On the scope and characteristics of relapse when treating severe destructive behavior

Abstract

Prior studies on treatment relapse have typically examined the prevalence of resurgence or renewal of target behavior (e.g., destructive behavior) in isolation. This study analyzed both types of relapse during 25 consecutive treatments involving functional communication training during worsening reinforcement conditions for alternative behavior (i.e., schedule thinning) or following context changes. We also examined disruption of alternative behavior (i.e., functional communication requests, compliance). Resurgence and renewal of destructive behavior occurred in 76% and 69% of treatments, respectively, and in approximately a third of changes in reinforcement or context. Relapse of destructive behavior predicted alternative-response disruption and vice versa; the co-occurrence of these two events always exceeded the background probabilities of either event occurring in isolation. General reductions in treatment efficacy occurred across changes in reinforcement or context, with no apparent decrease in likelihood in later transitions. We discuss implications of our findings with respect to future studies examining treatment durability.

The durability of behavior-analytic interventions has become of increasing interest to researchers and practitioners alike, with the goal of reducing treatment relapse. Treatment relapse is a phenomenon in which previously reduced target behavior like severe destructive behavior (e.g., aggression, self-injurious behavior) recurs after a period of successful intervention (e.g., Wathen & Podlesnik, 2018). One necessary component for understanding and mitigating treatment relapse is to thoroughly define the conditions under which treatment relapse occurs with commonly prescribed interventions. Traditional descriptions of treatment maintenance have focused on the continuation of treatment effects when implementing the intervention under similar conditions (i.e., with high fidelity by the same therapists in the same context). However, recent examinations of treatment maintenance have focused on the persistence of treatment effects during common treatment challenges (Nevin & Wacker, 2013). Two such treatment challenges shown to produce relapse of destructive behavior include a worsening of reinforcement conditions (Briggs et al., 2018; Fisher et al., 2020; Muething et al., 2021; Wacker et al., 2011) and contextual changes (Falligant et al., 2021; Muething et al., 2020; Saini et al., 2018). These two forms of treatment relapse are called resurgence and renewal, respectively (Wathen & Podlesnik, 2018).

A common intervention examined in previous papers on resurgence and renewal is functional communication training (FCT; Carr & Durand, 1985). FCT uses differential reinforcement of alternative behavior to teach functional communication responses (FCRs; e.g., card exchanges), providing an appropriate way of requesting the reinforcers maintaining destructive behavior. To facilitate FCT treatment effects, behavior analysts can program extinction for destructive behavior along with reinforcement of FCRs. Although FCT with extinction is effective at reducing destructive behavior and teaching FCRs (Hagopian et al., 2011), resurgence may occur when making the treatment more practical for stakeholders by thinning the schedule of reinforcement for the FCR (e.g., increasing periods of nonreinforcement from 30 s to 60 s; Briggs et al., 2018; Muething et al., 2021). Similarly, behavior analysts might observe renewal of destructive behavior when implementing FCT in a novel context (e.g., transitions from a therapist implementing FCT in a clinic to caregivers implementing it in the individual’s home; Muething et al., 2020; Saini et al., 2018).

Treatment relapse may occur in clinical care because behavior analysts routinely program treatment challenges that could produce resurgence in an effort to make FCT more practical (e.g., thinning reinforcement for alternative behavior) or produce renewal when evaluating generalization of treatment effects (e.g., caregiver- or teacher-implemented sessions in the home or school). Indeed, four recent studies used consecutive-controlled-case-series (CCCS) designs to investigate the prevalence of resurgence or renewal during routine treatment of severe destructive behavior in clinics specializing in its assessment and treatment. As described by Hagopian (2020), CCCS designs may help minimize publication bias by reporting on the use of a treatment with a specific individual, for a certain function, or applied to a given situation, regardless of the treatment outcome and as long as the individual’s data set or treatment procedure meets criteria for scientific rigor.

Briggs et al. (2018) used a CCCS design to examine the prevalence of resurgence when conducting reinforcement schedule thinning with discriminative stimuli across 25 consecutive FCT treatments. Behavior analysts tend to employ multiple or chained schedules to thin reinforcement schedules within FCT (Greer et al., 2016; Hagopian et al., 2011; Hanley et al., 2001). These compound schedules involve at least two schedule components (often a fixed-ratio 1 schedule and extinction for FCRs) that alternate quasirandomly and are correlated with unique stimuli, such as different colored cards. Typically, the extinction component ends and produces the availability of reinforcement following the passage of time (e.g., 30 s) in multiple schedules or following compliance with a requisite number of instructions (e.g., completion of three math problems) in chained schedules. Behavior analysts usually thin these schedules by gradually increasing the amount of time that passes in multiple schedules (e.g., transitioning to a 60 s extinction interval) or increasing the required instances of compliance in chained schedules (e.g., completion of six math problems). We refer to these schedule changes as one type of transition.

