Further evaluation of treatment duration on the resurgence of destructive behavior

Abstract

Translation of promising procedures for mitigating treatment relapse has received considerable attention recently from researchers across the basic–applied continuum. One procedure that has demonstrated mixed support involves increasing the duration of treatment as a strategy for blunting resurgence. In a recent translational study, Greer et al. (2020) failed to detect a mitigation effect of increased treatment duration on the resurgence of destructive behavior. However, design limitations may have been responsible. The present study corrected these limitations by (a) employing a sequential design to decrease the possibility of multiple‐treatment interference, (b) evaluating more treatment durations, (c) arranging treatments of fixed durations, and (d) conducting treatments of more extreme duration in a different clinical sample. Despite these improvements in experimental rigor and the testing of more extreme boundary conditions, the present study also failed to detect a mitigation effect of increased treatment duration. Likely explanations are discussed.

Both existing quantitative theories of resurgence (i.e., Behavioral Momentum Theory [BMT; Shahan & Sweeney, ²⁰¹¹; see also Nevin & Shahan, ²⁰¹¹], Resurgence as Choice in Context [RaC²; Shahan, Browning, & Nall, ²⁰²⁰; Shahan & Craig, ²⁰¹⁷; see also Greer & Shahan, ²⁰¹⁹]) predict that increased exposure to extinction‐based treatments will mitigate the recurrence of target responding when rates of alternative reinforcement (e.g., differential reinforcement of alternative behavior; DRA) are thinned or discontinued. However, the two models differ substantially in the degree to which treatment duration is predicted to affect resurgence. Whereas BMT predicts robust resurgence mitigation with small to moderate increases in the duration of treatment (see Figure 1 in Greer et al., ²⁰²⁰ and related simulations in Greer, Fisher, Romani, & Saini, ²⁰¹⁶), RaC² predicts very little resurgence mitigation with similar increases in treatment duration (Shahan, Browning, & Nall, 2020).

Figure 1.

Resurgence Evaluations for Adonis

Note. Adonis's evaluations of resurgence following short (top panel), moderate (middle panel), and extended (bottom panel) functional communication training (FCT). The parenthetical numbers refer to the order in which Adonis experienced the experimental sequences.

Empirical support for increased treatment duration mitigating resurgence has been mixed, with some evidence of such a relation (Fisher, Greer, Fuhrman, et al., 2018; Leitenberg et al., ¹⁹⁷⁵; Shahan, Browning, & Nall, ²⁰²⁰; Smith & Greer, ²⁰²²; Sweeney & Shahan, ^2013a; Wacker et al., ²⁰¹¹) and other investigations failing to detect an effect (Greer et al., 2020; Nall et al., ²⁰¹⁸; Winterbauer et al., ²⁰¹³). At least four of the studies that provided evidence that increased treatment duration mitigated resurgence included confounds, which further clouds conclusions regarding the relation. Specifically, Leitenberg et al. (1975) and Winterbauer et al. (2013) did not control for rates of alternative reinforcement, which itself has been shown to modulate resurgence (Sweeney & Shahan, ^2013b). This same confound was also present in two translational investigations (Fisher, Greer, Fuhrman, et al., 2018; Wacker et al., ²⁰¹¹) involving destructive behavior that similarly provided support for the relation. Further, two of the studies (Sweeney & Shahan, ^2013a; Wacker et al., ²⁰¹¹) also included repeated exposures to alternative reinforcement presence versus absence while increasing treatment durations, a manipulation now known to mitigate resurgence itself (e.g., Schepers & Bouton, ²⁰¹⁵; Shahan, Browning, & Nall, ²⁰²⁰; Trask et al., ²⁰¹⁸).

Whether extended treatment durations alone might mitigate resurgence is an important question for applied behavior analysis because the quantitative models upon which these predictions are based are directly relevant to common intervention procedures. DRA‐based treatments for destructive behavior are common in clinical practice (Tiger et al., 2008) and are highly efficacious when function‐based (Greer, Fisher, Saini, et al., 2016; Lindgren et al., ²⁰¹⁶; Rooker et al., ²⁰¹³). However, they can be prone to treatment relapse in the form of resurgence (Briggs et al., 2018; Mitteer et al., ²⁰²²; Muething et al., ²⁰²¹), especially with larger downshifts in the availability of alternative reinforcement (Falligant et al., 2022; Shahan & Greer, ²⁰²¹). If an increased duration of treatment protects against the resurgence of destructive behavior, clinicians would be encouraged to persist with current intervention procedures for longer than current standards of care typically require even though no changes in responding occur, meaning that the clinical benefit of increased treatment duration would be undetectable to the clinician based on rates of responding alone. In other words, rates of destructive and appropriate behavior (i.e., the data upon which clinicians would ordinarily base their decisions) may be poor indicators of when reinforcement schedule thinning should occur. Because quantitative models of resurgence have shown good predictive validity when applied to clinically relevant populations and socially significant behavior (Fisher, Greer, Craig, et al., 2018; see also Fisher et al., 2022), exploring the translational feasibility of potential relapse‐mitigation techniques in the clinic is clearly warranted.

