A single case experimental design study using an operationalised version of the Kaufman Speech to Language Protocol for children with childhood apraxia of speech

Abstract

Purpose

A Phase I study was conducted to examine the treatment effectiveness of the Kaufman Speech to Language Protocol using a research-operationalised protocol. It was hypothesised that articulatory accuracy would improve as a result of the treatment and that these improvements would be maintained after treatment was discontinued.

Method

A single case experimental design was used to evaluate the effectiveness of the Kaufman Speech to Language Protocol. Four children with a confirmed childhood apraxia of speech diagnosis were included in this study. Each child received 12 individual 1 hr treatment sessions that each consisted of an approximation setting phase and a practice phase. Probe data was collected during treatment and at post-treatment time points to measure treatment effectiveness and to measure changes in the untreated words. Untreated (control) sounds were included to test whether recorded improvements in articulatory accuracy could be attributed to the Kaufman Speech to Language Protocol.

Result

Two of the four children demonstrated a response to the intervention and maintenance of these changes, while the two remaining children demonstrated some generalisation in the absence of improved target (treatment) words. No specific child factors were clearly associated with positive treatment outcomes.

Conclusion

This study replicated the findings of an earlier pilot study and found that the operationalised protocol for the Kaufman Speech to Language Protocol is effective in improving articulatory accuracy for some children with childhood apraxia of speech. Additional replication with a further refined treatment protocol and a larger sample size is needed to support a recommendation of clinical use of the Kaufman Speech to Language Protocol.

Keywords:

• childhood apraxia of speech
• treatment
• Kaufman
• dyspraxia
• speech therapy
• single case experimental design

Introduction

Childhood apraxia of speech (CAS) is a neurological motor speech disorder where the overt speech characteristics of inconsistent production of consonants and vowels, errors at the level of coarticulatory transitions between sounds and syllables, and prosodic errors are attributed to disrupted motor speech planning and programming (American Speech-Language-Hearing Association [ASHA], 2007). Although a number of theories exist to account for CAS (see, for example, Namasivayam et al., 2019), one current theoretical understanding of CAS is that these observable speech features result from unstable or underspecified representations of speech motor plans, which interfere with the feedforward control system (Terband & Maassen, 2010). Using this psycholinguistic understanding of CAS, researchers and clinicians can look for treatments that address either the surface motor speech features described by ASHA (2007) or the underlying processes and pathways described in the theoretical literature (van der Merwe & Steyn, 2018).

In the period 2008–2020 and following the publication of the consensus features in the ASHA (2007) technical report described above, a number of peer-reviewed papers described original treatment research in children with CAS. Many of these described experimental treatments (e.g. Rapid Syllable Transition therapy or ultrasound biofeedback) that are reported to not be widely used in clinical practice (Gomez et al., 2019; Randazzo, 2019). A recent Cochrane review identified that limited evidence existed in the literature to support the clinical use of Nuffield Dyspraxia Programme-Third Edition (NDP-3) and Rapid Syllable Transition treatment (ReST; Morgan et al., 2018). Morgan et al. (2018) also observed that there was no high level evidence to support other interventions commonly used to treat CAS. While such robust research (e.g. randomised controlled trials and meta-analyses) is being developed, we need to consider lower levels of evidence to direct treatment decisions.

Only a few studies have examined treatments that clinicians report using in therapy such as the Nuffield Dyspraxia Programme (Murray et al., 2015), which is widely used in Australia and New Zealand (Gomez et al., 2019). PROMPT© (Dale & Hayden, 2013), Dynamic Temporal and Tactile Cueing (Strand, 2020), and more recently the commercially available Kaufman Speech to Language Protocol (K-SLP; Gomez, McCabe, & Purcell, 2018) are used more frequently by clinicians to treat CAS in the US and Canada (Gomez, McCabe, & Purcell, 2018). Although the K-SLP is a commonly used intervention for CAS in the US (Gomez, McCabe, & Purcell, 2018), to date only two studies have evaluated the K-SLP (i.e. Gomez, McCabe, Jakielski, & Purcell et al., 2018; Tierney et al., 2016). Only the Gomez, McCabe, Jakielski, & Purcell (2018) study used a standardised and operationalised protocol to ensure consistent application of the intervention that was repeated across multiple participants.

In order for treatments to be research evidence-based, they need to go through a clinical trial process, which necessitates standardising an intervention that otherwise may be used more eclectically or without a strict operationalised procedure in clinical settings (see Gomez et al., 2019). In the current study, standardisation was achieved through implementing a consistent method that addressed how the stimuli would be selected for each treatment session; a structured cueing hierarchy and feedback system (designed following consultation with the author of the K-SLP; N. Kaufman, personal communication, May 22, 2015); a method for determining the best approximations; baseline and experimental testing; a distinct phase to establish the best approximation for treatment words; and a distinct practice phase. Most of these elements were either loosely structured in the original K-SLP treatment manual (e.g. cueing hierarchy, feedback) or unspecified (distinct phases to establish approximations and practice these approximations; Kaufman, 2013). Standardisation can have an effect of distancing the treatment from the reality of clinical practice (see Wears, 2015), however, it is a beneficial step in demonstrating the value of treatment. Tierney et al. (2016) reported a case report utilising an eclectic approach combining elements of non-speech oral motor therapy, sign language (i.e. unaided alternative communication), and the K-SLP. Unfortunately, the protocol for the intervention was not described and therefore does not allow for replication.

Interventions for CAS can be grouped into motor-based interventions, linguistic interventions, and alternative and augmentative communication (AAC; see Murray et al., 2014, for a discussion of the categories of CAS interventions). Unlike DTTC, PROMPT©, and the NDP-3, which are all motor-based clinically derived interventions, the K-SLP is a hybrid motor-linguistic-based intervention that uses progressive approximations of adult word forms as therapy targets to teach children with CAS. It accomplishes this overarching goal by applying developmentally common phonological processes to simplify the target word, so that the target word falls within the child’s zone of proximal development (Kaufman, 2013; Vygotsky, 1978). The resulting treatment target approximation is frequently the best production the child is able to attempt, but it is not necessarily error-free (Gomez, McCabe, Jakielski, & Purcell, 2018; Kaufman, 2013) as it contains phonological simplifications that are intentionally integrated into practice to both linguistically and motorically simplify adult word forms. Linguistic-gestural and articulatory-phonology models of speech motor control also propose that these “errors” (referred to as simplifications in this article) occur to motorically simplify the original word (see Namasivayam et al., 2019). Although the theoretical underpinning for changes resulting to speech using the commercialised K-SLP are unclear, the K-SLP is an approach that appears to challenge the motoric system in a developmentally sequenced way by gradually increasing the complexity of the target word approximation at all levels of the intervention (e.g. word, carrier phrase, sentence; Kaufman, 2014).

These targets are repetitively practiced in therapy using a quasi-errorless therapy model (i.e. “error reducing” techniques, see Fillingham et al., 2003), by anticipating the child’s difficulties with any given target—whether it’s the adult target of the word or the child’s best approximation. The goal of treatment is to then provide cues that will maximise success and minimise erroneous productions, which in the case of the K-SLP is defined as any production that does not correspond to the model provided to the child by the clinician (a predetermined approximation or the adult form). Even with the use of anticipatory cues to facilitate production, true errorless learning may be difficult to achieve as the child with CAS may still incorrectly imitate the target (see Fillingham et al., 2003, for a general discussion on errorless learning).

