A Clinical Demonstration of Individualized Restraint Fading

Abstract

Individuals with intellectual and developmental disabilities who engage in severe problem behavior may often cause significant harm to themselves as well as their caregivers. It follows that severe problem behavior can interfere with skill acquisition and access to educational and community-based services. Behavior-analytic interventions primarily apply antecedent and consequence reinforcement-based treatments, which have substantial efficacy research supporting their application. However, at times, severe problem behavior does not respond to these approaches alone without the concurrent application of more restrictive interventions. Research featuring severe clinical cases is relatively rare. It follows that there appears to be a relative dearth of work highlighting effective and efficient fading of restrictive supports. The current report is a clinical demonstration of the systematic fading of restrictive support for a large adolescent male client who engaged in highly dangerous, pervasive problem behavior. The paper showcases creative clinical work that resulted in safely and successfully fading restraint application to zero with sustained zero application rates at a 4-month follow-up, wherein the restraint device was faded entirely from the setting. We also report participant engagement throughout the fading process (as a secondary dependent variable), something seldom done in restraint fading research. Innovative research and clinical processes are described in detail, as well as a call for more collaboration to address existing restraint–fading research gaps.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40617-025-01074-9.

A Clinical Demonstration of Individualized Systematic Restraint Fading

It is well-established that problem behavior can substantially impact the health and well-being of individuals with intellectual and developmental disabilities (Kurtz et al., 2020). The risk of harm increases substantially when problem behavior is considered severe. Occurring in up to 14% of individuals who exhibit problem behavior, problem behavior may be categorized as severe when it (a) causes more than minor injury to self and others, (b) results in the destruction of one’s environment, and (c) necessitates physical intervention at least once per week, wherein intervention requires more than one caregiver to manage the individual during an episode (Emerson et al., 2011; Poppes et al., 2010). Severe problem behavior can take many forms (i.e., topographies) and results in adverse consequences for the individual and their caregivers, including, but not limited to, (a) exclusion from community settings, (b) an inability to learn adaptive skills, and (c) an increased likelihood of experiencing treatments that are commonly associated with adverse side effects, like psychopharmacological interventions (Deb et al., 2023; Kurtz et al., 2020). This latter issue may be especially problematic because persons with intellectual and developmental disabilities may be more likely than other clinical populations to experience significant adverse side effects commonly associated with psychotropic medication use (e.g., adverse cardiac events, insomnia, erectile dysfunction, extreme weight gain; Scheifes et al., 2016).

In applied behavior analysis (ABA), antecedent and consequence reinforcement-based strategies are well established and recommended as the primary method to decrease problem behavior (Kurtz et al., 2020; Lloyd et al., 2016). However, there is evidence that intervention effectiveness for these approaches may wane when supporting clients with highly dangerous and destructive behavior (for some examples, see Foxx, 2003; Thillainathan et al., 2024) or treatment-resistant problem behavior possibly maintained by an automatic function (Gover et al., 2019; Verriden & Roscoe, 2019). In these circumstances, the ethical application of restrictive procedures that can produce an immediate decelerative effect (e.g., restraints) may be considered (Behavior Analyst Certification Board [BACB], 2020, p. 12). Restraints have been described as physically holding or securing an individual, either temporarily to interrupt or intervene in severe problem behavior or for a prolonged duration, using mechanical devices to prevent uncontrollable behavior that could cause serious harm (e.g., extreme property destruction, significant self-injury causing tissue damage/lacerations, aggression causing bodily harm; Vollmer et al., 2011).

Therapeutic Restraints

The field of ABA often defines the term punishment as a stimulus change that decreases the future frequency of a response (Fontes & Shahan, 2021; Mitteer et al., 2015). Therapeutic restraints have been described as “a strategic intervention procedure that, when implemented as one component of a comprehensive behavior support plan, is intended to have a therapeutic effect” (Luiselli, 2012, p. 243). This approach may be distinct from emergency restraints because emergency restraints are typically applied in the moment to manage an unexpected severe problem behavior outburst, particularly when there is an immediate risk of injury present (Ayvaci et al., 2025; Luiselli, 2012). Therapeutic restraints may function as a punisher if its application decreases the frequency of problem behavior in the future. Owing to a tumultuous history of misapplication and misuse, punishment is associated with significant stigma (Reed & Lovett, 2007). A widely-held stigma may promote the avoidance of therapeutic restraint application and can detract from the effective results that these procedures can have rapidly in the context of truly severe behavior (Foxx, 2003). This stigma may be detrimental to some clients, given they have the right to the most effective, least intrusive intervention (Van Houten et al., 1988). Importantly, in some cases, least intrusive may include strategies that can impart immediate decelerative effects (like therapeutic restraint) because the alternative (solely reinforcement-based approaches) may reduce problem behavior but not to zero (e.g., Functional Communication Training [FCT]; Ghaemmaghami et al., 2021). Or, reinforcement-based strategies may be ineffective because problem behavior is so sensitive to the contingencies maintaining it that it occurs before any adaptive, alternative behavior can be taught. Additionally, because some highly dangerous behaviors can be lethal (e.g., pica of glass, metal, or poisonous objects; Busch et al., 2018), they cannot be permitted to occur—even at low levels. In turn, clinicians may need to intervene to stop the behavior from occurring (i.e., applying therapeutic restraints) to circumvent these dire outcomes. In these cases, multifaceted programming comprising concurrent reinforcement and adaptive skill-building programs, paired with therapeutic restraint application, may be warranted (Foxx, 2003; Heyvaert et al., 2014; Thillainathan et al., 2024). However, behavior reduction to zero or near-zero rates does not signal the end of treatment. That is, in situations where restrictive programming helped facilitate zero problem behavior rates, this approach should then be “reduced and eliminated when possible” (Vollmer et al., 2011, p. 107).

Clinical Demonstrations & Research on Fading Intrusive Interventions

Despite the great strides made in behavior-analytic treatment, likely afforded by the development and ongoing refinement of functional analysis technology, the fact remains that restraint application is still being applied frequently across sectors (Fitton & Jones, 2020). This suggests that evaluating and disseminating restraint fading approaches may be warranted to develop a comprehensive evidence base on the topic.