Greer et al. (2020) recently reported on outcomes from a preliminary study designed to isolate the effects of treatment duration when evaluating the resurgence of destructive behavior in a clinic. In that study, the investigators arranged two experimental contexts that differed only in color. Specifically, sessions occurred in a blue context in which the therapist wore a blue scrub top and a blue light filter cast a blue hue throughout the session room or a yellow context in which these stimuli were yellow. Therapists alternated between these contexts in a quasirandom order throughout phases of baseline in which only destructive behavior produced reinforcement, functional communication training (FCT) in which only a newly trained functional communication response (FCR) produced reinforcement, and a resurgence test in which both destructive behavior and the FCR resulted in extinction. One of the two contexts was associated with a long exposure to treatment, and the other context was associated with a short exposure to treatment. The total number of treatment sessions differed across participants; however, within participant, long exposures to treatment always consisted of conducting triple the number of treatment sessions relative to the short exposure to treatment. Treatment continued until destructive behavior decreased to clinically meaningful rates across the two contexts. Despite programming three times the number of treatment sessions in one of the two contexts, the investigators failed to detect a reliable effect of treatment duration on the resurgence of destructive behavior across their six participants.

Greer et al. (2020) discussed a few possibilities for their results. One possibility was that their participants failed to discriminate between the two treatment durations due to the multielement design used in that study, despite the programming of multiple contextual stimuli to establish unique contexts. That is, participants may have responded as though the designed contexts constituted a single context such that baseline, treatment, and the resurgence‐test phases were simply one longer exposure to each set of contingencies. An alternative design that programs the same three‐phase progression sequentially should not suffer from the same potential of multiple‐treatment interference.

A second possibility for the results of Greer et al. (2020) is that the treatment durations arranged were insufficiently dissimilar from one another, potentially masking the mitigating effects of increased treatment duration on the resurgence of destructive behavior. Across participants, treatment durations were between two to five sessions for the short‐treatment duration and between six and 15 sessions for the long‐treatment duration. The tripling of sessions between these two treatment durations was informed by the quantitative predictions of BMT. It is important to note that a prior clinical study from the same research team, using many of the same experimental procedures, found a robust and consistent resurgence‐mitigation effect when similar treatment durations were programmed along with other procedures predicted by BMT to lessen resurgence (i.e., lean schedules of reinforcement in baseline and in treatment; Fisher, Greer, Fuhrman, et al., 2018).

Since the publication of Greer et al. (2020), Shahan, Browning, and Nall (2020) reported the results of a highly powered and well‐controlled evaluation involving a range of five different treatment durations on subsequent resurgence of target responding by rats. Under these conditions, Shahan et al. identified a small, but statistically reliable, effect of treatment duration on resurgence, a relation that was well characterized by RaC², but not by BMT (see also Smith & Greer, ²⁰²²). Whether a clinic‐based study informed by the procedures used by Shahan et al. could similarly detect an effect of treatment duration on resurgence remains to be seen.

The present study aimed to address the limitations of Greer et al. (2020) described above by (a) employing a different experimental design to decrease the possibility of multiple‐treatment interference, (b) evaluating more than two treatment durations, (c) arranging treatments of fixed durations within and across participants, and (d) conducting treatments of more extreme duration in a different clinical sample.

Design of the present study and that of Shahan, Browning, and Nall (2020) were coordinated. Thus, aspects of the present study procedures (e.g., specific treatment durations evaluated, criteria used for phase termination, type of reinforcement schedule programmed) were determined in advance of either study and were jointly informed by both clinical considerations regarding typical durations of, and procedures for, implementing FCT with extinction, as well as the quantitative predictions of RaC². A primary goal of this careful coordination of studies was to produce optimal translations of prospective, basic research findings on potential relapse‐mitigation procedures suitable for clinical application (see Greer et al., ²⁰²², for discussion).

