Feasibility and Preliminary Efficacy of Direct Instruction for Individuals With Autism Utilizing Speech-Generating Devices

Abstract

Direct instruction (DI) is an evidence-based approach to education that has been shown to be effective across a wide variety of student populations. Growing evidence suggests that DI may be an efficacious strategy for individuals with autism spectrum disorder (ASD). The current study aimed to evaluate the feasibility of using DI with students with ASD who utilize speech-generating devices (SGDs); 3 students with ASD whose primary mode of communication was an SGD were exposed to the Language for Learning Curriculum, Lessons 1–10. Student performance on pre- and posttests was measured, as well as student performance on exercises within each lesson. The average time to complete an exercise, number of repetitions, number of terminated sessions, and student affect were also evaluated. Results indicated that all 3 students could participate and complete exercises with some modifications to support SGD use. The students demonstrated improved performance, positive affect, and overall timely completion of exercises. Taken together, these findings suggest that DI may be feasible for some students with ASD who utilize SGDs.

Direct instruction (DI) is a comprehensive approach to education focused on promoting student mastery through active responding and interactions with the teacher (Bereiter & Engelmann, 1966; Engelmann, 1980). DI incorporates both a curriculum (i.e., what to teach) and procedures (i.e., how to teach). The curriculum is designed to be implemented by a trained teacher or paraprofessional in a classroom setting with groups of students. DI curricula include scripts for the instructor to follow when presenting exercises and a fully developed scope and sequence of skills, reducing teacher preparation time. DI also incorporates a general case model of instruction such that appropriate discriminations are developed (O’Neill, 1990). For example, multiple exemplars of visual stimuli (e.g., a red hat in one exercise and a blue hat in another), as well as a systematic arrangement of irrelevant features, are included. DI curricula have been developed across a variety of topic areas (e.g., reading, math, writing).

Instructional procedures within DI typically include (a) orientation, (b) presentation, (c) practice, (d) feedback, and (e) independent practice (Engelmann, 1980). The orientation component consists of the teacher describing the content of the upcoming exercise to the learners. The content will be linked to previously mastered skills or concepts. The presentation component consists of the teacher describing or modeling the desired response. The practice component consists of opportunities for students to respond individually or chorally as a group. The feedback component consists of praise for correct responses and error correction following incorrect responses. Teachers are encouraged to continually assess student responding and learning through these response and feedback interactions (Engelmann, 1980).

To date, research suggests that DI is one of the most effective instructional approaches identified in educational research (Hattie, 2009), with an average effect size of 0.59 across 304 studies and over 42,000 students. However, the effectiveness of DI with children with autism spectrum disorder (ASD) has yet to be firmly established (Wong et al., 2015). At the time of the review by Wong et al. (2015), additional studies from varied institutions were required for DI to meet the criteria for consideration as evidence based. Additional publications since that time indicate these criteria may now be met (e.g., Shillingsburg, Bowen, Peterman, & Gayman, 2015).

DI curricula focused on teaching language (Engelmann & Osborn, 1976, 1999) may be particularly useful for educators serving students with language delays associated with ASD. The DI Language for Learning (DI-LL) includes 150 lessons divided into groups of 10, with exercises within each lesson designed to target language skills related to basic vocabulary, common concepts, and grammatical structures up to a second-grade level. Ganz and Flores (2009) used components of the DI-LL to teach vocabulary skills to three children with ASD. Results indicated the intervention was effective for all participants, with one student demonstrating generalized responding across settings and instructors. Shillingsburg et al. (2015) extended this research by investigating the DI-LL curriculum in its entirety with 18 children with ASD. All children showed individual gains postintervention, which were sustained up to 8 months over preintervention levels. Additionally, the experimental group showed significant improvements compared to the wait-list control, ruling out the effects of maturation alone as responsible for increased language skills.

Of note, Shillingsburg et al. (2015) incorporated several modifications to the DI-LL curriculum to support its implementation with children with ASD. Sessions were conducted one on one in a therapy room, rather than in a group setting in a classroom. With these changes, some exercises were modified to include picture stimuli rather than three-dimensional objects (i.e., a picture of a flag instead of an actual flag). To promote responding, structured preference assessments were used to select tangible reinforcers (Hagopian, Long, & Rush, 2004), and more specific schedules of reinforcement, which included toys, snacks, and breaks from instruction, were programmed into the sessions. Last, the lesson posttests were used as pretest measures in order to evaluate progress. Although the effects of these modifications cannot be specifically evaluated for their individual impact on performance, they may be considered part of an effective treatment package in combination with the DI-LL.

