm of this study had been included in a previous study and therefore were excluded. i Although there were two arms in the design of this study, in the paper the data was combined. j One participant had insufficient data to be included in this analysis. Characteristics of included papers. Distributed versus intensive η 2 = .108 Use of ultrasound η 2 = .052 4960 * , c 4760 * , d 6720 * , c 4240 * , d MSTP e (low intensity) Distributed d1 = 1.05–1.28; Massed d1 = 2.39–2.65 High frequency d 2 = 1.32; Low frequency d 2 = 0.83–2.55 Massed d 1 = 1.47–3.50; Distributed d1 = 0.61–3.46 Notes. min = minutes. Cumulative intervention intensity (minutes) was calculated as Dose (minutes per session) x Dose frequency (sessions per week) x Total intervention duration (total number of weeks) x number of treatment phases. IS/DTTC = Integral Stimulation/Dynamic Temporal and Tactile Cueing; K‐SLP = Kaufman Speech to Language Protocol; MSTP = Motor Speech Treatment Protocol; NDP3 = Nuffield Dyspraxia Programme, Third Edition; ReST = Rapid Syllable Transition Treatment; SMC = Speech Motor Chaining. r = Pearson's r; d = Cohen's d; d2 = Robey & Beeson's d2 ; C‐IRD = composite Improvement Rate Difference (IRD); TC‐IRD = total IRD (C‐IRD) and total IRD (TC‐IRD = t). Some group studies made comparison between two intervention approaches. In such cases, both approaches were listed separately. Titrated frequency of 10 times (twice a day for five days) in the first week, three times a week for two weeks, and two times a week for two weeks giving a total of 20 sessions. Mean and SD of participants who received SMC via face‐to‐face intervention. Mean and SD of participants who received SMC via telepractice intervention. Studies included two groups using the same intervention with different dosage characteristics. Initial dose reported was number of trials. Dose information in minutes obtained via personal correspondence with the author. Where an actual mean total trials was provided in the paper it is reported (#), where it was not reported the estimated mean was calculated by multiplying mean session trials by number of sessions or the target number of total productions is reported (*). Five of the children in the ReST arm of this study had been included in a previous study and therefore were excluded. Although there were two arms in the design of this study, in the paper the data was combined. One participant had insufficient data to be included in this analysis. Across the 20 SCED studies, there were 71 unique participants (range = 1–7, M = 3.6, SD = 1.6). The treatments used in the studies were: DTTC or integral stimulation (DTTC/IS, n = 9), ReST (n = 5), K‐SLP (n = 3), ultrasound biofeedback (n = 2), video self‐modelling (n = 1). A number of SCED studies had more than one intervention phase. Specifically, 9 studies had 2 phases including studies on IS (n = 7), ultrasound biofeedback (n = 1), and video self‐modelling (n = 1). While the number of treatment phases for the SCED studies is included in Table 2 for completeness, it is not further analysed since analysing each phase separately would duplicate participants, adding potential of bias and conflicting with an a priori decision to include each participant only once. Therefore, each of these studies is considered from pre‐treatment to post‐treatment across the whole study rather than pre‐treatment to post‐treatment across each phase. The included studies came from the following countries: US (n = 12), Australia (n = 9), Canada (n = 3), and Italy (n = 1) and were written about children who spoke English (n = 23), Spanish (n = 1) and Italian (n = 1). The therapy agent was primarily a qualified SLT (n = 14), or a student clinician (n = 11) with one study each respectively using a teaching assistant (n = 1) or caregivers (n = 1). Two papers reported both SLTs and student clinicians as agents. Across studies, a range of primary outcome measures were reported including: Whole word accuracy (binary, n = 9); Whole word accuracy (3‐point scale, n = 4); Percent Phonemes Correct (PPC, n = 5); PPC of target sounds (n = 3); Multidimensional scoring (n = 2); Percentage syllables correct (n = 1); and intelligibility (n = 1). Dose Characteristics Reported in Included Papers As shown in Table 2, the number of sessions ranged from 10–100 (M = 21.8, SD = 21.4, mode = 12 sessions); session duration ranged from 15–60 min (M = 56.4, SD = 23.3, mode = 60 minutes); sessions per week (dose frequency) ranged from 1–10 (M = 3.3, SD = 2.2, mode = 2 sessions); the number of weeks (total intervention duration) ranged from 3–12 (M = 5.3, SD = 2.5, mode = 3 weeks); the cumulative intervention intensity (minutes) ranged from 450–1920 (M = 972, SD = 384, mode = 720). Number of trials was reported in less than half of the studies (11/25, 44%) therefore it was not meaningful to use cumulative intervention intensity of production trials across included studies. When it was reported, the cumulative number of treatment trials ranged from 1175–8760 (M = 2780.3, SD = 467.3, median = 