Original Reports|Articles in Press

# Indoor or outdoor? Generalization of costly pain-related avoidance behavior to conceptually related contexts

Open AccessPublished:November 15, 2021

## Highlights

• Concept-based avoidance generalization to contexts was tested using an operant task
• Avoidance generalized to novel exemplars of the learned threat context category
• Generalization effects to novel contexts are limited to contingency-aware subjects
• Contingency-aware subjects also generalized pain-expectancy and pain-related fear
• Avoidance generalization in patients might emerge as result of category belonging

## Abstract

When pain persists beyond healing time and becomes a “false alarm” of bodily threat, protective strategies, such as avoidance, are no longer adaptive. More specifically, generalization of avoidance based on conceptual knowledge may contribute to chronic pain disability. Using an operant robotic-arm avoidance paradigm, healthy participants (N=50), could perform more effortful movements in the threat context (e.g. pictures of outdoor scenes) to avoid painful stimuli, whereas no pain occured in the safe context (e.g. pictures of indoor scenes). Next, we investigated avoidance generalization to conceptually related contexts (i.e. novel outdoor/indoor scenes). As expected, participants avoided more when presented with novel contexts conceptually related to the threat context than in novel exemplars of the safe context. Yet, exemplars belonging to one category (outdoor/indoor scenes) were not interchangeable; there was a generalization decrement. Posthoc analyses revealed that contingency-aware participants (n=27), but not non-aware participants (n=23), showed the avoidance generalization effect and also generalized their differential pain-expectancy and pain-related fear more to novel background scenes conceptually related to the original threat context. In contrast, the fear-potentiated startle response was not modulated by context.

## Introduction

In current fear-avoidance models, pain-related fear and avoidance behavior are put forward as key determinants in the transition from acute to chronic pain
• Crombez G
• Eccleston C
• Van Damme S
• Vlaeyen JWS
• Karoly P.
Fear-avoidance model of chronic pain: the next generation.
,
• Leeuw M
• Goossens MEJB
• Linton SJ
• Crombez G
• Boersma K
• Vlaeyen JWS.
The fear-avoidance model of musculoskeletal pain: Current state of scientific evidence.
,
• Vlaeyen JWS
• Linton S.
Fear-avoidance and its consequences in chronic musculoskeletal pain: a state of the art.
,
• Vlaeyen JWS
• Linton S.
Fear-avoidance model of chronic musculoskeletal pain: 12 years on.
. Generalization of fear and avoidance behavior becomes a maladaptive learning process when extending to safe situations or safe movements following a better safe than sorry strategy
• Van den Bergh O
• Brosschot J
• Critchley H
• Thayer JF
• Ottaviani C.
Better Safe Than Sorry: A Common Signature of General Vulnerability for Psychopathology.
,
• Vlaeyen JWS
• Crombez G.
Behavioral Conceptualization and Treatment of Chronic Pain.
. Avoidance behavior can generalize to other movements if it is expected that similar movements will have the same protective effect in that context
• Glogan E
• Gatzounis R
• Meulders M
• Meulders A.
Generalization of instrumentally acquired pain-related avoidance to novel but similar movements using a robotic arm-reaching paradigm.
. In addition, the same behavior can generalize to new contexts in order to ensure safety in novel environments
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
. Further, generalization can happen based on perceptual similarities
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
,
• Meulders A
• Harvie DS
• Moseley LG
• Vlaeyen JWS.
Generalization of Pain-Related Fear Using a Left–Right Hand Judgment Conditioning Task.
,
• Meulders A
• Vlaeyen JW.
The acquisition and generalization of cued and contextual pain-related fear: an experimental study using a voluntary movement paradigm.
or based on conceptual relatedness or category membership
• Dunsmoor JE
• Murphy GL.
Stimulus Typicality Determines How Broadly Fear Is Generalized.
,
• Dunsmoor JE
• Murphy GL.
Categories, Concepts, and Conditioning: How Humans Generalize Fear.
,
• Glogan E
• van Vliet C
• Roelandt R
• Meulders A.
Generalization and extinction of concept-based pain-related fear.
,
• Meulders A
• Bennett MP.
The Concept of Contexts in Pain: Generalization of Contextual Pain-Related Fear Within a de Novo Category of Unique Contexts.
,
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
. Although some studies have identified perceptual (fear) generalization
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
,
• Meulders A
• Harvie DS
• Moseley LG
• Vlaeyen JWS.
Generalization of Pain-Related Fear Using a Left–Right Hand Judgment Conditioning Task.
,
• Meulders A
• Vlaeyen JW.
The acquisition and generalization of cued and contextual pain-related fear: an experimental study using a voluntary movement paradigm.
, little systematic attention has been paid to conceptual generalization in the pain field. Yet, conceptual generalization may play a role in the development of chronic pain. For example, if a person experiences pain whilst sitting in a beer garden with friends, this person might avoid going with them to a club the next weekend as he/she expects pain to occur during social activities in general. Having a drink with friends can – based on conceptual knowledge – be considered a social activity, to which also belong game nights, dinner parties or going to a club with friends. In the same vein, conceptual generalization might result in the loss of a value-oriented life (e.g. social connectedness) in favor of the fear-avoidance vicious cycle
• Vlaeyen JWS
• Linton S.
Fear-avoidance and its consequences in chronic musculoskeletal pain: a state of the art.
,
• Vlaeyen JWS
• Linton S.
Fear-avoidance model of chronic musculoskeletal pain: 12 years on.
. There is first evidence for this type of fear (but not avoidance) generalization in the context of pain
• Glogan E
• van Vliet C
• Roelandt R
• Meulders A.
Generalization and extinction of concept-based pain-related fear.
,
• Meulders A
• Bennett MP.
The Concept of Contexts in Pain: Generalization of Contextual Pain-Related Fear Within a de Novo Category of Unique Contexts.
,
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
. However, pain-related fear and avoidance behavior do not necessarily converge considering that avoidance is more subject to motivational aspects
• Claes N
• Karos K
• Meulders A
• Crombez G
• Vlaeyen JWS.
Competing goals attenuate avoidance behavior in the context of pain.
,
• Wiech K
• Tracey I.
Pain, decisions, and actions: a motivational perspective.
. Indeed fear of pain might induce distress, but disability and associated costs only arise when (valued) activities are avoided as well
• Meulders A.
From fear of movement-related pain and avoidance to chronic pain disability: A state of the art review.
. In comparison to perceptual generalization, in which a generalization gradient as a function of perceptual similarity to the original stimuli can be observed
• Ghirlanda S
• Enquist M.
A century of generalization.
,
• Meulders A
• Harvie DS
• Moseley LG
• Vlaeyen JWS.
Generalization of Pain-Related Fear Using a Left–Right Hand Judgment Conditioning Task.
,
• Meulders A
• Vandebroek N
• Vervliet B
• Vlaeyen JW.
Generalization gradients in cued and contextual pain-related fear: an experimental study in healthy participants.
, in conceptual generalization the same reaction for trained and novel stimuli is expected when equivalent members of one category are used
• Dunsmoor JE
• Murphy GL.
Categories, Concepts, and Conditioning: How Humans Generalize Fear.
.
Importantly, besides response generalization to novel stimuli, contexts may have a modulatory function in operant conditioning
• Bouton ME
• Todd TP.
A fundamental role for context in instrumental learning and extinction.
. An acquired avoidance response might be required in one context, but not in another context. There is recent limited evidence for contextual modulation of pain-related avoidance behavior and its generalization to novel contexts based on perceptual similarities
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
. In a within-subjects design, black and white backgrounds were used as contexts during acquistiton. In the threat context, pain could be prevented by performing more effortful movements, whereas in the yoked (control) context, pain occurred independently of participants´ movements. Subsequently, responses were assessed in two novel generalization contexts (different shades of grey varying between the original threat and yoked context). In the scope of this study we adapted the robotic arm-reaching paradigm
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
,
• Meulders A
• Franssen M
• Fonteyne R
• Vlaeyen JWS.
Acquisition and extinction of operant pain-related avoidance behavior using a 3 degrees-of-freedom robotic arm.
and focused on avoidance generalization to novel contexts based on conceptual information. Participants had to infer overarching categories (indoor vs. outdoor) dividing different background pictures instead of perceptually simply noticing either black or white, which better reflect the complexity of real-life conditions. Further, instead of a yoked context a safe (control) context was used. We hypothesized that (1) novel exemplars of the threat context would elicit more avoidance behavior than novel exemplars of the safe context, (2) generalization of differential pain-expectancy and pain-related fear (verbal and psychophysiological measures) would be modulated by context, and (3) acquisition and novel (generalization) exemplars would be interchangeable/equivalent members of the learned category.

