Prader-Willi syndrome (PWS) is a neurodevelopmental disorder caused by the deletion or mutation of genes in the chromosome 15q11-q13 region (Butler 2011). Estimates suggest between 1 in 16–25,000 individuals are diagnosed with PWS each year in the United States (e.g., Burd et al. 1990; Butler 1990). Many individuals with PWS, specifically those in the deletion subgroup, have a preoccupation with food and an “insatiable appetite” which can lead to overeating, severe obesity (Ho and Dimitropoulos 2010; McAllister et al. 2011), and other associated health risks (e.g., Bianchine et al. 1971; Johnston and Robertson 1977).

Potentially harmful food-seeking behaviors exhibited by individuals with PWS can include obsessive compulsive behaviors, hoarding food, stealing food, rapidly consuming food, and pica (Ho and Dimitropoulos 2010; Page et al. 1983a, Page et al. b; Johnston and Robertson 1977; McAdam et al. 2004). These behaviors could lead to life-threatening health risks such as puncture or blockage of the digestive tract and gastric necrosis from pica, choking and stomach rupture from rapid eating behaviors, type II diabetes, respiratory problems, cardiovascular disease, and obesity (Bianchine et al. 1971; Ho and Dimitropoulos 2010; McAdam et al. 2004; McAllister et al. 2011; Page et al. 1983a, b; Johnston and Robertson 1977). In some cases, food-related problem behaviors lead to death. Death rates are higher for individuals with PWS than any other intellectual disability (Einfeld et al. 2006).

Obesity is a health problem for many Americans (Ogden et al. 2014) and is particularly prevalent among individuals with intellectual disabilities (Fox and Rotatori 1982; Rimmer and Yamaki 2006), including those with PWS (Butler 2011). Individuals with PWS often present with low muscle tone, contributing to low metabolic rate and decreased levels of physical activity (Butler 2011), providing a clear rationale as to why healthy food choices and consumption patterns are important for this population. Individuals with intellectual disabilities, especially individuals with PWS, are at risk of accumulating chronic health conditions due to food-related behaviors (Butler et al. 2002).

Currently, there are few pharmacological agents known to effectively treat the excessive hunger of individuals with PWS (Butler 2011; Griggs et al. 2015) and experts suggest caregivers keep small amounts of food in the home, lock refrigerators, pantries, and cabinets, and supervise food consumption (Butler 2011; Griggs et al. 2015; Ho and Dimitropoulos 2010). Although supervising the consumption of food is not an inherently instructional activity (McAllister et al. 2011), it often requires caregiver attention and effort and represents a context in which instruction could easily be provided. For example, while supervision is occurring, caregivers could intentionally arrange contingencies of reinforcement designed to establish healthy consumption patterns. Published examples of this include differential reinforcement of other behavior (DRO) in hospital settings for children with PWS to decrease food stealing behaviors (Page et al. 1983a, b) and differential reinforcement of low-rate behavior to slow the pace of eating for two adults with moderate intellectual disability in a group home setting (Echeverria and Miltenberger 2013).

Although effective, the interventions in these studies were not informed by functional analyses (FA; Iwata et al. 1982/1994) and did not explicitly incorporate functional reinforcement into programmed contingencies. Additionally, evidence of generality of therapeutic outcomes to different settings was limited (although see Page, Stanley et al.). Thus, we sought to contribute research demonstrating that an FA-informed intervention could decrease the food stealing of a young female diagnosed with PWS in both clinical and home settings by imbedding differential reinforcement procedures into mealtime routines.

Method

Participant and Setting

Leah was a 7-year-old female diagnosed with PWS. Leah was non-ambulatory without assistance but could move independently by “scooting” across the floor. She had a limited communication repertoire, consisting of some signs and one- to two-word phrases. Leah was referred to a university-based behavior clinic for a history of challenging behavior during mealtimes including self-injurious behavior (SIB), tantrums, and aggression. Per parent report, Leah consumed her food more rapidly than those around her, presenting the opportunity for food-stealing behaviors. During initial interviews with parents, it was reported that Leah did not have free access to food, had a strict diet, and consumed all meals in isolation from peers and family members with the assistance of one adult. Initial assessment and intervention trials were conducted in a clinic room equipped with a small table, chairs, and a one-way mirror. Generalization trials occurred at the kitchen table in Leah’s home.

Dependent Variables and Response Measurement

Primary dependent variables included food stealing, tokens earned, and token-board exchanges. We defined food stealing as an instance of Leah touching or grasping an edible item situated on another person’s plate and bringing it toward her mouth. During token-board training, we scored a token earned each time Leah independently satisfied contingencies for earning a token independent of physical prompting. We defined a token-board exchange as lifting a filled token board from off the table and placing it into the hands of another person (e.g., mother) in an appropriate manner (e.g., not forceful) independent of physical prompting.