Given prospective experimental research opportunities on this topic may be limited, clinical demonstrations may be especially important in working toward refining restraint fading techniques. Further, disseminating clinical demonstrations can help by showcasing clinician problem-solving approaches when target behavior does not respond favorably (e.g., initiating restraint fading coincides with problem behavior resurgence) on the first attempt. Finally, conducting applied research (e.g., clinical demonstrations) has been identified as one possible solution to addressing the “research to practice” gap (Valentino & Juanico, 2020).

Despite persistent restraint application prevalence, there are relatively few contemporary papers that have examined and reported on programming elements that may facilitate individualized mechanical restraint fading in the context of severe aggression toward others (e.g., Fisher et al., 1997). A comprehensive review of empirical research and related works (e.g., chapters) describing individualized restraint fading goes beyond the scope of the current study. However, to our knowledge, peer-reviewed mechanical restraint fading research that has concurrently reported on client engagement in academic and activities of daily living rates across restraint fading phases appears to be scarce. Further, most of the recently published mechanical restraint fading works feature participants who engaged in self-injurious behavior (e.g., Deshais et al., 2015; Morgan et al., 2017). A research dearth on restraint fading in the context of severe aggression may be problematic given up to 8.3% of individuals with intellectual and developmental disabilities engage in severe forms of aggression (e.g., aggression causing lacerations, tissue damage, and/or broken limbs; Bowring et al., 2017).

Collaborating to Disseminate Clinical Demonstration Outcomes

To shed light on socially important, relatively understudied topics, researchers may collaborate with agency partners (e.g., Thillainathan et al., 2024). For projects like the current one, this may include researchers partnering with agencies tasked with supporting high-risk clients engaging in severe problem behavior. Through these alliances, researchers may offer additional personnel (at no cost to the agency) to support treatment integrity and interobserver agreement (IOA) calculation, as well as other facets that may be too resource-intensive for the agency to enact (e.g., IOA on dependent variables). Other expertise offered by researchers may be to apply creative analytic approaches to analyze resultant data sets (e.g., daily data collected over years; Cox et al., 2022) as well as to augment research processes that may fall short of publication expectations (Critchfield & Reed, 2017).

Project Objectives

The purpose of the current project was twofold. First, to add to the limited therapeutic mechanical restraint fading literature base by outlining the approaches employed to systematically fade therapeutic restraints that had been applied to safely support a large adolescent male who engaged in highly dangerous, pervasive problem behavior. Second, to showcase creative solutions that served to augment clinical-research protocols that may not have aligned precisely with “the de facto gold standard for identifying good [emphasis added] work in applied behavior analysis” (Critchfield & Reed, 2017, p. 123).

Method

This project received clearance from the collaborating researchers’ University Ethics Board. Specifically, researchers were permitted to (a) compile existing clinical data (for secondary use), (b) conduct treatment fidelity, and (c) calculate IOA on target behaviors and treatment fidelity for the purposes of dissemination.

Participant, Setting, & Staff

Relevant Participant History

John was an autistic 15-year-old who lived in his family home with his mother, father, and younger sibling. He was 1.83 m tall and weighed approximately 113.40 kg. He took the same dosage of cannabidiol oil (i.e., CBD oil) at bedtime throughout the clinical demonstration timeline. He performed some skills at the 2-to-3-year-old level (e.g., drinking from a cup, removing clothing, selecting the correct object from an array) and other skills at the 1-to-2-year-old level (e.g., performing simple motor actions on command, stacks blocks, imitates gross motor actions). John had no vocal-verbal behavior, and he had been largely unsuccessful in gaining a large, robust functional communication repertoire despite years of teaching by highly qualified augmentative communication specialists who were also Board-Certified Behavior Analysts (BCBAs).

Owing to John’s size, strength, and the nature and pervasiveness of problem behavior, he was expelled from public school at the age of 12. As such, he attended a private school designed for students with special needs, where the current programming took place. Historically, problem behavior resulted in staff requiring emergency room medical care (e.g., suspected broken/sprained limbs) and several weeks of modified duties until physician clearance to resume full duties was received. These problem behaviors also precluded him from participating in school activities with peers, given the danger his problem behavior posed to others. At home, problem behavior of similar topographies and intensity occurred even though his parents had already (a) eliminated all instructions, (b) honored all reasonable requests, and (c) largely permitted destructive behavior to occur uninhibited (e.g., dumping all milk and cereal onto the kitchen floor) unless it became a safety risk (e.g., attempting to throw the family cat against a wall or down the stairs). Despite these adjustments, John continued to frequently engage in aggression (e.g., hitting, pushing), especially toward his younger sibling, who was the target of John’s unpredictable and dangerous problem behavior. For example, when John’s sibling was 4 years old, he pushed them down a full flight of stairs.

Over the years, a group of community-based consulting professionals (e.g., developmental pediatrician, dually certified licensed psychologist/BCBA-D, BCBA, frontline staff) comprised John’s treatment team. John had also been assigned a Ministry-provided Case Supervisor. With the developmental pediatrician on the team, potential medical factors were ruled out (BACB, 2020, p. 12). Repeated functional analyses indicated problem behavior maintained by social negative and social positive reinforcement, which resulted in a corresponding series of intervention approaches across several years (see Table 1 and Supplementary Materials for a programming timeline briefly outlining interventions attempted). As illustrated in Table 1, John’s team—collaborating with his parents and key local and federal stakeholders (BACB, 2020, p. 9)—revised and enhanced various function-based and other less restrictive interventions, both with and without the use of physical restraint (BACB, 2020, p. 12). However, these treatment plans failed to produce meaningful reductions in severe problem behavior. Fortunately, the most recent approach, a multifaceted program that included the application of therapeutic mechanical restraints, resulted in sustained, near-zero levels of problem behavior, which prompted the team to begin executing the featured restraint fading protocol.

Table 1.

