Speech in Ten-Minute Sessions: A Pilot Randomized Controlled Trial of the Chaining SPLITS Service Delivery Model

Abstract

Purpose:

This study evaluates the initial efficacy of Chaining SPeech Lessons in Intensive Ten-minute Sessions (SPLITS), an alternative service delivery model for the Speech Motor Chaining treatment approach. We hypothesized that Chaining SPLITS would result in improvements in /ɹ/ accuracy on syllables and untrained words when compared to a no-treatment condition.

Method:

Within a randomized controlled trial, thirteen 7–9-year-old children with difficulty producing /ɹ/ were randomized to receive treatment either immediately or after an 8-week delay. Treatment sessions were conducted 4 times a week over 8 weeks. Syllable and word-level recordings were collected at the baseline, 8-week, and 16-week time points. Recordings were rated along a 5-point scale by three masked, independent listeners. The primary outcome was changed score from the baseline to 8-week time point (Interval 1) and 8-week to 16-week time point (Interval 2).

Results:

Linear mixed modeling revealed that Chaining SPLITS led to significantly greater improvement in /ɹ/ accuracy at the syllable level for active treatment compared to a period of no treatment. This improvement was replicated in both groups. There was not sufficient evidence of an effect of Chaining SPLITS on untrained words after 8 weeks of treatment.

Conclusions:

The current study provides initial evidence of the effectiveness of 8 weeks of Chaining SPLITS on improvement in /ɹ/ accuracy in syllables. Short, frequent sessions may be a viable approach to promote acquisition of /ɹ/ among school-age children; however, longer courses of treatment may be needed to observe further improvement at the word level.

Supplemental Material:

https://doi.org/10.23641/asha.27868800

Nearly all school-based speech-language pathologists (SLPs) regularly provide services to children with speech sound disorders (SSDs), who represent a significant proportion of school caseloads (American Speech-Language-Hearing Association [ASHA], 2020). SSDs affect 11.4% of 8-year-old children, with about 70% of cases involving common clinical distortions of sounds such as /ɹ/ (Wren et al., 2016). Typical service delivery for students with SSDs consists of one or two 20–30-min group sessions weekly (Brandel & Frome Loeb, 2011), using a traditional Van Riper articulation approach (Brumbaugh & Smit, 2013; Cabbage et al., 2022). Concerningly, about a quarter of these children demonstrate inadequate progress by the time they are dismissed from services (Mullen & Schooling, 2010). It is necessary to critically evaluate current treatment approaches and parameters of service delivery in order to maximize the effectiveness and efficiency of SSD treatment within the logistic constraints of the school environment (e.g., Individualized Education Program eligibility requirements, scheduling challenges, removing students from instructional time).

Alternative service delivery models involving shorter, more frequent sessions (Brosseau-Lapré & Greenwell, 2019; Kuhn, 2006) have been proposed as a potential solution to address some of the logistical challenges of school-based speech services. While service delivery models using short, frequent sessions are theoretically grounded in frameworks of learning and retrieval (Donovan & Radosevich, 1999) and likely feasible given the time constraints on school-based SLPs, the efficacy of these models has not been widely studied. The present study advances this need, reporting a preliminary investigation of the efficacy of an alternative service delivery model using short, frequent speech sessions compared to a no-treatment period. Sessions use Speech Motor Chaining (SMC), a treatment approach grounded in principles of motor learning (Preston et al., 2019) for children with SSDs affecting distortion of /ɹ/, the most prevalent—and most persistent—speech error in General American English.

Service Delivery Models for SSDs

Traditional Service Delivery

The term “service delivery” includes reference to session frequency, duration, and format (e.g., group, individual, in-person, telepractice). Brandel and Frome Loeb (2011) indicate that 60%–70% of school-based SLPs treating SSDs implement 20–30-min in-person group sessions, 1 to 2 times a week. Group sizes typically range from two to four students (Brandel & Frome Loeb, 2011; Mullen & Schooling, 2010). In-person treatment has been the traditional format for SSDs, but telepractice has become more common since the COVID-19 pandemic. Telepractice offers several advantages, specifically for communities that have historically faced barriers due to geographic location or general access and availability. Studies report similar outcomes for in-person and teletherapy modalities (Coufal et al., 2018; Grogan-Johnson et al., 2011, 2013) and parents generally perceive telepractice sessions as convenient, accessible, and effective (Fairweather et al., 2016; Lincoln et al., 2015).

It is possible that individual student factors, such as severity of the disorder, factor into service delivery decisions; however, given the limited variation in service delivery configurations, SLP decisions may also be constrained by external factors such as caseload size and scheduling (Brandel & Frome Loeb, 2011). While group sessions may alleviate some issues related to servicing large caseloads, they do not likely result in efficient courses of treatment for individual students and run counter to teachers' needs for students to miss as little instructional time as possible (Kuhn, 2006). In school-based therapy sessions (with an average of two children per session), children achieve an average of 50 production trials in each session (Farquharson et al., 2022), already placing them at risk of not meeting the optimal dose of 50–100 production trials per session that appears to be necessary for effective SSD treatment (Kaipa & Peterson, 2016). Farquharson et al. (2022) found that for each additional student added to a treatment group, practice dosage decreased by 13 trials per student. These issues highlight the need for investigations of service delivery models alternative to the traditional model that preserve the opportunity for children to be seen individually and achieve high doses of practice, while missing less in-classroom instruction and fitting within SLPs' busy schedules and caseload demands.

Alternative Service Delivery Models

Multiple studies have shown that more practice trials per session tend to yield better learning outcomes for SSD therapy, broadly (Edeal & Gildersleeve-Neumann, 2011; Williams, 2012). The frequency of sessions throughout the week also likely has an impact on learning outcomes. In one randomized controlled trial, Allen (2013) found that a more frequent schedule of speech sessions (three sessions per week for 8 weeks) resulted in significantly greater treatment outcomes than an equivalent amount of time provided with more space between sessions (one session per week for 24 weeks) for 3–5-year-old children with phonological difficulties. This support for more frequently scheduled sessions has been observed in other speech treatment studies with preschoolers with phonological difficulties (Page et al., 1994) and with 9–17-year-old children with childhood apraxia of speech (Preston et al., 2024).