## Methods and materials

### Participants

Fifty-three pain-free, healthy volunteers gave written informed consent to participate in this study. The sample size was based on previous related research
• Meulders A
• Franssen M
• Fonteyne R
• Vlaeyen JWS.
Acquisition and extinction of operant pain-related avoidance behavior using a 3 degrees-of-freedom robotic arm.
. Volunteers were recruited using the departmental Experiment Management System (EMS) of the KU Leuven, through flyer-distribution and word-of-mouth. Psychology students were compensated either with course credits or a monetary compensation of €12; other volunteers always received financial compensation. Individuals were excluded from participating if they reported to suffer from chronic pain, pain or problems located at the wrist, elbow, or shoulder, any heart- or cardiovascular problem(s), neurological problem(s), other severe medical problem(s), clinical depression, panic or anxiety disorder, any other psychiatric problem (at present or in the past), and when they were asked by general practitioner to avoid stressful situations, had any type of electronic implant (e.g. pacemaker), were left-handed or pregnant. Three participants were excluded due to technical problems. The final sample included 50 participants (15 men; mean ± SD age = 23.31 years ±8.42, range = 17-53 years). The experimental protocol was approved by the Social and Societal Ethics Committee of KU Leuven (registration number: G-2017 11 990).

### Design

The experiment was carried out in a single test session and took around 90 minutes. The test session included 2 prepatory phases (the practice phase and the startle habituation phase, see supplementary material) and 5 main experimental phases: acquisition phase, acquisition test, reminder-of-acquisition phase, generalization phase, and generalization test (see Figure 1). We used a within-subjects design in which participants during the acquisition phase performed arm-reaching movements with a robotic arm both in a threat context and a safe context. In the threat context, participants could learn to avoid painful stimuli, whereas no painful stimuli were delivered in safe contexts and thus avoidance behavior was not necessary. Context was manipulated by using different background pictures of indoor and outdoor scenes. The assignment of scenes to the safe and threat context was counterbalanced across participants. Based on underlying categories (and not perceptually similarities) participants had to infer response-outcome contingencies in both contexts. During the generalization phase, we tested to what extent avoidance behavior generalized to novel contexts that were conceptually either related to original threat context exemplars or original safe context exemplars. Besides avoidance behavior, we measured pain-expectancy and pain-related fear (psychophysiological measure and self-report). Contexts were always presented in blocks of 12 trials (and corresponding background screens). Each block comprised either indoor or outdoor pictures and the sequence of blocks was semi-randomized, with the restriction that no more than two blocks of the same context could appear in a row.

### Stimulus material, apparatus and experimental task

#### HapticMaster

The Haptic Master (HM) is a 3-degrees of freedom haptic interface (FCS Robotics; Moog Inc, East Aurora, New York). The robotic arm is programmable and reacts to the force acted upon the sensor by the user in terms of a corresponding relocation. Since the HM is able to perform smooth movements, using the HM results in a very natural feeling. The HM is capable of tracking forces and movements (displacement and velocity) accurately during performed action. Additionally, it allows us to use this information (coordinates) to feedback into other systems, e.g. deliver painful stimuli or probes based on where participants are moving/positioned in the 3-dimensional space at a certain time. Even though the theoretically available range of motion of the HM is 3-dimensional, we determined a 2-dimensional horizontal movement area (plane of 36 × 36 cm) for this experiment.