1262). This data was heavily biased by the 7 ReST studies and 2 NDP3 studies which each aimed for 1200 trials. This highly skewed data is not analysed further. Effect Sizes The SMD effect sizes are shown in Figures 3 and 4. In each figure, panel A shows the effect size for treated items and panel B shows the effect size for generalisation items. Overall, the effect size for treated items in studies with three or more participants was 1.22 (CI: 0.76‐1.67), a large effect, and for generalisation items was 0.66 (CI: 0.46‐0.85), a medium effect. In studies with one or two unique participants, the effect size for treated items was 2.34 (CI: 1.60‐3.09) and for generalisation items was 1.97 (0.26‐3.69). No validated interpretation scale is available for effect sizes calculated on fewer than three participants. Forest plots of effect sizes (SMD) for treated and generalisation items (group studies and SCEDs with three or more unique participants).Note: More studies reported generalisation data than reported treatment data and all ns are unique children with duplicates removed. IS = Integral Stimulation, including DTTC; K‐SLP = Kaufman Speech to Language Protocol; MSTP = Motor Speech Treatment Protocol; NDP3 = Nuffield Dyspraxia Programme 3rd Edition; ReST = Rapid Syllable Transition Treatment; SMC = Speech Motor Chaining; VSM = video self‐modelling; dis = distributed; in = intensive; RE Model = Restricted Maximum Likelihood (REML) random‐effects model. Forest plot of effect size (SMD) for treated and generalisation items (studies with one or two unique participants). Note: More studies reported treatment data than generalisation data and all ns are unique children with duplicates removed. IS = Integral Stimulation, including DTTC; K‐SLP = Kaufman Speech to Language Protocol; ReST = Rapid Syllable Transition Treatment. Kovacs (2017) mass = massed; Kovacs (2017) dis = distributed. Lim (2019) P1 = participant identifier “Brian”; P2 = participant identifier “Nicholas” (NB: very different cumulative dose across the two participants). RE Model = Restricted Maximum Likelihood (REML) random‐effects model. Two SCED studies were outliers in effect size, one DTTC (SMD = 17.9; Lim et al. 2019) and one ReST (SMD = 5.5; Scarcella et al. 2021). However, preliminary analyses with and without these two studies showed no difference and therefore they were included in the effect size calculations. Relationship Between Dose and Generalisation Effect Size for Studies With Three or More Participants As shown in Table 3, of the components that had sufficient data, only weeks of therapy showed a significant, moderate correlation with SMD using a weighted Pearson's correlation. TABLE 3Principal components correlations between dose measures and effect size for study main outcome # .Dose measureWeighted Pearson correlation with SMD1. Total minutes of therapy−0.022. Number of sessions−0.033. Weeks of therapy * 0.394. Sessions per week−0.261 & 2−0.021 & 3−0.021 & 4−0.022 & 3−0.152 & 4−0.063 & 4 * 0.371 & 2 & 3−0.021 & 2 & 4−0.021 & 3 & 4−0.032 & 3 & 4−0.151 & 2 & 3 & 4−0.03 # study main outcome could be either treatment or generalisation items.* p < 0.10 (two tailed). Principal components correlations between dose measures and effect size for study main outcome # . study main outcome could be either treatment or generalisation items. p < 0.10 (two tailed). Omnibus Treatment Effect Sizes and GRADE Levels In calculating overall effect size for each intervention, most included intervention types were represented in fewer than three studies and therefore omnibus effect sizes are only calculated for DTTC/IS, ReST, and ultrasound biofeedback (Table 4). Across these three intervention types, treated items’ effect sizes ranged from 0.94 to 1.70 and generalisation items’ effect sizes ranged from 0.35 to 0.81. Using the GRADE approach (Schünemann et al. 2019), six interventions were rated as high, indicating for these treatments (DTTC/IS, K‐SLP, NDP3, ReST, SMC, and ultrasound biofeedback) clinicians can be very confident that the reported results are in the range indicated in the studies. Two interventions (MSTP, SMC + ultrasound biofeedback) were rated as moderate confidence, and one (video self‐modelling) was rated as low confidence. TABLE 4CAS interventions omnibus effect sizes from group studies and SCEDs with three or more participants and GRADE levels of evidence.Intervention approachTotal number of papersNumber of treatment arms across papers (group studies + n≥3) Unique participants (group studies + n≥3) Whole of approach level Omnibus effect size d [95%] a GRADE level of evidence (based on all papers)TreatmentGeneralisationDTTC/IS45180.94 [024, 1.63]0.49 [0.14, 0.84]HighK‐SLP229Not calculableNot calculableHighMSTP1233Not calculableNot calculableModerateNDP32227Not calculableNot calculableHighReST66341.70 [1.19, 2.21]0.35 [0.14, 0.56]HighSMC1233N/ANot calculableHighSMC + Ultrasound biofeedback1215N/ANot calculableModerateUltrasound biofeedback3316N/A0.81 [0.30, 1.32]HighVideo self‐modelling113Not calculableNot