In addition to the K-SLP, there is one other intervention that has been adapted for CAS treatment that uses a child’s best production to facilitate improved motor planning and programming for speech—the modified core vocabulary therapy (mCVT; Iuzzini & Forrest, 2010). Here, best approximation is the child’s production not a predetermined approximation contained within the program as it is in K-SLP. Iuzzini and Forrest (2010) demonstrated that an approach that uses a child’s best approximation (even if it does not correspond to the adult target of the word), coupled with stimulability exercises (e.g. see Miccio & Elbert, 1996) can in fact result in improved speech accuracy and improved consistency of speech production. Preliminary results for K-SLP in a 2018 study by Gomez et al. demonstrated that practicing successively more accurate phonological forms of an adult target word might lead to improvements in speech accuracy.

Treatments designed to improve the consistency of word production have been sporadically reported in the literature in relation to children with speech disorders (e.g. Forrest & Elbert, 2001), however, only Iuzzini and Forrest (2010) used such an approach with children with CAS. We posit that similar to the mCVT treatment, the K-SLP may act to create a stabilising effect on the child’s speech system by allowing the child to establish consistent, accurate motor forms based on the adult model provided, which may include erred targets (i.e. approximations). This may ultimately improve the child’s ability to create and use stable motor plans/programs. We further posit that this stability is achieved by repetitive practice of developmentally simplified motor and linguistic plans, allowing for a stable feedforward signal to be established for future productions of the same word (see Terband et al., 2009). Given that the K-SLP aims to establish approximations that successively more closely resemble adult word forms, with each approximation being stabilised before a new one is added (i.e. a stable motor program for a given adult target), we propose that, if effective, the K-SLP may have a stabilising effect on the child’s speech system.

Gomez and colleagues (Gomez, McCabe, Jakielski, & Purcell, 2018) reported a Phase I pilot study with two children and reported that the operationalised K-SLP was effective with one of the children. Early studies (such as Phase I studies) are necessary to establish and refine a treatment protocol, to test for safety, and to identify whether there is a treatment effect. Given that Phase I studies focus on developing and refining treatment protocols, one can expect multiple iterations of the protocol before it is finalised and ready to be used in larger-scale studies (Robey, 2004). Although the K-SLP is a widely used, commercially available program with a manual designed to support implementation of this treatment, Gomez, McCabe, Jakielski, et al. (2018) were the first researchers to operationalise the treatment approach for a research context. Gomez, McCabe, Jakielski, & Purcell (2018) identified several limitations with the operationalised protocol that needed to be refined for improved research fidelity and to make the protocol more user-friendly. Specifically, the researchers identified the need to address vowel transcription reliability, the complexity of the standardised K-SLP methodology, and a more purposeful establishment of experimental control. Despite the clear lack of research evidence supporting its use, the K-SLP is commonly used in the US and Canada (Gomez, McCabe & Purcell, 2018; Randazzo, 2019), suggesting that clinicians are relying on practice-based evidence (see Dunst, 2009) to inform clinical practice. The effectiveness of the K-SLP deserves to be investigated so that diverse stakeholders can be confident in its effectiveness, efficacy, and efficiency.

Early efficacy studies constitute the foundation of larger and more rigorous research designs that evaluate an intervention in both research (Phase III) and community (Phase IV) contexts (Robey, 2004). At this stage, the K-SLP has demonstrated a therapeutic effect for one child in a small-scale Phase I pilot study; however, further empirical research evidence is needed to determine if positive treatment effects can be replicated across a larger number of children with CAS.

The current study aimed to test the effectiveness of a refined K-SLP protocol with a larger sample of children with CAS. We hypothesised that intervention with the K-SLP would result in:

1. improved articulatory accuracy for target words used in treatment, measured in percent phonemes correct (PPC);

2. maintenance of the articulatory accuracy of target words in treatment PPC; and

3. no improvements in the accuracy of unrelated, untrained speech sounds or words.

Method

This study was approved by The University of Sydney Human Research Ethics Committee (2015/361) and was registered with the Australia and New Zealand Clinical Trials Registry (ACTRN12615000695505) as a clinical trial.

Inclusion criteria and recruitment

The inclusion criteria for this study were: children aged 3;0–5;11 (years; months) with suspected CAS; normal (or adjusted to normal) vision; normal (or adjusted to normal) hearing acuity; English as their first language, along with at least one parent who spoke English as their first language; no known concomitant diagnoses of neurodevelopmental disorders including autism spectrum disorder, neurological deficits, intellectual impairment, and attention-deficit/hyperactivity disorder; and no structural/oral motor anomalies.

Children who met these criteria were assessed to determine if they displayed the ASHA (2007) consensus features for CAS. Specifically, the children were required to have measurable difficulties with coarticulatory transitions (perceptually realised as sound or syllable segregation or inappropriate pausing in phrases), prosodic errors (measured by the presence of stress pattern errors in words or phrases), and inconsistent production of consonants and/or vowels in repeated productions of words.

Research participants were recruited through digital media (e.g. Facebook, Twitter, the university website), flyers posted at the university’s Communication Disorders Treatment & Research Clinic, and emails sent to speech-language pathologists (SLPs) and parents who registered to receive notifications about research occurring through The University of Sydney. Five children were initially recruited for this study; however, only four children met the inclusion criteria after telephone screening.

Participants

Four male children received a CAS diagnosis and were enrolled in the research study. Pseudonyms are used to protect the identity of the children. The children and their ages were: Billy (4;6), Sam (4;6, Billy’s monozygotic twin), Max (4;8), and James (4;5). No other speech therapy was received by the children from the start of the baseline pre-treatment period until after the 1 month post-treatment follow-up probe.

Assessments

Each child participated in an initial evaluation across at least two sessions. During the assessment, each participant had their hearing screened based on an adapted version of the American Speech-Language-Hearing Association (ASHA; 1997) audiological screening protocol to account for the low levels of ambient noise in the clinic room. The children were required to have either normal or adjusted to normal hearing in at least one ear.

If the child had not completed a language assessment within the last 12 months, the subtests required to calculate the receptive language index of the Clinical Evaluation of Language Fundamentals-Preschool Second Edition (CELF-P2; Wiig et al., 2006, Australian standardisation) were administered. The Peabody Picture Vocabulary Test-Fourth Edition (PPVT-4; Dunn & Dunn, 2007) was administered to assess receptive vocabulary skills, while the Oral and Speech Motor Control Protocol (Robbins & Klee, 1987) was conducted to rule out any structural anomalies.

The assessments used to confirm a CAS diagnosis included the Diagnostic Evaluation of Articulation and Phonology (DEAP) inconsistency subtest (Dodd et al., 2002) to assess token-to-token phonemic inconsistency at the word level¹. We used the criterion established by the DEAP inconsistency subtest where an inconsistency of at least 40% (i.e. 10/25 words) was indicative of “inconsistent speech”. A polysyllabic words task (Gozzard et al., 2006), the DEAP inconsistency subtest, (Dodd et al., 2002), the Goldman Fristoe Test of Articulation-Second Edition (GFTA-2; Goldman & Fristoe, 2000), and the diadochokinetic speech tasks from the Oral and Speech Motor Control Protocol (Robbins & Klee, 1987) were used to perceptually assess difficulties with coarticulatory transitions and the presence of prosodic errors, specifically inaccurate stress patterns, sound/syllable segregation, and syllable deletion/addition. Given that the gold-standard for CAS diagnosis is expert opinion, no verified benchmarks were available to guide the clinicians in determining when an ASHA consensus feature was established. With the exception of the inconsistency feature, the remaining CAS features were considered established when the first and second author, both experienced clinicians in diagnosing and treating CAS, agreed that evidence of that feature was displayed twice in at least two of the speech tasks described in Table I. Additionally, the children assessed demonstrated multiple additional features of CAS including: idiosyncratic substitution errors, omissions, distortions, assimilation errors, and voicing errors, with relatively fewer typical phonological substitution errors. To operationalise the assessment procedure, the authors reported the occurrence of prosodic and lexical stress errors, syllable segregation, and inconsistency.