Method

Participants, Setting, and Materials

Five individuals participated in the study, four of whom completed at least one resurgence evaluation. All participants were children referred to a university‐based clinic specializing in the assessment and treatment of severe destructive behavior. They attended appointments for 3 hr or 6 hr a day, 5 days per week. Please see Table 1 for participant demographics. The study was approved by the university institutional review board, and we obtained informed consent from each child's caregiver prior to initiating study procedures. We withdrew one participant prior to the start of the study because a functional analysis indicated that their destructive behavior was maintained by automatic reinforcement.

Table 1.

Participant Information

Participant	Age	Autism diagnosis	Target behavior	Function(s)	FCT sequence
Adonis	8	Yes	Aggression	Tangible	Moderate, short, extended
Arlo	7	No	Aggression, property destruction, SIB	Tangible	Moderate, extended, short
Craig	10	No	Aggression, property destruction	Tangible	Short, moderate, extended
Trevor	10	Yes	Aggression, property destruction, SIB	Attention + tangible	Extended, moderate, short *

Note: FCT = functional communication training; SIB = self‐injurious behavior.

^*Trevor experienced only the extended‐FCT evaluation prior to withdrawal from the study.

Sessions occurred in approximately 3‐m by 3‐m therapy rooms, which included floor and wall padding. Each room had a one‐way observation mirror for unobtrusive data collection, a two‐way intercom system, and a table and chairs sized suitably for each participant. Each room contained the contextual materials correlated with experimental conditions (described below in Preliminary Procedures), FCR materials during relevant phases, and other materials necessary to conduct sessions.

Response Measurement and Interobserver Agreement

Trained data collectors recorded participant destructive behavior and FCRs, as well as therapist reinforcer deliveries using BDataPro (Bullock et al., 2017). Destructive behavior included aggression (e.g., hitting, kicking, biting), property destruction (e.g., breaking objects, ripping materials, overturning furniture), and self‐injurious behavior (e.g., self‐biting, self‐hitting). Data collectors scored FCRs when participants extended FCR cards toward the therapists as part of a card exchange and reinforcer deliveries when therapists provided access to preferred leisure items (Adonis, Arlo, and Craig) or attention and leisure items (Trevor).

A second independent data collector scored 80%, 63%, 78%, and 53% of sessions for Adonis, Arlo, Craig, and Trevor, respectively. Interobserver agreement sampling occurred across phases and experimental conditions, and at least half of each participant's interobserver agreement data were collected by an observer blinded to the study aims and hypotheses. BDataPro computed exact agreement for destructive behavior and FCRs by scoring if each data collector marked the same count of behavior within each 10‐s interval. The number of agreements was then divided by the total number of intervals (i.e., 60) and converted to a percentage. Mean agreement coefficients for destructive behavior were 94% (range, 65%–100%) for Adonis, 99% (range, 83%–100%) for Arlo, 99% (range, 81%–100%) for Craig, and 99% (range, 90%–100%) for Trevor. Mean agreement coefficients for FCRs were 99% (range, 86%–100%) for Adonis, 99% (range, 90%–100%) for Arlo, 96% (range, 81%–100%) for Craig, and 95% (range, 80%–100%) for Trevor. We used a similar method for reinforcer deliveries, except that BDataPro computed proportional interobserver agreement within 10‐s intervals because bouts of reinforcement were also measured as a duration. This calculation entailed BDataPro dividing the smaller value from one data collector's 10‐s interval by the larger value of the other data collector's 10‐s interval to produce a ratio and then averaging these ratios and converting the average to a percentage. Mean agreement coefficients for reinforcer deliveries were 96% (range, 93%–100%) for Adonis, 95% (range, 83%–100%) for Arlo, 93% (range, 84%–100%) for Craig, and 91% (range, 81%–100%) for Trevor.