The results of previous studies suggest that the DI-LL can be effective in teaching language skills to individuals with ASD (e.g., Ganz & Flores, 2009; Shillingsburg et al., 2015). However, research has yet to determine for which individuals this approach will be effective. Shillingsburg et al. found that the DI-LL was effective for students demonstrating basic vocal skills at enrollment (e.g., echoing instructors, using single words only to request). Given the language impairments prevalent in ASD, more research is needed to determine if this curriculum can be adapted for learners with greater language challenges without sacrificing efficacy.

Research suggests that as many as 30% of children with ASD do not acquire functional speech (Rose, Trembath, Keen, & Paynter, 2016) and may require the support of augmentative and alternative communication (AAC; Prizant & Wetherby, 2005). AAC is communication through any other means than speech. It can be used in the support or replacement of speech. AAC systems are categorized as either no-tech, low-tech, or high-tech systems. No-tech systems are those that require nothing except a communicator’s body, such as facial expressions, gestures, or signs. Low-tech systems are composed of paper-based or tangible materials, such as objects, photographs, text, or communication boards or books. High-tech systems are electronic systems with digitized voice output, such as single-message voice-output buttons or speech-generating devices (SGDs).

SGDs are high-tech voice-output devices that provide the AAC user with opportunities to emit a variety of communicative responses, such as greetings, asking and answering questions, and commenting. SGDs may be “dedicated” devices such as tablets or computers produced exclusively for use as communication devices or commercially available tablets equipped with aftermarket speech-generating applications. Most SGDs offer touch screens and have dynamic screen displays, meaning that the user’s selection of one button causes the display to change or navigate to other screens with additional folder options. This can facilitate faster communication and allows for a very large number of icons to be provided in one system. For many AAC users, the ultimate goal is the production of novel phrases. Therefore, SGDs are beneficial in that available vocabulary can be robust enough to provide thousands of word combinations. This expansive vocabulary allows the AAC user to emit communication responses with a high degree of specificity.

There are many factors that can determine the successful use of an SGD, and these must be considered when teaching new language skills to individuals with ASD. In order to create an utterance (i.e., speech output via device), an AAC user must have access to the necessary icons representing words (i.e., vocabulary). AAC vocabulary words are often categorized into two types: core vocabulary words and fringe vocabulary words (Yorkston, Dowden, Honsinger, Marriner, & Smith, 1988). Core words are the words most frequently used and consist mainly of verbs, pronouns, and descriptors. Fringe words are more specialized and less frequently used and consist mostly of nouns. Another important factor is the symbol used, which could include text, numbers, icons, or pictures.

Another factor as important as the symbols selected for the programmed vocabulary is ensuring the icons are arranged in a manner that is accessible for each specific user (Beukelman & Mirenda, 2013). A symbol display is the way symbols are organized on an SGD. There are different types of displays, and these have an impact on the type of utterances the AAC user can produce (Da Fonte & Boesch, 2019). For instance, an activity grid display contains vocabulary for one specific activity. Although this may make communication easy and accessible in that activity, it may make it harder for the communicator to generalize responses to other contexts. Similarly, visual scenes, nongrid pictures, or drawings of specific locations with embedded speech-output hot spots may allow in-depth and accurate communication about one location or routine while limiting communication about other locations or topics. However, a semantic-syntactic display is a grid-based display that organizes grammatical elements in a typical syntactic order. Although this display may not be as convenient for one specific activity, it allows the user to create grammatical responses across many contexts and activities. Many SGDs allow for the combination of more than one type of display.

To date, no studies have evaluated whether students with ASD who use SGDs could benefit from a structured curriculum like the DI-LL. The purpose of the current study was to evaluate the feasibility of implementing the DI-LL with children with ASD who use SGDs as their primary mode of communication. In other fields of research, feasibility studies are used to develop the parameters that will be needed to demonstrate an effect on a larger scale. Feasibility studies do not necessarily aim for an experimental demonstration of efficacy. Rather, a feasibility study may be used to determine how many participants are needed to have a sufficient sample size, whether participants can be randomized into experimental and control groups, whether the researchers can recruit sufficient participants who meet their inclusion criteria, whether the procedures can be followed with sufficient rigor, and whether the dependent measures are valid and appropriately calibrated (Arain, Campbell, Cooper, & Lancaster, 2010). Given the novelty of the combination of DI-LL and SGD use, we elected to adopt the framework of a feasibility study before developing a larger scale study. Thus, the aims of the current study were to evaluate whether (a) we could identify and enroll students who were appropriate for participation, (b) the students willingly participated in the sessions, (c) the length and frequency of sessions would be feasible in a non-public school setting serving students with ASD, and (d) the students could complete the exercises with reasonable modifications. We also included measures of efficacy, which were evaluated with the use of a single-subject design.