Programming timeline

Year	Setting	Procedure	Outcomes
Assessment baseline	School	Task instructions, honor requests, choice, pairing	Average rate of high-risk PB was 4.1/h (range, 1.0–10.0)
1	School	FBA (including FA)	Primary behavior function was escape, with indirect measures suggesting possible social positive functions (attention, tangible)
		FCT and NCR	Average rate of high-risk PB was 8.9/h (range, 0.3–35)
	Home	FBA (including FA)	Behavior function was multiply controlled escape and attention
		FCT and NCR	Program terminated due to high-risk PB during morning routine interfering with FCT trials
		Zero baseline	Increase in interfering idiosyncratic repetitive behavior
	School	Zero baseline	Average rate of high-risk PB was 14.7/h (range, 5.6–40.9)
		Direct observation	Hypothesis was action of hitting instructor may be reinforcing
		Response-to-behavior protocol adjustment	Average rate of high-risk PB was 1.9/h (range, 0–7.5). Acquisition or learning tasks could not be reliably nor safely presented. High-risk PB was 7.5/h
2	School	Indirect assessment and direct observation	Increase in repetitive behavior. Average rate of high-risk PB increased to 5.7/h (range, 0–23.0). Average instruction rate was 11.0/h (0–25.0)
		Multi-faceted behavioral program	Average rate of high-risk PB was 2.4 (range, 0.2–8.0). Active engagement rate was 256 min/day (range, 177–402)
	School (post-COVID-19 closure)	Multi-faceted behavioral program with increased staff: student ratio	Escape-extinction element of program became difficult to safely implement with physical containment due to John’s increased size and strength
3	School	Zero baseline	Average rate of high-risk PB was 2.0/h (range, 0.2–2.4). Average instruction rate was 11.0/h (0–26.0)

Please see Supplementary Materials for detailed descriptions of procedures and outcomes. COVID-19 coronavirus disease 2019; FA functional analysis; FBA functional behavior assessment; FCT functional communication training; NCR noncontingent reinforcement; PB problem behavior

Staff

All programming and corresponding restraint fading were supervised by a senior BCBA and consulting BCBA-D. The senior BCBA had been certified for 2 years, was a certified special education teacher, and had been working with autistic persons for 20 years. The consulting BCBA-D had been certified for almost 10 years and had developed extensive clinical and research expertise over the years on this and related topics (e.g., Cox & Virues-Ortega, 2022; Cox et al., 2022; Thillainathan et al., 2024).

John’s frontline team comprised eight therapists who rotated through all therapist roles. All team members received pre-service training in physical management (i.e., Safety Care Crisis Prevention Training; https://qbs.com/safety-care-crisis-prevention-training/), ABA skill acquisition, and behavior management techniques. The BCBA and BCBA-D, who supervised John’s case, provided ongoing training via behavior skills training (Parsons et al., 2012), including regular treatment integrity checks (see Method section below). Eighty-eight percent of the frontline team were educated at the college or university level (13% at the master’s level, 63% at the bachelor’s, and 13% had obtained a college diploma). Most staff were female (81%). Staff’s previous experience in the field ranged from 0–5 years. John’s team changed over the fading period as required (e.g., staff turnover).

Setting & Protective Equipment

By the time the team initiated restraint fading, John had learned to accurately request primary reinforcers (i.e., food items), label known locations and people, and answer some personal information questions (e.g., “Where do you live?” and “What is your sibling’s name?” using a speech-generating device [SGD]; i.e., Proloquo2Go on an iPad). The restraint fading phases featured in the current report were enacted with a 2:1 frontline staffing complement and took place in John’s treatment room, as well as different areas across the center as dictated by fading phases (described in detail below). The treatment room was approximately 5.28 m with a long 1.22 m × 0.61 m wooden table that John used as a worktable. He had a chair (either the containment chair or a crate chair) to sit on while working, while the instructor sat across from him in a desk chair, and the physical prompter sat approximately 0.91 m to his side when prompting was not required. All programming materials were neatly organized in bins (without lids) for the team to easily access during teaching.

The frontline team and any clinical supervisors interacting with John were required to wear personal protective equipment (see Supplementary Materials for equipment information).

Response Measurement, Interobserver Agreement, and Treatment Integrity

To ensure restraint fading protocols did not interfere with programming success and/or compromise client and staff safety, the team recorded problem behavior and engagement across restraint fading phases.

Response Measurement and Interobserver Agreement

The primary dependent variable was problem behavior, which consisted of aggression and property destruction. Aggression was operationally defined as hitting, kicking, or pushing others, as well as forcibly throwing objects at others. Property destruction was operationally defined as throwing objects, pushing objects forcibly, hitting or kicking objects. During sessions, John’s instructors recorded instances of aggression and property destruction on separate mechanical clicker devices. These frequencies were summed to generate total problem behavior instances and converted to rate (responses per hour). Total problem behavior rates informed restraint fading backstep and mastery criteria and were established before each phase was enacted in accordance with cumulated data trends (see Fading Phases described below for criteria specifics). Data were graphed daily and informed the clinical supervisors’ (BCBA and BCBA-D) intervention decisions (BACB, 2020, p. 12).

The secondary dependent variable was engagement. This consisted of the total number of successful token boards completed. Tokens were awarded contingent on John independently (or prompted to support skill acquisition, where appropriate) engaging in a wide variety of academic tasks, activities of daily living, and leisure activities. These tasks included those that were completed while sitting (e.g., putting together Mechano®, building Lego from the instruction booklet), also called seated tasks, as well as those completed while standing (e.g., sorting clothing, sweeping, loading the dishwasher), also called standing tasks. Specifically, tokens were delivered on a variable ratio 3 schedule (VR-3). As such, the number of tasks John completed was calculated by multiplying the number of tokens delivered over the course of the session by three. This value was then converted to rate (engagement per hour).

We calculated IOA for problem behavior and engagement for 25% of sessions across 45% of the featured phases. That is, the team calculated IOA across phases B, Ai, D, and Ei. We used Google Randomizer to randomly select which video recordings the research assistant coded. The team employed total count IOA, wherein the total number of responses recorded was independently tallied, and the smaller total was divided by the larger total, multiplied by 100. Percentage agreement averaged 92% (range 44–100%) for problem behavior and 99% for engagement (range 83–100%).