Table I. Summary of pre-treatment speech, language, and hearing findings.

Pre-treatment assessments
Assessment	Billy (4;6)^+			Sam (4;6)			Max (4;8)			James (4;5)
	Score	Percentile	Result	Score	Percentile	Result	Score	Percentile	Result	Score	Percentile	Result
GFTA 2^a	^64	4	**LNR	*88	21	*LNR	*89	23	*LNR	*56	1	**LNR
DEAP inconsistency assessment^b	80	−	IS	76	_	IS	76	_	IS	80	_	IS
Polysyllables word test^c
% Consonants correct	48	−	−	51	_	_	49	_	_	30	_	_
% Vowels correct	65	−	−	70	_	_	60	_	_	69	_	_
% Phonemes correct	55	−	−	59	_	_	54	_	_	46	_	_
Words with stress pattern errors^d	32/50	−	−	30/48			32/49			25/50
Syllable segregations^e	44	−	−	41			18			43
Oral and Motor Speech Protocol^f
Structure	24	−	WNL	24		WNL	24	_	WNL	24	_	WNL
Function	94	−	**LNR	90		** LNR	88	_	**LNR	88	_	**LNR
CELF-P2^g
Receptive language index	92	30	WNL	104	61	WNL	83	13	*LNR	103	58	WNL
Expressive language index	−	−	−	11	63	WNL	_	_	_	77	6	**LNR
PPVT-4^h	117	87	WNL	126	96	WNL	89	23	WNL	116	86	WNL
Hearing screen^i	Passed binaurally			^#Passed binaurally			Passed binaurally			Passed binaurally

^aGFTA 2: Goldman Fristoe Test of Articulation-Second Edition (Goldman & Fristoe, 2000).

^bDEAP: Diagnostic Evaluation of Articulation and Phonology (Dodd et al., 2006).

^cGozzard et al. (2006).

^dIdentified perceptually as inaccurate stress pattern (mismatch between modelled target and child’s production).

^eIdentified perceptually as a pause between sounds and/or syllables within a word.

^fRobbins and Klee (1987).

^gCELF-P2: Clinical Evaluation of Language Fundamentals-Preschool Second Edition (Wiig et al., 2006).

^hPPVT-4: Peabody Picture Vocabulary Test-Fourth Edition (Dunn & Dunn, 2007).

ⁱHearing screen adapted from ASHA (1997).

WNL = within normal limits; LNR = lower than normal range; IS = inconsistent speech.

^{^}Assessment was completed by the child’s private speech pathologist less than 12 months before treatment was administered.

^#Passed on the second test that was administered (date). Failed the first attempt (date).

⁺ = years; months reported.

* = mild.

** = severe.

The Goldman-Fristoe Test of Articulation-Second Edition (GFTA-2; Goldman & Fristoe, 2000) was also administered to provide a means of comparing each child’s speech skills to that of their peers. A summary of the assessment results are shown in Table I and a description of each child’s assessment results follows.

All children passed their hearing screen binaurally and demonstrated age-appropriate receptive vocabulary skills. Each child had oral structures that were typical for their age, although their oral and speech motor skills were lower than normal. Billy, Sam, and James all had age-appropriate receptive language skills, while Max’s receptive language score was consistent with a mild developmental language disorder (receptive).

Intervention

Design

This study used a single case across multiple behaviours experimental design. Each child participated in three pre-treatment probes to establish a baseline followed by 12 individual 1 hr treatment sessions across three consecutive weeks (four per week). Three post-treatment follow-up sessions were used to assess maintenance and generalisation of treatment gains. Frequency of treatment was designed to replicate the previous K-SLP study (Gomez, McCabe, Jakielski, & Purcell, 2018) and to be similar to other recent CAS treatment studies (e.g. Murray et al., 2015). Following the initial K-SLP study (i.e. Gomez, McCabe, Jakielski, & Purcell, 2018), several changes to the protocol were made, which are shown in Table II.

Table II. Summary of changes to the K-SLP protocol following Gomez, McCabe, Jakielski, Purcel (2018).

Protocol changes	Gomez, McCabe, Jakielski, Purcel, 2018	Modified K-SLP	Rationale for changes
Training	Training provided through models provided by an experienced K-SLP trained clinician, along with limited videos of the treatment process made available by the Kaufman Children’s Centre (refer to Supplemental Material). Minimal training provided to develop proficiency transcribing severely disordered speech.	Enhanced training through additional treatment video examples from the Phase I trial (Gomez, McCabe, Jakielski, & Purcell, 2018), along with role playing between clinicians to practice the K-SLP process and techniques. Training and transcription practice provided to transcribers until a high level of inter-reliability obtained (target 80–85%).	Training was enhanced to improve treatment fidelity and reliability.
Protocol	Wording within the protocol and therapy criterion (i.e. production to time-based criterion) changed during the therapy, reflecting a pilot design (Gomez, McCabe, Jakielski, & Purcell, 2018).	Improved wording of instructions for better clarity. Protocol and therapy criterion stable (i.e. time-based criterion implemented from the first treatment session). The protocol used for this research study is available as Supplementary Material.	We sought to improve treatment fidelity by improving the clarity of instructions in the treatment protocol. Additionally, modifying the within-therapy criterion was important to better engage the children in therapy and to reduce the cognitive burden on the treating clinician.
Therapy progression	To progress from word to phrase level, 80% accuracy had to be achieved on the child’s current level of approximation on at least 12 productions of a word at the level of “independent production” in the cueing hierarchy (c.f. Gomez, McCabe, Jakielski, & Purcell, 2018).	To progress from word to carrier phrase level, the child had to achieve accurate productions on at least 5/6 productions on their current level of approximation with the use of the “oral posture” cue.	These changes were made to reflect the relatively fast progression from word to carrier phrase level in the original K-SLP manual (Kaufman, 2014).
Treatment word selection	Ten words were randomly selected from possible 20 words (two word shape goals) each day.	Five words were randomly selected from each of the two word shape goals, thus ensuring that both word shape goals were targeted each day.	This was done to ensure equal treatment dosage across the two word shape goals.
Feedback	Feedback was faded across each treatment session (i.e. 100% feedback on the first 10 trials, 90% feedback on the subsequent 10 trials, etc.). The average feedback across each session was 55%.	Feedback was faded across each treatment week (i.e. every four sessions). Average feedback across the 12 sessions was 80%.	This change was made to simplify the feedback schedule for the treating clinicians.