Method

Participants

Participants all attended a private school serving students with ASD. Due to severe problem behavior or the degree of educational need, all participants had been placed in the private school by their respective individualized education program teams. Participants received their typical educational services in a self-contained classroom with a ratio of approximately two students to one instructor. Participants were included in the study if they used an SGD as their primary form of communication and demonstrated at least single-word mands and tacts for at least 10 items. Participants were also required to follow a minimum of 10 single-step instructions (e.g., “Touch your nose.”) and remain seated for 80% of a 15-min session when interacting with the clinician and accessing preferred tangible items. Consent was sought from all caregivers, and participants assented to participation in each session by willingly walking with the experimenters to the session rooms.

Participants were nominated for participation in the study by a speech-language pathologist (SLP), the third author. The SLP was involved in educational services for the students and was familiar with the students’ SGD layouts and their overall fluency using their devices. Participants included Pablo, a 10-year-old Hispanic male; Brian, a 13-year-old Caucasian male; and Madison, a 15-year-old Caucasian female. None of the participants had previous experience with the DI-LL. Once caregivers consented to participation, the SLP conducted the Peabody Picture Vocabulary Test (PPVT; Dunn & Dunn, 2007) and the Expressive Vocabulary Test (EVT; Williams, 2007) individually with each participant. Of note, on these tests, the participants were required to respond using their SGDs, though some participants occasionally emitted vocal approximations. Results of these tests are displayed in Table 1.

Table 1.

Participant Characteristics

Name	Age	Diagnosis	Peabody Picture Vocabulary Test Score	Expressive Vocabulary Test Score
Pablo	10	ASD	26	53
Brian	13	ASD and unspecified intellectual disability	36	38
Madison	15	ASD	25	26

Brian and Pablo both used dedicated communication systems with chat software called NovoChat 8 (R). Madison used the TouchChat (R) app on an iPad Mini (R). All three participants used the WordPower 42 (R) vocabulary. The WordPower 42 app is a text- and icon-based vocabulary with a semantic-syntactic grid display. All three students’ systems had been personalized and modified with unique fringe words to reflect their interests and personal information. All three students used text buttons, icons, and the keyboard function to communicate. The SLP (to be referred to as the “clinician”) administered the DI-LL placement test, and all students placed into Lesson 1.

Setting and Materials

Sessions took place in varied locations throughout the school, depending on the availability of space. The locations were quiet rooms with sufficient space for the participant and clinician to both sit at a table and for a second observer to collect data. During all sessions, the participants were seated in proximity to the clinician so that he could easily model responses on the SGDs and prompt responses as needed. The DI-LL teacher presentation book and the participant’s SGD were also used in all sessions.

The DI-LL curriculum is designed for use with a group of children in a classroom setting, with some targets requiring the participants to tact classroom furniture and the gender of peers. As all sessions were conducted one on one in a private room, some modifications were made to the materials and targets. When the exercise called for the participant to tact the gender of a classmate, the clinician presented a picture of a boy or girl displayed in a book or on an iPhone screen. Different exemplars of each picture were used for each exercise to maintain the variety that would be expected in a classroom setting. When the exercise called for the participant to tact furniture or an environmental item, if the particular item was not available in the space (e.g., a flag), it was substituted for another piece of common furniture in the room. For Pablo, preferred edible items were used in the sessions to maintain his motivation to engage in the task. Pablo chose a desired item from an array of other edibles in his classroom, then brought the item to the session room.

Dependent Variables

The primary dependent variable was the percentage of correct responses on the DI-LL Assessment Test 1. Assessment Test 1 is typically administered as a posttest measure following completion of Lessons 1–10. Like Shillingsburg et al. (2015), we included the test as a baseline measure, as well as a posttest and maintenance measure. To be considered correct, responses had to be initiated independently within 5 s of the clinician’s question or signal as suggested by Shillingsburg et al. This response interval was intended to promote fluency in responding (Shillingsburg et al.), which is mentioned throughout the DI-LL guidebook but without a specific response criterion described. Of note, as the participants were responding using SGDs, we only required responding to be initiated—but not necessarily completed—within the interval. Use of an SGD with multiple categories necessarily requires navigation between several pages. Thus, we only required that responding be initiated within that time period and completed with independence. Responses had to meet the minimum criteria for length as indicated for each target in the DI-LL test (Engelmann & Osborn, 1999). For example, the response “hat” would be scored as incorrect when the target response was “a hat.” However, if a participant exceeded the length requirement, we still considered this a correct response when the utterance was grammatically correct. For example, the response “This is a hat” would be scored as correct when the target response was “a hat.” The test includes 25 total items; thus, all scores were calculated by counting the number of correct responses, dividing by 25, and then multiplying by 100.