There is strong theoretical justification for providing frequent treatment sessions. The learning and retrieval literature supports the spacing effect: a phenomenon in which spacing out or distributing practice over time benefits learning. Results of several studies have suggested that individuals are more likely to learn and retain information if they acquire it over multiple, distributed sessions rather than a single, massed session (Donovan & Radosevich, 1999). The perceived mechanism behind the spacing effect is that distributed learning allows for more opportunities for memory consolidation and frequent retrieval. However, too much space between sessions can also be detrimental, leading to forgetting information and negatively affecting learning of new skills (Allen, 2013; Bahrick & Hall, 2005; Donovan & Radosevich, 1999; Maas et al., 2019; Preston et al., 2024).

Alternative service delivery models using short, frequent speech treatment sessions for children with SSDs have been incorporated by some school districts to integrate with tiered intervention systems (e.g., response to intervention [RTI]) and/or to reduce the amount of time children are removed from academic instruction. The “Speedy Speech” (Kuhn, 2006) and “Quick Articulation!” (Brosseau-Lapré & Greenwell, 2019) models, which both incorporate short (5–10 min), frequent speech sessions, are examples of service delivery models developed specifically for children with mild SSDs.

Despite the uptake of these models, strong theoretical support, and growing anecdotal endorsements, empirical evidence is currently lacking. Two studies have compared the traditional model of one (Byers et al., 2021) or two (Rehfeld & Sulak, 2021) 30-min weekly group sessions to an experimental model of shorter, more frequent individual sessions (5 min, 3 times weekly, Byers et al., 2021; 15 min, 4 times weekly, Rehfeld & Sulak, 2021) for school aged (5–9 years, Byers et al., 2021) children with SSD. Neither study found evidence of a difference between the traditional and experimental service delivery conditions. However, these studies are limited in their interpretation, as they lack control for maturation and cannot be generalized to other SSD treatments outside of the two investigated (traditional/Van Riper and multiple oppositions). By contrast, the current study uses a well-controlled experimental design that accounts for maturation to investigate the efficacy of an alternative service delivery model using SMC (an evidence-based treatment for SSDs; Preston, Leece & Storto, 2019), compared to a no-treatment period.

SMC

SMC is a motor-based treatment approach intended for school-aged children with SSDs (Preston et al., 2019) that directly incorporates the principles of motor learning (Maas et al., 2008) and aspects of the challenge point framework (Guadagnoli & Lee, 2004) in an adaptive structure to support both acquisition and generalization of new, accurate motor skills. This contrasts with the traditional, Van Riper (1939) treatment approach, which targets phonemes one at a time, beginning with the establishment of correct production of the target phoneme in isolation through phonetic placement cues, followed by generalization of the sound through practice trials advancing along a vertical hierarchy of linguistic complexity until the correct version of the sound is used by the speaker naturally in all communication contexts. Both SMC and the traditional approach incorporate a systemic increase in stimulus complexity; however, this increase occurs quicker (i.e., within sessions) in SMC than in the traditional approach. Additionally, unlike SMC, the traditional approach is not operationalized with respect to feedback type and frequency, prosodic variability, and practice schedule. Therefore, there is likely much variability in the implementation of the traditional approach across SLPs.

SMC includes a prepractice phase, in which the SLP provides detailed instructions and feedback to elicit accurate productions of the target sound(s), a structured chaining phase, in which the child advances through a series of “chains” containing the target sounds(s), and a random practice phase, in which the child practices the target sound(s) in random order at the highest level of the chains that they produced accurately during structured chaining. Chains from structured chaining are founded upon a syllable (e.g., /ɪɹ/); they gradually build in complexity to allow for practice of that syllable in monosyllabic words, multisyllabic words, phrases, and self-generated sentences. Children are able to advance to higher levels of the chain if they meet a clinician-determined threshold for accuracy in lower levels of the chain. A key feature of SMC is the ability to adapt the task difficulty within a session in response to the child's performance. Early on in the chains, practice occurs with simple targets, and knowledge of performance (KP) feedback (i.e., specific feedback that references articulator movement; e.g., “lift the tip of your tongue higher”) is provided frequently. These conditions promote the acquisition of correct speech motor movements. To promote generalization as the child advances, variable practice, complex targets, and less frequent, knowledge of results (KR) feedback (i.e., feedback that exclusively references degree of correctness; e.g., “not quite”) are utilized. This modification in task difficulty as the child's performance improves ensures that the child receives an adequate level of challenge, in order to promote motor learning.

Previous studies have examined the efficacy of SMC with school-aged children who demonstrate residual speech sound errors or childhood apraxia of speech, and results indicate that SMC can lead to improvements in speech sound production accuracy (Preston, Leece, & Maas, 2016, 2017; Preston, Leece, McNamara, & Maas, 2017; Preston, Maas, Whittle, Leece, & McCabe, 2016; Preston et al., 2014). Although many of these studies included the use of ultrasound technology as biofeedback within SMC, the study by Preston, Leece, and Maas (2017) found that children who received SMC without biofeedback also demonstrated improvements in speech sound production accuracy. Positive outcomes have also been observed with children receiving SMC via telepractice (Preston et al., 2024). In terms of service delivery models, evidence from SMC research—like most research investigating SSD treatment approaches (Baker & McLeod, 2011; Sugden et al., 2018)—largely comes from studies with individual treatment sessions held in-person in university clinics, provided in 30–60 min sessions 2 or 3 times a week. This model does not reflect current clinical practice, which could result in “watered down” effects if SSD treatment is delivered in a traditional group-treatment setting in the schools. To date, no study has investigated the use of SMC within an alternative service delivery model. Given the improvements observed for children with SSDs following motor-based interventions (e.g., Preston, Leece, & Maas, 2017), the incorporation of motor-based treatment within an alternative service delivery model is warranted.