#### Stimulus material

A 2-ms painful electrocutaneous stimulation was used as painful stimulus. It was controlled through a commercial constant current stimulator (DS7A; Digitimer, Welwyn Garden City, England) using a reusable bar electrode (8mm electrode diameter; 30mm distance between electrodes) attached on the triceps tendon above the elbow of the right arm. The intensity of the electrocutaneous stimulus was determined individually. Participants were given a series of stimuli of rising intensity and were requested to rate the respective intensity on a scale from 0 to 10 (0 = You feel nothing; 10 = This is the worst pain imaginable). Participants were encouraged to select a stimulus that is “painful and needs some effort to tolerate” roughly corresponding to an intensity of 8 on the pain calibration scale. The mean physical intensity of the electrocutaneous stimulus that was selected during the calibration phase was 36.51 mA (SD = 23.37, range = 8-99.99 mA).
The acquisition context was manipulated by using exemplars of two real-life categories, i.e. different pictures of indoor and outdoor scenes as background screens (see Figure 1). In total, acquisition contexts consisted of 12 pictures showing outdoor settings, e.g. nature or urban scenes and 12 pictures representing indoor scenes, e.g. private rooms or public places (CTXs). Taken together, generalization contexts (GTXs) in the generalization and generalization test phase consisted of 48 unique and novel background pictures of both real-life categories (24 indoor, 24 outdoor). It was ensured that the main similarity between acquisition context exemplars and generalization context exemplars was the membership to the indoor or outdoor category.

While holding the handgrip/sensor of the HM in the right hand, participants were asked to complete an arm-reaching task in the 2-dimensional horizontal movement area. They were provided with real-time visual feedback, which tracked the location of the sensor, on a 46-inch LCD screen (36PFL3208K/12; Koninklijke Philips N.V., Amsterdam, the Netherlands). The objective of the reaching task was to move the “green ball” from the starting point through one of three available mid-plane arches to the “target” (a green arch) at the end of the movement area. The arches represented three possible movement trajectories (T1-3) to reach the target (also see methods figures of Meulders, Franssen, Fonteyne, and Vlaeyen
• Meulders A
• Franssen M
• Fonteyne R
• Vlaeyen JWS.
Acquisition and extinction of operant pain-related avoidance behavior using a 3 degrees-of-freedom robotic arm.
). The movement was successfully completed when the ball went through the target arch and one unit was added to the overall trial counter located on top of the screen. The HM is programmed in a way that there was a linear relationship between the lateral displacement of the HM and the tractive force. Hence, T1 is the easiest trajectory to perform as no tractive force is applied by the HM, while trajectories T2 or T3, respectively, are accompanied by moderate and high tractive force. A visual signal together with an auditory signal were presented in order to indicate the beginning and end of each trial. The latter stayed on-screen to remind participants not to touch the robotic arm when it was repositioning to the start location. Back in its initial position the HM remained fixed for 2 s (intertrial interval; ITI) plus the time needed for ratings if there were instructions or startle probes before the beginning of the next regular trial.

#### Software

The experiment was programmed in C/C++, for which the Microsoft Integrated Development Environment Visual Studio (Microsoft Corporation Redmond, WA) and the development platforms OpenGL for graphical assistance was used. To control the HM the provided HM Application Programming Interface was used.

#### Experimental setting

Throughout the whole experiment the experimenter was present in an adjacent control room to allow communication with the participant if needed. Participants’ actions and their physiological responses were observed continuously via computer monitors and live broadcast through a webcam (Logitech v-uooo3) in the experimenter room.

### Primary outcome measures

#### Behavioral avoidance

To operationalize avoidance we used the maximal deviation of the shortest trajectory. The HM logged the deviation automatically. Larger deviations reflect more avoidance.

#### Anticipatory fear of movement-related pain and pain-expectancy ratings

Participants provided pain-related fear and pain-expectancy ratings (“To what extent are you afraid to perform this movement?” and “To what extent do you expect an electrocutaneous stimulus when performing this movement?”) during each phase of the experiment except the startle habituation phase (two ratings for each trajectory per block). Participants were instructed to answer on an 11-point rating scale with the scale ends 0 = “not at all” to 10 = “very much” by means of foot pedals (Windows 7 compatible triple foot switch, USB-3FS-2; Scythe, Tokyo, Japan), which was selected to limit interference with the arm-reaching task.

#### Fear-potentiated startle measures

Besides verbal ratings of fear of movement-related pain, we recorded the eyeblink startle (EMG) response as a psychophysiological index of conditioned fear. The startle response is a full-body reflexive reaction, which is present across species, and is part of a defensive response mobilization
• Lang PJ
Measuring emotion: Behavior, feeling, and physiology. In: Cognitive neuroscience of emotion. Series in affective science.
. It is facilitated when a person is in a fearful state and therefore reflects fearful arousal (i.e. fear-potentiated startle response). In this study, the startle reflex was evoked by acoustic startle probes (unexpected bursts of white noise, 50 ms duration and presented at 100 dBA), which were presented binaurally using headphones (HF92 Stereo headphones, Hoher, Sennheiser, Germany). In order to amplify the raw signal of the startle response a Coulbourn isolated bioamplifier (Coulbourn Instruments, Whitehall, PA, USA) was used. The eyeblink startle response was recorded on the orbicularis oculi muscle by surface electromyography (EMG). 3 Ag/AgCI Sensormedics electrodes (4mm), which were filled with electrolyte gel, were attached beneath the left eye and on the forehead after peeling the skin with exfoliating cream to improve the conductivity
• Blumenthal TD
• Cuthbert BN
• Filion DL
• Hackley S
• Lipp OV
• van Boxtel A.
Committee report: Guidelines for human startle eyeblink electromyographic studies.
. A bioamplifier with bandpass filter was set at 13 - 500 Hz (Lablinc V75-04 Isolated Bioamplifier with Bandpass Filter, Coulbourn Instruments, Pennsylvania, USA). The EMG signal was smoothed and rectified by a 4 channel integrator (LabLinc V76-24), analog input was digitized at 1000 Hz. The peak amplitude was defined as the maximum response within 21-175ms after the probe onset minus the EMG baseline score (average response of the first 20ms after probe onset).