calculableLow Notes: As some papers compared two types of intervention and/or two different dosage variables, the total number of papers presented here is greater than the 19 included studies (group or SCED studies with three or more participants). a Due to limited studies/treatment arms, only interventions with sufficient studies for calculation (i.e., represented in three or more studies/arms) have the standard deviation and range presented.Abbreviations: SDTTC/IS = Dynamic Temporal and Tactile Cueing (Strand, 2020)/ Integral Stimulation; K‐SLP = Kaufman Speech to Language Protocol (Namasivayam et al. 2024); MSTP = Motor Speech Treatment Protocol (Namasivayam et al. 2015); NDP3 = Nuffield Dyspraxia Programme, Third Edition (Murray et al. 2015); ReST = Rapid Syllable Transition Treatment (McCabe et al. 2020); SMC = Speech Motor Chaining (Preston et al. 2019).min = minutes, M = mean, SD = standard deviation, N/A = not applicable as data on treated words was not reported. CAS interventions omnibus effect sizes from group studies and SCEDs with three or more participants and GRADE levels of evidence. Unique participants (group studies + n≥3) Omnibus effect size d [95%] a Notes: As some papers compared two types of intervention and/or two different dosage variables, the total number of papers presented here is greater than the 19 included studies (group or SCED studies with three or more participants). Due to limited studies/treatment arms, only interventions with sufficient studies for calculation (i.e., represented in three or more studies/arms) have the standard deviation and range presented. Abbreviations: SDTTC/IS = Dynamic Temporal and Tactile Cueing (Strand, 2020)/ Integral Stimulation; K‐SLP = Kaufman Speech to Language Protocol (Namasivayam et al. 2024); MSTP = Motor Speech Treatment Protocol (Namasivayam et al. 2015); NDP3 = Nuffield Dyspraxia Programme, Third Edition (Murray et al. 2015); ReST = Rapid Syllable Transition Treatment (McCabe et al. 2020); SMC = Speech Motor Chaining (Preston et al. 2019). min = minutes, M = mean, SD = standard deviation, N/A = not applicable as data on treated words was not reported. Discussion The aim of this study was to understand the relationship between intensity variables and treatment outcome in treatments used for children with CAS. All included studies show positive treatment effects despite some children in individual studies not improving. In examining the relationship between dose variables and effect size, the headline result is that the number of weeks of therapy has a moderately positive correlation with generalisation effect size and that none of the other dose variables are related to treatment outcome. At first glance it is surprising that no other intensity variable correlated with effect size, as this goes against the received wisdom that more frequent therapy is better for children with CAS. In the final analysis of 25 included papers, some treatments and dose characteristics were more frequently included than others. For example, 9 of 25 reported some form of integral stimulation (e.g., DTTC, ASSIST, etc.) and 7 of 25 included studies about ReST (but no other syllable sequencing studies). In terms of dose characteristics, most reported cumulative doses were above 12 sessions and 720 min of therapy and all bar one study (Thomas et al. 2024) had a dose frequency of two or more sessions per week. This was a result of both the inclusion and quality processes. Therefore, the available evidence does not allow us to make comments about doses less than approximately 1200 trials, dose frequencies below twice per week or above twice per day, or cumulative intervention intensities less than 720 min. These limitations are represented in the lack of correlation, where a lack of diversity in data reduces the likelihood of a correlation. Overall, across treatments and studies, the effect sizes for treated words are roughly double those for generalisation words. This is not a surprising result given treated words are the focus of the intervention and generalisation words show a spread of effect. However, it has been argued that generalisation effect is the more desirable metric as it measures the efficiency of the intervention (e.g., McCabe et al. 2023). Again, unsurprisingly, studies with 1–2 participants tended to show larger effect sizes than those with 3 or more participants. This is likely to be due to (1) less variability in the study (and therefore a lower standard deviation in calculating effect size) and (2) selection bias as smaller studies are often proof of concept or pilot studies where researchers are likely to select suitable children, compared with, for example, an RCT where children are randomised to a group without regard to whether it is the best intervention for them at that time, and (3) the use of smaller within‐person standard deviation as the denominator of SMD, rather than the larger between‐person standard deviation of n > 2 studies. The effect sizes reported in this paper may however be useful as a preliminary