Purpose

The alarming number of school-aged children with SSDs who are dismissed from caseloads with unresolved errors warrants a challenge to the “traditional” model used by most school-based SLPs and an investigation of treatments that are effective within alternative models and can be easily implemented by SLPs. Although research on alternative service delivery models for treatment of SSDs is limited, the results of available studies suggest that treatment schedules with short, frequent sessions may provide equal levels of improvement of target sounds in isolation (Rehfeld & Sulak, 2021) and in untreated words (Byers et al., 2021) when compared to treatment schedules with longer, less frequent sessions. Indeed, service delivery models incorporating short frequent sessions such as “Speedy Speech” and “Quick Articulation!” appear to have several clinical advantages, especially for children with mild SSDs. However, no study to date has examined alternative service delivery models within motor-based interventions. Because the optimal treatment intensity likely varies for different approaches (Kaipa & Peterson, 2016), the current study examines the efficacy of the SMC treatment approach within an alternative service delivery model of four 10-min sessions provided weekly for school-aged children with difficulty producing /ɹ/. The current study incorporates telepractice sessions for all participants, in order to include children from a wider range of regions and contribute to the literature on the use of telepractice for children with SSDs. This innovative combination of an alternative service delivery model and motor-based intervention is referred to as Chaining SPeech Lessons in Intensive Ten-minute Sessions (SPLITS). We hypothesize that Chaining SPLITS will result in improvements on /ɹ/ compared to ratings after an equivalent period of time without treatment. The primary outcome measure is performance on perceptually rated probes containing syllables and untrained words. We also explore participant and parent experiences and opinions regarding the Chaining SPLITS model, through the analysis of questionnaires.

Method

The current study compared immediate treatment and delayed treatment conditions within a randomized group design to evaluate the efficacy of Chaining SPLITS. This project was approved by the Syracuse University Institutional Review Board (#22-148). All parents/guardians of participants provided oral consent, and all participants provided oral assent. Participants were recruited through posts in SLP social media groups. A three-stage eligibility procedure was implemented, consisting of an online prescreening survey, a 1.5-hr evaluation session, and a 30-min dynamic assessment visit to verify adequate stimulability.

Eligibility Procedure

Prescreening Survey

First, all potential participants completed an online prescreening survey administered via the REDCap software to determine eligibility for the next two stages. The online prescreener verified that participants (a) were between the ages of 7;0 (years;months) and 9;11; (b) spoke English as a dominant language and began learning English prior to age 3 years; (c) spoke a rhotic dialect of English; (d) experienced difficulty producing the American English /ɹ/; (e) had no known permanent hearing loss; (f) had no reported neurodevelopmental diagnosis (e.g., Down syndrome, autism, cerebral palsy), with a reported diagnosis of attention-deficit/hyperactivity disorder being admissible if the participant met all other inclusionary criteria; (g) had no reported epilepsy or other neurological disorder; (h) had no prior history of major brain injury, surgery, or stroke; (i) had wired broadband internet connection at home; and (j) had access to a Windows or Apple desktop or laptop computer with the ability to join a Zoom session. All participants who passed the prescreening survey were invited to attend an initial Zoom call with the first author. During this initial Zoom call, oral consent and assent were obtained and the technology requirements for the study were reviewed.

Evaluation

The next step of the eligibility process consisted of an evaluation session via Zoom. During the evaluation session, the first author administered a brief oral mechanism examination (in which participants were asked to move some of their articulators within view of their computer's camera to demonstrate adequate lip seal, lip spreading, and lip rounding and tongue protrusion, tongue elevation, and tongue range of motion), the Goldman–Fristoe Test of Articulation–Third Edition (GFTA-3; Goldman & Fristoe, 2015), the Peabody Picture Vocabulary Test–Fourth Edition (PPVT-4; Dunn & Dunn, 2007), the Recalling Sentences subtest of the Clinical Evaluation of Language Fundamentals–Fifth Edition (CELF-5; Wiig et al., 2013), the LinguiSystems Articulation Test (LAT) Inconsistency Screener (Bowers & Huising, 2011), the Syllable Repetition Task (SRT; Shriberg et al., 2012), a 45-item /ɹ/ syllable probe, and a 125-item /ɹ/ word probe. The LAT Inconsistency Screener and SRT criteria were included to rule out childhood apraxia of speech; however, it is important to note that these two criteria alone are not sufficient to determine a childhood apraxia of speech diagnosis (Murray et al., 2021). To confirm the presence of an SSD, children were required to score below the 7th percentile on the GFTA-3. Eligible participants were also required to demonstrate errors on no more than two sounds other than /ɹ/ on the GFTA-3. Additional inclusionary criteria included a standard score ≥ 80 on the PPVT-4, a scaled score ≥ 6 on the Recalling Sentences subtest of the CELF-5, less than three inconsistent productions on the LAT Inconsistency Screener, less than four additions on the SRT, and an oral mechanism examination within functional limits. The Elision and Phoneme Isolation subtests of the Comprehensive Test of Phonological Processing–Second Edition (Wagner et al., 2013), a growth mindset questionnaire (Benway et al., 2021), and a speech perception task (using an approach similar to Preston, Hitchcock, & Leece, 2020) were also administered during the evaluation session for (noneligibility) descriptive purposes.

Dynamic Assessment

Regardless of the evaluation results, all participants were invited to a 30-min dynamic assessment visit with the first author via Zoom. During this visit, the clinician used a PowerPoint presentation to demonstrate and discuss proper tongue placement for /ɹ/. A script adapted from the tutorial by Preston, Benway, et al. (2020) was followed, with emphasis on tongue root retraction, lateral bracing, and raising the tip/blade. The PowerPoint slides included MRI images as well as articulatory animations (Benway & Preston, 2022). The clinician set a 30-min timer at the beginning of the dynamic assessment session; any additional time following the presentation was used as an opportunity for unstructured practice to elicit /ɹ/ through phonetic cueing and shaping, with maximal feedback provided by the clinician.

Immediately after the 30-min dynamic assessment visit with each participant, the clinician conducted both the syllable and word probes a second time. To be eligible for the study, participants were required to demonstrate less than 50% accuracy on the /ɹ/ word probe and accurate /ɹ/ production on at least five out of 45 syllables (our definition of “stimulable”) on the syllable probe from either the evaluation or dynamic assessment visit. The participant's productions on the stimulability and word probes from both the evaluation and dynamic assessment sessions were audio recorded by parents and perceptually rated by the first author. In cases where the first author rated the productions within 10% of the eligibility criteria (n = 16), the second author independently rated the productions. The average rating was then used to determine eligibility.