Despite FCT being highly efficacious at treating destructive behavior (Greer et al., 2016; Muharib et al., 2021), Briggs et al. (2018) found that destructive behavior resurged at least once in 76% (19 of 25) of treatments and in 42% (47 of 111) of schedule-thinning steps. When resurgence occurred, it exceeded baseline levels of destructive behavior at least once in about a third of the treatments (6 of 19). When the authors analyzed the data without regard to baseline (i.e., by comparing rates of destructive behavior prior to and immediately following a schedule-thinning transition), destructive behavior increased by an average 508%.

Recently, Muething et al. (2021) replicated Briggs et al. (2018) with 32 additional participants from a similar but distinct clinical program. Muething et al. found that destructive behavior resurged at least once in 90% of participants (29 of 32) and in 40% (97 of 239) of transitions when using similar types of treatments. Taken together, the findings of Briggs et al. and Muething et al. suggest that resurgence is common even when reinforcement schedule thinning occurs with well-trained therapists in highly controlled and specialized clinics. Further, when resurgence occurs, its magnitude can exceed baseline levels of responding.

Researchers have used similar methods to evaluate the effects of contextual changes during routine clinical care. Muething et al. (2020) used a CCCS design to evaluate the prevalence of renewal across 67 participants by examining increases in destructive behavior when implementing treatment with a novel adult or in a novel setting. Muething et al. found that problem behavior renewed in 67% of participants (45 of 67) and in 42% of transitions (77 of 182) from one setting or therapist to another. The authors also examined the magnitude and duration of renewal in addition to its prevalence. Muething et al. found that there was a three-fold increase in problem behavior (relative to the preceding treatment session) during the first context change. Recently, Falligant et al. (2021) replicated Muething et al. by analyzing participants from a separate clinic and observed renewal in 59% of participants (20 of 34) and 24% of transitions (47 of 200). Although these research groups found differing levels of renewal at the transition level, the prevalence of renewal at the participant level was high for both studies, suggesting that renewal occurs regularly, even with highly trained therapists in rigorously controlled clinics.

Based on these CCCS studies, resurgence and renewal seem to occur at least once with most participants or in many treatments, and relapse magnitude may be substantial initially. However, there are two primary limitations to these studies that warrant additional research: first, none of the CCCS studies permitted a within-subject evaluation of treatment relapse. By analyzing individual patterns of relapse across schedule-thinning steps or context changes, one may be able to examine the relevance of purported mitigation strategies. For example, a within-subject analysis may detect less resurgence at the end of an individual’s schedule thinning progression (i.e., extended treatment exposure; Leitenberg et al., 1975; cf. Greer et al., 2020) or minimal renewal after experiencing FCT with several novel implementers (i.e., multiple context training; Bernal-Gamboa et al., 2020). In this study, we used within-subject evaluations of treatment relapse to demonstrate the prevalence, magnitude, and persistence of resurgence and renewal across 25 consecutive FCT treatments for severe destructive behavior at the individual participant level. Second, studies have yet to examine degradation of alternative behavior during relapse tests, which is a clinically relevant, but sometimes overlooked, aspect of the treatment-relapse literature. Thus, for all 25 consecutive FCT treatments in this evaluation, we measured the covariation between relapse of destructive behavior and disruption of alternative behavior.

Method

Participants and Setting

We reviewed clinical records of 15 discharged individuals ranging in ages from 3–19 years old who received services from a large center-based treatment facility between 2017–2018.¹ We included individuals referred for the assessment and treatment of severe destructive behavior from an intensive-outpatient or day-treatment admission. We defined a participant as an individual who met four inclusion criteria. First, the individual’s destructive behavior was maintained by socially mediated consequences as indicated by a functional analysis. Second, the individual experienced at least one treatment relevant to our evaluation of resurgence and renewal. Specifically, we defined a treatment as an application of FCT with (a) multiple or chained schedules, (b) extinction programmed for destructive behavior, and (c) at least one opportunity to identify resurgence or renewal. Third, interobserver-agreement data were collected for at least 25% of all treatment sessions. Fourth, the individual was discharged from the clinical program as scheduled (i.e., not early due to poor attendance). Some participants had separate treatments to address different functions or referral concerns in isolated sessions (e.g., one treatment for escape-maintained destructive behavior and another for attention-maintained destructive behavior). Please see Table 1 for participant and treatment numbers and demographic information. No participants or treatments overlapped with Briggs et al. (2018).