Treatment Integrity and Interobserver Agreement

Treatment integrity data were collected by the onsite clinical supervisor, an additional instructor (when available), or by trained research assistants via video recordings. Critical treatment integrity components included: (a) the instructors’ management of teaching trials (e.g., prompt levels, proportion of seated and standing tasks), (b) restraint chair application (e.g., application contingent on target behavior, correct number and order of strap application, release procedure, break procedure), and (c) application of Safety Care approaches (Interested readers may acquire treatment integrity forms by emailing the corresponding author). We revised integrity forms to reflect restraint fading adjustments accordingly. Because John was in session five days per week for 6 h per day, integrity data were collected one time across one of the five-session days each week. This resulted in approximately 20% integrity data collection (e.g., 1 out of 5 days). On average, instructors implemented restraint fading elements with 88% accuracy (range 74–100%).

The team calculated IOA for 30% of treatment integrity data. The clinical supervisor or an additional instructor coded live sessions for brief session periods while trained, independent raters (i.e., research assistants), watched video recordings of treatment sessions (i.e., when John entered the treatment room to when John left the treatment room for the day) and recorded probe data across 6 h sessions. That is, each item on the checklist was scored as either a correct or incorrect application in accordance with the first response opportunity of the day. Agreement was defined as both raters scoring the correct or incorrect application across checklist items. The team enacted trial–by–trial IOA, which was computed by dividing the number of item agreements by the number of item agreements plus item disagreements, multiplied by 100. On average, IOA was 93% (range 75–100%).

Clinical Procedures

General Clinical Procedures

Several practices were held constant across all clinical procedures described in this manuscript, including those programming elements described in Tables 1–3 and Supplementary Materials. These practices included (a) ongoing stakeholder meetings, (b) comprehensive assessments, (c) ongoing health and well-being monitoring by the clinical team (e.g., frequent body check forms completed by the clinical team), (d) repeated medical appointments with consulting developmental pediatrician, and (e) daily data collection and adherence to legal and ethical standards (BACB, 2020, p. 9, 12; Vollmer et al., 2011).

Table 3.

Description of phase E

Subtle fading step	Lap belt fading		Generalization
	Straps per board	% TB	Tasks	% TB
1	Lap + shoulder	30	2 standing tasks in classroom and 1 standing task in hall (just outside room)	100
	Lap (fully applied)	30
	Lap (1-inch slack)	40
2	Lap + shoulder	20
	Lap (fully applied)	30
	Lap (1-inch slack)	30
	Lap (2-inch slack)	20
3	Lap + shoulder	10
	Lap (fully applied)	20
	Lap (1-inch slack)	30
	Lap (2-inch slack)	20
	Lap (4-inch slack)	20
4	Lap + shoulder	*0
	Lap (fully applied)	10
	Lap (1-inch slack)	30
	Lap (2-inch slack)	20
	Lap (4-inch slack)	20
	Lap (8-inch slack)	20
5	Lap (fully applied)	*0
	Lap (1-inch slack)	20
	Lap (2-inch slack)	20
	Lap (4-inch slack)	20
	Lap (8-inch slack)	20
	Lap belt over/not attached	20
6	Lap (1-inch slack)	10
	Lap (2-inch slack)	10
	Lap (4-inch slack)	20
	Lap (8-inch slack)	20
	Lap belt over/not attached	20
	Crate chair	20
7	Lap (1-inch slack)	10	2 tasks in classroom and 1 task in hall (just outside room) 2 tasks in classroom and 1 task in another classroom	80 20
	Lap (2-inch slack)	10
	Lap (4-inch slack)	10
	Lap (8-inch slack)	20
	Lap belt over/not attached	20
	Crate chair	30
8	Lap (1-inch slack)	10
	Lap (2-inch slack)	10
	Lap (4-inch slack)	10
	Lap (8-inch slack)	20
	Lap belt over/not attached	10
	Crate chair	30
	Crate chair in another classroom	10
…	-	-	-
18	Lap belt over/not attached	20	2 tasks in classroom and 1 task in hall (just outside room)	20
	Crate chair	30	2 tasks in classroom and 1 task in another classroom 2 tasks in room and 1 task outside	20 40
	Crate chair in another classroom	40	2 tasks in room and 1 in kitchen	20

TB token board. The ellipses signify steps 9–17. These steps each reduced a level of strap application in a manner similar to steps 1–8. Interested readers may email the corresponding author for the full descriptions of the steps

Programming Details Preceding Phase A

Importantly, our primary intention was to disseminate the restraint fading iterations described below, which culminated in safely removing the restraint device from the setting entirely without compromising John’s active engagement and staff safety. However, we felt it may also be prudent to briefly describe the development and evaluation of intervention packages that immediately preceded phase A. So, in addition to Table 1 and Supplementary Materials briefly outlining programming over the years, we felt readers may benefit from the provision of additional information as it relates to intervention iterations that were required to arrive at Phase A: Pre-restraint Fading. Interested readers may contact the first author to obtain Figures and corresponding Results associated with these programming aspects.

Arriving at phase A took time. Initially, protocols preceding phase A comprised tasks, token board and associated response cost, break delivery, and mechanical restraint application, where applicable. During this time, we executed an A-B-A-B design to examine which release criterion (fixed-time release versus behavior-contingent release; Luiselli et al., 2006) may coincide with the shortest restraint application duration and/or fewest restraint applications. That is, we attempted to leverage the release criterion to help us minimize how long and/or often restraint was applied. Unfortunately, across this assessment period, we observed an upward trend in problem behavior, with the final phase coinciding with high rates of problem behavior per hour (M = 18.18; range 13.32–22.93) paired with moderate, variable engagement per hour (M = 56.23; range 36.65–95.0). Given how dangerous John’s problem behavior was, the team (with the support of John’s parents) proceeded with completing a punisher assessment (as per Verriden & Roscoe, 2019). This was done to determine whether contingent positive punishment might be an appropriate addition to the treatment package. The assessment was conducted over three days, during which we compared the effectiveness of three caregiver-informed non-preferred edible stimuli presented contingently on a fixed-ratio 1 schedule (FR 1). Although the separate presentation of all three stimulus variations coincided with reduced problem behavior rates compared to the preceding condition wherein FR 1 contingent positive punishment was absent, the stimulus associated with lowest problem behavior rates (M = 5.91, range 0–7.50) and consequently less time spent in restraint, was the application of a single droplet of an edible bitter solution (commonly used and available over the counter in any pharmacy) to John’s lower lip paired with a neutral, corrective statement. Upon adding this program element (on an FR 1 schedule), the team observed an immediate decrease in problem behavior, which also meant that the duration of mechanical restraints application was much reduced. That is, John was quickly able to meet antecedent application release criteria (see below). Within three sessions of this addition, the team was able to reintroduce acquisition tasks without corresponding increases in high-risk problem behavior—something that they had not been able to do reliably for ~ 18 months. Further, within 15 sessions of this addition, the team made its first attempt to fade restraints (see phase B below). Of note, during phase A (described below), John only contacted the contingent positive punishment element during six out of 21 sessions (29%; see Fig. 1, phase A). In four out of the six contacts, he only experienced the stimulus once per 6-h session. Overall, only 14 drops were applied across 126 session hours (0.11 drops/hour) during phase A. The team monitored John for any emotional and physical side effects following contingent positive punishment application. No concerns were documented, and John’s independent engagement rate remained high (M = 82.63; range 64.29–95.02).