benchmark for future CAS treatment studies, should those studies use SMD as the reported effect size. In a number of interventions, including ultrasound biofeedback, ReST, and SMC, both the number of trials reported and those measured for fidelity is the number in the practice phase and may not include any trials that occurred in any pre‐practice phase(s). This is different to other types of therapy such as DTTC/IS, where, if the number of trials is reported, it is for the entire session. This means that reports of the dose per session are not always equivalent across different treatments. No attempt has been made here to look at the nature of the treatments or their active ingredients. For example, it is beyond the scope of this study to examine factors other than treatment type and dose characteristics which relate to change due to therapy. Factors such as severity and previous treatment history are no doubt important but require detailed individual participant data which was not universally available. For example, we could predict that generally the children in ultrasound biofeedback studies would have less severe speech impairment than the children in integral stimulation studies due to the age of the children and the nature of the intervention, but this is beyond the scope of this study. This paper also does not explore the nature of the treatment stimuli or how the stimuli are presented in a session. For example, whether the stimuli are designed to be within the zone of proximal development (ZPD) and relatively straight forward or alternately bootstrapped and therefore more difficult, while some of this is inherent in some treatments (e.g., DTTC has ZPD targets; ReST has bootstrapped ones), other interventions (e.g., concurrent therapy) choose targets with diverse complexity. There is no examination of the method of stimuli presentation which varies from game supported (e.g., NDP3) to intensive drill (e.g., SMC). Likewise, various principles of motor learning which may contribute to the dose decisions in a study, such as whether the trials were blocked or randomised, are not explored. Limitations The recent CAS research priorities report (McCabe et al. 2024) recommended that research about people with CAS should include people with lived experience of CAS at most or all stages of the research. It is a significant limitation of this paper that no person with lived experience of CAS was included in the research team to help guide development, interpretation, and recommendations. While a health sciences librarian was consulted, the trial pre‐registered, the grey literature included where possible, and all measures undertaken by two or more independent assessors, as with any systematic review and meta‐analysis there are limitations to the completeness and interpretation of the results presented. These arise from the inclusion criteria and the quality of the included papers with potential sources of bias in five main categories: Included papers; trustworthiness of CAS diagnosis; quality of study design; weighting of included papers, cumulative dose and dose frequency, and types of treatments; and availability and nature of the included data. A number of limitations surround discussion of the included papers. These are the exclusion of papers which were not written in English (12 papers), the exclusion of papers which had duplicate outcome data (e.g., repeated speech outcome data; Kovacs 2017; Welsh 2017), the exclusion of any manuscript which was a conference presentation or abstract (e.g., Amponsah et al. 2022; Baas et al. 2020), and the exclusion of papers that lacked details necessary to understand the method (e.g., case reports such as Catrini and Lier‐DeVitto 2019) or intensity (Martin et al. 2016). The quality of papers retained from the search process into the systematic review (84) varied considerably and is a source of bias in these results. In particular, since the reporting standards and definitions of CAS have changed over the years, all but one of the papers from before 2010 (Rosenthal 1971) were excluded based on our confidence in the CAS diagnosis. This includes some highly cited but older papers (e.g., Strand et al. 2006). Likewise, standards for reporting of SCED studies have been formalised in the last 10 years and 33 earlier papers published in 2015 or earlier lacked sufficient detail to be included despite representing interesting and important developments in the field (e.g., Rosenthal 1994). With the exception of papers from individual research teams where some standardisation of outcome measures was evident, there was very little homogeneity across papers. This is not a new issue and has been debated since before the ASHA (2007) technical paper with continuing calls for standardisation in recent papers (e.g., Maas 2024; McCabe et al. 2024). While this paper analysed the primary speech outcome measure from each study, regardless of the measure, and calculated standardised effect sizes for comparison, it is possible that in doing so, the best we