Table 1.

Participant Demographics

Participant	# Treatments (Resurgence)	# Treatments (Renewal)	Age	Diagnoses a	Target Behavior b	Communication Modality
1	2	2	5	ABCD	Aggression, PD, SIB	Card Exchange
2	2	0	11	B	Aggression, PD, SIB	Card Exchange
3	3	0	6	B	Aggression, PD, SIB	Card Exchange
4	2	0	11	A	Aggression, PD, SIB	Card Exchange
5	3	4	19	B	Aggression, PD, SIB	Vocal
6	1	2	3	A	Aggression, PD, SIB, NV	Card Exchange
7	1	0	6	AB	Aggression, PD	Card Exchange
8	0	1	10	B	Aggression, PD	Card Exchange
9	1	0	9	AE	Aggression, PD	Card Exchange
10	3	2	11	AEF	Aggression, PD, Button Removal	Card Exchange
11	2	1	8	B	Aggression, PD, SIB	Card Exchange
12	2	0	9	AB	Aggression, PD, SIB	Card Exchange
13	2	2	12	E	Aggression, PD	Vocal
14	0	1	5	AB	Aggression, PD, SIB, NV	Card Exchange
15	1	1	7	G	Aggression, PD, SIB	Speech-Generating Device

Note. Target-behavior topographies are the destructive behavior addressed by the programmed treatment. Communication modality is the topography of the programmed functional communication response during the treatment.

^aA = autism spectrum disorder; B = unspecified disruptive, impulse-control, and conduct disorder; C = stereotypic movement disorder with self-injury; D = pica; E = oppositional defiant disorder; F = deafness; G = disruptive mood dysregulation disorder.

^bPD = property destruction; SIB = self-injurious behavior; NV = negative vocalizations. Button removal involved attempts to pull out a gastrostomy button used for tube feedings.

General Assessment and Treatment Procedures

For each participant, we conducted a functional analysis (similar to the procedures described by Iwata et al., 1982/1994) to determine the variables maintaining destructive behavior. Across participants, destructive behavior included topographies of aggression (e.g., hitting, kicking, biting others), self-injurious behavior (e.g., self-hitting, self-biting, headbanging), and property destruction (e.g., breaking objects, overturning furniture). Other target behavior measured alongside these primary topographies included negative vocalizations (e.g., screaming, yelling) and gastrostomy-button removal (i.e., attempts to pull out a gastrostomy button used for tube feedings).

We taught an FCR using a combination of progressive prompt delays and differential reinforcement while programming extinction for destructive behavior. Across participants, FCRs included topographies of nonvocal (e.g., card exchange) and vocal (e.g., “Play, please”) behavior. We began FCT with a fixed-ratio 1 reinforcement schedule for the FCR. After the participant acquired the FCR, we evaluated the efficacy of FCT as a treatment for destructive behavior using an ABAB design prior to reinforcement schedule thinning or programming contextual changes.

Following this evaluation, we began to increase the time that the reinforcer was unavailable. As in Briggs et al. (2018), we thinned reinforcement during FCT by incorporating discriminative stimuli into a compound schedule of reinforcement (i.e., a multiple or chained schedule) to signal periods of reinforcement (S^D) and extinction (S^Δ) for the FCR. We progressively increased the duration of the S^Δ interval (e.g., from 30 s to 60 s) relative to that of the S^D duration until we reached the caregivers’ preferred terminal schedules (e.g., 600 s S^Δ, 150 s S^D). In general, the common practice in the clinical setting where these cases were conducted is to thin reinforcement gradually (e.g., increasing the extinction component by 10 s or requiring an additional two instances of compliance) following low levels of destructive behavior and high levels of alternative behavior for at least two consecutive sessions. However, some deviations occurred for practical reasons such as limited time with a participant. Similar to other papers on FCT (e.g., Greer et al., 2016), we defined correct FCRs as FCRs occurring independently and exclusively during the reinforcement component of the multiple or chained schedule. We computed the percentage of correct FCRs by dividing the number of independent FCRs during the reinforcement component by the sum of the number of FCRs during the extinction component (and prompted FCRs during the reinforcement component, if these occurred); then, we multiplied this quotient by 100.