Fig. 1.

Rate of problem behavior per hour and engagement per hour. Note. This graph depicts the rate of problem behavior per hour on the left y-axis and the rate of engagement per hour on the right y-axis across restraint fading sessions on the x-axis. The closed circles represent the rate of problem behavior per hour, and the open circles represent the engagement rate per hour. The dotted vertical lines signify phase changes from phases A–F and follow-up

Phase A: Program Status Prior to Restraint Fading (i.e., Pre-restraint Fading Phase)

During this phase, the team implemented a function-based multifaceted program that resulted in near-zero rates of problem behavior. This program included tasks, a token board and associated response cost, break delivery, contingent positive punishment, and therapeutic mechanical restraint application, as required.

Antecedent Application

During phase A, John began his day (i.e., session) with a noncontingent restraint application because of his size and previous aggressive incidents that resulted in staff emergency room visits and modified duties due to injuries sustained. It was also hypothesized that this noncontingent restraint application would support John in starting his day successfully. The noncontingent restraint application consisted of securing a lap belt, shoulder strap, and arm strap, with one arm unrestrained for task completion. After three consecutive successful token boards, the team removed the other arm strap. After the fourth successful token board, only the lap belt was applied. When John completed the fifth token board, he earned his first break. Notably, a token board interval involved 10 tokens, wherein the lead instructor placed a series of known (mastered) and acquisition tasks across learning domains (e.g., expressive, receptive, leisure, household chores), delivering one token on a VR 3 schedule. Upon earning the seventh token (out of a possible 10), straps were removed, and John participated in a series of “standing tasks” to earn his final three tokens (e.g., cleaning windows, wiping down tables, sweeping, changing the garbage across the center). A break constituted removing all limb straps and free access to preferred items for the break duration. For all subsequent token boards (six through 10), the team only applied the lap belt, and a break was provided after each successful token board unless problem behavior occurred. If problem behavior occurred, the team reapplied restraint straps (i.e., contingent restraint application), and criteria were reset (i.e., return to the first step of the day’s session). Table 2 outlines the steps described above.

Table 2.

Phases A, B, C and D descriptions

Phase A: Antecedent application
	No. of TB	No. of straps	Upon successful TB
Beginning of day	First three TB	3	Continue to next TB
	Fourth TB	2	Continue to next TB
	Fifth TB	1	John earned a break
Rest of the day, unless PB occurred	Single TB	1	John earned a break
If PB occurred	Reset to beginning of day steps
Phase B: Initial restraint fading
	Restraint fading
	Step	No. of TB	No. of straps	Mastery criteria
Beginning of the day (initially – then decreased by 1 each time mastery criteria was met)	1	4	3	PB < 1 per hour, two consecutive days
	2	3	3
	3	2	2
	4	1	1
	5	1	Less secure lap belt
	6	1	No lap belt
After successful TB		Continue to next TB until break is earned	Reduced by 1
Rest of the day, unless PB occurred		Single TB	1
If PB occurred		Reset to beginning of day steps
Phase C: Indiscriminate restraint fading across days
	Restraint fading
	Step	No. of TB	No. of straps	Mastery criteria
Beginning of day	1	4 or 5	3	PB < 1 per hour, over four consecutive days and no more than two consecutive days > 1 per hour in last 10 days
	2	3 or 4	3
	3	2 or 3	2 or 3
	4	1 or 2	1 or 2
	5	1	1 or loosened strap
	6	1	Loosened strap or no strap
After successful TB		Continue to next TB until break is earned	Reduced by 1
Rest of the day, unless PB occurred		Single TB	1 or loosened strap or no strap
If PB occurred		Reset to beginning of day steps (or at minimum 3 TBs, 3 straps)
Phase D: Indiscriminate of the lap belt across days
	Lap belt application
	Step	Application	Mastery criteria
Beginning of day: A single TB was required to earn a break	1	Full	PB < 1 per hour, over four consecutive days and no more than two consecutive days > 1 per hour in last 10 days
Phase D: Indiscriminate of the lap belt across days
	Lap belt application
	Step	Application	Mastery criteria
Beginning of day: A single TB was required to earn a break	2	1-inch slack	PB < 1 per hour, over four consecutive days and no more than two consecutive days > 1 per hour in last 10 days
	3	2-inch slack
	4	4-inch slack
	5	Attached but not tightened (approximately 8-inch slack)
	6	Placed over, not attached
If PB occurred		Reset to minimum 3 TBs, 3 straps

TB token board. PB problem behavior

Restraint Fading

The multifaceted program elements that had produced the near zero rate of problem behavior (see phase A) were ongoing while the team faded restraints. Specifically, tasks, a token board and associated response cost, break delivery, contingent positive punishment, and therapeutic mechanical restraint application were applied across phases B, C, D, and E as required (see below). Importantly, token boards and strap applications were varied in accordance with fading phase protocols. Moreover, phase F depicts contingent positive punishment reversal, and as such, this programming element was absent from this phase. In total, the team enacted four unique fading protocols to eliminate restraint application (phases B, C, D, and E). Table 2 summarizes the relevant fading information across phases A, B, C, and D. Table 3 summarizes relevant fading information across phase E.