A doctoral‐level behavior analyst analyzed BDataPro files to assess procedural fidelity of reinforcer deliveries. BDataPro files displayed a second‐by‐second account of all keys pressed by the data collector during the session, including when an observer scored therapist reinforcer delivery and participant behavior. The analyst randomly selected one‐third of baseline and FCT sessions for each treatment condition (i.e., short, moderate, extended) for each participant. For sessions assessed, the analyst identified all reinforcer deliveries. They then measured accuracy of reinforcer deliveries by examining any participant behavior that occurred within 3 s of each reinforcer delivery. During baseline, the analyst marked a correct reinforcer delivery if targeted destructive behavior occurred within 3 s of the reinforcer delivery. If the time between destructive behavior and reinforcement was longer than 3 s, the analyst scored this as an incorrect reinforcer delivery. During FCT, the analyst coded a correct reinforcer delivery if the participant's FCR occurred within 3 s of the reinforcer delivery without co‐occurring destructive behavior during that same time window. If the time between the FCR and the reinforcer delivery was longer than 3 s or if destructive behavior occurred within the 3 s preceding the FCR (i.e., an incorrect changeover delay), the analyst scored this as an incorrect reinforcer delivery. Correct reinforcer delivery across all sessions averaged 98% (range, 80%–100%). When analyzed by condition, correct reinforcer deliveries across participants averaged 97% (range, 94%–100%) in baseline and 97% (range, 94%–99%) during FCT. Nearly all incorrect deliveries involved reinforcers delivered outside the 3‐s window. No reinforcers were delivered directly following destructive behavior during FCT. We similarly analyzed all resurgence‐test sessions, and no reinforcers were delivered during this condition for any participant.

Procedures

We used a three‐phase resurgence progression (i.e., baseline, FCT, resurgence test) to evaluate treatment relapse for each level of the independent variable (i.e., time in treatment). For the within‐subject comparison of treatment duration on resurgence, we applied each of the three‐phase arrangements (i.e., baseline, short FCT, resurgence test; baseline, moderate FCT, resurgence test; baseline, extended FCT, resurgence test) sequentially with each participant. However, we quasirandomly assigned the order of these arrangements across participants to detect potential sequence effects (e.g., less resurgence following short and moderate FCT after experiencing extended FCT). Regardless of the sequence used, all transitions from FCT with extinction to the following resurgence test occurred within the same appointment for each participant. See Table 1 for each participant's experimental sequence.

We programmed uniquely colored stimuli during each three‐phase resurgence arrangement. For example, during Arlo's short‐FCT evaluation, therapists wore purple scrub tops and overlaid purple light filters on the therapy room's fluorescent bulbs to create a purple context. The therapists did the same with green and red materials for the moderate‐ and extended‐FCT evaluations, respectively. All sessions lasted 10 min except for three sessions for Trevor that the experimenters terminated early due to high‐intensity destructive behavior (see below).

Doctoral‐ and master's‐level board certified behavior analysts met with each treatment team and observed sessions daily to monitor safety concerns. To minimize the risk of injury to participants and staff, therapists wore protective equipment (e.g., arm guards, padded helmets) and incorporated additional padded materials (e.g., small blocking pads) as needed. Finally, we established individualized termination criteria for each participant to determine when to conclude study sessions. Only Trevor's destructive behavior met these criteria, resulting in withdrawing him from the study following an increase in the frequency and intensity of his aggression.

Preliminary Procedures

Functional Analysis

With each participant, we conducted a multielement functional analysis based on the procedures of Iwata et al. (1982/1994) to identify the variables maintaining their destructive behavior. Only individuals with socially reinforced destructive behavior (e.g., aggression maintained by access to leisure items) participated in the procedures noted below. See Table 1 for the functions targeted for each participant in this study. If destructive behavior also served another function other than the targeted one, it was treated separately outside of the context of this study.

FCT Evaluation

Following the functional analysis, we examined the use of FCT with extinction as a treatment for destructive behavior. This FCT evaluation occurred in a neutral context (i.e., without condition‐correlated stimuli). Following a baseline phase, we pretrained the FCR (i.e., card touch or card exchange) using a progressive prompt delay (0 s, 5 s, 10 s, 20 s) with specific prompting techniques (e.g., vocal, model, physical) and modalities (e.g., card touch, card exchange) tailored to the participants' individual abilities and selected in collaboration with their caregivers. We increased the prompt delay following two consecutive sessions without destructive behavior until FCRs occurred with at least 90% independence across two 10‐trial sessions in the absence of destructive behavior. We then evaluated FCT with extinction within a reversal design, with baseline phases identical to the functional analysis condition that maintained destructive behavior (i.e., fixed‐ratio [FR] 1 for destructive behavior) and FCT phases involving FR 1 reinforcement of FCRs and extinction for destructive behavior. Thus, the sequence of phases comprising each participant's FCT evaluation consisted of baseline, FCT with extinction pretraining, FCT with extinction, followed by a return to baseline, and then followed by a return to FCT with extinction. This evaluation was generally completed within two calendar days (M = 15 sessions), and FCT with extinction was demonstrated to be an efficacious treatment for each participant's destructive behavior.