For many participants, a primary treatment goal was increasing compliance with adult instructions. We typically defined compliance as completing an adult’s instruction (e.g., “Clean up”) within 5 s of the adult’s initial vocal prompt or follow-up model prompt (e.g., “Clean up…like this” while modeling how to clean up toys) without co-occurring destructive behavior. If the participant did not respond to either the vocal or model prompt, the adult physically guided the participant to complete the response. This guided response was not scored as compliance. We measured the percentage of instructions with compliance by dividing the number of instructions with compliance by the total number of instructions and multiplying the quotient by 100.

Interobserver Agreement (IOA) and Procedural Integrity

For all participants, a second data collector simultaneously but independently collected data on the three primary dependent variables (i.e., destructive behavior, FCRs, and compliance) with 100% of resurgence participants (13/13) and 80% of renewal participants (8/10). Data collectors recorded the frequency of destructive behavior and FCRs during the S^D interval separately from the S^Δ interval and recorded each topography of destructive behavior separately (e.g., aggression, self-injurious behavior). For each dependent measure, the clinical teams calculated IOA coefficients using exact agreement within 10 s intervals. To be conservative and detect potentially low IOA, we selected the lowest IOA coefficient of all dependent measures in a given session to serve as the IOA coefficient for destructive behavior or FCRs for that session. We then averaged the mean IOA coefficients for each treatment to produce grand means for destructive behavior, FCRs, and compliance across treatments. We obtained the following grand means for resurgence treatments: destructive behavior = 99% (range, 99%–100%); FCRs = 98% (range, 95%–100%); compliance = 94% (range, 87%–100%). We obtained the following grand means for renewal treatments: destructive behavior = 99% (range, 95%–100%); FCRs = 99% (range, 95%–100%); compliance = 94% (range, 82%–100%).

We calculated procedural integrity on at least 30% of transitions included in the analysis of resurgence (44/144) and renewal (16/53), sampling across transitions with and without relapse. We calculated procedural integrity by analyzing raw data streams to determine if the therapist or caregiver delivered reinforcement precisely according to the programmed schedules. We defined a reinforcer to be delivered correctly in two ways. First, we scored correct delivery when the therapist or caregiver delivered the reinforcer within 5 s of an FCR during the S^D interval if destructive behavior did not immediately precede the FCR. Second, if destructive behavior did co-occur with an FCR during the S^D interval, we scored correct delivery if the therapist or caregiver waited at least 3 s following destructive behavior prior to reinforcing a subsequent FCR. This latter procedure involved the therapist or caregiver implementing a 3 s changeover delay during FCT with multiple and chained schedules (e.g., Greer et al., 2016) to minimize adventitious reinforcement of destructive behavior. We calculated procedural integrity by dividing the number of correct reinforcer deliveries by the sum of correct and incorrect reinforcer deliveries and multiplying this quotient by 100. We then averaged the mean procedural-integrity coefficients across treatments to calculate the following grand means: resurgence treatments = 98% (range, 75%–98%); renewal treatments = 92% (range, 50%–100%). The few instances of lower procedural integrity appeared to be coding errors in which the data stream indicated that the S^D interval and item engagement (e.g., playing with the programmed tangible item) continued following FCRs but the specific reinforcement key was not turned on. Because these integrity issues appeared to be minor coding errors, we included those transitions in our analysis.

General Record Review Procedures

Study authors reviewed all records from 2017–2018 for individuals who met the inclusion criteria. Each treatment consisted of the final set of procedures recommended to the participant’s caregivers. We included all treatments for each participant who met inclusion criteria until we identified 25 treatments that allowed at least one opportunity for resurgence or renewal. Once we identified 25 treatments for either the resurgence or renewal evaluation, we no longer considered additional participants or treatments for inclusion. These procedures led to 41 total treatments (25 treatments with at least one opportunity for resurgence and 16 treatments with at least one opportunity for renewal) across 15 participants.

Next, we analyzed all treatment sessions for opportunities for resurgence or renewal, which we labeled a transition. We defined a transition with an opportunity for resurgence or renewal as a progression from a previous condition (hereafter referred to as Condition A) to a subsequent condition (hereafter referred to as Condition B) that entailed schedule thinning or a context change, respectively. See the response-measurement section below for details regarding these transition types. We included the final two sessions of Condition A and the first three sessions of Condition B in all analyses, similar to the studies by Briggs et al. (2018) and Muething et al. (2020).