Graduate students trained to fidelity using the video-based observer-training protocol outlined by Dempsey et al. (2012) collected data on all dependent variables and on therapist procedural fidelity (measures described below). During the FA, treatment, withdrawal, schedule thinning, and generalization conditions, observers used electronic software to collect data on latencies from trial onset to the first instance of food-stealing (all conditions) and token-board exchanges (treatment, schedule thinning, and generalization). During token-board conditioning (only), observers used paper-and-pencil data sheets to collect occurrence/non-occurrence data on tokens earned and token-board exchanges during trials partitioned into blocks of five (see rationale below).

Interobserver Agreement

A second trained observer independently collected data on all dependent variables across all conditions of this study. We calculated interobserver agreement (IOA) during each condition by comparing agreements between observers and dividing by the sum of agreements plus disagreements. With the exception of token-board conditioning, we scored an agreement for each active variable when latencies fell within ± 5 s. Otherwise, we scored a disagreement. During token-board conditioning, we scored agreements using a point-by-point method. Agreement between observers was 100% during Leah’s FA and was calculated for 55.5% (five of nine) of FA trials. During treatment, withdrawal, schedule thinning, and generalization conditions, agreement between observers was 91% and was calculated for 39.1% (18 of 46) of trials. During token board conditioning, mean IOA was 99.4% (range 90–100%) and was calculated for 100% (27 of 27) of blocks of trials.

General Procedures

Prior to assessment, researchers interviewed Leah’s mother. Per parent report, Leah engaged in challenging behaviors during meals (i.e., grabbing or stealing food items from plates, tantrums, SIB, or aggression) if she was around others who still had food after she had finished her meal or did not have immediate access to her food. Based on this description and observations during clinical appointments, we determined that behavioral escalations associated with food removal often started with attempts to steal food. Thus, we targeted food stealing for this intervention (Smith and Churchill 2002). During a second (non-food related) FA, we established a functional relation between aggression and parent attention. Treatment for aggression [i.e., serial functional-communication training (Lambert et al. 2015, 2017)] was unrelated to treatment for food stealing. However, we interspersed aggression treatment sessions with food-stealing treatment trials throughout the duration of each of 2 weekly 2 h clinical appointments that spanned the duration of this study (i.e., approximately 8 weeks).

Because a single instance of food stealing during meal times was problematic for Leah’s family, we conducted a latency-based FA (Thomason-Sassi et al. 2011) of food stealing and used relative increases and decreases in latencies across time to monitor progress and evaluate treatment efficacy (Caruthers et al. 2015). Thus, the entire treatment-validation process was conducted in a trial-based format. Each trial began following the onset of food-stealing’s establishing operation (i.e., access to food was removed) and ended after a target response occurred (i.e., food stealing or token-board exchange), or after a fixed period of time elapsed with no responding.

Functional Analysis

FA trials alternated between 2 min test and control phases which were individualized based on parent report (Hanley et al. 2014; Schlichenmeyer et al. 2013). During both phases, Leah was seated at a table with two confederate therapists who had plates full of puffed rice cereal (selected because it was a low-calorie snack which did not interfere with Leah’s dietary restrictions). To begin each trial, a third researcher placed a plate in front of Leah while data collectors began a timer. During the test phase, Leah’s plate was empty. During the control phase, Leah’s plate was full and was refilled continuously with puffed rice cereal as it was consumed. Trials ended (plates were removed) when Leah engaged in food stealing, or after 2 min had elapsed. Leah was allowed to consume “stolen” food.

Intervention

Leah’s intervention included response blocking and incorporated functional reinforcers into a conjunctive-schedules paradigm (Ferster and Skinner 1957) that entailed both DRO (i.e., the absence of food stealing) and differential reinforcement of alternative behavior (i.e., token-board exchanges). Intervention phases included token-board conditioning (during which we did not provide Leah with opportunities to steal food from other people’s plates), treatment (during which confederates with full plates of food were seated next to Leah), a quick withdrawal (to establish experimental control over treatment effects), schedule thinning (to increase the utility of treatment effects), and generalization (to increase the social validity of treatment effects). Intervention entailed an ABAB withdrawal design with an embedded changing-criterion design, during which we used data from the latency-based FA as baseline during our first “A” condition (Caruthers et al. 2015).