Phase B: Antecedent Application Therapeutic Restraint Fading

 The initial restraint fading phase reduced the number of token boards required for John to earn his first break by one each time he met the mastery criteria. Liberal mastery criteria were set at two consecutive days with less than one instance of problem behavior per hour. Once he required only two token boards to earn his first break of the day, the number of secured straps John started his day in was reduced by one. Thus, a set of two consecutive days with less than one instance of problem behavior per hour resulted in one less token board completion and one less secured strap applied at the beginning of the day. This continued until John began starting his day with the lap belt not applied. If the team observed one of the following, the revision criterion would have been considered fulfilled: (1) an increasing trend over four days in problem behavior rate or (2) no decrease in problem behavior rate from week 1 to week 6 from when a program phase was initiated. In response to John fulfilling the revision criterion, we discontinued phase B and reinstated phase A.

Phase C: Antecedent Application with Indiscriminate Therapeutic Restraint Fading Across Days

During this phase, the team applied balanced randomization by using the Excel RAND function to sort the options across days to determine the number of straps John began his day in and the number of successful token boards required to earn the first break of the day. For example, during the first step of fading in this phase, John was expected to complete either four or five token boards with the shoulder strap and lap belt secured. When the revised mastery criteria (see below) were met, the required number of token boards to earn the first break was reduced to either three or four, with two or three straps applied at the beginning of the day in accordance with the random schedule. Subsequent steps were faded in the same manner until John began intermittently starting his day with no straps applied.

Revised Mastery Criteria

John’s data revealed the average number of consecutive days the team observed less than one instance of problem behavior per hour was 3.4 days. Therefore, one element of the revised mastery criteria was that John engage in less than one instance of problem behavior per hour across four consecutive days. A second element was also added. Specifically, in the last 10 days, there could not be two or more consecutive days with more than one instance of problem behavior per hour. If either of these criteria were unmet, then John would not progress to the next fading step.

Phase D: Antecedent Application with Indiscriminate Fading of the Lap Belt Across Days

 In response to fulfilling the revision criterion, fading was revised again. However, mastery and revision criteria were unchanged from phase C. Phase D was conducted in the same manner as phase C, except lap belt tightness was adjusted, and completion of only one token board in the absence of problem behavior was required to earn a break. Specifically, at the start of each day, the lap belt was applied with varying degrees of tension.

Phase E: Antecedent Application with Indiscriminate Fading of Straps including Lap Belt Slack Across Token Boards

 In response to fulfilling the revision criterion, fading was revised. This phase was conducted in the same manner as phase D, except balanced randomization informed the application of straps across token boards rather than across days. That is, at the start of each token board interval (i.e., work interval), the shoulder belt and lap belt were applied, or just the lap belt with varying degrees of tension. Further, phase E coincided with the addition of generalization steps. That is, John began to complete some token boards in a crate chair instead of the restraint chair. Some token board intervals were carried out in an adjacent classroom. Standing tasks, which were completed at the end of the token board interval, were sometimes delivered in other locations around the school. These additional steps were implemented with the same balanced randomization and subtlety as the lap belt fading.

Phase F: Drop out Component Analysis plus Restraint Fading Paused

 For this phase, the team paused fading at the current step (phase E; step 2, see Table 3) and withdrew the contingent positive punishment element of the multifaceted program. This was done to assess whether this programming element was still producing the desired outcome (i.e., problem behavior suppression).

Return to Phase E: Reinstating Antecedent Application with Indiscriminate Fading of the Lap Belt Across Token Boards

 In phase F, problem behavior rates increased immediately. Thus, the team reinstated phase E, but upon reinstating this phase, the team reset John to step 1. John progressed through the rest of the fading steps (up to step 18), at which point data collection for the study was discontinued.

Follow-Up

Procedures during this phase included tasks and corresponding token board and break delivery (i.e., one token board equals one break) with an emphasis on skill acquisition and generalization (e.g., locations, staff). John was completing programming around the center and had been re-integrated into shared classroom spaces. Contingent positive punishment and therapeutic restraint application were no longer programming elements. Importantly, the restraint device was no longer in the setting. Follow-up data were provided by the clinical data staff, who continued to collect it to monitor John’s progress. The data sample featured in this phase (Fig. 1) were collected four months after the last data point in return to phase E.

Results

Problem Behavior Rates Across Restraint Fading Phases

Figure 1 depicts the problem behavior rate (per hour) and engagement rate (per hour) across the last 20 sessions of the multifaceted treatment program (phase A) and all subsequent fading phases (phases B, C, D, E, F, and follow-up). These data span 251 full-day sessions across approximately 14 consecutive months of full-time school attendance (6 h per day).1 The team’s first attempt to fade restraints (phase B) appeared to be successful until an upward trend in problem behavior became apparent starting at session 164. This session coincided with the first day of step 6 (i.e., no straps applied; see Table 2). At session 170, problem behavior rates peaked at 8.0 instances per hour and remained quite variable until restraint fading was discontinued after session 183. That is, the mean problem behavior rate across the last 12 sessions in this phase was 3.13 (range 0–5.0), compared to a mean rate of 0 across the first 12 sessions in this phase. Response patterns thus suggested that sustained near-zero rates (as those observed earlier in the phases) could not be re-established with the current fading approach in place.

A return to phase A coincided with a modest, steady declining trend that was observed immediately across sessions 184 to 202. At this point, near-zero rates were re-established and largely persisted across the remaining sessions in this phase. John’s performance across the final 11 session days met mastery criteria (M = 0.26, range 0.0–0.8). At this point, the team initiated a revised therapeutic restraint fading protocol (i.e., phase C). This indiscriminate fading approach was informed by a hypothesis established by John’s history of aggression founded in an insistence on prescribed conditions (i.e., sameness, rigidity). We felt John might have been experiencing salient fading changes (e.g., starting his day without restraints applied) as less preferred, and as a result engaged in a higher rate of problem behavior (see Discussion for details). The team’s second attempt to fade therapeutic restraints (phase C) appeared relatively promising at first. However, sustained near-zero rates did not persist, and session 273 marked the beginning of a relatively steep accelerating problem behavior trend. As with phase B, this accelerating trend coincided with the introduction of step 6 (see Table 2); that is, shifting from starting his day with “one or loosened strap” to “loosened strap or no strap.” In response, phase D was enacted. As in earlier fading attempts, the shift appeared to coincide with zero to near-zero problem behavior rates across the first 13 sessions. However, problem behavior increased across the phase, with moderate variability observed beyond session 294. Unlike phases B and C, this upward trend did not coincide with the introduction of no lap belt but instead coincided with a shift from step 2 to step 3. The mean problem behavior rate across the last 12 sessions in the phase was 0.96 (range 0–2.13) compared to 0.21 (range 0–0.73) across the first 12 in this phase.

A shift to phase E coincided with comparatively low rates of problem behavior for the first 13 sessions of this phase. Specifically, mean problem behavior rates were lower across the first 13 sessions of phase E at 0.60 (range 0–1.5) compared to mean problem behavior rates across the final 13 sessions in phase D (M = 1.1; range 0–2.13). However, a modest increasing trend began to emerge beyond session 316, with the problem behavior rate peaking at 4.76 instances per hour.

In reflecting on past data trends, it seemed that returning to phase A did not appear to facilitate sustained fading progress in subsequent phases (i.e., phase C). Moreover, shifting immediately from one fading approach (phase C) to another (phase D) did not coincide with re-instating near-zero rates of problem behavior. Thus, the team paused at fading step 2 and withdrew the contingent punishment programming element (i.e., phase F). Removing this programming element coincided with a marked increase in problem behavior rates, peaking at over 15 instances per hour. Reinstating phase E coincided with an immediate level change in problem behavior rates, followed by a modest declining trend across the first three sessions in the condition. Problem behavior rates of zero largely persisted after the sixth session in this condition. Follow-up data collected over four months later (87 sessions) showed problem behavior had maintained at zero. Of note was that the team was able to fade the restraint device entirely from the setting.

Engagement across Restraint Fading Phases

Phase A depicts participant engagement across the final 20 sessions of the multifaceted program just prior to initiating restraint fading. Perhaps not surprisingly, the team largely observed an inverse relationship between problem behavior and engagement. That is, when problem behavior rates were higher, engagement was lower; thus, there was modest to large variability in engagement rates across phases B, C, and D. Importantly, mean engagement was lowest during phase F (M = 13.79; range 0–48.01) while engagement in the return to phase E was much higher and more stable (M = 57.58; range 37.2–72.0). Average engagement at follow-up (M = 47.00) was slightly lower but less variable (range 45.05–48.00) compared to engagement across the preceding phase E (M = 57.58, range 37.20–72.00). This may not have been unexpected given that the tasks John completed in follow-up may be described as more complex and, therefore, took longer to complete (e.g., chores vs. simple matching tasks). Despite some variability across fading phases (phases B, C, D), outcomes suggest John continued to participate throughout sessions even when problem behavior rates were higher. Finally, acquisition tasks mastered per session in the absence of fading (phases A and Aⁱ) was approximately 1.4. Tasks mastered per session across fading phases was approximately 1.6. This could suggest that despite restraint fading revisions, active client engagement may not have been negatively affected. John continued to engage in academic and daily living tasks independently even as the team worked toward fading restraints.

Discussion

In this paper, we described a large, high-needs adolescent who caused frequent injuries to others, had been isolated from peers, and experienced many physical holds and ineffective interventions to manage the severe problem behavior he engaged in. Some restrictive practices, despite seeming intrusive, were thoughtfully applied to eliminate severe problem behavior and then faded entirely from the intervention plan. All the while, John continued to build on the gains he had made. These gains included active participation in learning activities and fulsome reintegration into shared classroom spaces in his school, wherein he only required 1:1 support to help with ongoing skill building, as well as to successfully navigate his classroom and self-care needs (e.g., changing his brief). Moreover, reports from caregivers indicate that high-risk problem behavior rates remained at zero, and restricted practices were never required again, even years later. Although rarely discussed and seldomly researched, restrictive procedures can be important elements of a treatment plan when conducted by qualified, experienced behavior analysts (Vollmer et al., 2011). Moving forward, there were several other noteworthy outcomes. First, it appeared the salient fading (phases B, C, D) may have alerted John to intensive programming changes (i.e., restraint fading), which coincided with problem behavior rate increases. That is, for two of the fading phases (phases B and C), problem behavior rates began to increase when the team tried to introduce “no strap” sessions at the beginning of his day. Notably, contingent application elements (i.e., reapplication contingent on problem behavior) remained consistent and discernable, as with the reset criteria. The way he started his day and proceeded through fading steps (see Table 3) may be unique in that approaches to restraint fading suggest teaching clients clear requirements for meeting restraint removal criteria. In the current case, given John’s history of aggression founded in an insistence on prescribed conditions (i.e., sameness, rigidity), the team hypothesized that he experienced a salient fading change (e.g., starting his day without restraints applied) as less preferred and as we observed increased problem behavior rates. It is possible that part of the reason the return to phase E was successful is that he was less able to distinguish the noncontingent restraint application changes easily and, therefore, did not respond to reinstitute sameness (i.e., resist progressing through fading steps). Given that rigidity may be a common characteristic observed in autistic persons (Weyman et al., 2022), clinicians struggling to fade restraints may consider revising fading steps so that they are less salient to the client. It may be that for some clients, salient fading steps may be part of the reason past restraint fading efforts have failed.

An alternative explanation for the current outcomes may be that the team observed treatment relapse, specifically renewal across phases B, C, and D (Wathen & Podlesnik, 2018). Renewal may be described as behavior that has been reduced in one context and re-emerges in another because of a context change (Podlesnik et al., 2017). In the literature on this phenomenon, context may describe a variety of stimuli conditions, including physical stimuli. It is possible that because of how different John’s sessions started out (i.e., out of therapeutic restraints versus in therapeutic restraints) across steps in phases B, C, and D, this was enough to trigger renewal. By contrast, the subtle changes across token boards associated with phase Eⁱ steps could have loosely resembled context fading. That is, the initial start of his token boards was similar. In other words, progressing to the next fading step looked very similar to the way sessions played out in earlier fading steps.