We included only those schedule-thinning steps or context changes for which destructive behavior remained at or below an 85% reduction from baseline levels during both sessions of Condition A. Although this differed from previous CCCS studies, this criterion ensured that increases in destructive behavior during Condition B would be meaningfully higher than our clinic’s standard for acceptable levels of destructive behavior. We also excluded all transitions for which (a) fewer than two sessions were available for analysis in Condition A and (b) when more than one change occurred (e.g., an increase in response requirement and the introduction of a novel therapist; a change in therapist and setting). For the included participants and treatments, these procedures led to the inclusion of 197 transitions. Data on alternative behavior during Condition A were excluded from analysis if it fell below criterion levels used clinically within our program (i.e., correct FCRs ≥ 80%; compliance ≥ 70%).

Response Measurement

We labeled transitions with opportunities for resurgence as schedule-thinning steps, which involved increases in duration of the S^Δ interval in multiple schedules, increases in the response requirement for compliance of chained schedules, or decreases in duration of the S^D interval of either treatment type. We classified 144 of the transitions as schedule-thinning steps. The remaining 53 transitions were opportunities for renewal, or context changes, which involved a novel therapist or caregiver implementing the treatment procedures or a familiar therapist or caregiver conducting treatment in a novel setting (e.g., a new location within the clinic, at home or school).

We used the same criteria as Briggs et al. (2018) and Falligant et al. (2021) to determine whether resurgence or renewal of destructive behavior occurred during each included transition. Specifically, destructive behavior in at least one session of Condition B had to exceed responding during either session of Condition A to be considered an instance of resurgence or renewal. Inversely, we considered alternative responding to be disrupted if levels of alternative responding (percentage of correct FCRs or compliance) in at least one session of Condition B fell below those in either session of Condition A.

We computed conditional probabilities to determine the covariation between the recurrence of destructive behavior and disruption of alternative behavior. We determined the unconditional probabilities of resurgence or renewal of destructive behavior and alternative-behavior disruption by counting the number of transitions that met the relapse or disruption criteria, respectively, independent of the other dependent measures. We determined conditional probabilities by counting the number transitions in which destructive behavior recurred or alternative behavior was disrupted when both responses had the opportunity to occur. Although all transitions had an opportunity to measure the recurrence of destructive behavior, not every transition had an opportunity to measure alternative-behavior disruption due to the above criteria for analyzing alternative behavior.

Reliability and Interrater Agreement

A second author independently reviewed all participants, treatments, and transitions initially deemed appropriate for inclusion. Transcription of the data from clinical records always entailed (a) one author reading aloud the raw data while the other author transcribed the data or (b) having a second author review the raw data transcribed by another author. Finally, two authors reviewed at least 33% of all included participants to ensure that summary spreadsheets matched the clinical data. For all these steps, any disagreements were discussed until consensus was reached.

Results

Our record review resulted in an analysis of datasets for 15 participants. For the resurgence evaluation, we analyzed data for 13 participants (e.g., an individual named James who may have one or more FCT treatments), 25 treatments (e.g., James’s multiple schedule for attention and chained schedule for escape), and 144 transitions within those treatments (e.g., thinning the extinction interval from 30 s to 60 s). For the renewal evaluation, we analyzed data for 10 participants, 16 treatments, and 53 transitions.

Figure 1 shows the number of treatments with and without relapse of destructive behavior by function(s). If the functional analysis identified more than one variable maintaining destructive behavior, we counted each treatment as targeting multiply maintained destructive behavior. The top panel shows treatments for which there were opportunities for resurgence, and the bottom panel shows treatments for which there were opportunities for renewal. Resurgence occurred with 100% (7/7) of treatments targeting escape functions, 75% (3/4) of treatments targeting attention functions, 36% (4/11) of treatments targeting tangible functions, and 33% (1/3) of treatments targeting multiply maintained destructive behavior. Renewal occurred with 80% (4/5) of treatments targeting escape functions, 0% (0/2) of treatments targeting attention functions, 80% (4/5) of treatments targeting tangible functions, and 75% (3/4) of treatments targeting multiply maintained destructive behavior. In sum, treatment relapse (i.e., resurgence or renewal) occurred across all functions addressed. For treatments targeting an attention function, however, we only observed relapse in the form of resurgence.

Figure 1. Number of Treatments with Relapse by Function(s).

Note. Bars are separated by function of destructive behavior identified in the functional analysis. Bar height indicates the number of treatments with either a schedule-thinning step (top panel) or a context change (bottom panel). The filled portion of each bar represents instances of resurgence (top panel) or renewal (bottom panel). If the functional analysis indicated that a participant’s destructive behavior was maintained by more than one variable (e.g., escape, attention), we counted each treatment toward the “multiply maintained” bar.