### Experimental phases

#### Acquisition phase

The acquisition phase consisted of four blocks of training in the safe and four blocks in the threat context (see Figure 1). The movement-pain contingencies varied along with the background contexts. Participants only received painful stimuli in the threat context but not in the safe context. One category of background pictures (indoor vs. outdoor scenes) served as the safe context, the other as the threat context. In the threat context, the shortest and easiest movement trajectory was always accompanied by a painful stimulus (T1 = 0% negative reinforcement; no resistance), which was administered when approximately two-third of the trajectory was completed. Participants were given the opportunity to prevent painful stimuli by deviating from the shortest trajectory and choosing a more effortful movement (T2 = 50% negative reinforcement; moderate deviation and resistance; T3 = 100% negative reinforcement; largest deviation and strongest resistance).

#### Acquisition test

During the acquisition test, participants were probed to take a certain trajectory (by coloring the arch of the movement path of interest in blue) to test differential acquisition of the fear-potentiated startle response. The order was randomized. A startle probe was presented on each trial, 1 s after the start signal appeared.

#### Reminder-of-acquisition phase

This phase was inserted to remind participants of the original movement-pain contingencies that they acquired previously using all original background pictures. As in the acquisition phase, participants were provided the possibility to choose trajectories freely.

#### Generalization phase

Besides original threat and safe contexts, this phase consisted of generalization contexts with novel exemplars of both categories (GTX_threat for the threat category and GTX_safe for the safe category). These novel and unique exemplars were never followed by painful stimulation. Original background pictures were still reinforced in the same way as during the acquisition phase in order to prevent extinction. The presentation order of the blocks was randomized.

#### Generalization test

The generalization test was largely similar to the acquisition test as participants were probed which trajectory they were requested to perform. Again, the order of signaled trajectories was randomized. As in the generalization phase, again novel and unique exemplars of the indoor or outdoor categories were presented besides the original exemplars with their original reinforcement schedule. Additionally, on all trials startle probes were presented 1 s after start signal was given.

### Contingency awareness

Upon completion of the experiment, participants were requested to fill out a post-experimental questionnaire about their overall experience (see supplementary material for details). We used this questionnaire to assess participants contingency awareness as based on Lovibond and Shanks
• Lovibond PF
• Shanks DR.
The role of awareness in Pavlovian conditioning: Empirical evidence and theoretical implications.
, contingency awareness might be an important prerequisite for conditioned responding and thus also generalization effects. Participants were asked to indicate movement-pain contingencies for each trajectory (from 1 = least to 3 = most chance to be followed by an painful stimulus; not all numbers had to be used) for one exemplar of the threat context. Additionally, participants were asked if they noticed different background screens, if they could put them into categories and how they would describe these categories. Accordingly, contingency awareness was defined as awareness of movement-pain contingencies in the threat context (T1>T3) and awareness of categories (defined as being able to verbally describe the different categories). Based on this criterion, n=27 participants were contingency-aware (aware) and n=23 participants were non-contingency aware (non-aware).

### Data analysis strategy

To examine acquisition and generalization of avoidance behavior, pain-related fear, and pain-expectancy a series of repeated measures (RM) ANOVAs were run. Additionally, we computed planned comparisons to test our a priori hypotheses. As we assumed to find the largest difference between the trajectory with the highest pain probability and the trajectory with the lowest pain probability, we focused on the difference between T1 and T3, also see procedure of Meulders, Franssen and Claes
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
. To test the first hypothesis, we compared the avoidance data of GTX_threat with GTX_safe. Subsequently, to evaluate our second hypothesis, and to provide evidence of context modulation in the generalization of differential pain-expectancy and pain-related fear, the Trajectory x Context interaction and subsequent follow-up tests of GTX_threat and GTX_safe were analyzed. Finally, to test our third hypothesis and to evaluate if there was no generalization gradient, we compared responses of CTX_threat with responses of GTX_threat and responses of CTX_safe with responses of GTX_safe. Further, we computed secondary analyses in which we included contingency awareness as group factor in the main analyses and computed subgroup analyses. For multiple planned comparisons Holm-Bonferroni corrections were applied. Additionally, if sphericity was violated, Greenhouse-Geisser corrections were applied. For each significant effect uncorrected degrees of freedom and corrected p-values along with ε and the effect size $ηP2$ are reported. Statistical analyses were run on IBM SPSS software.

## Results

### Manipulation check: Acquisition of avoidance behavior, pain-expectancy and pain-related fear

Analyis of the avoidance behavior during acquisition showed that avoidance behavior gradually emerged in the threat context, but not in the safe context (see Figure 2A). Further, there was differential acquisition of pain-expectancy and pain-related fear, which was greater in the threat context than in the safe context. However, we found differential acquisition for pain-expectancy and pain-related fear in the safe context as well, that is participants also reported more fear and expected pain to occur more for T1 compared to T3 in the safe context. There was no contextual modulation in the fear-potentiated startle response. Detailed statistical analyses can be found in the supplementary material.