Data were collected during intervention sessions to monitor progress and determine the feasibility of implementing the DI-LL with this population (Smith et al., 2007). The DI-LL suggests that exercises be repeated until fluency is observed. Thus, we tracked the number of exercises per lesson that was completed correctly and independently on the first attempt. To be considered correct, responses within each exercise had to be initiated within 5 s of the presented question or signal. Relatedly, we tracked the total number of repetitions per exercise. If correct responding was not achieved after three attempts to complete the exercise, the exercise met termination criteria. These criteria were in place to ensure participants progressed through the curriculum and were not held back by the failure to demonstrate correct responding on particular exercises. Given the systematic, programmed multiple exemplars throughout the curriculum, moving on from an exercise still ensured concepts and skills were practiced in later exercises. We also recorded the duration of each exercise, as well as participant affect as a social validity measure, which was evaluated using a behaviorally anchored rating scale adapted from Koegel, Vernon, and Koegel (2009; see Table 2). Termination criteria could also be met if the duration of an exercise exceeded 15 min or challenging problem behavior was observed. These criteria were in place to promote participant and instructor safety during the sessions.

Table 2.

The Behaviorally Anchored Rating Scale Used to Measure Participant Affect

Affect	Score	Description
Low interest	0	The individual looks bored and attempts to leave the area of the activity. The individual may attempt to avoid or escape the task by engaging in problem behaviors such as running away, throwing or destroying materials, crying, or refusing to perform the task. The individual appears unhappy (e.g., frowning, negative vocalizations, crying) during most of the session.
	1	The individual remains in the area of the activity but looks bored and is uninvolved. The individual may spend much time looking around and little time attending to the task. The individual may engage in behaviors unrelated to the activity. The individual appears unhappy (e.g., frowning, negative vocalizations, crying) during some of the session.
Moderate interest	2	The individual generally complies with the instructions but does not appear eager to participate. There may be moments of staring or inattention. Prompting the task and assessment of motivation were needed on a frequent basis to regain compliance with the activity. The individual infrequently shows brief signs of unhappiness (e.g., frowning, negative vocalizations, crying).
	3	The individual complies with the instruction but does not appear eager to participate in the activity. The individual generally focuses on the clinician and stimulus materials. Prompting the task and assessment of motivation were needed occasionally to regain compliance with the activity. The individual infrequently shows signs of minor unhappiness (e.g., frowning, whining) but also demonstrated signs of happiness (e.g., smiling, laughing, showing humor).
High interest	4	The individual attends and responds to the task readily. The individual is fairly alert, eager, and involved in the activity and frequently attends to the clinician and the stimulus materials during the trial. The individual appears happy during some of the session (e.g., smiling, laughing, showing humor).
	5	The individual attends readily to task and responds readily and willingly. The individual is alert, eager, and involved in activity. The individual attends to the clinician and the stimulus materials intently during the trials. The individual appears happy during most of session (e.g., smiling, laughing, showing humor).

Note. Ratings completed at the end of each session

Interobserver Agreement and Treatment Fidelity

For pre- and posttest sessions, primary data were collected in vivo by the first author. Reliability data were collected in vivo by the clinician conducting the sessions. Interobserver agreement (IOA) for responses to pre- and posttests was calculated by counting the number of agreements, dividing by the total number of test items (25), then multiplying by 100. For Pablo, reliability data were collected on 67% of sessions with an IOA of 100%. For Brian, reliability data were collected on 75% of sessions with an IOA of 100%. For Madison, reliability data were collected on 80% of sessions with an IOA of 100%.