Probes

Children attended probe sessions to measure articulatory accuracy (i.e. PPC) for baseline (three visits), acquisition (two visits—treatment week 2 and week 3), maintenance (three post-treatment visits), and generalisation (measured in all probe visits). The child’s individualised list of words comprised 30 treated words that were selected as per the method described in the next section (goal selection), and included 20 word shape goals and 10 functional words. There were also 35 untreated words, which were matched to the treated targets in the word shape and phonemes (to measure generalisation), 10 treated carrier phrases (common two- to three-syllable phrases), 10 untreated phrases (matched based on the number of syllables and the phonotactic structure of the phrases), and five control words containing two- and three-element clusters that were not expected to change. Each word was elicited once, unless the production was unclear or if there was loud background noise. The word order was randomised for each probe session and each child’s productions were transcribed using broad phonemic transcription by a speech-language pathologist who was experienced in transcribing disordered speech and who was not either the assessing or treating clinician. PPC was calculated for each word individually and then combined to represent the overall PPC for each separate goal (e.g. CVCVCV words, where C represents consonants and V represents vowels). To control for confounds from increased familiarity with children’s speech, the SLP was blinded to the sequence of the probe sessions but was not blinded to the study or the participants.

Target word selection

Three separate goals were established for each child, including two goals (10 words per goal, 20 words in total) with specific word shapes that were frequently produced in error during the pre-treatment assessment, irrespective of the child’s ability to accurately produce individual sounds. The third goal was a functional word list (10 words in total), which were included in the 2 minute break activity. These functional words reflected words that occur frequently in speech and that can be easily elicited during play (e.g. stop, finish, play). Ten carrier phrases were predetermined to be used at the phrase level if the child met the step-up criteria of 5/6 trials of a word matching the adult model at the oral posture stage of the cueing hierarchy (discussed in the Treatment section). Additionally, 10 corresponding untreated carrier phrases were developed to measure generalisation at this level. No child progressed to the carrier phrase level over the course of the 12 treatment sessions. Treated goals for each child have been included in Table III. Target words for each session varied as only 10 words were randomly selected for each session (i.e. five words from each word shape goal), thereby incorporating elements of both variable and distributed practice.

Table III. Treated words used in individual therapy sessions.

	Drill therapy targets (words)		Play therapy targets
Billy	Bisyllabic + CV(C)	Polysyllabic	Functional words
	Bottle top	Animal	Stop
	Muddy boot	Domino	Show
	Honey pot	Piano	Outside
	Pony ride	Tomato	Help
	Panda bear	Banana	Push
	Dirty hat	Potato	Open
	Paddle boat	Lavender	Purple
	Nobody	Table top	Don’t
	Tummy hurt	Tornado	Finish
	Happy day	Bicycle	Sleeping
Sam	Bisyllabic + CV (C)	Polysyllabic	Functional words
	Bottle top	Animal	Open
	Muddy boot	Domino	Stop
	Honey pot	Piano	Outside
	Pony ride	Tomato	Help
	Panda bear	Banana	Little
	Dirty hat	Potato	Purple
	Paddle boat	Lavender	Yellow
	Table top	Nobody	Kick
	Tummy hurt	Tornado	Catching
	Happy day	Bicycle	seven
Max	Bisyllabic	Polysyllabic	Functional words
	Napkin	Animal	Stop
	Candle	Domino	Shut
	Tablet	Piano	Close
	Finger	Tomato	Slowly
	Music	Banana	Faster
	Chicken	Potato	Yellow
	Panda	Lavender	Please
	Handle	Nobody	Play
	Pumpkin	Tornado	Finish
	Pencil	Bicycle	Five
James	Bisyllabic	Polysyllabic	Functional Words
	Napkin	Animal	Out
	Candle	Domino	Off
	Tablet	Piano	Open
	Finger	Tomato	Shake
	Music	Banana	Purple
	Chicken	Potato	Yellow
	Panda	Lavender	White
	Handle	Nobody	Play
	Pumpkin	Tornado	Help
	Pencil	Bicycle	Show

Note. Drill therapy refers to the words included in the 8 minute portion of the practice phase, while the play therapy words refer to the targets included during the 2 minute play time (at the end of each 8 minute drill therapy block).

Target words

Most of the target words along with their respective approximations were sourced from the Kaufman (K-SLP) Treatment Kits 1 and 2, although some approximations were modified to reflect Australian English vowels. Some words that were inappropriate in the Australian cultural context (e.g. Idaho) or that had multiple possible realisations in Australian English (e.g. petunia) were also replaced with phonotactically similar targets (e.g. lavender). Consistent with the previous research (see Gomez, McCabe, Jakielski, & Purcell, 2018), the standardised protocol recognised that prosodic errors are prevalent in children with CAS (ASHA, 2007; Shriberg et al., 1997). We therefore did not model inaccurate prosody such as syllable/sound segregation or inaccurate syllabic stress patterns to the children during therapy, nor did we accept productions with prosodic errors as a matching production either during the approximation setting or practice phases of the therapy.

Treatment

The treatment protocol from the Gomez et al. (2018) study was updated to improve clinician training, protocol instructions, the selection of daily therapy targets, therapy progression, and clinician feedback (see Supplemental Materials for the full treatment protocol). These changes are detailed in Table II. Two speech-language pathology students were trained to administer this revised K-SLP, with each student being assigned two of the four child participants. They were supervised by both the first and second authors to ensure fidelity to the standardised protocol. Treatment sessions were divided into two phases—the approximation setting phase and the practice phase.

Approximation setting phase

This phase of the treatment was designed to allow the clinician to establish the targets for each treated word for each week of therapy. Each target word was included in the approximation setting phase once per week, to establish the child’s highest level of approximation for that week—i.e. the approximation the child could say that most closely resembled the adult form of a word (e.g. /æmʊ/ for /ænəməl/). Each target word had between three and five approximations, which were simplified by applying phonological processes (see Kaufman, 2013). The clinicians were instructed to use any combination of cues (e.g. tactile hand cues, oral posture cues, etc.) contained in the treatment protocol to elicit the highest level of approximations. They continued to simplify the target until the child was able to correctly produce one of the pre-set approximations with clinician assistance. An approximation was selected for each word once the child was no longer stimulable for the subsequent (higher) levels of approximation with any combination of cues from the clinician. If during the approximation setting phase the child was not stimulable for any higher approximation (above their spontaneous production), the clinicians were instructed to automatically start at the next highest level of approximation above their spontaneous production for that word to ensure that there was a therapeutic benefit to the words practiced in therapy. The length of the approximation setting phase varied by session, as each target word was included in this phase only once per week when it was randomly selected for inclusion in the therapy session. Consequently, the last treatment session of each week generally had fewer novel words. The approximation setting phase did not have any time limits imposed and therefore varied by session and by child, although the clinicians were advised to restrict it to no more than 30 minutes. Once the highest approximation was established for each relevant randomly selected word, the child entered the practice phase of that session.

Practice phase

The practice phase had a time-based criterion of 8 minutes of drill therapy and 2 minutes of play therapy, which was repeated until the end of the session (see below). During play, the treating clinicians had a set of targets (i.e. functional words) they were expected to elicit and shape using a clinician-client driven play-based format. This play-based goal was designed with the intention of incorporating the dynamic and play-based nature of the K-SLP. This time-based cycle continued until the child either produced 100 trials (of drill therapy targets only) during a combined approximation setting and practice phase, or had completed a 60 minute session (inclusive of approximation setting phase).