Prior to each trial (or block of trials), therapists conducted a one-trial preference assessment using a three-choice array that included edible items that aligned with Leah’s dietary restrictions and that had either been previously identified through a more comprehensive preference assessment (Cheerios and raisins) or by parental input (crisped rice cereal). Each item offered was easily consumed and low in calories.

Token-Board Conditioning

The token board was a half sheet of laminated paper with three Velcro® circles to attach plastic tokens. We required Leah to wait 3 s to earn each token. Thus, earning three tokens required Leah to wait for 9 s total, which was slightly above the average latency to food stealing during the FA.

To teach Leah how to use her token board, we employed a systematic fading procedure using a 5 s constant time delay across three stages of instruction (one token, two tokens, three tokens). During “one token”, Leah’s token board contained two out of three tokens and she was only required to earn one. During “two tokens”, Leah’s token board contained one out of three tokens and she was required to earn two. During “three tokens”, the token board did not contain any tokens and Leah was required to earn all three of her tokens before exchanging it for food.

Because we anticipated Leah might require a considerable amount of practice with programmed contingencies before the token system would exert control over food-related responding, we reasoned that a single-trial approach to mastery and data display would neither be advantageous nor efficient at this stage of instruction. Thus, we conducted token-board conditioning across blocks of five trials. As food removal (via consumption) naturally occurred following periods of reinforcement (setting the occasion for the next instructional trial), trials within blocks were conducted in rapid succession.

Initially, trials began with the primary therapist simultaneously placing the token board and a plate with a small amount of food (e.g., one raisin) in front of Leah, while another (prompt) therapist began a 3 s visual timer that Leah could see. At the end of a 3 s interval (when the timer beeped), the primary therapist said “token”, placed a token on Leah’s token board, and either restarted the timer (for Leah to earn another token) or (when the token board was full) pushed the board toward Leah, said “it’s full!”, and held out her hand. Contingent upon Leah handing over a full token board, the therapist allowed her to consume the food on her plate.

During the first block of trials at each stage of instruction, a prompt therapist held Leah’s hands down to ensure that she waited, without grabbing her food, until the primary therapist could deliver all relevant tokens. The prompt therapist then immediately manually guided a token-board exchange and allowed Leah to consume the food on her plate. During subsequent blocks of trials, the prompt therapist did not touch Leah’s hands while Leah waited for token intervals and waited 5 s from the delivery of the final token before manually guiding Leah to exchange the full token board. If Leah attempted to grab her food at any time before successfully exchanging a full token board, the prompt therapist reset the token timer and immediately manually guided all remaining desired responses (i.e., hands down while the rest of the tokens were delivered, token-board exchange after it was filled). This was done to ensure Leah could earn all relevant tokens without enduring extended periods of time without contacting reinforcement.

To progress to new a stage of intervention (i.e., two tokens, three tokens), we required Leah to independently exchange her token board during 100% of opportunities across two consecutive blocks of trials (this entailed never attempting to grab her food before exchanging a full token board across 10 consecutive trials). After Leah mastered three tokens, she advanced to treatment.

Treatment

The clinic room was arranged similarly to the way it was during the FA. Confederates were seated at the table next to Leah with plates full of puffed rice cereal. Trials began with confederates placing their full plates and Leah’s empty plate on the table while the primary therapist placed the empty token board in front of Leah and began the timer. Therapists then followed procedures similar to those described during the “three tokens” stage of token-board conditioning. Notable differences between these conditions were; (1) Leah did not have food on her plate; (2) the primary therapist manually guided completion of all remaining steps of the response chain contingent upon Leah attempting to steal food off of others’ plates (instead of her own); and (3) instead of permission to eat food already on her plate, token-board exchanges now produced a small amount of food on her previously empty plate (equal in quantity to amounts delivered during token-board conditioning and those available on the plates of confederates) that she could eat immediately. Unless she attempted to steal food, Leah was not prompted to exchange full token boards at this stage of intervention. Data were collected on latency to food-seeking behaviors (i.e., food stealing and independent token-board exchanges) and all trials ended with Leah earning food after satisfying token-board requirements (i.e., attempts to steal food were always followed by the manually-guided satisfaction of DRO requirements and token-board exchanges).

Withdrawal

These trials were identical to FA-test condition procedures in which full plates of food were placed in front of confederates and an empty plate was placed in front of Leah. Her token board and timer were not present. Trials ended following the first occurrence of food stealing, or after 2 min elapsed. Leah was allowed to consume “stolen” food.