For the intervention phase, primary data were collected by the first author, observing in vivo. In this phase, the clinician was implementing the teaching procedures; therefore, intensive data collection interfered with the pacing of the exercise presentations. Thus, a proportion of the intervention exercises was recorded on video, and these videos were later scored by a trained second observer. At the completion of each exercise, data collectors recorded a “yes” if responding across the entire exercise was completed independently and correctly, with all responses initiated within 5 s. If any one of these criteria was not met, they scored a “no.” IOA was calculated by comparing these scores for each exercise. If both observers scored “yes,” this was a score of 100% for the exercise. If scores differed, the score for the exercise was 0%. These scores were averaged across scored exercises for the reliability coefficient. For Pablo, reliability was scored on 25% of exercises with an IOA of 94% (range 0%–100%). For Brian, reliability was scored on 17% of exercises with an IOA of 91% (range 0%–100%). For Madison, reliability was scored on 11% of exercises with an IOA of 100%. IOA was calculated across exercises on participant affect by dividing the lowest affect score by the highest affect score, then multiplying by 100. For Pablo, reliability was scored on 25% of exercises with a mean IOA of 92% (range 75%–100%). For Brian, reliability was scored on 17% of exercises with a mean IOA of 100%. For Madison, reliability was scored on 11% of exercises with a mean IOA of 85% (range 60%–100%).

During 100% of test and intervention sessions, the first author recorded procedural fidelity data, in vivo, to ensure the components of the DI-LL and specific modifications were implemented correctly. A checklist was created with items detailing the key procedural elements, and if the item was implemented correctly, a “+” was scored. If implementation deviated from the item in any way, it was scored a “−.” The percentage of steps implemented with fidelity was calculated by recording the total number of “+” signs and dividing by the total number of scored items, then multiplying by 100. If consistent errors were observed during a lesson, feedback was given to ensure errors in implementation did not persist throughout the procedure. The first author verbally reminded the clinician to engage in the desired behavior (e.g., “Remember to signal before each instruction.”) once the lesson was completed. The mean percentage of steps implemented with fidelity was 98% (range 94%–100%) for Pablo, 97% (range 80%–100%) for Brian, and 91% (range 78%–100%) for Madison.

Experimental Design

A concurrent multiple-probe design across participants was used to evaluate the effects of the DI-LL. The DI-LL Assessment Test 1 was administered individually to each participant before and after the intervention. Participants were randomly assigned to be first, second, or third by drawing names from a hat. All participants completed one baseline pretest, and then the first participant began the intervention sessions. Once Lessons 1–10 were completed by the first participant, Assessment Test 1 was repeated, serving as a posttest for the first participant and an additional baseline measure for the second and third participants. Then the second participant began the intervention, and following the completion of Lessons 1–10, the test was administered again to the second participant, as a posttest, and to the third participant, as a final baseline measure. The third participant completed the posttest following Lessons 1–10. For all participants, a maintenance check was completed 4 weeks following their completion of their posttest with no additional exposure to the DI-LL.

General Procedures

To maintain high levels of participant engagement with the clinician, the researchers asked the participants’ teachers what strategies were typically used to promote participation in the classroom and what schedules of reinforcement were in place for skill acquisition programming. For Pablo, the teacher reported that edible or socially interactive (i.e., tickling) reinforcement was typically provided on a schedule of a variable ratio of three. For Brian, the teacher reported that he would clap for himself and look at the teacher as a means to mand that the teacher should clap for him as well, after almost every response. For Madison, the teacher reported that she would demonstrate sustained engagement in activities for approximately 15 min before accessing her leisure iPad for 5 min. These procedures were adopted for the test and intervention sessions for each participant, with the addition of praise following all correct responses. If participants manded for these or any other reinforcers in the session, they were intermittently provided with the desired item or activity immediately. Otherwise, they were told to “work first, then we can have [reinforcer].” These procedures mimicked the responses typically given during academic work sessions in the classroom.

If termination criteria were reached due to time (i.e., 15 min on one exercise), the participant was given a minimum of a 2-min break from the DI-LL session. If time in the school schedule allowed, the session was resumed after the break, but otherwise the participant returned to his or her classroom. If termination criteria were reached due to repetitions of a particular exercise, the exercise was ended, and intervention began for the next exercise. If termination criteria were reached due to challenging behavior, the session was ended, and the participant returned to class. During the next session, the exercise was restarted from the beginning for continuity.

As the participants used SGDs, rather than vocal speech, to respond, some additional modifications were made to the DI-LL procedures. For some exercises, participants had to touch one of their body parts, then tact what they were touching. As the participants could not both respond on the SGD and touch their body part simultaneously, the clinician would keep his own hand on his own body part throughout the trial as a constant model of the action. The participant could then reference the clinician as he or she scrolled and selected on the SGD. The clinician reviewed each participant’s device prior to exercises to ensure the icons required for upcoming exercises were available. As all participants had moderately robust vocabulary layouts programmed into their devices, edits were not frequently required.