During both the drill and play based therapy segments of the practice phase, accuracy was judged based on whether the child’s production matched the clinician’s model, meaning that if the clinician modelled an approximation (e.g. /matoʊ/ for tomato) and the child’s production matched the clinician’s model in terms of sounds, prosody, and smooth transitions (i.e. no segregated syllables/sounds), then the production was considered accurate. The cueing hierarchy used during the drill portion of the practice phase is outlined in the Supplemental Material (Figure 4 in the Supplemental Material). To progress through the cueing hierarchy, children were required to accurately produce the target three consecutive times at any given stage of the hierarchy. If the children were unsuccessful, they were then given additional cues in line with the errorless learning approach to reduce their chance of repeating any inaccurate productions of the target word approximation. Each child’s productions during the approximation setting and practice phases were transcribed by the treating clinician using broad phonemic symbols to assist them with judgement accuracy.

The K-SLP encourages the integration of a number of motor learning principles (Maas et al., 2008) that we have operationalised in this protocol, including high repetition drill-based therapy, incorporating faded feedback, to reduce feedback frequency; incorporation of distributed practice across time; and the integration of a random practice schedule to promote retention of newly instated speech motor skills (see Maas et al., 2008, for a discussion of motor learning principles; see Kaufman, 2013, for discussion of the motor learning principles included in the K-SLP). The operationalised K-SLP treatment utilises 100 trials per session (high rate of production), fading feedback across the 3 weeks of therapy (week 1 = 100%, week 2 = 80%, week 3 = 60%) with an overall average of feedback of 80% across the three weeks, and randomly selecting 10 of 20 available stimulus words for each treatment session (random and distributed practice). Knowledge of results and knowledge of production feedback were incorporated into both phases, consistent with the protocol reported by Gomez, McCabe, Jakielski, & Purcell (2018).

Outcome measures

Consistent with some research in CAS treatment (see for example Gomez, McCabe, Jakielski, Purcel, 2018; Iuzzini & Forrest, 2010; McNeill et al., 2009), the primary outcome measure for this study was PPC on the correct adult form (e.g. tomato instead of approximations) of treated and untreated stimuli to allow for a measure of both phoneme inventory expansion (Iuzzini & Forrest, 2010) and improved articulatory accuracy for targeted word shapes, despite the changing realisation of adult targets. This is reported separately for the three goals for each individual child to allow us to explore discrete changes across each of the three goals. Secondary outcome measures included the standard score from the GFTA-2 to measure improved general articulatory accuracy inventory expansion, PPC from a polysyllabic word list (to measure general articulatory accuracy in longer words; Gozzard et al., 2006), and percent inconsistency from the DEAP inconsistency subtest (to measure improved token to token consistency; Dodd et al., 2002).

Data analysis

The probe data (i.e. PPC for each goal) were visually inspected to identify any visual trends, then statistically tested to measure whether there were any significant changes in PPC across time. In keeping with a single case experimental design format, individual effect sizes were calculated for each child to measure clinical change. The effect sizes reported in this study include the improvement rate difference (IRD; Parker et al., 2009) and percentage of non-overlapping data (PND; Scruggs et al., 1987). The current literature does not point to a single effect size measure that is better suited for CAS research, consequently both IRD and PND measures were included as these measures complement each other (Rakap, 2015) while also allowing us to demonstrate whether the results of this Phase I study are robust. IRD calculates the improvement rate (IR) for the baseline, treatment, and maintenance phases (number of improved data points/total number of data points). To calculate the IRD treatment effectiveness (i.e. A1B1, where A1 = all three baseline measures and B1 = two treatment points), the IR_treatment − IR_baseline was computed; while treatment maintenance (i.e. A1A2, where A2 = all three post-treatment measures) was calculated by computing IR_baseline − IR_{post-treatment}. PND (Scruggs et al., 1987) was also calculated as a comparison measure using the following formula: number of B1 (treatment) and A2 (post-treatment) data points above the highest baseline data point are divided by the total number of data points across the B1 (treatment) and A2 (post-treatment) phases. The values generated for the IRD and PND were interpreted using the benchmarks reported in Rakap (2015) where: (a) PND and IRD of ≤50 is considered “ineffective,” (b) PND 51–69 is “questionable,” (c) PND 70–89 and IRD 51–69 is “effective”, and (d) PND ≥90 and IRD >70 is “very effective.”

Reliability and fidelity

Inter-rater reliability was completed on the segmental accuracy of a randomly selected 10% of the probe sample (i.e. nine targets per probe session) by an independent rater who was experienced in transcribing disordered speech and yielded an average reliability of 88% (SD = 5) across participants. Inter-rater reliability focused on the judgement of segmental accuracy of the probe words, as this was the primary outcome used to measure the effectiveness of the intervention.

The first author examined a randomly selected 10% of the practice phase associated with each treatment session to compute fidelity. Fidelity was not calculated on the approximation setting phase as clinicians were instructed to use any cues typically used in articulation therapy and/or the K-SLP intervention to elicit a child’s best approximation. With respect to the practice phase fidelity, the clinicians were provided with contemporaneous instructions regarding feedback accuracy and the feedback schedule for each week. Clinicians were advised to adhere to the specified treatment trials that were to receive feedback; however, if they failed to accurately adhere to the session feedback schedule, they could compensate for that error on a subsequent trial. Clinicians were assessed on how closely they adhered to these specific instructions using a fidelity form/checklist (see Appendix A). The average fidelity for feedback accuracy was 84.2% (SD = 3.3), while the average feedback provided by the clinicians was 88% (SD = 8; target 100%) for week 1, 81% (SD = 7; target 80%) for week 2, and 65% (SD = 8; target 60%) for week 3, with an overall feedback percentage of 78% (overall target 80%). The clinicians generally adhered to the feedback frequency each week (within a 10% margin), except for the first week of therapy. The overall feedback percentage, however, was within the target range. Average feedback frequency (across the 12 sessions) was statistically compared among the four children using the non-parametric Kruskal-Wallis Test and the differences were not significant (p = 0.336). The average number of productions per session was 98 (SD = 23; target 100), as recorded by the clinicians within the treatment session using therapy data forms. A one-way analysis of variance (ANOVA) was conducted to determine if each child received an equivalent treatment dosage (measured by total number of trials per session). There was a significant difference in terms of the number of trials (approximation setting and practice phases combined) among the children (F = 5.481, p = 0.003), with Max receiving overall a significantly reduced number of trials than both Billy (p = 0.012) and Sam (p = 0.011). There was also a significant difference in the number of words targeted during the practice phase of the sessions between the participants (statistically tested using the Kruskal-Wallis test, χ² = 7.833, p = 0.050), with James practicing significantly fewer unique words (different and/or repeated) than Billy (p = 0.037) and Sam (p = 0.011).

Result

Billy

Billy had stable baselines for his first treated goal, meaning that PPC across the three baseline probes did not vary by more than 10% (see Figure 1; Iuzzini & Forrest, 2010). The remaining two treated goals varied at baseline by up to 18%.

Figure 1. Billy’s percent phonemes correct (PPC) for experimental probe data.

Billy’s average baseline accuracy (i.e. PPC) for Goal 1 (bisyllabic word + Consonant [C] Vowel [V] C, where the post-vocalic C was included in all but one word) was 82% (SD = 2), which increased to 90% during treatment and 91% post-treatment (SD = 1). The average accuracy for Goal 2 (polysyllabic words) was 58% (SD = 7) at baseline, 66% during treatment, and 73% (SD = 2) during the post-treatment phase. PND and IRD were calculated. For Goal 1, PND = 100% and IRD (A1B1 [baseline and treatment phase comparison], A1A2 [baseline and post-treatment phase comparison]) = 100. For Goal 2, PND = 80% and IRD (A1A2) = 100. Positive treatment outcomes generalised to untreated bisyllabic word + CV(C) targets (PND = 100%, IRD [A1A2] = 100) and untreated polysyllabic words during the treatment phase (IRD [A1B1] = 100). See Table IV for a summary of all PND and IRD calculations for treated, untreated, and control words, along with an interpretation of these results.