Schedule Thinning

During schedule thinning, we systematically increased the total “wait” time required to fill the token board (9, 15, 21, 32, 48, 72, & 108 s) by increasing the time required to earn individual tokens (3, 5, 7, 11, 16, 24, & 36 s). Increases occurred every time Leah independently exchanged her token board, without engaging in food-stealing behavior, for three consecutive trials. Mastery entailed Leah independently exchanging her token board during four consecutive trials without food stealing at the terminal schedule value (i.e., 36 s per token). We set our terminal value at 36 s per token because we anticipated that this schedule would keep Leah on pace to finish her meal at approximately the same time as the rest of her family during their typical meal routine.

Generalization

The final phase consisted of generalization across people, contexts, and food. Leah’s mother served as the primary therapist for trials in the clinic and then served as primary therapist during trials at home. Trials in the clinic continued with the same procedures previously described. Trials in the home were conducted with similar procedures; however, Leah’s mother replaced snacks used at the clinic (e.g., puffed rice) with small portions of typical snack and dinner foods Leah normally ate at home. Leah’s other family members were also present during these trials. Therapists remained present to provide coaching and feedback as needed.

Procedural Fidelity

Observers used yes/no checklists to evaluate fidelity to programmed procedures during all conditions. Checklist items during the FA included correct materials present/absent, trial duration (± 5 s of prescribed time), correct motivating operations (Laraway et al. 2003), and correct consequences for problem behavior. Checklist items during intervention included correct materials present/absent, correct intervals enforced for each token delivery, correct prompts delivered, correct consequences (e.g., token delivery) at prescribed times (e.g., end of an interval), and reinforcers delivered following token-board exchange. We calculated fidelity by dividing the number of “yes” (marked when a procedure was implemented as programmed) by the sum of “yes” and “no” and multiplying by 100. Mean FA fidelity was 100% and was calculated for 44.4% (four of nine) of FA trials. Mean fidelity during token-board conditioning was 99% (range 92–100%) and was calculated during 40.7% of blocks of trials. Fidelity during all other treatment conditions was 100% and was calculated during 50% (23 of 46) of trials.

Results

We completed the FA of food stealing in nine trials, during one clinical appointment. We completed token-board conditioning after 27 blocks of trials conducted across seven clinical appointments, with a mean of 3.9 blocks of trials (range 2–6) conducted per appointment. We completed the treatment phase of this study after 46 trials conducted across five appointments, with a mean of 9.2 trials (range 3–14) conducted per appointment. Results of the FA are shown in Fig. 1 and depict latency to food stealing during both test (plate empty) and control (plate full) conditions. Food stealing occurred during every test trial with a mean latency of 6.3 s (range 4.8–8.7 s) and only once during control trials (with a latency of 70.8 s); indicating that food stealing was maintained by contingent access to food.

Fig. 1
figure 1

Results of latency-based FA of food stealing. Bx behavior

Full size image

Figure 2 shows the average number of tokens Leah independently earned (left Y-axis) and the percentage of token-board exchanges she independently emitted (right Y-axis) during token-board conditioning. Due to a 0 s prompt delay, Leah never had an opportunity to demonstrate independence during the first block of trials of each stage of training. However, across all three training phases, Leah quickly satisfied schedule requirements (15 blocks of trials during phase 1; nine blocks of trials during phase 2; three blocks of trials during phase 3). That is, she waited appropriately and independently to earn all tokens and then independently exchanged her token board before attempting to consume the food on her plate.

Fig. 2
figure 2

Results of token board conditioning procedure. Closed triangles are plotted using the left y-axis and open squares use plotted using the right y-axis

Full size image

Intervention results are shown in Fig. 3 and depict latency to both appropriate and inappropriate food-seeking behavior (i.e., food stealing and token-board exchanges). During the first treatment phase, food stealing occurred only one time (with a latency of 6.5 s). By the final three trials, food stealing was not observed and independent token exchanges consistently occurred with a mean latency of 17.6 s (range 12.1–23.6). During the final three trials of Leah’s return to baseline, the average latency to food stealing dropped back down to 4.8 s (range 3–6.8) but was quickly eliminated again during the second treatment condition. Afterward, food stealing only occurred one time (Trial 31), when token requirements increased from 16 s to 24 s. During schedule thinning, token requirements increased from 3 to 5 s, 7 s, 11 s, 16 s, 24 s, and 36 s and mean latencies to token exchanges during the final three trials of each stage were 16.1 s (range 15.4–17.1), 27.4 s (range 22.2–35.7), 32.2 s (range 28.7–37.1), 43.3 s (range 40.3–46.3), 66.9 s (range 64.3–68.2), 88.2 s (range 81.8–100.1), and 120.5 s (range 116.5–127.7), respectively. Treatment effects generalized to Leah’s mother in both clinic and home settings across a variety of preferred foods, with both confederates and family members next to Leah with full plates of food. Mean latencies to token-board exchanges during the final three trials of generalization in the clinic and home were 125 (range 119–130.3) and 126 (range 118.7–133.2).