Pre- and Posttests

The pretest and posttest sessions were identical, following the procedures described in the DI-LL guidebook with the modifications mentioned in the general procedures. The clinician was instructed to present each item as directed on the test form. On test items with questions (e.g., “What is your name?”), the clinician read the text and then waited up to 5 s for a response to be initiated. On test items that included listener instruction, the clinician read the instruction and then snapped as a signal to respond (e.g., “Touch your nose.” + snap). For test items requiring the participant to tact pictures, the pictures provided in the DI-LL presentation book were presented, the clinician pointed to the picture, and then read the question (e.g., point + “What is this?”). No prompts or error corrections were provided following incorrect responses, though praise and edible or interactive reinforcement were provided (Pablo only) for correct responses.

Intervention

All participants were exposed to Lessons 1–10 of the DI-LL. Intervention sessions followed procedures described in the DI-LL guidebook with the modifications described in the general procedures and a modified error correction procedure. If an incorrect response occurred or no response was initiated within 5 s, the clinician immediately prompted the correct response on the participant’s SGD. A gestural prompt was used, and if no response to the gesture occurred within 5 s, gentle physical guidance was used. After the prompted response, the clinician evaluated the response without a prompt. When an incorrect response occurred, the error correction was conducted, and then the clinician continued the exercise. When the exercise was completed, it was repeated. If an error occurred during the second attempt, the same procedures were followed, and the exercise was repeated a final time. If, on this third attempt, the participant still did not complete the exercise correctly and independently, termination criteria were met for that exercise. The clinician was instructed to present each exercise as scripted in the DI-LL guidebook, with the noted modifications to accommodate SGD use. For exercises requiring a signal, a snap instead of a point or hand-drop cue was used, as the participant could hear the signal while simultaneously looking down at the SGD. Additionally, when the script indicated the clinician was to vocally model a response, he would do so while also physically modeling the response on the participant’s SGD (e.g., selecting the icons). Of note, the clinician would clear the screen after modeling so that the participant could not refer to the product of the modeled response.

Results

Results are displayed in Fig. 1 for all participants. During the pretest, Pablo scored 20%, responding correctly when instructed to complete target actions (e.g., sit down, touch your nose) and when asked his name. In Lesson 1, Pablo did not initially complete any exercises independently on the first attempt. His performance gradually improved, with an average of 46% of exercises completed independently on the first attempt (range 0%–71%). During the posttest, Pablo scored 84% correct. During the 4-week maintenance probe, his performance decreased to 64%. During the intervention phase, approximately one additional repetition was needed per exercise across all lessons (M = 0.85, range 0.42–1.6), and termination criteria were met due to repetitions on an average of one exercise per lesson (M = 1.2, range 0–2). Pablo most frequently met termination criteria due to repetitions on exercises related to tacting body parts. On average, exercises lasted 4 min and 21 s (range 2 min, 54 s, to 8 min, 40 s). Throughout the intervention, his affect was rated as positive (M = 4.35, range 3–5). The entire intervention phase for Lessons 1–10 required 17 sessions lasting between 10 and 24 min for a duration of 2–4 days per week.

Fig. 1.

The percentage of correct responses emitted across conditions for all participants. Performance on Assessment Test 1 is shown with black squares. The percentage of exercises completed correctly on the first attempt during Lessons 1–10 are shown with gray circles

Brian’s performance was similar to Pablo’s, with scores of 28% correct on both Pretest 1 and Pretest 2. Across the intervention phase, Brian independently completed an average of 74% of exercises on the first attempt (range 40%–100%). During the posttest, Brian scored 96% correct. However, when the 4-week maintenance probe was conducted, his score decreased to 36%. Of note, Brian’s primary iPad (used during intervention) had been broken and was not available. Thus, he was using a borrowed iPad with the same WordPower 42 vocabulary but with several key icons in different locations. Given the estimated time for his primary iPad’s repair, we went forward with the maintenance check. However, these results should be taken with caution. During the intervention phase, repetitions were needed approximately once per five exercises (M = 0.2, range 0–0.5), but Brian’s responding never met termination criteria. He completed exercises faster than Pablo did—on average, 2 min and 30 s (range 1 min, 51 s, to 3 min, 30 s)—and was rated with a highly positive affect score (M = 4.94, range 4–5). The entire intervention phase for Lessons 1–10 required 11 sessions lasting between 10 and 20 min for a duration of 2–4 days per week.