### Generalization of avoidance behavior, pain-expectancy and pain-related fear to novel contexts

To assess whether avoidance behavior generalized to novel contexts, we computed a RM ANOVA with Context (CTX_threat, GTX_threat, GTX_safe, CTX_safe) on the mean maximal deviation from the shortest trajectory in the generalization phase (see Figure 2B). As expected, this analysis yielded a significant main effect of Context, F(3, 47) = 15.30, p < .001, ε = .80, ηp2 = .238, indicating that avoidance behavior indeed varied between contexts. To further identify relevant context differences, we computed follow-up comparisons. These analyses showed that there was still a significant difference in avoidance behavior between the CTX_threat context and the CTX_safe context, F(1, 49) = 23.57, p < .001, ηp2 = .325, in the generalization phase, corroborating a successful transfer of selective avoidance behavior from the acquisition phase. Most crucially, avoidance behavior in the GTX_threat context was higher than in GTX_safe context, F(1, 49) = 5.34, p < .05, ηp2 = .099. This provides evidence for our central hypothesis that learned avoidance behavior would selectively spread to conceptually related new exemplars of the threat context, but not the safe context. However, planned comparisons demonstrated a significant decrement of avoidance behavior from the CTX_threat context to the GTX_threat context, F(1, 49) = 20.84, p < .001, ηp2 = .298, indicating that participants avoided less when they were presented new exemplars of the threat context than when old exemplars of the threat context were used as background sceneries. In contrast, there was no significant difference between avoidance behavior of the CTX_safe context and the conceptually related GTX_safe context, F(1, 49) = 3.97, p = .05.
To test whether pain-expectancy generalized to conceptually related, but novel contexts of the generalization phase, we carried out a 3 × 4 (Trajectory [T1-3] x Context [CTX_threat, GTX_threat, GTX_safe, CTX_safe]) RM ANOVA (see Figure 3A). There was a significant Context x Trajectory interaction, F(6, 44) = 10.42, p < .001, ε = .51, ηp2 = .175, suggesting that pain-expectancy ratings for the three available trajectories differed between contexts. For a better understanding of this interaction, planned contrasts were performed. To analyze whether the acquired differential pain-expectancy was still present in the generalization phase, we checked if there was a difference in pain-expectancy ratings between the CTX_threat and the CTX_safe context. Planned comparisons revealed that the mean T1-T3 difference in the CTX_threat context was still higher than in the CTX_safe context, t(49) = 4.01, p < .001, confirming that acquisition was still present. For a better understanding of responses in the generalization contexts, we compared the differential response in the GTX_threat context with the differential response in the GTX_safe context. Participants reported a higher pain-expectancy for T1 than for T3 not only in the GTX_threat context, t(49) = 6.95, p < 0.001, but also in the GTX_safe context, t(49) = 6.46, p < 0.001. Furthermore, there was no significant difference of pain-expectancy ratings for mean T1-T3 ratings between the GTX_threat and GTX_safe context. These results suggest that acquired pain-expectancy for different trajectories generalized to novel, but conceptually related contexts, but that differential pain-expectancy also generalized to the the safe context. Further planned comparisons revealed that the T1-T3 difference in the original CTX_threat context was larger than in the generalization GTX_threat context, t(49) = 3.97, p < 0.001. Yet, the T1-T3 difference did not differ significantly comparing the CTX_safe and the GTX_safe context.
To test pain-related fear generalization to novel contexts of the generalization phase we performed a similar 3 × 4 (Trajectory [T1, T2, T3] x Context [CTX_threat, GTX_threat, GTX_safe, CTX_safe]) RM ANOVA (see Figure 3B), which yielded a significant Context x Trajectory interaction, F(6, 44) = 6.86, p < .001, ε = .52, ηp2 = .123. To disentangle the effect we computed planned contrasts. Mean differential fear ratings (T1-T3) in the CTX_threat context were still higher than in the CTX_safe context, t(49) = 3.51, p < .001, confirming that acquisition was still present. Further analyses corroborated differential pain-related fear for T1 vs. T3 in the GTX_threat context, t(49) = 6.52, p < .001. However, participants also feared performing T1 more than T3 in the GTX_safe context, t(49) = 6.25, p < .001. Furthermore, mean differential fear ratings (T1-T3) did not differ between the GTX_threat and the GTX_safe context. Taken together, these results provide limited evidence for fear generalization to novel, but conceptually related contexts, as differential pain-related fear is shown in the GTX_safe context as well. Furthermore, the mean T1-T3 difference of fear ratings was higher in the CTX_threat context than in the GTX_threat context, t(49) = 3.26, p < .01. Yet, the mean T1-T3 difference of the CTX_safe and the GTX_safe context did not differ significantly.

### Secondary analyses comparing generalization effects in contingency-aware vs. non-contingency aware participants

To investigate the role of contingency awareness during generalization we first ran our main analyses again with contingency awareness as an additional between-subjects factor and then computed subgroup analyses for the crucial contrast between GTX_threat and GTX_safe. We computed a 2 × 4 RM ANOVA with Contingency Awareness (aware, non-aware) as between-subjects variable and Context (CTX_threat, GTX_threat, GTX_safe, CTX_safe) as within-subjects variable on the mean maximal deviation from the shortest trajectory (see Figure 4). This analysis revealed a main effect of Context F(3, 46) = 14.79, p < .001, ε = .81, ηp2 = .24, confirming the already reported effect that contexts in the generalization phase differed significantly. There was no main effect of Contingency Awareness nor an interaction of Context and Contingency Awareness. Nevertheless, we continued testing our crucial contrasts for both subsamples to further unravel potential influences of contingency awareness. Indeed, aware participants avoided significantly more in the GTX_threat than in GTX_safe context, F(1, 26) = 7.49, p < .05, ηp2 = .22, whilst non-aware participants did not differ in their avoidance behavior comparing the two generalization contexts.
To test the influence of contingency awareness on the generalization of differential pain-expectancy, we first performed a 2 × 4 [Contingency Awareness (aware, non-aware) x (CTX_threat, GTX_threat, GTX_safe, CTX_safe)] RM ANOVA on the mean T1-T3 anticipatory pain-expectancy as dependent variable (see Figure 5). This analysis revealed a Context x Contingency Awareness interaction, F(3, 46) = 3.62, p < .05, ηp2 = .19, indicating that aware participants and non-aware participants had different pain-expectancies for T1-T3 in the four contexts. To further disentangle this interaction, we computed our crucial contrasts in both subsamples. In the aware subsample, a significant mean T1-T3 pain-expectancy difference was found comparing the GTX_threat context with the GTX_safe context, t(26) = 2.80, p < .05. This contrasts the finding in the total sample, in which we did not find differential generalization based on conceptual relatedness of the generalization contexts with the acquisition contexts. Furthermore, non-aware participants did not differ in their mean T1-T3 pain-expectancies between the GTX_threat and the GTX_safe context, corrobating that contingency awareness influenced the generalization of responses to novel contexts.
We computed similar analyses for the generalization of pain-related fear. A 2 × 4 [Contingency Awareness (aware, non-aware) x (CTX_threat, GTX_threat, GTX_safe, CTX_safe)] RM ANOVA on the mean T1-T3 pain-related fear revealed a significant effect of Context, F(3, 46) = 7.68, p < .001, ε = .80, ηp2 = .14, and of Contingency Awareness, F(1, 48) = 6.98, p < .05, ηp2 =.13, confirming that aware participants and non-aware participants differed in their differential fear responses (see Figure 5). Yet, no significant interaction effect emerged for the interaction of Contingency Awareness and Context, F(3, 46) = 2.09, p = .114, ηp2 = .12, when all four contexts were taken into account. Nevertheless, we continued testing the crucial contrast between generalization contexts in further subsample analyses. In line with findings of pain-expectancy, in aware participants the difference between T1 and T3 was larger in the GTX_threat than in the GTX_safe context, t(26) = 2.27, p < .05, while in non-aware participants context did not modulate the magnitude of this difference.
Together, these results suggest that aware participants showed increased avoidance as well as stronger differential T1-T3 pain-expectancy and pain-related fear when presented with novel exemplars of the threat context compared to novel exemplars of the safe context, whilst non-aware participants did not.