We chose to include only children who were stimulable for multiple reasons, including (a) for homogeneity of the small sample; (b) for practical considerations, as service delivery models such as Chaining SPLITS are typically reserved for children with milder SSDs; and (c) theoretically, children who are not yet stimulable likely need more detailed instruction and elicitation strategies, which may require longer, massed sessions. This contrasts with children who are stimulable, who need to stabilize a skill they are beginning to acquire. Furthermore, SMC has previously been implemented only once children show some degree of stimulability (Preston et al., 2019). Thus, we speculated that children who were stimulable were most suitable for the treatment provided by the Chaining SPLITS model.

Figure 1 below displays a CONSORT table outlining recruitment, testing, and treatment of the 13 participants (nine boys, four girls) who met the eligibility criteria and enrolled in the study. Thirty participants were considered ineligible either because they did not meet the stimulability criteria, received too high of score on the word probe, demonstrated more than two sounds other than /ɹ/ in error, or a combination of the criteria. Of the 13 eligible participants, seven participants were randomized to receive immediate treatment and six were randomized to receive delayed treatment. Characteristics of the 13 enrolled participants are included in Table 1. Nonparametric analyses revealed that the groups did not differ in sex (X²(1, N = 13) = 0.03, p = .85), age (X²(1, N = 13) = 0.06, p = .8), speech (GFTA-3: X²(1, N = 13) = 0.68, p = .41), or language (PPVT-4: X²(1, N = 13) = 0.1, p = .75) skills. Individual participant characteristics can be found in Supplemental Material S1.

Figure 1.

CONSORT diagram depicting participant enrollment, randomization, and follow-up.

Table 1.

Group characteristics.

Variable	Immediate group	Delayed group
Sex	2 female	2 female
	5 male	4 male
Age (years)	8.61 (0.99)	8.65 (1.34)
GFTA-3 Standard Score	50.43 (8.87)	49.17 (13.09)
PPVT-4 Standard Score	112.57 (16.05)	115.17 (3.19)
CELF-5 Recalling Sentences Scaled Score	12.57 (2.44)	12.67 (3.14)
LAT Inconsistent Productions	0.86 (0.69)	0.67 (0.52)
Additions on SRT	0.43 (0.53)	0.5 (0.84)

Note. Values reported as mean (standard deviation). GFTA-3 = Goldman Fristoe Test of Articulation–Third Edition; PPVT-4 = Peabody Picture Vocabulary Test–Fourth Edition; CELF-5 = Clinical Evaluation of Language Fundamentals–Fifth Edition; LAT = LinguiSystems Articulation Test; SRT = Syllable Repetition Task.

Eligible participants represented a variety of states, including Georgia, North Carolina, New York, California, Pennsylvania, Maryland, Massachusetts, and Connecticut. Twelve participants identified as White, and one participant identified as more than one race. Previous speech therapy experience ranged from no prior speech therapy experience to 7 years of speech therapy experience. In regards to speech therapy for /ɹ/ specifically, experience ranged from no prior treatment targeting /ɹ/ to 4 years of treatment targeting /ɹ/. Enrolled participants were asked to pause outside speech therapy targeting /ɹ/ for the duration of the study.

Study Design

The treatment phase of the study consisted of thirty-two 10-min telepractice Chaining SPLITS sessions provided at a frequency of four sessions per week for 8 weeks. Participants were randomized to receive the 8 weeks of Chaining SPLITS either immediately after eligibility was determined or after an 8-week delay, as shown in Figure 2 below. The treating clinician was blinded to the randomization procedures. Randomization blocks were generated using random.org to generate a batch of 10 assignments with five in the immediate group and five in the delayed group. A second batch of 10 assignments was also generated, although the study stopped recruiting when the project timeline ended with a total of 13 participants randomized to a treatment condition (seven immediate, six delayed).

Figure 2.

Design overview for N = 13 participants. SPLITS = SPeech Lessons in Intensive Ten-minute Sessions.

Probe sessions for all participants were scheduled at 3 times: at baseline, after 8 weeks, when the immediate treatment group had completed Chaining SPLITS and when the delayed treatment group had a period of no treatment, and after 16 weeks, when the delayed treatment group had completed Chaining SPLITS and when the immediate treatment group had a period of no treatment.

Treatment Program

Treating clinician. The first and second author conducted all treatment sessions. Both the first and second authors are ASHA-certified SLPs with prior experience working with school aged children via telepractice. The first author was the primary treating clinician for 11 participants (six immediate, five delayed), and the second author was the primary treating clinician for two participants (one immediate, one delayed).

Treatment targets. The Chaining SPLITS model included thirty-two 10-min treatment sessions targeting /ɹ/, provided 4 times a week over 8 weeks. Treatment targets included two /ɹ/ onset chains (e.g., “ree” ➔ “read” ➔ “reading” ➔ “reading a good book” ➔ self-generated sentence for “reading”) and two /ɹ/ coda chains (e.g., “are” ➔ “car” ➔ “cartoons” ➔ “cartoons on TV” ➔ self-generated sentence for “cartoon”). Targets were individually selected by the treating SLP (i.e., the first or second author) based on the results of the syllable probe from the evaluation and dynamic assessment visits, such that each participant was stimulable for at least two of the four practiced syllables.

Treatment procedures. Treatment procedures for this study were guided by the SMC website (chaining.syr.edu). As described by Preston, Leece, and Storto (2019), each chain consisted of five levels, beginning with syllables (e.g., “ree”) and ending with self-generated sentences. Each level of the targeted chain was practiced in blocks of six trials, with participants needing to demonstrate accurate productions on at least five of the six trials to proceed to the next level of the chain. If this step-up criterion was not met, the next block of trials began with a different syllable. In regards to clinician feedback, the SMC website prompted the SLP on each trial to provide either KP feedback (e.g., “I think you need to lift the front of your tongue up higher”), KR feedback (e.g., “That was a good one!”), or no feedback. Additionally, for 50% of trials in each block, the participant was also asked to rate the accuracy of their own production of /ɹ/ (e.g., the clinician asked, “What did you think of that one?”). An example video of the treatment is available in Supplemental Material S2.