Fig. 3
figure 3

Results of function-based intervention for food stealing. Bx behavior

Full size image

Discussion

Prior to intervention, Leah quickly finished her own plate of food during mealtime and immediately attempted to steal food from others’ plates. When attempts were blocked, Leah would often escalate to severe challenging behavior. Solutions previously attempted at home and school focused on eliminating free access to food and strictly controlling Leah’s diet and meal routines such that Leah did not eat meals with her family or peers at school and was fed by one adult. Whereas such solutions have been suggested in the literature (Griggs et al. 2015), they do not aim to teach new skills for continued success during mealtimes. The need for strict food monitoring for PWS in general can limit independence (McAllister et al. 2011), can limit social interactions with family members and peers during meals, and can affect family dynamics due to separately scheduled meals. By contrast, teaching Leah to appropriately wait for small portions of food across increasingly longer periods of time allowed Leah’s caregivers to extend the amount of time it took her to finish her own meal, thus eliminating food stealing and creating opportunities for Leah to interact positively with family members during meals. Social validity questionnaires administered upon discharge indicated that Leah’s mother was highly satisfied with the services provided and that she believed that both Leah and her family benefited from the interventions Leah received.

There are several limitations of this study. First, Leah was the only participant. Although experimental control was established over treatment effects in a variety of ways, future research may wish to expand these findings to a larger sample of participants. Next, our data collection system was incapable of detecting multiple instances of behavior within each trial. Thus, it is possible that Leah emitted considerably more instances of food stealing than what is shown in our graphs. Notwithstanding, previous research has demonstrated an inverse relation between rates and latencies during the assessment and treatment of problem behavior (Caruthers et al. 2015; Thomason-Sassi et al. 2011) and suggests it is valid to use changes in latencies across time as an indicator of treatment efficacy.

Although effects generalized across people (graduate students, family), contexts (clinic, home) and food preferences (snack, dinner), this research did not test effects in school settings. Additionally, this intervention included multiple components (i.e., differential reinforcement, a token board, response blocking) that may not have been necessary. Even though programmed components were primarily reinforcement-based, it is possible that response blocking functioned as a punisher. Previous research (e.g., Maglieri et al. 2000) has demonstrated that punishment-based procedures can suppress food stealing. Thus, future researchers might conduct a component analysis to determine which aspects of our intervention were effective.

Our study contributes evidence of the flexibility and utility of latency-based FAs by successfully addressing the food-stealing behavior of a child with PWS using a trial-based treatment-validation model in which interventionists reacted to response latencies (instead of rates) when engaging in data-based decision making. Additionally, it extends the literature by demonstrating that function-based interventions can help to mitigate some of the collateral effects of a genetic condition (i.e., PWS) and promote adaptive patterns of behavior that facilitate inclusion into typical family routines. Future researchers might evaluate the degree to which treatment gains from interventions like this could contribute to long-term positive health outcomes for individuals with PWS (cf. Page et al. 1983b).

Interestingly, our intervention shares points of contact with behavioral interventions for decreasing impulsivity [defined as selecting smaller sooner rewards (SSR) over larger later ones (LLR) when given a choice, see Madden and Johnson 2010]. Specifically, in a concurrent–operant arrangement in which participants could either choose SSRs or LLRs, Dixon et al. (1998) described a progressive-delay training procedure in which LLR choices were initially delivered immediately. Then, when participants consistently chose larger rewards over smaller ones, Dixon et al. gradually increased delays to LLRs until achieving a terminal value that suggested participants had learned to engage in patterns of choice making that were reflective of self-control. We too used a progressive delay training procedure that produced response patterns that might be interpreted as instances of self-control. However, our study differed from Dixon et al. in that food stealing was purportedly placed on extinction (and potentially punished) by the response blocking procedure and delayed reinforcers were not larger in magnitude than those available for food stealing. Notwithstanding, by the end of this study, our intervention shaped patterns of behavior that produced delayed reinforcers (i.e., “waiting” and token-board exchanges) to the full exclusion of behavior that had historically produced immediately reinforcement (i.e., food stealing). Future researchers may wish to evaluate whether treatment effects might generalize to a self-control paradigm by including a condition in which food stealing is no longer blocked and token-board exchanges produce a delayed but quantitatively (and/or qualitatively) superior reinforcement option.