## Discusssion

To our knowledge, this is the first study to investigate generalization of a learned pain-related avoidance behavior to novel contexts based on conceptual information. Besides self-reported fear and pain-expectancy ratings, we measured participants’ fear-potentiated startle response as a psychophysiological correlate of conditioned fear. We expected that (1) novel exemplars of the threat context would lead to more avoidance than novel exemplars of the safe context, that (2) generalization of differential pain-related fear and pain-expectancy would be modulated by context, and that (3) original and novel exemplars of both contexts would be equivalent members of the used categories.
First, we tested for acquisition and generalization of avoidance behavior. In line with our expectations, we were able to establish learning of pain-related avoidance behavior based on conceptual knowledge about contexts. Participants avoided more in novel contexts related to the threat context than in novel contexts belonging to the safe category, confirming our main hypothesis that avoidance behavior generalizes to novel contexts based on conceptual information, and in particular category membership. This result supports previous research, showing that acquisition of avoidance behavior can be modulated by context
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
. We extended these findings for a safe instead of a yoked context and for conceptual (indoor vs. outdoor) instead of perceptual (black vs. white) context modulation. Further, this study is the first to show generalization of pain-related avoidance behavior to novel contexts.
Second, to test whether participants differed in their differential pain-related fear and expectancy response comparing novel contexts of both categories, we tested differential pain-expectancy and pain-related fear ratings. Differential acquisition can be confirmed as context modulated the extent of differential pain-related fear and pain-expectancy. However, and contrary to our expectations, participants fear-potentiated startle response did not differ between the threat and the safe context. Discrepancies between different response systems have been reported before
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
,
• Soeter M
• Kindt M.
Dissociating response systems: Erasing fear from memory.
. One consideration previously made for conceptual fear generalization
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
is that the startle index as a short-latency psychophysiological response might not have been sensitive enough to identify small differences in a complex design, in which complex information might need more time to be processed. More importantly and contrary to our expectations, participants’ fear and expectancy ratings did not differ significantly between novel contexts of both categories. This is in contrast with previous studies
• Bennett M
• Meulders A
• Baeyens F
• Vlaeyen JWS.
Words putting pain in motion: the generalization of pain-related fear within an artificial stimulus category.
,
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
,
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
,
• Wong AHK
• Lovibond P.
Breakfast or bakery? The role of categorical ambiguity in overgeneralization of learned fear in trait anxiety.
, in which fear generalization based on higher order cognitions about underlying categories was found. In comparison to the study of Meulders et al.
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
differential responses in the present study were already generalized to the safe context during the acquisition phase. Contexts used in this study were more complex and needed elaboration on a higher cognitive level to be distinguished as safe or threatening. In addition to response-outcome contingencies contextual information had to be integrated. Possibly, participants were distracted from the diverse and maybe interesting contexts and therefore employed less exploratory behavior. Indeed, only 54% of participants were able to explicitly name both categories and movement-pain contingencies in the threat context. Interestingly, secondary analyses revealed that aware participants showed significant less T1-T3 differential pain-expectancies and pain-related fear in novel contexts related to the safe context than in novel contexts of the threat category. Non-aware participants did not differ in their mean T1-T3 ratings comparing novel examples of both categories.
Third, to test our last hypothesis, we compared acquisition contexts with novel (generalization) contexts. Contrary to our expectations, we found a generalization decrement of avoidance behavior in novel exemplars of the learned threat category compared to original exemplars of the threat context. In line with this, participants´ ratings indicated less T1-T3 differentiation of pain-expectancy and pain-related fear in novel contexts of the threat category compared to original threat contexts. However, there was neither an avoidance decrement nor a decrease of T1-T3 differential pain-expectancy and pain-related fear comparing original and novel contexts of the safe category. Thus, our last hypothesis was only partly supported. Original and novel examples of the safe category were interchangeable, whereas contexts of the threat category were not. As our sample consisted of healthy subjects, they might have come to explore contingencies in the novel contexts, which might have led to a general decrement of avoidance behavior and differential fear responses for novel contexts of the threat category. Safe contexts might have become interchangeable due to a floor effect. Another possibility is that the generalization gradient occured, because exemplars were not equal members of the category, as we assumed them to be. Typicality of contexts might have influenced how broadly avoidance generalized
• Dunsmoor JE
• Murphy GL.
Stimulus Typicality Determines How Broadly Fear Is Generalized.
. As a canary is a better exemplar of the category bird than an ostrich, a beautiful landscape might be a more typical exemplar of the category outdoor than an ordinary city capture. Trained and novel category exemplars might become interchangeable in de novo learned networks
• Meulders A
• Bennett MP.