Figure 2 shows the number of resurgence (top panel) and renewal (bottom panel) evaluations with and without relapse of destructive behavior and disruption of alternative behavior across participants, treatments, and transitions. Destructive-behavior relapse and alternative-behavior disruption occurred in all situations we analyzed. Resurgence of destructive behavior (the first bar in each trio of bars on the top panel) occurred with 85% (11/13) of participants, 76% (19/25) of treatments, and 38% (55/144) of transitions. Renewal of destructive behavior (the first bar in each trio of bars on the bottom panel) occurred with 90% (9/10) of participants, 69% (11/16) of treatments, and 30% (16/53) of transitions. Thus, destructive behavior was similarly likely to recur during opportunities for resurgence and renewal at each level of analysis. The prevalence of resurgence was similar to previous CCCS studies (Briggs et al., 2018; Muething et al., 2021), whereas the prevalence of renewal was most similar to Falligant et al. (2021) at the transition level but higher than Falligant et al. or Muething et al. (2020) at the participant level.

Figure 2. Relapse by Dependent Variable Across Participants, Treatments, and Transitions.

Note. Bars are separated by dependent variable and grouped by level of analysis (i.e., participant, treatment, or transition). The filled portion of each bar represents resurgence (top panel) or renewal (bottom panel) at the respective level of analysis.

During opportunities for resurgence, disruption of correct FCRs (second bar in each trio of bars on the top panel) occurred with 91% (10/11) of participants, 79% (13/18) of treatments, and 33% (27/82) of transitions whereas disruption of compliance (third bar in each trio of bars on the top panel) occurred with 89% (8/9) of participants, 91% (10/11) of treatments, and 60% (34/57) of transitions. During opportunities for renewal, disruption of correct FCRs (second bar in each trio of bars on the bottom panel) occurred in 67% (4/6) of participants, 44% (4/9) of treatments, and 14% (5/35) of transitions, whereas disruption of compliance (third bar in each trio of bars on the bottom panel) occurred in 71% (5/7) of participants, 75% (6/8) of treatments, and 47% (16/34) of transitions. In sum, changes in reinforcement seemed to disrupt alternative behavior slightly more than contextual changes, and compliance with adult instructions was more likely to be disrupted than FCRs by either type of treatment challenge.

Figure 3 shows the magnitude and persistence of destructive behavior relapse across transitions with resurgence (top panel) or renewal (bottom panel). Each line represents a transition in which we identified relapse, with the weighted red line indicating mean responding at each session. We converted each session’s rate of destructive behavior from Conditions A and B into a proportion of baseline responding by dividing the raw value (e.g., two responses per minute at Session B1) by the average of the participant’s final five baseline sessions (e.g., four responses per minute during FCT baseline, resulting in a proportional value of 0.5). Proportional values above 1.0 indicate that response rates exceeded the participant’s baseline average. Such a measure allows for comparisons of relapse across participants who may have markedly different baseline rates of destructive behavior. The general patterns of resurgence and renewal were similar, with both transition types showing substantial increases in destructive behavior during some instances of reinforcement schedule thinning or context changes. Plotting the transition data in this way also shows that average levels of destructive behavior continued to increase or remained elevated throughout Condition B during the resurgence evaluations, suggesting that analyzing only three sessions of posttransition data may be insufficient to fully capture treatment relapse during such transitions in the clinic.

Figure 3. Magnitude and Persistence of Destructive-Behavior Relapse.

Note. Resurgence (top panel) and renewal (bottom panel) of destructive behavior expressed as a proportion of baseline responding across transitions with relapse. The weighted, red line indicates mean responding at each transition session. A1 and A2 represent sessions immediately preceding the transition. B1, B2, and B3 represent sessions immediately following the transition. To facilitate inspection of all transitions, we nudged duplicate values across transitions by up to +/−0.05. However, mean lines reflect the computation of the actual values for each transition.

Figure 4 shows the magnitude and persistence of alternative-behavior disruption during reinforcement schedule thinning or context changes. The data are arranged similarly to those in Figure 3. However, these panels show transitions in which FCRs (left panels) or compliance with adult instructions (right panels) were disrupted by opportunities for resurgence (top panels) or renewal (bottom panels). Similar to the results of Figure 3, disrupted alternative behavior did not appear to stabilize to pretransition levels by the end of Condition B in many transitions. Aside from compliance during context changes, schedule thinning or context changes occasionally disrupted the percentage of correct FCRs or compliance to near-zero levels in certain transitions, meaning that some participants did not emit an FCR or comply with any adult requests in a given transition session.

Figure 4. Magnitude and Persistence of Alternative-Behavior Disruption.