can hope for is a description of fruit rather than the apples, oranges, blueberries and mangosteens that are being reported. Likewise, the primary outcomes reported for some papers were both treatment and generalisation data (e.g., McKechnie et al. 2020; Murray et al. 2015) while for others only treatment (e.g., Olivares 2020) or only generalisation data (e.g., McCabe et al. 2023; Preston et al. 2024) was reported, again potentially reducing the validity of the overall findings. Even in included papers, detailed information about individual participants and methods was often missing, despite being “available on request”. Use of standard data repositories such as PhonBank (https://talkbank.org/phon/) or Open Science Framework (https://osf.io/) could have increased the number of studies which were included in the primary analysis. Finally, when we registered this study with Prospero, we believed we would be able to answer the questions in the title: “Does more speech therapy lead to better results in Childhood Apraxia of Speech?” Unfortunately, the data included in the published papers does not allow us to conclusively answer the question. However, we have identified the dose components that need to be reported in future studies, in order for such a question to be answered. Recommendations for Data Collection and Reporting in Treatment Studies One of the difficulties in conducting this review was comparison across studies, as not all studies reported all intensity characteristics. Researchers are asked to report all dose characteristics: duration of sessions, number of trials (dose), number of sessions, frequency of sessions, length of the block. There is also evidence that it is informative to collect data on the time use within sessions, such as the amount of time spent on active therapy compared with the amount of time encouraging a child to participate (Abu Awwad et al. 2026). This is important as the data from child language therapy suggests that school aged children receiving group therapy are actively involved in treatment for only 2.9 min per therapy session (Schmitt 2020). Allowance for counting number of trials from recordings should be made in writing of funding applications. For future meta‐analysis, researchers are encouraged to report or make publicly available individual participant data including the mostly unreported variable of previous intervention. Likewise, researchers are encouraged to use the RoBiNT and PEDro‐P scales in designing their studies, especially when reporting treatment design details (e.g., dosage information and outcome measures), which will help improve the design of treatment studies. Implications for Practice Clinicians and families can be confident that if one of the included treatments is selected to meet the child's current speech needs and provided at least twice per week for at least 12 sessions for at least 45 min, measurable change can be expected in treated items, with generalisation to untreated items and that the more weeks of therapy that are provided, the more change will be seen in the child's speech. These results highlight the importance of ensuring that children receive a sufficient duration of intervention. The number of sessions, both expressed as weeks of therapy and total number of sessions, in any study is not a random variable nor does it represent the best possible duration or intensity of a block of therapy in clinical practice. Rather, in most cases, it is determined by factors such as previous research, available funding, or convenience (e.g., a summer break program). Therefore, clinicians should consider that even when children demonstrate significant improvement within a treatment study after, for example, 12 or 24 sessions, therapy may need to continue until the child has achieved their functional communication goals across relevant activity and participation contexts. In many cases, achieving these outcomes will require therapy to be delivered over multiple years, although longitudinal data supporting this assertion is thin. An additional caveat needs to be included, which is there is no evidence from this or any other study that therapy needs to be continuous and in fact a few studies show that following sufficient therapy, as described in the included studies, a number of children may continue to improve in their speech in a non‐treatment period (e.g., Murray et al. 2015). For several of the included studies, dose was reported as number of speech trials per session as well as number of minutes per session (e.g., Gomez et al. 2024; Preston et al. 2017). Number of production trials provides a much more accurate measure of therapy intensity than minutes alone. Clinicians therefore should deliberately and conscientiously attend to the dose of their treatment (number of treatment trials per session and possibly the number of minutes), following the recommended dose where it exists and the dose reported in the literature. Not only does each session require a focus on dose, the cumulative intervention intensity (the dose, session frequency, and duration of the therapy block) needs to be sufficient. Providing fewer than 10 sessions has not been tested in any of the included literature and therefore the common pattern of therapy once per week across a 10‐week school term (e.g. Gomez et al. 2018), for example, is not current evidence‐based practice. Where service constraints to sufficient cumulative dose cannot be eliminated completely, clinicians could consider condensing treatment into shorter, more intensive blocks. One clinical model (Murray 2018) involves pairing two children with CAS or other severe speech sound disorder. Rather than each child receiving 12 weeks of once‐weekly therapy, each receives a 6‐week block of twice‐weekly therapy, alternating with the other child. This approach maintains the same total number of sessions per child, without reducing occasions of service or billable hours. Another alternative is to provide interventions which have been shown to be effective when delivered by trained student SLTs, teacher's aides or speech and language therapy assistants (SLT‐As) or to share the weekly dose across two SLTs or an SLT and an SLT‐A (e.g., Lim et al. 2019). Not all treatments have been shown to be appropriate for delivery outside the SLT environment (e.g., ReST; Thomas et al. 2018), so it is important to consider the interaction between the child's needs, the treatment approach, the delivery agent, and training requirements. Therapy for CAS is effective when provided as described in the literature, with the overall effect size for CAS treatment and generalisation in the large and medium ranges respectively. The following treatments (in alphabetical order) can therefore be used with moderate to high confidence of an effect for individual children, noting that not all treatments are suitable for all children and professional advice is required to select an intervention: Integral stimulation including DTTC, K‐SLP, MSTP, NDP3, ReST, SMC, ultrasound biofeedback, SMC + ultrasound biofeedback. Families of children with CAS have expressed a desire to understand more about CAS and its treatment so they can actively contribute to decision making regarding their child's care (Dawoud et al. 2025; Highman et al. 2025; Thomas et al. 2025). Clinicians, therefore, should talk with families about the dose evidence, indicating that speech production treatment has not yet been shown to be efficacious at a frequency of less than two sessions per week and that the number of weeks of therapy is an important factor in treatment effectiveness. We know that CAS treatment is a rapidly evolving field with several close‐to‐publication studies which could not be included in this systematic review, for example, from a recent as yet unpublished RCT (Iuzzini‐Seigel et al. 2025), showing that DTTC may have similar efficacy when 24 h of therapy are provided across either two or four sessions per week. Clinicians and researchers alike need to be open to new intensities and service delivery models which may emerge, as has happened with the recent Preston et al. (2024) paper which examined a titrated dose, with an intensive start and diminishing intensity over the course of treatment, in effect a ‘jump start’. How these novel approaches change intervention will be interesting to watch, certainly, the 1–2 sessions per week for school terms often found in clinical practice (e.g., Gomez et al. 2019, Gomez et al. 2022; Wong et al. 2023) is being challenged by these models. Conclusion Overall, this systematic review shows that treatments for CAS are effective when delivered with the intensity typically reported in the literature, with weeks of therapy emerging as the only dose variable consistently associated with treatment outcome. These results highlight both the importance of sustained intervention and the need for more standardised reporting and comprehensive data to clarify optimal treatment intensity. Conflicts of Interest The authors declare no conflicts of interest. Generative AI Statement No AI was used in the creation of this article. Supporting information Supporting Information: jlcd70337‐sup‐0001‐SuppMat.docx Supporting Information: jlcd70337‐sup‐0002‐SuppMat.docx Acknowledgements Thanks to authors of treatment papers who responded to our requests for additional information about their work. Thanks to speechBITE for sharing PEDro‐P scores and specifically to Emily Yun for support with training in the rating scale. Open access publishing facilitated by The University of Sydney, as part of the Wiley ‐ The University of Sydney agreement via the Council of Australasian University Librarians Data Availability Statement The data for this paper is available from: https://ses.library.usyd.edu.au/handle/2123/35577 References Associated Data Supplementary Materials Supporting Information: jlcd70337‐sup‐0001‐SuppMat.docx Supporting Information: jlcd70337‐sup‐0002‐SuppMat.docx Data Availability Statement The data for this paper is available from: https://ses.library.usyd.edu.au/handle/2123/35577