The Concept of Contexts in Pain: Generalization of Contextual Pain-Related Fear Within a de Novo Category of Unique Contexts.
, whereas this might not hold for pre-existing categories.
Even though, descriptively fear and avoidance results evolved in the same direction, we only found a significant generalization effect in the behavioral index, but not in self-reported pain-expectancy or pain-related fear ratings. In addition and contrary to our expectation, fear also generalized differentially to the safe context. Secondary analyses gave us better insight into underlying processes. These analyses illustrated very clearly that contingency-aware participants differentiated in pain-related fear, pain-expectancy and avoidance responses between novel contexts of both categories, while non-contingency aware participants did not. Therefore, it seems that contingency-aware participants are responsible for the significant difference of avoidance behavior in the generalization contexts in the entire sample. However, findings related to contingency awareness shed light on the sensitivity of generalization responses. An important practical implication, which could be drawn from the finding of our study, is that people who are unaware of contingencies and contexts might generalize their avoidance behavior excessively representing one possible pathway to chronic pain disability. Not explicitely recognizing signals that predict safety (i.e. context) can lead to sustained anxiety
• Grillon C
• Baas JMP
• Lissek S
• Smith K
• Milstein J.
Anxious Responses to Predictable and Unpredictable Aversive Events.
. Participants who were unaware did not pick up the predictors for safety in the experiment and therefore might have differentially responded to safe contexts as well. Elevated, albeit unnecessary differential fear in novel contexts can be paralleled with what occurs in sustained anxiety and seems to be a factor in pathology. In comparison to anxiety, avoidance behavior does not get disconfirmed and consequently is prone to even further spread.
A few limitations should be addressed in combination with implications for future research. Despite thought-provoking evidence on the importance of cognitions in generalization, caution in the interpretation of the main results is warranted as only about half of the participants were contingency-aware. Besides possible implications on the absent context modulation of the fear-potentiated startle response and the fear generalization, context ambiguity increases the ecological validity of this paradigm. Due to its correspondence with the clinical pathology model
• Beckers T
• Krypotos A-M
• Boddez Y
• Effting M
• Kindt M
What's wrong with fear conditioning?.
,
• Lissek S
• Pine DS
• Grillon C.
The strong situation: A potential impediment to studying the psychobiology and pharmacology of anxiety disorders.
, this design conveys clinical relevance. Restrictively, contingency awareness in the post-experimental questionnaire was not only assessed with context exemplars from the acquisition phase but also from the generalization phase, implying that the obtained contingency awareness information should be interpreted cautiously. In addition, our study was not preregistered. A further source of contamination is that there were perceptual similarities not only in the task but also in the used contexts (e.g. sky outdoor, ceiling and walls indoor). Yet, colors, arrangements and structures overlapped between contexts and contexts from one category were perceptually very diverse. For instance, exemplars of the indoor category consisted of pictures of a concert hall, a kitchen, a sports hall, a living rooms and a bedroom to only name a few. Finally, our sample comprised mostly young and healthy students and therefore generalization of these learning processes to the entire population and especially to chronic pain patients is limited. Particulary the avoidance decrement might be smaller in participants with higher degrees of inflexibility who might show less exploratory behavior as it has been outlined for a similar paradigm before
• Glogan E
• Gatzounis R
• Meulders M
• Meulders A.
Generalization of instrumentally acquired pain-related avoidance to novel but similar movements using a robotic arm-reaching paradigm.
. Further, high levels of trait anxiety might lead to increased avoidance in safe contexts
• Hulsman AM
• Kaldewaij R
• Hashemi MM
• Zhang W
• Koch SBJ
• Figner B
• Roelofs K
• Klumpers F.
Individual differences in costly fearful avoidance and the relation to psychophysiology.
.
In conclusion, the results of this study indicate that pain-related avoidance behavior can be modulated by conceptually related contexts and that this modulation generalizes to novel exemplars of the learned categories. Interestingly, posthoc analysis revealed striking differences in generalization effects based on explicit contingency awareness. The aware subsample differentiated in avoidance behavior and differential pain-expectancy and pain-related fear between the generalization contexts of both categories, while the non-aware subsample did not. Importantly, this study is the first to demonstrate generalization of pain-related avoidance behavior to novel contexts based on category membership and thereby provides insights into underlying mechanisms of the spreading of avoidance behavior which is known to be a crucial contributor to chronic pain. Thus, exposure-based therapies could benefit from taking contexts and their conceptual belonging into account. To endorse these findings for chronic pain populations future studies with clinical samples are clearly recommended.

## Disclosures

The contribution of Ann Meulders was supported by a Senior Research Fellowship of the Research Foundation Flanders (FWO-Vlaanderen), Belgium (grant ID: 12E3717N) and by a Vidi grant from the Netherlands Organization for Scientific Research (NWO), The Netherlands (grant ID 452-17-002). The authors have no conflict of interests to report.

## Acknowledgements

The authos wish to thank Mathijs Franssen and Juliane Traxler for their technical assistance, Stien Meulders for assisting in the data collection, and the individuals who participated in the experiment. Part of the data was presented as at the German Pain Congress 2020, Mannheim, Germany, October, 2020 and at the International Congress of Behavioral Medicine 2021, Glasgow, June, 2021.