Table IV. Summary of percentage non-overlapping data (PND) and improvement rate difference (IRD) as measures of effect size for treated, control, and untreated words.

Goals	Analysis	Billy (%)	Sam (%)	Max (%)	James (%)
Treated and control words
Goal 1	PND	**100	20	**100	20
	IRD (A1B1)	**100	−17	**100	−67
	IRD (A1A2)	**100	−100	**100	−67
	IRD (B1A2)	17	−100	−50	−67
Goal 2	PND	*80	40	**100	40
	IRD (A1B1)	17	−33	**100	17
	IRD (A1A2)	**100	0	**100	0
	IRD (B1A2)	**100	−33	**100	−100
Goal 3	PND	40	20	**100	40
	IRD (A1B1)	−100	−17	**100	−17
	IRD (A1A2)	0	−100	**100	−67
	IRD (B1A2)	−33	−100	−17	−67
Control	PND	20	60	20	0
	IRD (A1B1)	−33	17	0	0
	IRD (A1A2)	0	33	33	0
	IRD (B1A2)	*67	−17	33	0
Untreated words
Goal 1	PND	**100	60	**100	0
	IRD (A1B1)	**100	−17	**100	−100
	IRD (A1A2)	**100	−33	**100	−100
	IRD (B1A2)	−100	−100	−33	−100
Goal 2	PND	*80	60	60	60
	IRD (A1B1)	**100	**100	−50	**100
	IRD (A1A2)	0	33	33	−33
	IRD (B1A2)	−67	−100	−17	−100
Goal 3	PND	40	*80	**100	0
	IRD (A1B1)	−67	−50	**100	−100
	IRD (A1A2)	−33	**100	**100	−100
	IRD (B1A2)	−33	−17	−67	17

Note. Positive scores are indicative of improvement, whilst negative scores are indicative of deterioration of some to most data points to baseline levels of accuracy.

A1B1 compared the pre-treatment baseline accuracy to treatment phase accuracy.

A1A2 compared the pre-treatment baseline accuracy to the post-treatment phase accuracy.

B1A2 compared the treatment phase accuracy to the post-treatment phase.

**Very effective intervention effect; *effective intervention effect (Rakap, 2015).

The average baseline accuracy for Goal 3 (functional words) was 64% (SD = 4), increasing to 67% during the treatment phase and 73% (SD = 6) during the post-treatment phase. The effect size measures were PND = 40% and IRD (A1A2) = 0.

The untreated control sounds/structures for Billy were two- and three-element clusters. The average baseline accuracy was 7% (SD = 12, range = 0–20%), which reduced to 0% during treatment and then increased to 20% (SD = 20) during the post-treatment phase. The PND was 20%, IRD for A1B1 was −33, and IRD for A1A2 = 0. Overall, the changes in the control words followed a different trajectory than the treated words.

Sam

Sam had stable baselines for treated Goal 1 (bisyllabic word + CVC, where the post-vocalic C was included in all but one word) and treated Goal 3 (functional words; see Figure 2), while treated Goal 2 (polysyllabic words) varied at baseline by 22%. Sam’s average baseline accuracy for Goal 1 was 90% (SD = 1), and increased to 93% during the treatment phase and 88% (SD = 1) during the post-treatment follow-up. The average baseline accuracy for Goal 2 was 60% (SD = 7), which increased to 65% during the treatment phase and 71% (SD = 9) during the post-treatment phase. The average baseline accuracy for Goal 3 was 68% (SD = 3), which increased to 70% during the treatment phase and dropped to 64% (SD = 10) during the post-treatment phase. PND and IRD (A1A2) measures were calculated. For Goal 1, PND = 20% and IRD = −100. For Goal 2, PND = 40% and IRD = 0. For Goal 3, PND = 20% and IRD = −100. Positive treatment outcomes were recorded for untreated Goal 2 (PND = 60%, IRD [A1B1] = 100) and untreated Goal 3 (PND = 80%, IRD [A1A2] = 100).

Figure 2. Sam’s percent phonemes correct (PPC) for experimental probe data.

Sam’s control words contained two and three-element clusters. The accuracy of these control words was variable across the three phases (i.e. baseline, treatment, and post-treatment). The average baseline accuracy was 7 (SD = 12, range = 0–20%), which increased to 20% during the treatment phase and 33% (SD = 31) during the post-treatment phases. Measures of effect size were calculated as PND = 60% and IRD = 33.

Max

Max had stable baselines for treated Goal 1 (bisyllabic words; see Figure 3) while treated Goals 2 (polysyllabic words) and 3 (functional words) varied at baseline by 13%. Max’s average baseline accuracy for Goal 1 was 61% (SD = 3), which increased to 79% during the treatment phase and then 82% (SD = 3) at the post-treatment time-point. The average baseline accuracy for Goal 2 was 57% (SD = 4), which increased to 65% during the treatment phase and 78% (SD = 2) during post-treatment follow-up phase. The average baseline accuracy for Goal 3 was 59% (SD = 4), increasing to 70% during the treatment phase and 74% (SD = 5) during the post-treatment phase.

Figure 3. Max’s percent phonemes correct (PPC) for experimental probe data.

Measures of effect size were calculated (IRD = A1A2). For Goal 1, PND = 100% and IRD = 100. For Goal 2, PND = 100% and IRD = 100. For Goal 3, PND = 100% and IRD = 100. Positive treatment outcomes generalised to both untreated Goal 1 (PND = 100%, IRD = 100) and Goal 3 (PND = 100%, IRD = 100).

The untreated control words for Max comprised two- and three-element clusters in word initial (four words) and word medial (one word) positions within polysyllabic words. The mean baseline and treatment accuracies were 0%, while the average accuracy during the post-treatment phase was 7% (SD = 12). The calculated effect sizes for these words were PND = 20% and IRD = 33.

James

James had stable baselines for treated Goal 1 (bisyllabic words; see Figure 4), while treated Goals 2 (polysyllabic words) and 3 (functional words) varied at baseline by 18%. James’ average baseline accuracy for Goal 1 at both baseline and during treatment was 50% (SD = 2 during the baseline phase), while the post-treatment mean accuracy was 50% (SD = 7). The average baseline accuracy for Goal 2 was 57% (SD = 6), which increased to 68% during the treatment phase and then dropped to an average accuracy of 64% (SD = 6) during the post-treatment follow-up phase. For Goal 3, the average baseline accuracy was 55% (SD = 6) and increased to a mean of 61% during the treatment phase and fell to an average accuracy of 59% (SD = 13) during the post-treatment phase.

Figure 4. James’s percent phonemes correct (PPC) for experimental probe data.

PND and IRD (A1A2) calculations were as follows. Goal 1: PND = 20% and IRD = −67. Goal 2: PND = 40% and IRD = 0. Goal 3: PND = 40% and IRD = −67. There were also no recorded changes in the untreated goals associated with Goal 1 (PND= 0%, IRD = −100) and Goal 3 (PND = 0%, IRD = −100). Some treatment effect was detected for untreated Goal 2: PND = 60%, IRD (A1B1) = 100; however, these positive effects were not maintained.