Progressive‐Interval Assessment

We replicated the procedures of Fisher, Greer, Fuhrman, et al. (2018) for determining a lean variable‐interval (VI) reinforcement schedule for each participant for use during baseline and treatment that did not extinguish responding or generate adverse effects (e.g., a burst of destructive behavior or negative vocalizations). VI schedules of reinforcement are ideal for studying resurgence because they set an upper limit on the number of available reinforcers per unit of time (e.g., per 10‐min session), which helps to control rates of reinforcement across conditions, thus minimizing a known threat to internal validity when reinforcement rate is not the independent variable under investigation (cf. Sweeney & Shahan, ^2013b). The single‐session, progressive‐interval assessment involved delivering reinforcers according to a sequence of increasing fixed‐interval (FI) reinforcement schedules for destructive behavior (e.g., FI 2 s, 4 s, 8 s) until the researchers observed (a) three instances of destructive behavior within 5 s, (b) 5 s of continuous negative vocalizations or crying, or (c) two reinforcer deliveries at an FI 180‐s schedule, whichever came first (see also Miller et al., ²⁰²²). Results of this open‐ended session were used to select each participant's VI schedule of reinforcement for the remainder of the study. This resulted in a VI 2 s for Adonis, Craig, and Trevor and a VI 8 s for Arlo, each of which used the constant‐probability distribution described by Fleshler and Hoffman (1962).

Baseline

During baseline, the therapist delivered 20 s of reinforcement for destructive behavior according to the VI schedule that was identified by the progressive‐interval assessment. As in previous resurgence evaluations (Fisher, Greer, Fuhrman, et al., 2018; Greer et al., 2020), baseline ended after obtaining (a) at least five sessions, (b) a flat or upward trend in the rate of destructive behavior, and (c) a standard deviation for the last five sessions that was less than 50% of the mean for those five sessions. We did not include FCR materials during baseline.

FCT

The therapist delivered 20 s of reinforcement for FCRs according to the same VI schedule used in baseline. Destructive behavior was placed on extinction, and the therapist implemented a 3‐s changeover delay for FCRs that co‐occurred with destructive behavior. That is, if the participant engaged in destructive behavior immediately before emitting an FCR, the therapist waited for a subsequent FCR without co‐occurring destructive behavior prior to delivering reinforcement. Implementation of the changeover delay had no effect on the VI schedule (i.e., the timer continued). Short, moderate, and extended FCT durations ended following four, eight, and 32 sessions, respectively.

Resurgence Test

During the resurgence test, the therapist implemented extinction for both FCRs and destructive behavior for a minimum of three sessions. This phase ended following two consecutive sessions with at least an 85% reduction in destructive behavior from baseline or a maximum of 10 sessions, whichever came first.

Results

Figures 1, 2, 3, 4 display the participants' rates of destructive behavior and FCRs across baseline, FCT, and resurgence‐test phases. Panels within each figure show the results of one of the three progressions. The panels are ordered identically across participants (i.e., short, moderate, and then extended exposure to treatment) to facilitate comparison both within and across the participants; however, the order in which each progression occurred differed across participants. Numbers inset in each panel indicate the temporal order in which each progression occurred. Table 1 also displays this information for each participant. Table 2 displays information on the reinforcers delivered per phase for each participant.

Figure 2.

Resurgence Evaluations for Arlo

Note. Arlo's evaluations of resurgence following short (top panel), moderate (middle panel), and extended (bottom panel) functional communication training (FCT). The parenthetical numbers refer to the order in which Arlo experienced the experimental sequences.

Figure 3.

Resurgence Evaluations for Craig

Note. Craig's evaluations of resurgence following short (top panel), moderate (middle panel), and extended (bottom panel) functional communication training (FCT). The parenthetical numbers refer to the order in which Craig experienced the experimental sequences.

Figure 4.

Resurgence Evaluation for Trevor

Note. Trevor's evaluation of resurgence following extended functional communication training. ^aBaseline session terminated after 8 min. ^bResurgence‐test session terminated at 48 s. ^cResurgence‐test session terminated at 33 s; Trevor was withdrawn from the study after two consecutive sessions with termination criteria met.

Table 2.