## References

• Beckers T
• Krypotos A-M
• Boddez Y
• Effting M
• Kindt M
What's wrong with fear conditioning?.
Biological Psychology. 2013; 92: 90-96
• Bennett M
• Meulders A
• Baeyens F
• Vlaeyen JWS.
Words putting pain in motion: the generalization of pain-related fear within an artificial stimulus category.
Frontiers in Psychology. 2015; 6: 520
• Blumenthal TD
• Cuthbert BN
• Filion DL
• Hackley S
• Lipp OV
• van Boxtel A.
Committee report: Guidelines for human startle eyeblink electromyographic studies.
Psychophysiology. 2005; 42: 1-15
• Bouton ME
• Todd TP.
A fundamental role for context in instrumental learning and extinction.
Behavioural Processes. 2014; 104: 13-19
• Lang PJ
Measuring emotion: Behavior, feeling, and physiology. In: Cognitive neuroscience of emotion. Series in affective science.
Oxford University Press, New York2000: 242-276
• Claes N
• Karos K
• Meulders A
• Crombez G
• Vlaeyen JWS.
Competing goals attenuate avoidance behavior in the context of pain.
The Journal of Pain. 2014; 15: 1120-1129
• Crombez G
• Eccleston C
• Van Damme S
• Vlaeyen JWS
• Karoly P.
Fear-avoidance model of chronic pain: the next generation.
The Clinical Journal of Pain. 2012; 28: 475-483
• Dunsmoor JE
• Murphy GL.
Stimulus Typicality Determines How Broadly Fear Is Generalized.
Psychological Science. 2014; 25: 1816-1821
• Dunsmoor JE
• Murphy GL.
Categories, Concepts, and Conditioning: How Humans Generalize Fear.
Trends in Cognitive Sciences. 2015; 19: 73-77
• Ghirlanda S
• Enquist M.
A century of generalization.
Animal Behaviour. 2003; 66: 15-36
• Glogan E
• Gatzounis R
• Meulders M
• Meulders A.
Generalization of instrumentally acquired pain-related avoidance to novel but similar movements using a robotic arm-reaching paradigm.
Behaviour Research and Therapy. 2020; 124103525
• Glogan E
• van Vliet C
• Roelandt R
• Meulders A.
Generalization and extinction of concept-based pain-related fear.
The Journal of Pain. 2019; 20: 325-338
• Grillon C
• Baas JMP
• Lissek S
• Smith K
• Milstein J.
Anxious Responses to Predictable and Unpredictable Aversive Events.
Behavioral neuroscience. 2004; 118: 916-924
• Hulsman AM
• Kaldewaij R
• Hashemi MM
• Zhang W
• Koch SBJ
• Figner B
• Roelofs K
• Klumpers F.
Individual differences in costly fearful avoidance and the relation to psychophysiology.
Behaviour Research and Therapy. 2021; 137103788
• Leeuw M
• Goossens MEJB
• Linton SJ
• Crombez G
• Boersma K
• Vlaeyen JWS.
The fear-avoidance model of musculoskeletal pain: Current state of scientific evidence.
Journal of Behavioral Medicine. 2007; 30: 77-94
• Lissek S
• Pine DS
• Grillon C.
The strong situation: A potential impediment to studying the psychobiology and pharmacology of anxiety disorders.
Biological psychology. 2006; 72: 265-270
• Lovibond PF
• Shanks DR.
The role of awareness in Pavlovian conditioning: Empirical evidence and theoretical implications.
Journal of Experimental Psychology: Animal Behavior Processes. 2002; 28: 3-26
• Meulders A.
From fear of movement-related pain and avoidance to chronic pain disability: A state of the art review.
Current Opinion in Behavioral Sciences. 2019; 26: 130-136
• Meulders A
• Bennett MP.
The Concept of Contexts in Pain: Generalization of Contextual Pain-Related Fear Within a de Novo Category of Unique Contexts.
Journal of Pain. 2018; 19: 76-87
• Meulders A
• Franssen M
• Claes J.
Avoiding Based on Shades of Gray: Generalization of Pain-Related Avoidance Behavior to Novel Contexts.
The Journal of Pain. 2020; 21: 1212-1223
• Meulders A
• Franssen M
• Fonteyne R
• Vlaeyen JWS.
Acquisition and extinction of operant pain-related avoidance behavior using a 3 degrees-of-freedom robotic arm.
Pain. 2016; 157: 1094-1104
• Meulders A
• Harvie DS
• Moseley LG
• Vlaeyen JWS.
Generalization of Pain-Related Fear Using a Left–Right Hand Judgment Conditioning Task.
Behavior Therapy. 2015; 46: 699-716
• Meulders A
• Vandael K
• Vlaeyen JWS.
Generalization of Pain-Related Fear Based on Conceptual Knowledge.
Behavior Therapy. 2017; 48: 295-310
• Meulders A
• Vandebroek N
• Vervliet B
• Vlaeyen JW.
Generalization gradients in cued and contextual pain-related fear: an experimental study in healthy participants.
Frontiers in human neuroscience. 2013; 7
• Meulders A
• Vlaeyen JW.
The acquisition and generalization of cued and contextual pain-related fear: an experimental study using a voluntary movement paradigm.
Pain. 2013; 154: 272-282
• Soeter M
• Kindt M.
Dissociating response systems: Erasing fear from memory.
Neurobiology of Learning and Memory. 2010; 94: 30-41
• Van den Bergh O
• Brosschot J
• Critchley H
• Thayer JF
• Ottaviani C.
Better Safe Than Sorry: A Common Signature of General Vulnerability for Psychopathology.
Perspectives on Psychological Science. 2020; 16: 225-246
• Vlaeyen JWS
• Crombez G.
Behavioral Conceptualization and Treatment of Chronic Pain.
Annual Review of Clinical Psychology. 2020; 16: 187-212
• Vlaeyen JWS
• Linton S.
Fear-avoidance and its consequences in chronic musculoskeletal pain: a state of the art.
Pain. 2000; 85: 317-332
• Vlaeyen JWS
• Linton S.
Fear-avoidance model of chronic musculoskeletal pain: 12 years on.
Pain. 2012; 153: 1144-1147
• Wiech K
• Tracey I.
Pain, decisions, and actions: a motivational perspective.
Frontiers in Neuroscience. 2013; 7: 46
• Wong AHK
• Lovibond P.
Breakfast or bakery? The role of categorical ambiguity in overgeneralization of learned fear in trait anxiety.
Emotion. 2020; 21: 856-870