Weekly treatment structure. The general weekly treatment structure of sessions within the Chaining SPLITS model are displayed in Figure 3 below. The first 10-min session in the first week of treatment consisted of 10 min of prepractice, in order to remind participants of the strategies and tongue placement necessary for /ɹ/. For Weeks 2 through 8, the first 10-min session (Session A) consisted of a 1-min review of proper tongue placement for /ɹ/ using articulatory images and animations (Benway & Preston, 2022), followed by 9 min of structured practice guided by the SMC website. In structured practice, the website identified and adapted the target and feedback assigned to each trial and modified practice/feedback conditions (e.g., feedback type, prosodic variation, target complexity) based on the participant's performance. Sessions B and C included 10 min of structured practice, resuming where the previous 10-min session left off. If the participant advanced to the monosyllabic word level on at least three out of the four chains targeted in that week, the final session of the week (Session D) consisted of 10 min of random practice. The website identified the highest level that the participant achieved at least five out of six accurate productions for each target, and those levels were included in random practice. Participants who did not meet the criteria to advance to random practice proceeded with 10 min of continued structured practice in Session D.

Figure 3.

Weekly treatment structure for Chaining SPeech Lessons in Intensive Ten-minute Sessions (SPLITS).

Changing targets. At the end of each week, the treating clinician considered the participant's accuracy within sessions to determine if targets needed to be changed. Similar to Preston et al. 2024, if the participant accurately produced at least five out of six self-generated sentences for a given chain, the clinician modified the monosyllabic word included in that chain in the following week (e.g., the chain “row” ➔ “roast” ➔ “roasted” ➔ “roasted chicken” ➔ self-generated sentence for “roasted” was modified to “row” ➔ “road” ➔ “rodeo” ➔ “rodeo show” ➔ self-generated sentence for “rodeo”). If, in the following week, the participant achieved at least five out of six accurate productions at the self-generated sentence level on the chain again, a new syllable target was chosen (e.g., “row” was switched to “rah”).

All sessions were conducted using a HIPAA-compliant Zoom account. The clinician shared a portion of their screen with the participants during all treatment sessions, so that the participants were able to see the target assigned to each trial without being able to view the clinician's perceptual ratings. Treatment was initiated in July of 2022 and concluded in November of 2023.

Fidelity

Sessions were audio recorded, and screen-recordings of the clinician's SMC window (i.e., not including the child's face) were collected. For each participant, the third author measured fidelity to the treatment protocol for 25% of each participant's sessions, to describe how often the treating SLP administered treatment aspects that were theorized to be clinically important. Treatment aspects measured for fidelity included feedback type and frequency, progression through the chains, and incorporation of the articulatory placement review and of random practice on six randomly selected sessions, including one of Session A, four of Session B/C, and one of Session D. Additionally, the third author tracked the time from the first model to the last instance of feedback for each session to ensure that sessions followed the correct timing, as the sessions are intended to be only 10 min in the Chaining SPLITS model. As shown in Table 2, the average treatment time per session was 10 min and 32 s for both groups. On average, both groups received slightly less than 32 sessions, with the immediate group receiving an average of 31.57 sessions (SD = 1.13) and the delayed group receiving an average of 31.33 sessions (SD = 1.63). Overall, treatment fidelity was high, with fidelity for any treatment aspect being no lower than 86%.

Table 2.

Fidelity for the treatment groups.

Variable	Immediate group (n = 7)	Delayed group (n = 6)
Average practice time per session (min)	10:32 (0:55)	10:32 (0:42)
Number of sessions received	31.57 (1.13)	31.33 (1.63)
Average number of Speech Motor Chaining structured practice trials per session	46.97 (14.74)	53.55 (15.8)
Percent of Session A's in which prepracticed was provided (rated as yes/no)	100% (0)	90.91% (30.15)
Percent of Session D's in which random practice was appropriately implemented (rated as yes/no)	100% (0)	100% (0)
Percent of sessions eligible for new chain and clinician implemented change (rated as yes/no)	100% (0)	100% (0)
Fidelity for percentage of trials per session in which the clinician provided a model matching the stimulus	99.58% (0.01)	99.36% (1.76)
Fidelity for percentage of trials per session in which the rater independently agreed with the clinician's rating	86.93% (0.05)	89.30% (0.05)
Fidelity for percentage of trials per session in which the clinician prompted the participant to self-evaluate when warranted by SMC website	99.89% (0.005)	99.94% (0.004)
Fidelity for percentage of trials per session in which feedback was provided when warranted by SMC website	99.69% (0.009)	99.49% (0.01)
Fidelity for percentage of trials per session in which the appropriate type of feedback was provided (i.e., KR/KP) as specified SMC website	99.02% (0.02)	99.13% (0.02)
Fidelity for percentage of trials per session in which prosodic cues were provided when warranted by SMC website	99.95% (0.004)	100% (0)
Fidelity for percentage of trials per session in which the clinician provided a prosodic variation as specified by SMC website	97.91% (0.04)	100% (0)

Note. 25% of each participant's sessions were reviewed for fidelity. Means are displayed in each cell with standard deviation in parentheses. SMC = Speech Motor Chaining; KR = knowledge of results; KP = knowledge of performance.

Outcome Measures

Probes

Syllable and word-level productions were recorded within the probe sessions. Because this study was completed entirely via telepractice, parents of participants recorded the probes at home using wired headphones with a built-in microphone sent to them by the research team, while the clinician provided direct models to the participant and recording instructions to the parents via Zoom. The syllable probe included 45 directly imitated /ɹ/ syllables, including /ɹ/ in both the onset (e.g., “ree”) and coda (e.g., “ear”) positions. The word probe consisted of 125 untreated words containing /ɹ/ in the onset singleton (e.g., “rabbit”), onset cluster (e.g., “pretzel”), nucleus (“bird”), and coda (e.g., “steer”) positions, with pictures as a reference.