Consistent with the other children, untreated control words for James consisted of two- and three-element clusters in polysyllabic words. The mean accuracy did not change over time (mean = 0%, SD = 0) and there was no treatment effect across time.

A number of speech assessments were readministered one month after the intervention was completed, including the GFTA-2, the same polysyllabic word list, and the DEAP inconsistency subtest. The results of these assessments are summarised in Table V. It is noteworthy that Billy and Max’s GFTA-2 standard scores increased, suggesting improved articulatory accuracy.

Table V. Summary of assessment results from baseline and 1 month post-treatment follow-up speech assessments.

	GFTA-2 (standard score/percentile)		Polysyllabic word list (PPC %)		Inconsistency (%)
Participant	Pre-treatment	Post-treatment	Pre-treatment	Post-treatment	Pre-treatment	Post-treatment
Billy	64/4	81/12	55	56	83*	75*
Sam	88/21	70/4	59	54	76	80
Max	83/14	85/16	54	57	76*	63*
James	56/1	57/2	46	45	80	44

Note. Inconsistency percentages would be expected to decrease over time with intervention to demonstrate improved and stabilised motor plans/programs for speech.

GFTA-2 = Goldman Fristoe Test of Articulation-Second Edition (Goldman & Fristoe, 2000); PCC = percent phonemes correct.

*Only 24 of the 25 Diagnostic Evaluation of Articulation and Phonology (DEAP; Dodd et al., 2002) words were administered due to the assessment clinicians’ oversight.

Discussion

This research was designed to evaluate the treatment effects of an operationalised version of the K-SLP. We hypothesised that treatment with the K-SLP would result in: (a) improved articulatory accuracy for targeted words used in treatment, as measured by PPC; (b) maintenance of treatment gains; and (c) no improvement in the articulatory accuracy of unrelated, untrained words.

Positive treatment outcomes resulted in measurable changes for Billy and Max, while Sam and James did not demonstrate measurable changes for treated words. Maintenance of the reported changes during the post-treatment phase were evident for both Billy and Max. Maintenance of treatment changes are generally used to establish that speech motor learning has occurred, meaning that the improvement has gone beyond just acquisition of a new skill during treatment (which may or may not be retained) and has become established as an enduring change in the child’s speech motor skill (Maas et al., 2008). The results therefore partially support the first and second hypotheses.

The results also partially supported our final hypothesis. There was no evidence of significant improvements in cluster accuracy (i.e. untreated control sounds/structures) from baseline to treatment/post-treatment phases for any of the children. This suggests that experimental control was established for each child and any positive outcomes may be attributed to the intervention; although given the severity of the speech disorders presented by the children in this study, it is difficult to ascertain whether the children had the capacity to improve cluster production with treatment.

This study replicated the findings from Gomez, McCabe, Jakielski, & Purcell (2018) by demonstrating that the K-SLP was effective in improving articulatory accuracy for some children who have CAS. Although the current study was unable to identify any significant baseline factors (e.g. receptive language skills, severity of the associated speech disorder) that were predictive of a child’s response to the K-SLP, it is noteworthy that the two children who did not demonstrate notable changes also demonstrated a higher proportion of syllable segregation during a polysyllabic word production task in the initial pre-treatment testing. This pattern was also noted in the Phase I study published by Gomez, McCabe, Jakielski, & Purcell (2018), where it was reported that the child with the reduced response to treatment also presented with a higher proportion of syllable segregation, as recorded by the pre-treatment speech assessments. A larger cohort of children may provide more insight into whether this feature is predictive of children’s response to the K-SLP and may guide clinicians when considering interventions for their individual clients with CAS. It is important to note that because there were only four children in the study, it was not possible to investigate whether any child factors were related to improved PPC and it would be helpful to investigate these within the context of a larger study.

It is also interesting to observe that the monozygotic twins (Billy and Sam), had vastly different responses to intervention. Their baseline profile was similar except for the severity of the speech disorder that was associated with their CAS diagnosis, with Billy demonstrating a more severe speech impairment at baseline and reduced accuracy with consonants and vowels in polysyllabic words. During therapy, Sam had more off-task behaviours compared with Billy and appeared to have reduced resilience in treatment as each week progressed.

This research represents a larger Phase I study (Robey, 2004) that aimed to build on the pilot study published by Gomez, McCabe, Jakielski, & Purcell (2018). The refinements in the protocol achieved improved transcription reliability (to 88% in the current study) and improved fidelity (for feedback frequency and accuracy) than were achieved in the pilot study (see Gomez, McCabe, Jakielski, & Purcell, 2018, for initial data on reliability and fidelity). Although the clinicians adhered to the protocol (based on the fidelity measures described), they were not given a strict protocol for either the length of the approximation setting phase (although they were advised not to exceed 30 minutes) nor the maximum number of attempts per word before proceeding to the next item within the treatment session. Consequently, the approximation setting component of each session varied in length and the number of trials attempted over the course of the study. James had fewer trials than the other children (i.e. reduced treatment dosage), which may have contributed to his reduced overall response to the intervention. As with all interventions, the K-SLP will be adapted across settings and clinicians, which may impact overall treatment intensity and client engagement, and ultimately may impact children’s response to therapy. These factors demonstrate the need to further refine the protocol to establish greater consistency between treatment delivery across children in research settings, particularly for elements that may be responsible for response to intervention.

The overall pattern of improvement from baseline to treatment/immediate post-treatment phases using the K-SLP intervention appears to follow a similar trajectory to other CAS interventions. Like the results obtained for the Nuffield Dyspraxia Programme, treatment using the K-SLP resulted in significant treatment effects between the baseline and immediate/1 month post-treatment time points (for treated words) for some children, while improvements in articulatory accuracy did not continue to improve with the removal of treatment (i.e. treatment versus follow-up phases;see Murray et al., 2015). This suggests that children who receive the K-SLP approach require clinician-directed feedback and cues to mediate ongoing improvements in motor planning and programming for specific sounds and syllable transitions (Murray et al., 2015). This study also demonstrated that individual response to therapy using the K-SLP varied. These findings are consistent with the results reported in Gomez, McCabe, Jakielski, & Purcell (2018), as well as other interventions used for CAS that have generally showed that children have a variable response to therapy, with some children demonstrating significant measurable changes while other children may show few to no changes with therapy (see Strand & Skinder, 1999; Maas et al., 2012).

The ability of children to generalise newly instated skills to related but untreated targets is mediated by a combination of factors. These may include the selection of a range of highly salient target skills (Ballard, 2001; Iuzzini & Forrest, 2010) and the implementation of various principles that are widely believed to promote long term motor learning (i.e. principles of motor learning; see Maas et al., 2008). The K-SLP advocates for the integration of both of these broad principles (see Gomez, McCabe, Jakielski, & Purcell, 2018). Given that the early goal of the K-SLP focuses on word shapes rather than segmental accuracy (Kaufman, 2013) and that the approximation levels are unique for each word, multiple sounds and word shapes are treated in a single session. All four children in this present study demonstrated improved articulatory accuracy for at least one set of untreated words (i.e. generalisation), thereby demonstrating some level of motor learning, although maintenance of these positive outcomes was variable. It is plausible that speech motor learning could have been further enhanced had the children received more than 12 treatment sessions, as the number of sessions administered in this study may not have provided sufficient opportunities for speech motor learning for the two non-responders. The results from this study encourage researchers to consider alternative methods (from DTTC, NDP-3 etc) to treating CAS; specifically using best approximations for salient target words to provide children with communication success, and to build their capacity to generate and store motor plans and programs.