Madison also emitted few correct responses during the pretests, showing no improvement over the three exposures (M = 25%, range 20%–28%). Madison independently completed a moderate proportion of exercises on the first attempt throughout intervention (M = 68%, range 42%–86%). During the posttest, Madison scored 72% correct, and during the maintenance probe, her performance decreased to 36% correct. During the intervention phase, repetitions were needed approximately once per two exercises (M = 0.5, range 0.3–0.8), and on average, termination criteria were met due to repetitions on one exercise per lesson (M = 0.9, range 0–1). Madison most frequently met termination criteria due to repetitions on exercises related to tacting body parts. On average, exercises lasted 3 min and 12 s (range 1 min, 43 s, to 3 min, 50 s), and her affect was rated as positive (M = 4.5, range 4–5). The entire intervention phase for Lessons 1–10 required 16 sessions lasting between 11 and 20 min for a duration of 2–4 days per week.

Discussion

The current study aimed to investigate the feasibility of using the DI-LL with students responding on SGDs to determine if a larger scale study could be undertaken. We evaluated feasibility based on the following aims: (a) enrollment, (b) willing participation, (c) duration and frequency of sessions, and (d) accommodations for SGD use. Regarding our first aim, the clinician (third author) identified three potential candidates for inclusion in the study, and the caregivers for all three participants provided informed consent for their participation. Thus, we successfully enrolled participants meeting our study criteria. Regarding our second aim, our results suggest that the intervention was acceptable to the participants, as they maintained a positive affect during the intervention sessions, and only one session in the study was terminated due to problematic behavior (Pablo). Regarding the feasibility of participants’ inclusion in a private school setting serving children with ASD, participants required a range of 11–17 sessions to complete Lessons 1–10. Sessions were completed in as few as 10 min, up to 24 min. These sessions were conducted in a one-on-one context, which was not the typical level of educational assistance provided to these students in their classrooms. Thus, these procedures were feasible in a research context, but the degree of feasibility in a classroom remains to be seen. Regarding our final aim, we found that with reasonable accommodations, our participants participated in Lessons 1–10, and we have some preliminary evidence of efficacy.

Results of the present study show improved performance on the DI-LL posttest for all three participants with ASD using an SGD. Pablo’s scores increased from an average of 20% to 84% correct, Brian’s scores increased from an average of 28% to 96%, and Madison’s increased from an average of 25% to 72%. This is a promising result, as this is the first application of the DI-LL curriculum with individuals with ASD who communicate primarily with an SGD. Of note, only Brian met the 90% mastery criteria determined in the DI-LL guidebook. The guidebook stipulates that exercises related to low scores on the posttest should be practiced again, and the posttest repeated. Given that this study was aimed at evaluating feasibility and preliminary efficacy, we did not follow this step as we wished to evaluate effects from only a standard “dose” of the curriculum. A larger trial will be necessary to fully evaluate the DI-LL program for this population and the participant characteristics that are predicted to align with the best outcomes. Regardless, all three participants demonstrated improved responding as a result of completing DI-LL lessons, suggesting that the curriculum can be effective with individuals who do not speak when they are provided with another communication modality.

To some extent, these findings replicate those from Shillingsburg et al. (2015), as all participants demonstrated improvement, and some participants demonstrated high levels of correct responding on the posttests. Further, as in Shillingsburg et al., maintenance for some participants was strong but diminished for others. These results are exciting, as they suggest that children with ASD who communicate primarily using an SGD may respond similarly to their vocal counterparts within the DI-LL. However, much more investigation is necessary to understand characteristics other than vocal or nonvocal communication that may be needed to benefit from this approach. The scores on the EVT and PPVT reported in Table 1 provide a preliminary glimpse at potentially necessary prerequisite language skills. A larger scale clinical trial will be needed to further understand the possible correlations between these variables and to pinpoint other characteristics of interest.

The number of repetitions per exercise remained relatively low across participants, though some patterns were observed within and across lessons. The exercises that required participants to either stand or sit and tact their body position were challenging for Pablo and Madison, resulting in frequent repetitions and some terminations. Of note, the students received discrete-trial instruction in their classrooms from a seated position. Maintaining a standing position during teaching sessions required frequent redirections, which were complicated by the students’ use of an SGD. Standing while touching one’s nose and typing on an iPad is not an easy task. Future studies should consider whether it is necessary for students that utilize SGDs to stand when responding or if the curriculum can be modified so the student can remain seated. Additionally, responses targeting body part identification required more repetitions, as the students could not both touch a body part and type on their device using their dominant hand. Even with the addition of the sustained model from the clinician, participants did not initially consistently refer to this model, which resulted in repetitions and some terminations. Future studies should consider alternate strategies for presenting these exercises—for example, potentially using picture stimuli that would be presented on a tabletop or board.