Admittedly, phase E did not coincide with sustained low problem behavior rates and corresponding progression through fading steps. However, John’s engagement outcomes may offer some insight. Resurgence has been defined as “the reappearance of an extinguished behavior when an alternative behavior reinforced during extinction is subsequently placed on extinction” (Shahan & Sweeney, 2011, p. 91). Some have hypothesized that skill repertoire size could offset resurgence (Fuhrman et al., 2022). As applied to the current case, John’s engagement may have facilitated learning an impressive array of new skills across restraint fading phases. This larger repertoire may have meant that by phase Eⁱ, he had established a larger collection of skills to draw from, thus offsetting resurgence. Presently, research outcomes on variables mitigating resurgence are mixed. We encourage researchers to consider reporting on overall skill acquisition rates in the context of restraint fading. This may be helpful because it could provide additional commentary on the occurrence of resurgence in clinical settings supporting truly severe cases while also demonstrating that skill acquisition targets need not be compromised in the interest of restraint fading.

Study Strengths & Limitations

For comprehensive commentary on this project, it may be prudent to draw attention to some project strengths and limitations. First, it may be important to draw attention to the resources and time associated with the current program versus the preceding programming. That is, Table 1 and corresponding Supplementary Materials offer a brief description of programming, which suggests the team had been working diligently without success for years with highly skilled oversight and up to a 3:1 staff ratio. Further, new and different trigger scenarios were emerging. It was also becoming apparent that there was no viable pathway to initiate fading 3:1 staffing ratio and reintroduce John back into shared classroom spaces—let alone his community through less restrictive approaches. Moreover, the Relevant Participant History section suggests these issues were long-standing and had not ever been effectively treated. By contrast, within one year of implementing the multifaceted behavioral program (phase A), the team was able to reinstate learning opportunities (as depicted by engagement), reintegrate John into shared classroom spaces in the school, fade staffing ratios from 3:1 to 1:1, and fade the restraint device entirely from the setting. Although the latter was resource-heavy initially, it appears to have created a viable pathway toward an effective treatment with a sustained impact (as evidenced by the follow-up data; BACB, 2020, p. 10).

One project limitation was that the dependent variable IOA and treatment fidelity IOA were not conducted across all featured fading phases. Regardless, we were able to collect adequate IOA and treatment fidelity samples across all other phases alongside. This, paired with the fact that our IOA percentages were adequate, suggests our results and associated interpretations may be viable. Another limitation was that the current work features only one participant, and inter-participant replication is absent. However, our primary intent was a clinical demonstration, which we felt was a reasonable way to disseminate outcomes in the context of a difficult-to-study phenomenon (mechanical restraint fading). Further, homogeneity among this clinical population may be rare. As a result, smaller agencies may be unlikely to support more than one severe complex case that presents similarly at the same time point. This makes inter-participant replication difficult, even with the flexibility of a nonconcurrent multiple baseline design. Consecutive controlled cases series may also be difficult to enact, given several years may need to elapse before clinicians compile enough case exemplars to produce a fulsome report that adheres to this design’s requirements (Hagopian, 2020). Despite this, the team was able to retain some internal validity by showing intra-participant reversals where clinically appropriate. Another data-related limitation is that details on staff injury rates across the years of supporting John were not retained in the clinical file, so we could not compare these rates across intervention approaches (e.g., treatment packages with vs. without restrictive programming elements). We encourage others tasked with supporting high-needs clients working on fading restrictive supports to consider including this variable should they elect to disseminate clinical case outcomes, as in the current paper. Finally, the University-affiliated researchers did not have ethics clearance to collect social validity data from stakeholders. Anecdotally, the team had been approached by staff, parents, and other stakeholders (e.g., allied health professionals), expressing how meaningful the changes they observed were. Frontline staff frequently shared their positive impressions of how well the entire program went and how happy they were to have been able to support John throughout. That said, a recent review by Leif et al. (2024) indicated that only 17.6% of papers published in the Journal of Applied Behavior Analysis include a measure of social validity. Thus, future research across the field should endeavor to address this.

Conclusions

Current demographic research suggests restraints are still a relatively common occurrence across settings (Fitton & Jones, 2020). Therefore, BCBAs may find themselves tasked with restraint fading (e.g., Fisher et al., 1997; Petursson & Eldevik, 2019). It is also possible that they may find themselves in critical situations where they have identified the need for restrictive interventions to prevent death, severe injury, or other dire outcomes (e.g., homelessness due to repeat evictions; Newcomb & Hagopian, 2018; Thillainathan et al., 2024). Regardless of the situation, if restraints have been part of a client’s program for an extended interval, it is ill-advised that they simply discontinue them and hope for the best. Instead, a precise, evidence-informed process may be preferable. The featured case was resource-intensive and required the highest level of clinical expertise (i.e., boundaries of competence; BACB 2020, p. 9) and commitment across stakeholders, a situation not all providers may find themselves in. Regardless, we hope it served to exemplify how researchers and clinicians may work together to circumvent the barriers that stand in the way of clinicians disseminating their outcomes to the field (see Critchfield & Reed, 2017) and how to systematically innovate for the purpose of fading mechanical restraint.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

All authors contributed to the study in various ways, including material preparation, data collection, and data analysis. The first draft of the manuscript was written by Alison D. Cox and Angeline Savard, while Asude Ayvaci and Thurka Thillainathan supported manuscript development (e.g., APA formatting, developing Tables and Figures).

Funding

This project was not funded.

Data Availability

Interested readers may contact the corresponding author to access the raw data (numerical) coinciding with Fig. 1 of the manuscript.

Declarations

Ethical Approval

This project received clearance through Brock University’s Research Ethics Board.

Consent to Participate

The participant’s parents provided written informed consent on his behalf and were informed of the right to withdraw at any time without penalty.

Approval Committee

This project received clearance through Brock University’s Research Ethics Board.

Competing interests

The first author served as a consultant to the collaborating agency, the second and fifth authors were clinical team members on the case. The remaining authors have no conflicts of interest to declare.