Several factors could contribute to the positive treatment effects associated with this operationalised version of the K-SLP. Firstly, repetitive practice facilitates motor planning and programming in the child’s zone of proximal development (Vygotsky, 1978), as achieved through the use of successive approximations, and may provide children with CAS with the practice needed to effectively establish and store a stable, albeit simplified, motor plan to be accessed during future attempts of the word (see Guenther, 2016; Schmidt, 1975). The motor plans may have been further established and stabilised through the implementation of high intensity practice, the use of cues to create a quasi-errorless learning environment, and the incorporation of faded feedback, as advocated in the K-SLP (Kaufman, 2013; see Maas et al., 2008, for a discussion of the principles of motor learning). Given the use of speech production simplifications (e.g. “phonological processes”) to develop approximations features prominently in the approach, the K-SLP may be described as a dual motor-linguistic based CAS intervention (see Namasivayam et al., 2019). The author of the K-SLP encourages the use of motor learning principles to instate these simplified motor speech plans and programs (Kaufman, 2014).

Similar to Dynamic Temporal and Tactile Cueing (DTTC; Strand, 2020), the Nuffield Dyspraxia Program (Murray et al., 2015), and the PROMPT © method (Dale & Hayden, 2013), the K-SLP integrates an element of quasi-errorless learning by implementing a bottom-up approach whereby targets are initially simple and gradually progress to more complex sounds and syllable combinations within the child’s zone of proximal development, to enable a child with CAS to experience success and produce accurate and stable motor programs for targeted words.

Although broadly speaking, the K-SLP is unique in its use of successive approximations, our data suggests that treatment success may be mediated by the child repeatedly practicing planning and programming sounds and sequences in specific word shapes. Children are therefore producing stable motor plans of otherwise errored productions (compared with the adult target of the word). Ultimately, this practice may help to establish a functional feedforward system (see Guenther, 2016) for syllables/words, which in turn may provide an overall stabilising effect to the motor system of a child. This may allow the child to replace grossly errored productions with productions that are a more accurate realisation of adult targets. Given that the overarching goal of therapy in CAS treatment is not simply the acquisition of targets in therapy, but a permanent change in the speech motor system (e.g. improved stability and motor planning; Case & Grigos, 2021), the K-SLP appears to be a promising intervention for children to achieve these goals.

Limitations and future directions

This study utilised trained student clinicians to deliver the therapy, which may have impacted the outcomes of the study relative to more experienced clinicians. Despite the intervention being replicated across four children in this study, a stronger, more rigorous design is needed to accurately evaluate the effectiveness of the operationalised K-SLP protocol used in this study. This could be achieved either through a bigger single case experimental study with a refined treatment protocol that incorporates a multiple baseline design, whereby each child is allocated a random number of pre-treatment probe sessions (Kazdin, 2011), or through use of a more rigorous research design, for example randomised controlled trials with large N studies where possible, along with N-of-1 studies (see Rvachew & Matthews, 2019, for an example of an N-of-1 study). Additionally, fidelity was only calculated on a limited number of features within the current protocol. Further standardisation of the protocol is recommended to avoid clinician effects and to ensure that each child receives equivalent therapy. Although the rater was blinded to the sequence of the probes, the rater was not blinded to the participants which may have introduced an observer bias.

Based on the results of this study, several recommended changes should be applied to the operationalised K-SLP protocol. Future research should consider how to operationalise the move from word to phrase level in a way that allows progression through the hierarchy at a rate that is generally observed clinically in the K-SLP approach. Additionally, the cueing hierarchy should be modified to provide clinicians with increased flexibility to differentially select the cues that meet the individual needs of each child based on their target words, and the cues to which they are most responsive.

A time limit could be imposed on the approximation setting phase. Considering that the practice phase has a standardised structure in terms of the amount of feedback, cueing hierarchy, and the number of trials, and that the approximation setting phase should only be used to identify the child’s best approximation for each word, therapy sessions should be designed to optimise the number of trials produced in the treatment phase of the session. In the current study, although the overall number of trials per session (inclusive of both the approximation setting and practice phases) was 98 (SD = 23), a number of treatment sessions (irrespective of the child) had instances where there were a larger number of trials during the approximation setting phase, at the expense of targeted/standardised practice during the practice phase.

Finally, more consistency is recommended in terms of the type of feedback given at each phase of the intervention, improved adherence to the feedback schedule, consistency with progression through the treatment hierarchy, and consistency with the number of trials elicited per word to negate any clinician effects.

This research only measured one aspect of CAS articulatory accuracy (as measured by PPC). In addition to measuring and reporting changes in PPC, it may be beneficial to establish a reliable way to evaluate word shape accuracy more comprehensively, along with the three core consensus features of CAS (ASHA, 2007; namely inconsistency, prosody, and lexical stress) to allow more subtle changes in a child’s speech to be detected and to determine, if possible, which specific CAS features the K-SLP targets.

Conclusions

Use of the operationalised K-SLP was an effective intervention for two of four children with CAS. Clinicians are cautioned against generalising the results of this study to the use of the commercially available K-SLP in the absence of data about how SLPs are using the commercially available K-SLP, as the data here does not represent clinical practice. This research demonstrates that an intervention using successive approximations is an effective alternative for some children to improve articulatory accuracy. At this stage, no individual child factors appear to predict which children are most likely to benefit from this operationalised version of the K-SLP. Further replication of these results is required using a larger sample size and a more rigorous research design, to both test the effectiveness of this operationalised version of the K-SLP and to more comprehensively evaluate which child factors may be likely to predict the suitability of this approach for individual children.

Acknowledgements

This research was partially funded by The University of Sydney Faculty Funding Scheme awarded to the first author. Parts of this research were presented at the 2016 ASHA Convention, and the 2017 Speech Pathology Australia Conference. We thank the parents and children who participated in this study, as well as the clinicians who assessed and treated these children—Pippa Evans, Tanya Price, Nicole Shaw, Olivia Vun, and Diana Irwin. The researchers are not aware of any conflicts of interest.

Declaration of interest

No potential conflicts of interest were reported by the authors. Some of these findings were presented at the American Speech-Language-Hearing Association National Convention (2016) and the Speech Pathology Australia National Conference (2017).

Additional information

Funding

This research was partially funded by The University of Sydney Faculty Funding Scheme to the first author.

Notes

1 This research pre-dates the Iuzzini-Seigel et al. (2017) article, which suggests that the DEAP is an unreliable diagnostic measure for CAS.

Appendix A.

Treatment fidelity data sheets

Client name ___________________________           Student Clinician   _________________________

Date   ___________________________             Session number    __________________________

Table

Trial #	Target from data sheet	Clinician judgement	Fidelity clinician judgement	Decision score (1/0)	Feedback provided?	Feedback score (1/0) i.e. was feedback scheduled for the trial?	Repetition elicited? 1 2		Repetition fidelity score (1/0) i.e. was repetition elicited when it was scheduled?
1					Y / N
2					Y / N
3					Y / N
4					Y / N
5					Y / N
6					Y / N
7					Y / N
8					Y / N
9					Y / N
10					Y / N
11					Y / N
12					Y / N
13					Y / N
14					Y / N
15					Y / N
16					Y / N
17					Y / N
18					Y / N
19					Y / N
20					Y / N