Perceptual Ratings

Productions of /ɹ/ from the probe recordings were rated using a 5-point scale of correctness by three independent SLPs or SLP graduate students who were provided with training and anchor stimuli. The treating clinicians (the first and second author) did not complete any ratings. All raters were masked to the treatment condition and time points at which these recordings were collected. Raters were assigned listening modules of 105 utterances (100 utterances, plus five utterances rated a second time for intrarater reliability purposes) through a full-stack, custom-built perceptual rating website. Utterances in each listening module were presented in a random order from a variety of participants, time points, and probe tasks. Raters listened to each utterance up to 5 times and rated /ɹ/ production along the following scale: (a) Deletion or Substitution of /ɹ/; (b) Obvious distortion, minimal/slight rhotic quality; (c) Very close, leaning distorted; (d) Very close, leaning correct, or correct /ɹ/ but slightly unnatural sound; (e) Fully correct, adultlike production. In cases of words containing more than one /ɹ/ (e.g., “strawberry,” “tractor”), raters were instructed to rate only the initial /ɹ/ within the cluster. Additionally, if a rater encountered an utterance in which the audio recording was of such poor quality that the /ɹ/ sound could not be rated (e.g., due to background noise), they were told to label that utterance as unusable audio so that it could be filtered out from the analysis.

Reliability. Spearman's ρ was used to evaluate intrarater reliability of the four SLPs and one SLP graduate student who completed ratings, with a total of 789 utterances rated a second time for intrarater reliability purposes. Intrarater reliability was good for all raters, ranging from 0.82 to 0.89. Average intrarater reliability across the five raters was 0.86. Spearman's ρ calculations were completed using Statistical Analysis Software (SAS, Version 9.4).

Posttreatment Parent/Child Questionnaires

Following treatment, participants and their parents were asked to complete questionnaires regarding their opinions and experiences with the treatment included in this study. The child questionnaire included three statements (“I know what to do to make a good R sound,” “I liked doing the short sessions,” and “The short sessions made my speech clearer”); the participants responded using a 5-item Likert scale ranging from strongly disagree to strongly agree. Children were also asked the open-ended question, “How did you feel about the short speech sessions?” To analyze the results of the child posttreatment questionnaires, the 5-point Likert scale responses were coded to a numeric scale ranging from −2 (strongly disagree) to 2 (strongly agree).

The parent questionnaire included multiple open-ended questions (e.g., “How did you feel about the short, frequent sessions that your child received in this study?” “How would you feel if your child was offered these short, frequent sessions in school?”). Additionally, the parent questionnaire included one multiple choice question (“If your child were to receive speech therapy in school, which of the following treatment schedules would you prefer?”), with answer choices of “One session a week for 60 minutes,” “Two sessions a week for 30 minutes each,” and “Five sessions a week for 10 minutes each.” Ten of the 13 parents filled out the parent questionnaire following treatment. OpenAI's ChatGPT was used to extract common themes from the parents' responses in an unbiased manner (OpenAI, 2024). All parent responses were deidentified prior to the use of ChatGPT.

Data Analysis

Utterances from the stimulability (n = 45) and word (n = 125) tasks were rated across the three time points, yielding a total of 510 rated utterances per participant. Three independent raters rated each utterance. We filtered out the utterances marked as unusable audio (n = 80) and the utterances that were rated a second time for intrarater reliability, leaving a total of 19,798 analyzed data points. The data points were averaged to find the listener-average perceptual rating for a given utterance at a given time point.

Linear mixed models were used to evaluate the efficacy of Chaining SPLITS compared to the no-treatment phase. Because differences in performance were noted at baseline between the two groups, the analysis was completed with change score as the outcome variable. Change score was determined by finding the difference between average perceptual ratings for each syllable or word across the different time points (with the difference between the 8-week and baseline time points identified as “Interval 1 Change Score” and the difference between the 16-week and 8-week time points “Interval 2 Change Score”). Using SAS (Version 9.4), we evaluated the Bayesian Information Criterion (BIC) of models with various fixed effects (Interval, Group, Task), random effects structures (utterance, participant-specific random slopes, and intercepts), and covariance structures. The model optimizing (lowering) BIC included fixed effects of treatment group (immediate or delayed), time interval (Interval 1 or 2), task (stimulability or word), the interaction between group and interval, and the interaction between group, interval, and task. Baseline was included in the final model as a covariate. The final model included participant-specific random slopes and intercepts, as well as the Heterogeneous Compound Symmetry covariance structure. Analyses were conducted with α = .05. The SAS code and de-identified data used for this analysis can be found in Supplemental Materials S3 and S4, respectively.

Additionally, we analyzed effect sizes to evaluate the clinical impact of Chaining SPLITS. Because the sample size was small (N < 20), we used Hedges's g to describe the magnitude of the effect of Chaining SPLITS on /ɹ/ production accuracy (Lakens, 2013; Turner & Bernard, 2006).

Results

Primary Analysis

We hypothesized that Chaining SPLITS would result in improvements in /ɹ/ accuracy compared to an equivalent period of time without treatment. Table 3 below displays raw mean scores across both tasks at the baseline, 8-week, and 16-week time points, as well as change score from Interval 1 and Interval 2, for each participant. Eleven of the 13 participants demonstrated improvement in /ɹ/ accuracy in response to the 8-week treatment period, based on perceptual ratings. Both groups showed similar levels of improvement in response to treatment, with the immediate and the delayed groups improving an average of 0.54 points on the 5-point rating scale.

Table 3.

Average ratings for each participant across two speech tasks.

Group	Participant	Baseline	8-week	16-week	Change score (Interval 1)	Change score (Interval 2)
Immediate	4412	2.85	3.68	3.66	+0.83	−0.02
	4426	2.18	2.06	1.83	−0.12	−0.23
	4439	2.05	2.48	2.18	+0.43	−0.3
	4452	2.45	4.4	4.46	+1.95	+0.06
	4454	2.05	1.86	2.28	−0.19	+0.42
	4456	2.38	2.99	3.46	+0.61	+0.47
	4458	2.55	2.85	3.04	+0.3	+0.19
	Group M (SD)	2.36 (0.29)	2.9 (0.9)	2.99 (0.94)	+0.54	+0.09
Delayed	4413	3.2	3.49	3.77	+0.29	+0.28
	4432	2.7	2.64	3.37	−0.06	+0.73
	4434	1.9	1.75	2.95	−0.15	+1.2
	4437	3.93	4	4.17	+0.07	+0.17
	4446	3.73	3.88	4.3	+0.15	+0.42
	4457	2.78	3.94	4.35	+1.16	+0.41
	Group M (SD)	3.04 (0.75)	3.28 (0.91)	3.82 (0.56)	+0.24	+0.54

Results of the linear mixed-effect model are shown in Table 4. The model revealed a significant three-way time-by-group-by-task interaction (p = .008), indicating that there were significant differences in change score between the two groups across time intervals between the two tasks. To deconstruct this three-way interaction, we examined the group-by-time interaction for each of the tasks. A significant group-by-time interaction was found specifically for the syllable probe task (p = .0133) but not the word probe task (p = .1315). The treated groups demonstrated more progress than the untreated groups on the stimulability task during the time interval in which they received treatment, showing a statistically significant effect of Chaining SPLITS at the syllable-level. There was not sufficient evidence to claim that the untreated word-level items were significantly improved with Chaining SPLITS in the study's time frame.