Note. Alternative behavior during transitions with schedule-thinning steps (top panels) or context changes (bottom panels) resulting in disruption to functional communication responses (FCRs; left panels) or compliance (right panels). The weighted, red line indicates mean responding at each transition session. A1 and A2 represent sessions immediately preceding the transition. B1, B2, and B3 represent sessions immediately following the transition. To facilitate inspection of all transitions, we nudged duplicate values across transitions by up to +/−2%. However, mean lines reflect the computation of the actual values for each transition.

Figures 5 and 6 show the scope and persistence of relapse and response disruption during opportunities for resurgence and renewal, respectively. The number at the end of each row displays the percentage of transitions for a given participant’s treatment that had relapse of destructive behavior or disruption of alternative behavior. Several findings may be gleaned from these figures. First, a disruption in treatment effects did not appear to decrease in likelihood across subsequent transitions. For example, resurgence occurred in two of three treatments for Participant 3, including during the final two schedule-thinning steps in each of those two treatments. Similarly, context changes renewed destructive behavior or disrupted compliance for Participant 8 (Treatment 9) during initial transitions, but even the final two context changes resulted in renewed destructive behavior and disrupted compliance.

Figure 5. Scope and Persistence of Relapse During Reinforcement Schedule Thinning.

Note. Symbols represent opportunities for resurgence of destructive behavior (squares) or disruption of functional communication responses (circles) or compliance (triangles). Closed symbols indicate resurgence of destructive behavior or disruption of alternative behavior. The final column depicts the percentage of transitions with resurgence or disruption for each response within the given treatment.

Figure 6. Scope and Persistence of Relapse During Contextual Changes.

Note. Symbols represent opportunities for renewal of destructive behavior (squares) or disruption of functional communication responses (circles) or compliance (triangles). Closed symbols indicate renewal of destructive behavior or disruption of alternative behavior. The final column depicts the percentage of transitions with renewal or disruption for each response within the given treatment.

A second finding related to Figures 5 and 6 is that the reinforcement-schedule thinning and context changes also disrupted alternative responding. Alternative-response disruption occurred both in conjunction with the recurrence of destructive behavior and in the absence of such recurrence. Resurgence or renewal occurred with the possibility to measure collateral response disruption of alternative responding in 46 and 14 transitions, respectively. Across these transitions, collateral response disruption occurred across 67% (or 31/46) and 64% (or 9/14) of transitions, respectively. These data suggest that across resurgence and renewal opportunities, observing collateral response disruption was highly likely when destructive behavior recurred. Conditional probabilities of alternative-response disruption given the recurrence of destructive behavior were consistently higher than the unconditional probabilities of alternative-response disruption (unconditional probabilities = 52% or 54/104 of transitions; 49% or 20/41 of transitions), suggesting that the recurrence of destructive behavior was a good predictor of alternative-response disruption.

A third and related finding from Figures 5 and 6 is that alternative-response disruption was a good predictor of destructive behavior recurrence. Alternative-response disruption occurred with the possibility to measure resurgence or renewal of destructive behavior in 54 and 20 transitions, respectively. Across these transitions, destructive behavior recurred across 57% (or 31/54) and 45% (or 9/20) of transitions, respectively. Conditional probabilities of the recurrence of destructive behavior given alternative-response disruption were consistently higher than the unconditional probabilities of destructive-behavior recurrence for every dataset (unconditional probabilities = 38% or 55/144 of transitions; 30% or 16/53 of transitions).

Discussion

Although researchers have documented both resurgence and renewal during routine, behavior-analytic service delivery (Briggs et al., 2018; Falligant et al., 2021; Muething et al., 2020; Muething et al., 2021; Saini et al., 2018), prior investigations have not included analyses of other clinically relevant responses that are important components of FCT (i.e., alternative behavior) or presented within-subject data for all transitions. We believe this is the first study to analyze alternative behavior during relapse arrangements and to present it alongside destructive behavior for each analyzed transition. The result is a broad, yet simultaneously detailed, picture of treatment relapse as it occurs when treating severe destructive behavior.

A few notable findings stand out. First, treatment relapse occurred across all functions of destructive behavior targeted. In fact, for only one function of destructive behavior (i.e., attention) did we not observe both resurgence and renewal. However, the limited number of treatments in which renewal of attention-reinforced destructive behavior could have occurred (n = 2) suggests that this may have resulted from a sampling issue rather than being representative of a broader finding. Additionally, all treatments for escape-reinforced destructive behavior showed resurgence, replicating the findings by Briggs et al. (2018). In fact, the overall prevalence of resurgence at the treatment level (76%) was identical to the prevalence data reported by Briggs et al., despite being from independent samples. To our knowledge, this is the first study to examine resurgence and renewal within the same sample. This afforded the opportunity to identify a handful of individuals who participated in both the resurgence and renewal evaluations, though it is an insufficient sample size to answer broad questions about within-subject prevalence of multiple forms of relapse. An interesting preliminary finding is that all seven individuals who participated in both resurgence and renewal evaluations displayed at least one form of relapse and 5 of 7 participants displayed both resurgence and renewal at least once. To examine this phenomenon more thoroughly, researchers should design a CCCS that requires an opportunity to evaluate multiple forms of treatment relapse (e.g., resurgence, renewal) as an inclusion criterion. Such a study would allow researchers to examine the co-occurrence of relapse phenomena within participants on a larger scale and determine if resurgence is predictive of renewal and vice versa.

Second, the recurrence of destructive behavior and the disruption of alternative responding often co-occurred, and in all situations analyzed, the probability of this co-occurrence equaled or exceeded the unconditional probabilities of either event occurring in isolation. This observation suggests more generally that the full extent to which treatment challenges disrupt treatment effects may be underrepresented in the applied literature when researchers report data only on destructive behavior. A more comprehensive view of treatment durability might mean that researchers report on both destructive and alternative behavior, especially when the alternative behavior is clinically important. Some schedule-thinning transitions resulted in near-zero levels of compliance, which could be especially troublesome when compliance is imperative (e.g., if the individual fails to respond to adult directions during an emergency). Furthermore, researchers may need to extend the temporal window with which they analyze treatment relapse. Data from the present study suggest that restricting the analysis to three sessions posttransition may compromise descriptions of the relapse effect. Clinically speaking, this implies that practitioners may need to commit to more than three sessions after making a change in reinforcement schedule or context before expecting behavior to return to pretransition levels if treatment effects indeed deteriorate.

Another implication of this finding highlights a limitation of the study. Although one can certainly envision situations in which increased destructive behavior competes with alternative responding, a strong correlation between the two should not necessarily imply that one event is the cause of the other. Our study lacks the sort of precision to answer questions related to behavioral processes, such as why the magnitude of treatment relapse and alternative-response disruption varied so widely across transitions (Figures 5 and 6, respectively), as well as why such decrements in performance often occurred across transitions with no apparent improvement in treatment maintenance during later transitions. However, we refer interested readers to Shahan and Greer (2021), which analyzes portions of our data and those of Muething et al. (2021) to determine how different proportional decrements in obtained alternative reinforcement rates from Condition A and Condition B affect the magnitude of resurgence. Such retrospective analyses and additional prospective experiments may help illuminate the processes influencing differential relapse outcomes.

Finally, our findings highlight the need to continue evaluating relapse-mitigation techniques. Researchers have used discriminative stimuli during FCT to mitigate resurgence of destructive behavior during prolonged extinction tests (Fisher et al., 2020; Fuhrman et al., 2016) or to transfer treatment effects to novel contexts without relapse (Fisher et al., 2015; Greer et al., 2019), with these cited papers involving patients from the same clinical program as used in Briggs et al. (2018) and in the current study. Yet, as Briggs et al. and this study show, the strategies demonstrated to mitigate relapse in experimental studies may be less robust when analyzed on a larger scale or with repeated treatment challenges (e.g., frequent schedule-thinning steps). Other suggested strategies for reducing relapse are not necessarily supported by our data. For example, multiple-context training, such as conducting FCT across several implementers or settings, has been described as a possible renewal-mitigation technique (Bernal-Gamboa et al., 2020; Podlesnik et al., 2017). Additionally, research suggests that increasing time in extinction (e.g., continued exposure to FCT) can reduce resurgence (e.g., Sweeney & Shahan, 2013). Both strategies were in place in our renewal and resurgence data sets, respectively.

We were able to evaluate these mitigation strategies in four renewal treatments and 11 resurgence treatments. In these treatments, there was an initial instance of renewal or resurgence during the first transition and then opportunities to evaluate renewal or resurgence in subsequent transitions. We observed continued renewal in 3 of 4 treatments after additional context changes and resurgence in 9 of 11 treatments despite extra exposure to FCT. Although these findings do not diminish the outcomes of well-constructed experimental preparations, they support investigating the generality of promising mitigation strategies within designs such as the CCCS. It may be that such techniques need refinement or to be combined to bolster their durability across treatment challenges during routine care.