Challenges with exercises in which participants tacted body parts could also be attributed to underlying challenges with conditional discrimination. In these exercises, the student is instructed to sit or stand, then asked, “What are you doing?” Once the student responded, he or she was given the next instruction to touch a body part, then asked, “What are you doing?” Upon this second presentation of the question, the student was engaging in two actions simultaneously (e.g., either sitting or standing and touching a particular body part), but the correct response pertained to tacting the most recently emitted action (e.g., touching a body part). Thus, the recency of the emitted action (e.g., standing/sitting vs. touching a body part) functions as a conditional stimulus establishing the discriminative function of the question “What are you doing?” Emitting a conditional discrimination under control of recency may require more intensive instruction for some learners with ASD. Modifying the question to “What are you touching?” might minimize the complexity of the instructional sequence. However, once modifications to question types are permitted, this seems to be a slippery slope for clinicians and may be an impediment to achieving consistent integrity across staff members. Rather, additional practice developing conditional discriminations outside of DI-LL sessions may be warranted for participants who demonstrate this response pattern. By tracking which exercises lead to terminations due to repetitions, clinicians could identify content areas requiring more intensive practice and address these areas outside of DI-LL sessions.

The duration of the exercises was relatively brief, suggesting they could be embedded into a school day with minimal interruption to the typical schedule. The present study did not program for maintenance or generalization to the classroom. Yet, in the absence of any additional practice, all participants scored above baseline in the 4-week maintenance probe. This finding suggests some degree of durability of the intervention effects. However, with only one posttest score and one maintenance score, we must be cautious in drawing strong conclusions from these findings. Of note, the DI-LL is intended to be implemented on an ongoing basis with progression to more advanced lessons as participants demonstrate mastery of earlier lessons. Thus, interruptions of the intervention lasting multiple weeks would be unusual in real-world applications, except for school holidays. To address these limitations regarding generalization and maintenance, future studies should consider using the student’s teacher as the primary interventionist in the student’s actual classroom, following the typical classroom schedule. This would allow for ongoing data collection as the student progresses through the curriculum and for evaluation in the more relevant educational context.

The lack of maintenance may indicate flaws within the progression and mastery criteria. We required correct responses to be initiated within 5 s in alignment with prior research (Shillingsburg et al., 2015) and consideration for SGD use. However, these criteria may not have produced what Binder (1996) described as fluency: “a fluid combination of accuracy plus speed that characterizes competent performance” (p. 164). Future studies may consider decreasing the interval to initiate the response (e.g., 3 s) and including criteria for response completion. These modifications may promote greater speed of responding, but additional measures to promote competence may still be required. As noted previously, remedial practice of those skills using discrete-trial methodology may be warranted. Practice outside of DI-LL sessions would be recommended until both correct (i.e., competent) and rapid responding is observed in a discrete-trial session. Upon meeting these criteria, these skills can be practiced again in the context of the DI-LL.

There are some limitations to consider with regard to the present study. The clinician leading the sessions was an SLP with many years’ experience using and programming SGDs for students with ASD. Although it was not necessary for every session, the SLP frequently needed to add vocabulary to the students’ devices so they could complete the exercises. This task was necessary for completion of the study. To assist anyone wishing to pursue similar research in the future without this degree of support from an SGD expert, we have included an outline of all needed vocabulary icons for Lessons 1–10 (see Figure 2). Additionally, these students were selected for participation in the study based on the SLP’s familiarity with them and their educational needs. Whether these results can be obtained with broader recruitment and randomized assignments to treatment remains to be seen.

Fig. 2.

Vocabulary icons needed for lessons 1-10 of Direct Instruction Language for Learning

Generally, this study offers preliminary support for the efficacy and feasibility of the DI-LL for children with ASD who utilize SGDs. The intervention was effective for all participants, producing increased performance from baseline, though only one participant met the criteria for mastery. This participant, Brian, also had the strongest overall scores on the EVT and PPVT. More research is needed to determine what types of prerequisite skills are predictive of success with the DI-LL for children with ASD. Future studies should consider using preference assessment procedures (e.g., Hagopian et al., 2004) to evaluate whether participants would choose the DI-LL intervention over traditional forms of language instruction. This may be considered a more direct measure of acceptability (Hanley, 2010) from the perspective of the participants. We hope this study offers an avenue of investigation for clinicians wishing to utilize this technology with their learners.

Compliance With Ethical Standards

Conflict of Interest

The authors have no declared financial conflicts of interest.

Ethical Approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Informed Consent

Informed consent was obtained for all individuals in the study.