Table 4.

Results of the linear mixed-effect model.

Term	Estimate	SE	Test statistic	p value
Intercept	1.37	0.43	3.19	—
Group (ref = Delayed)	−0.73	0.2	−3.69	.0002
Time Interval (ref = Interval 2)	−0.33	0.32	−1.03	.3149
Baseline	−0.26	0.13	−1.95	.0507
Task (ref = Stim)	−0.03	0.08	−0.41	.68
Group × Time Interval (ref = Stim)	1.08	0.44	2.48	.0133
Group × Time Interval (ref = Word)	0.64	0.43	1.51	.1315
Task × Group × Time Interval (ref = Word)	−0.28	0.11	−2.65	.008

Note. Interval 1 = Baseline to 8 weeks during which time the immediate group was treated; Interval 2 = 8 to 16 weeks during which time the delayed group was treated.

Figure 4 below displays the change scores for each participant and group averages across both time intervals for the stimulability and word probe tasks. After receiving treatment during Interval 1, the immediate treatment group improved an average of +0.78 points along the 5-point rating scale on the syllable probe and an average of +0.47 points on the word probe. In the same interval, when no treatment was provided, the delayed treatment group still demonstrated minor improvement on both the stimulability (+0.24) and word (+0.24) probes. At Interval 2, after 8 weeks of no treatment, the immediate treatment group showed maintenance of their performance (change scores of +0.02 on stimulability and +0.1 on word probes). The delayed treatment group demonstrated positive change scores on the stimulability (+0.57) and word probes (+0.57) after receiving treatment in Interval 2.

Figure 4.

Change scores for both groups on the stimulability and word probes. Blue and orange lines represent group means for the immediate and delayed treatment groups, respectively. Individual data points represent participant mean change scores during each time interval. Interval 1 = Baseline to 8 weeks (during which time the immediate group was treated); Interval 2 = 8 to 16 weeks (during which time the delayed group was treated).

Using Hedges's g, the effect size of the between-group differences in change score at Interval 1 for the syllable probe was 0.46, which indicates a small-to-medium effect of Chaining SPLITS at the syllable level. At the untreated word level, however, the between-group effect size at Interval 1 was 0.19, indicating a small effect. For the between-group differences in change score at Interval 2, we found a medium effect at the syllable level (g = 0.51) and a small-to-medium effect size at the untreated word level (g = 0.36). Overall, the magnitude of the effect of Chaining SPLITS appears to be greater at the syllable level than at the word level.

Analysis of Child/Parent Posttreatment Questionnaires

Average posttreatment responses to the statements, “I know what to do to make a good R sound,” “I liked doing the short sessions,” and “The short sessions made my speech clearer” were 1.15, 1.08, and 1.15, respectively, indicating that average response from the participants was “agree” for all three statements. Twelve participants responded positively to the open-ended question, “How did you feel about the short speech sessions?”, with many participants specifically mentioning the short length and helpfulness of the sessions. One participant stated both a positive and negative aspect of the study, with the negative aspect being that the study “took a long time.”

In regards to the parent questionnaires, the following ChatGPT-generated text was produced in response to the prompt, “Please summarize the common themes of the survey responses in two sentences.” Output was reviewed and edited slightly for clarity and accuracy: “The survey responses commonly praised the short, frequent sessions for their convenience and effectiveness in keeping children focused and motivated. However, one parent expressed concern about the sessions' brevity potentially limiting deeper exploration of topics” (OpenAI, 2024). In response to the multiple-choice question, “If your child were to receive speech therapy in school, which of the following treatment schedules would you prefer?”, eight of the 10 parents who responded to the questionnaire indicated that they would prefer their child to receive five 10-min sessions a week, rather than longer, less frequent sessions. Individual parent responses can be found in Supplemental Material S5.

Discussion

The purpose of this study was to determine the initial efficacy of the Chaining SPLITS model (i.e., four 10-min sessions per week over 8 weeks, provided with the SMC approach). Broadly, the SSD treatment literature supports frequent practice, (Allen, 2013; Page et al., 1994; Preston et al., 2024) and the mechanism for these improvements—the spacing effect (i.e., benefit of spaced/distributed vs. longer, massed sessions)—has been well established in the cognitive psychology literature (Donovan & Radosevich, 1999). The development of the Chaining SPLITS model was informed by this base of support. This study serves as the first investigation of an alternative service delivery model with a no-treatment control.

We hypothesized that after 8 weeks of Chaining SPLITS, 7–9-year-old children would demonstrate improvement in /ɹ/ accuracy on syllables and untrained words. The current study confirmed this hypothesis at the syllable level: we found that the use of Chaining SPLITS led to statistically significant improvements in /ɹ/ accuracy in syllables as a result of treatment, compared to a no-treatment condition. After receiving treatment during Interval 1, the immediate group demonstrated a significant increase in /ɹ/ production accuracy. The delayed group improved slightly in /ɹ/ accuracy during Interval 1 (after the no-treatment period) but demonstrated statistically significant improvement after the 8-week treatment period. The immediate group showed maintenance of treatment gains after an 8-week period without treatment (Interval 2). Based on responses to the posttreatment questionnaires, children liked the short speech sessions and felt that they were helpful in making their speech clearer.

It is important to note that the improvements observed were task-dependent; both groups improved significantly in /ɹ/ accuracy at the syllable-level, but improvements at the untreated word-level were not statistically significant. Overall, results from the current study provide further evidence to support frequently spaced sessions for children with SSDs. Results also contribute to the growing body of research on SMC, for which effectiveness has previously been established in the context of longer sessions (e.g., 60-min sessions provided in-person twice a week).