Reinforcement Schedules and Obtained Reinforcers per Session

		Obtained reinforcers per session M(SD)
		Baseline			FCT			Resurgence test
Participant	VI schedule	Short	Moderate	Extended	Short	Moderate	Extended	Short	Moderate	Extended
Adonis	2 s	23.83 (0.75)	23.00 (1.73)	24.60 (0.54)	18.25* (12.17)*	23.75 (0.88)	23.84 (0.91)	0 (0)	0 (0)	0 (0)
Arlo	8 s	10.71 (3.63)	11.57 (4.11)	11.87 (3.35)	12.50 (4.35)	13.14 (2.91)	9.08 (4.88)	0 (0)	0 (0)	0 (0)
Craig	2 s	19.80 (2.28)	21.60 (0.89)	22.20 (1.64)	20.75 (1.89)	23.37 (0.74)	22.71 (2.50)	0 (0)	0 (0)	0 (0)
Trevor**	2 s	–	–	8.50 (0.83)	–	–	9.21 (0.70)	–	–	0 (0)

Note. FCT = functional communication training; VI = variable interval.

^* These values were affected greatly by the first FCT session in which Adonis earned zero reinforcers. With this session omitted, the M(SD) was 24.33(0.57).

^**We withdrew Trevor from the study after his behavior met termination criteria across two consecutive resurgence‐test sessions.

Adonis (Figure 1) displayed high and generally stable rates of destructive behavior in each of the three baseline phases, and FCRs quickly replaced destructive behavior during each FCT phase, even though FCR rates were consistently lower during FCT than was destructive behavior in baseline. Resurgence for Adonis was most pronounced following a moderate exposure to FCT, which was the first progression he experienced. Only a single instance of destructive behavior occurred in either of the resurgence‐test phases that followed.

Arlo (Figure 2) displayed variable but moderate rates of destructive behavior in each of his three baseline phases. FCT suppressed Arlo's destructive behavior well across short and moderate exposures to treatment, but these gains became less consistent during the extended exposure. Arlo experienced the extended‐treatment duration second, and his performance during FCT was improved both before (moderate‐treatment duration) and after (short‐treatment duration) this progression. Arlo's extended exposure to FCT lasted three additional sessions (i.e., 35 rather than 32 sessions) so that the phase ended with a reasonable reduction in rates of destructive behavior relative to his immediately preceding baseline phase and to permit the identification of resurgence upon transitioning to the resurgence‐test phase. Arlo displayed resurgence of destructive behavior following all three treatment durations. Average rates of destructive behavior were lowest following the moderate‐treatment duration (0.4 responses per minute), followed by the extended‐treatment duration (0.8 responses per minute), and then followed by the short‐treatment duration (1.1 responses per minute). Resurgence became more pronounced with repeated testing for Arlo. Obtained reinforcer differences across the three progressions (see Table 2) would not have suggested this finding.

Craig (Figure 3) displayed moderate and generally stable rates of destructive behavior in each of his three baseline phases, and FCT produced a rapid and consistent decrease in destructive behavior with stable FCR rates across exposures. Craig displayed resurgence of destructive behavior following all three treatment durations. Like Adonis, resurgence for Craig was most pronounced following a moderate exposure to FCT; but, unlike Adonis, Craig experienced the moderate exposure to FCT second rather than first. For Craig, resurgence was lower, much higher, and again lower when examining his results temporally. Like Arlo, obtained reinforcer differences across the three progressions for Craig (see Table 2) would not have suggested this finding.

Trevor (Figure 4) displayed low to moderate rates of destructive behavior in baseline. Because Trevor engaged in self‐injurious behavior that produced immediate risk of tissue damage in his fourth baseline session, we ended that session after 8 min. No other baseline or FCT session required termination. Unlike the other participants, multiple sessions of FCT were required before Trevor's rates of destructive behavior decreased, and the consistency of the treatment effect was less than ideal, even though his FCR rates were generally stable throughout his extended exposure to treatment. Resurgence of Trevor's destructive behavior was very pronounced, with response rates of destructive behavior typically exceeding those in baseline. We ended the resurgence test and withdrew Trevor from the study after his behavior met session‐termination criteria in two consecutive sessions. Those two sessions ended at 48 s and 33 s, respectively, after Trevor violently flopped to the padded floor. Despite having recently experienced an extended exposure to FCT, the omission of reinforcers for Trevor's FCRs quickly produced high rates and dangerous topographies of destructive behavior.

Figure 5 compares raw response rates of destructive behavior across paired consecutive sessions of the resurgence test from each test phase (e.g., first resurgence‐test session following the short exposure to treatment with the first resurgence‐test session following the extended exposure to treatment). Within each pair of sessions, we then calculated the difference in rates of destructive behavior (e.g., rates following the short exposure minus rates following the extended exposure). This simple calculation allowed us to compare rates of destructive behavior directly across resurgence‐test phases that followed different durations of treatment. Figure 5 shows the results of three different comparisons (i.e., response rates following short‐ and extended‐treatment durations, response rates following short‐ and moderate‐treatment durations, and response rates following moderate‐ and extended‐treatment durations). The shaded areas within each panel correspond to the ordinal predictions of BMT and RaC², which, despite their many differences, both state that resurgence should be lower with increased treatment durations and that the differences in resurgence should be most pronounced with the largest discrepancy in treatment duration (i.e., response rates following short‐ and extended‐treatment durations). Thus, responding within and across participants is predicted to fall within the shaded areas of each panel, and this occurs when responding is lower in the condition with the longer of the two exposures to treatment. Although this pattern of responding occurred for some participants within certain comparisons, no participant showed a consistent effect of time in treatment on rates of destructive behavior across the resurgence tests. Additionally, no comparison showed a consist effect of time in treatment on rates of destructive behavior across participants. Responding was equally likely to fall outside the predicted area across comparisons.

Figure 5.

Response Rate Differences Across Resurgence‐Test Pairs

Note. Differences in paired resurgence‐test sessions across short, moderate, and extended exposures to treatment. Shaded areas show where differences in response rates are ordinally predicted to fall according to Behavioral Momentum Theory and Resurgence as Choice. ^aLargest predicted difference.

Figure 6 presents a summary comparison of destructive behavior in the final session of each FCT exposure and all sessions of the resurgence test that followed when expressed as a proportion of baseline responding. We calculated the mean rate of destructive behavior per baseline phase for each participant. We then divided the rate of destructive behavior in the final session of each treatment by the average rate of responding in the preceding baseline phase. We then did the same calculation for each session of each resurgence test. This common approach for analyzing relapse also failed to show a consistent effect of time in treatment on destructive behavior across the resurgence tests when analyzed within participant and across participants.

Figure 6.

Proportion of Baseline Responding for all Participants

Note. Destructive behavior during the final session of functional communication training (FCT) and all sessions of the resurgence test across short, moderate, and extended exposures to treatment are expressed as a proportion of baseline responding. Note that Trevor's y‐axis is scaled differently. ^aResurgence‐test session terminated at 48 s. ^bResurgence‐test session terminated at 33 s.

Discussion

We evaluated the resurgence of destructive behavior following four, eight, and 32 sessions of FCT with extinction to more thoroughly evaluate whether increases in treatment duration might mitigate this form of relapse in the clinic. In so doing, the present study corrected limitations of a preliminary study on this topic conducted in the context of clinical service delivery. Greer et al. (2020) failed to detect an effect of treatment duration within or across participants when they used a multielement design to test two durations of treatment, one that was short and another that was triple the number of treatment sessions conducted for the short exposure to treatment. Short and long exposures to treatment in the Greer et al. study varied across participants. To address these limitations and incorporate the quantitative predictions of RaC² along with the recent empirical findings of Shahan, Browning, and Nall (2020), the present study (a) employed a sequential design to decrease the possibility of multiple‐treatment interference, (b) evaluated three treatment durations in a counterbalanced sequence across participants, (c) fixed the treatment durations evaluated within and across participants, and (d) programmed more extreme differences in treatment duration with new participants. Despite these improvements in experimental rigor and the testing of more extreme boundary conditions, the present study also failed to detect a mitigation effect of increased treatment duration.

One limitation of the present study is that it included fewer participants than the one conducted by Greer et al. (2020). In the Greer et al. study, the investigators enrolled six participants and reported the data on all six participants. Each participant in that study experienced two treatment durations, producing 12 comparisons across participants. The present study enrolled four participants, but three of those four participants experienced three treatment durations, and the fourth participant experienced one treatment duration, producing 10 comparisons across participants. The clear lack of an effect of treatment duration on the resurgence of destructive behavior within and across participants in the present study made it difficult to justify continued enrollment for the project. That our prior study also failed to detect an effect of treatment duration on the resurgence of destructive behavior supported this decision. Nevertheless, had future participants shown the predicted effect, those results would not have invalidated the findings from the previous 10 combined participants.

A second limitation of the present study is that our goal of controlling treatment duration occasionally conflicted with our broader goal of progressing from FCT with extinction to the resurgence test only when responding clearly supported the transition. As mentioned above, we observed an unfortunate breakdown in treatment efficacy during Arlo's extended exposure to treatment. This same pattern of responding did not occur in similar treatment phases of different durations that preceded and followed this one. However, such breakdowns in treatment efficacy can cloud the precise evaluation of relapse and its mitigation. This tradeoff in experimental rigor gained by tighter control over one aspect of the study and the loss of similar control in another can be particularly problematic for translational research conducted in applied contexts. However, for the present study, even though the implication for doing so was well understood in advance, it was essential to control treatment duration, as this was our independent variable.

Another limitation of the present study is that we were unable to evaluate the resurgence of Trevor's destructive behavior following all three treatment durations. We ended Trevor's participation in the study after terminating two consecutive sessions of the resurgence test that followed an extended exposure to FCT with extinction. Although Trevor's participation in the study ended prematurely, we report his data for a few important reasons. First, we believe that studies on relapse mitigation should include a good faith effort to provide every relevant dataset to the scientific community so that complete transparency can be achieved regarding the viability of promising relapse‐mitigation procedures for improving the durability of common treatments for destructive behavior. We have taken this approach throughout our work but have rarely stated this fact explicitly. Reporting Trevor's data allows the present study to be classified as a prospective, consecutive controlled case series (Hagopian, 2020).

A second reason for including Trevor's data is that such high and unsafe rates of destructive behavior following an extended exposure to FCT with extinction are not only inconsistent with the quantitative predictions of BMT that such a history should greatly protect against resurgence, but the data raise a related problem for clinicians. The fact that resurgence occurred for Trevor despite conducting 32 sessions of treatment raises questions about how many treatment sessions are necessary to achieve mitigation. If even longer exposures to treatment could have produced a mitigation effect with Trevor, this raises the possibility that the requisite number of treatment sessions could be prohibitively large, limiting the clinical practicality of this tactic as a viable resurgence‐mitigation procedure.

An equally troubling possibility regarding Trevor's data is that the high and unsafe rates of destructive behavior we observed with him following an extended exposure to treatment constituted a blunted episode of relapse in relation to what would have occurred following shorter treatment durations. An interesting point related to this discussion is that Trevor obtained a lower rate of reinforcement in baseline and in treatment than did the other participants (see Table 2). Ironically, Trevor's leaner schedule of obtained reinforcement in baseline and in treatment should have helped to mitigate resurgence for him relative to the other participants (cf. Fisher, Greer, Fuhrman, et al., 2018; Fisher et al., ²⁰¹⁹; Shahan, Browning, Nist, & Sutton, ²⁰²⁰).

A reasonable question when refocusing on the main findings of the present study, along with those reported by Greer et al. (2020), is whether quantitative theories of resurgence (i.e., BMT and RaC²) correctly predict the effects of increased treatment duration on resurgence. Although our clinical studies and most basic studies on the topic have failed to show this relation (e.g., Nall et al., ²⁰¹⁸; Winterbauer et al., ²⁰¹³), there are strong theoretical and empirical reasons to believe that the “true” effects of treatment duration alone on resurgence are sufficiently small as to evade detection in all but only highly powered (i.e., large‐n group designs) and well‐controlled laboratory experiments (cf. Shahan, Browning, & Nall, ²⁰²⁰; Smith & Greer, ²⁰²²). Both laboratory experiments and clinical assessments appear to indicate that the substantial resurgence mitigation predicted by BMT is incorrect. Conversely, RaC² suggests, and Shahan, Browning, and Nall (2020) empirically confirmed, only a very small effect of treatment duration on resurgence. Although this relation was detectable under highly controlled laboratory conditions, such conditions are clearly very different than those typically present in clinical applications. Although it might be possible to statistically detect a small effect of increases in treatment duration on resurgence in clinic applications employing a large‐scale group design (and thus provide further support for this prediction of RaC²), such an effect would likely be of little‐to‐no clinical utility in mitigating resurgence. As a result, there would be little clinical justification for conducting such a study. This consideration raises an important point about translation of the predictions of quantitative models and basic research findings to clinical settings. Such models make many predictions, some of which are characterized by large changes in behavior with changes in an independent variable, and because of the precision allowed by the models, others of which are characterized by very small changes in behavior. Clinicians would do well to focus their translational efforts on manipulations predicted to have effects that are large enough to produce meaningful behavior change under typical clinical conditions (see Greer & Shahan, ²⁰¹⁹, for additional tactics and elaboration).