Although we found that Chaining SPLITS led to significant improvements at the syllable level, the null finding at the untreated word level suggests that 8 weeks of treatment provided in the Chaining SPLITS model may have been an insufficient amount of practice to promote improvements beyond the syllable level. However, when considering the participants' individual trajectories of improvement, two of the participants (4452 and 4434) improved their /ɹ/ productions at the untreated word level by over 1 point on the 5-point rating scale following their treated interval (+1.87, +1.37, respectively). This indicates that 8 weeks of Chaining SPLITS may in fact be sufficient to see significant word-level change for some, but not all children.

Interestingly, one of the participants in the delayed treatment group improved by 1.2 points during Interval 1, when they were not receiving treatment. It is possible that the 30-min dynamic assessment visit stimulated this improvement in the untreated condition; however, this study was not designed to test for improvements after the dynamic assessment visit alone. The remaining five participants in the delayed treatment group demonstrated change scores closer to zero in the no-treatment period. This observation suggests that a minority of children between the ages of 7 and 9 years may exhibit maturation in their /ɹ/ production accuracy without intervention. A strength of this study is that the experimental design provides scientific control for maturation, whereas interpretation of previous studies on alternative service delivery models (Byers et al., 2021; Rehfeld & Sulak, 2021) are limited by a lack of experimental control for maturation. Byers et al. (2021) and Rehfeld and Sulak (2021) both found no evidence of a difference between an alternative service delivery model and the traditional service delivery model; however, neither study included a control group to account for maturation.

Clinical Implications

The current study provides initial evidence of the effectiveness of Chaining SPLITS. Currently, some school-based SLPs already incorporate service delivery models similar to Chaining SPLITS in their clinical practice (Brosseau-Lapré & Greenwell, 2019; Kuhn, 2006). An advantage of service delivery models using short, frequent sessions is that they can be provided within the tiered intervention or RTI frameworks. The SLP's role in the RTI framework is especially important for children with milder SSDs, as they may not be eligible for speech therapy services through individualized education plans. In one school district, Kuhn (2006) described a model in which informal speech therapy services were provided 3 times a week for 5- to 7-min sessions. Similar to the Chaining SPLITS model, the model described by Kuhn was provided for 8-week increments. Kuhn reported that two students met their goals after the initial 8 weeks, but most students required 16 weeks of treatment to meet their goals. It is possible that implementing Chaining SPLITS with a similar length of treatment would result in more substantial improvement at the word level than we observed from 8-weeks of treatment in the current study.

Although they have not been widely studied, alternative service delivery models using short, frequent sessions offer many additional advantages in school settings. Because the sessions are shorter, any missed academic instruction time is less concentrated. The nature of short sessions can also make data collection and lesson planning relatively simple tasks for SLPs compared to the traditional longer, group sessions. This is especially true for sessions using SMC, as the website (chaining.syr.edu) automates many of the treatment decisions (e.g., target selection, type, and frequency of feedback) and keeps a record of the child's performance. With short sessions, there is greater likelihood for sustained attention to task, and therefore no need to plan and incorporate nonspeech activities, such as games or crafts. In the present study, children reported that they liked the short sessions, suggesting that drill-based speech practice without unrelated activities was not only tolerable, but an enjoyable experience for most children. Additionally, children likely have more practice opportunities in individual sessions than group sessions. Participants in the current study achieved an average of about 50 trials per 10-min session, compared to an average of 20 trials per goal achieved by children in 30-min three-person group sessions reported by Rehfeld and Sulak (2021). Opportunities to maximize dose (i.e., the number of trials per sessions) and dose frequency (i.e., the number of sessions per week; Warren et al., 2007) should be a key consideration of treatment for SSDs (Allen, 2013; Farquharson et al., 2022). The Chaining SPLITS model involving individual 10-min sessions 4 times per week appears to be an effective way to achieve many practice opportunities and is a promising model for efficient delivery of school-based services for children with mild SSDs.

Limitations

One limitation of the current study is that, due to its preliminary nature, the sample size was relatively small. The strict eligibility criteria related to stimulability was a primary reason why several potential participants were considered ineligible. However, we found this eligibility criteria necessary to ensure that the participants in this study were good candidates for treatment provided in the Chaining SPLITS model, as stimulability is a prerequisite for implementing structured SMC practice. While not systematically investigated in this study, differences between face-to-face treatment and treatment via telepractice are possible (although not regularly found; e.g., no difference between in-person and telepractice outcomes using SMC, Preston et al., 2024). The use of telepractice was a strength of this study, as it allowed us to include participants from a variety of states across the United States, thereby improving generalizability of the results.

The fact that the current study only treated /ɹ/ may also be considered a limitation, as it restricts our ability to generalize the results to other speech sounds. We chose to only treat /ɹ/ as a way to strengthen the external validity of the current study, given the prevalence of this sound error, and the internal validity of the current study, considering the complexity of this sound. Future studies are needed to determine if these results extend to other speech sounds.

Conclusions

The current study found that the Chaining SPLITS model, in which children receive 10-min speech sessions 4 times a week over 8 weeks, led to improvements in /ɹ/ accuracy at the syllable level when compared to a no treatment condition. At the untreated word level, however, there was no evidence of a positive effect of Chaining SPLITS relative to the no-treatment condition, suggesting that longer courses of treatment or different approaches may be necessary to result in significant word-level improvement. Alternative service delivery models such as Chaining SPLITS offer several clinical advantages, especially in school-based settings. Future studies, ideally with a greater sample size and a variety of targeted speech sounds, are necessary to document the generalizability of the results of the current study.

Data Availability Statement

De-identified data, data analysis code, and a treatment video are available within the supplemental materials of this article.

Artificial Intelligence Statement

ChatGPT was used in this study to summarize responses to the parent questionnaire in an unbiased manner, with the resulting text being revised by the authors prior to inclusion in the manuscript.

Funding Statement

The first author was supported by a fellowship provided by the Syracuse University College of Arts and Sciences. Data analysis for this study was supported by National Institutes of Health (NIH) Grant R01DC020959 (awarded to Principal Investigator: J.L. Preston). The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH.