Investigating whether socially acquired nocebo effects can spread to other treatments

Abstract

Observation of another’s treatment side effects can elicit side effects in the observer, even when the treatment is a placebo. This study investigated whether these socially acquired side effects can generalise to similar treatments. Healthy volunteers (N = 120) participated in a study ostensibly comparing the effect of two cognitive enhancers (placebos). Participants were randomised to one of four experimental groups. The three treatment groups comprised: social modelling of side effects associated with the same treatment; social modelling of side effects associated with the different treatment; and a verbal suggestion only group (i.e., no social modelling). The fourth group was a no-treatment control group. The primary outcome was severity of side effects reported. Groups that received placebos reported increased symptom severity, i.e., showed a nocebo effect. Surprisingly, primary outcome analysis revealed no significant enhancement of the nocebo effect due to social modelling. However, there was an additive effect of social modelling on general side effects (planned secondary outcome) and specifically for headaches and dizziness (exploratory analysis), both of which generalised across treatments. Therefore, preliminary findings suggest that socially induced nocebo side effects may not always occur, but when they do, they can generalise beyond identical treatments. This warrants replication and raises significant concern given the widespread sharing of treatment-related information, potentially contributing to the societal burden of nocebo effects.

Introduction

The nocebo effect refers to adverse health outcomes that cannot be attributed to the treatment itself but rather are due to the psychosocial context in which they occur^1,2. Growing evidence indicates that social information is a key contributor to the nocebo effect³. Simply observing another person (a ‘model’) experience treatment-related pain, nausea, or itch can elicit or exacerbate these symptoms when the observer subsequently undergoes the same treatment. These socially acquired nocebo effects can even be demonstrated when the treatment is itself a placebo ^4–6, and have been implicated in the formation of COVID-19 vaccine side effects^7,8. However, research has yet to determine if socially acquired nocebo effects can spread – or generalise – from one treatment to another. This is critical for understanding the full impact that socially acquired nocebo effects have, as well as for developing strategies to reduce their burden in healthcare and community settings.

Generalisation has been reliably demonstrated in the context of direct experiential learning both in animals and humans⁹. For example, rats trained to learn that a specific tone leads to shock will display a defence response to that specific tone and other similar tones¹⁰. Recent research has also shown that directly-conditioned placebo and nocebo effects generalise across treatment cues^11–13 and new environmental contexts¹⁴. However, there has been virtually no research on the extent to which socially acquired nocebo effects generalise. The only exception that we are aware of is a study by Saunders et al.⁶, which found that observing a model report cybersickness to one virtual reality (VR) setting (a rollercoaster), subsequently increased the observer’s nocebo cybersickness both to that setting and another (a VR flight simulation). This demonstrates that socially acquired nocebo effects can spread across contexts. However, to the best of our knowledge, there is no research examining whether socially acquired nocebo effects generalise across medical treatments (e.g., different types of treatments).

Generalisation of socially acquired nocebo effects across treatments is both highly concerning and relevant in the context of the significant proliferation of, often negative, information on social media. Seeing or hearing about another’s experience of side effects after one type of vaccination may not only lead to nocebo side effects for that specific vaccination—but could also increase nocebo side effects for other vaccinations. Previous research has shown that socially acquired symptoms arising from a placebo treatment can spread beyond the specific symptoms communicated by the model¹⁵. However, as models and observers received the same treatment, this is distinct from generalisation of socially acquired nocebo effects across treatments. As such there is a significant gap in our knowledge regarding the social transmission of nocebo effects when the model/observer intervention differs.

The current study examined whether nocebo side effects could be elicited via social modelling even when the model apparently received a different treatment. This involved simulating a clinical context by administering one of two placebo treatments (capsules) presented as cognitive enhancers, each of which was described as having a unique side effect profile. One was described as associated with “Headaches and Dizziness” and the other “Nausea and Stomach Discomfort”. Participants were randomised to one of four groups. The three treatment groups comprised: social modelling of side effects associated with the same treatment; social modelling of side effects associated with the different treatment; and a verbal suggestion only group (i.e., no social modelling). The fourth group was a no-treatment control group. We hypothesised that there would be an overall nocebo effect, with more side effects reported in the treatment groups than the control group. Most importantly, however, we hypothesised that social modelling would exacerbate this nocebo effect both when the same treatment was received by the observer and when a different treatment was received; the latter of which would indicate generalisation of socially acquired nocebo effects to a novel treatment with unique side-effects. Heart Rate Variability (HRV) and Electrodermal Activity were also collected to explore any effects of social modelling on physiological arousal. See Daniali et al. ¹⁶ for a review with respect to nocebo hyperalgesia.

Methods

The study design and analyses were preregistered (aspredicted.org #109597). Ethics approval was granted by the University of Sydney Human Research Ethics Committee (#2022/532), all methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants.

Participants

One hundred and twenty-one healthy volunteers (Female = 87, M_age = 19.52, SD = 2.28) from the University of Sydney participated and received course credit as compensation. Eligible participants were fluent in English without any known allergy to medication or lactose. Due to the gelatine capsules, those with dietary restrictions were advised not to participate. Participants were required to be in full health at the time of testing and were excluded if they exceeded pre-registered thresholds of physical symptoms at baseline.

Design

The study involved a single-blind one-way between-subjects design with four levels. Participants were told the study was investigating two cognitive enhancers (both actually placebos) of which they may receive one. Participants were warned that one of the treatments was associated with “headaches and dizziness” and the other with “nausea and stomach discomfort”. The two different side effect profiles, along with the supposed name and branding of the medication container served to distinguish the treatments as distinct. The critical manipulation was if the participant observed the confederate experience side effects, and for those who did, which treatment the confederate had supposedly taken, with participants randomly allocated to one of four groups. R 4.1.1¹⁷ was used as a random number generator to generate a random sequence of group assignment with which to allocate participants in the order in which they were tested. The Social Modelling Consistent group received placebos and observed side effects reported by a model who they believed had taken the same treatment as them. A Social Modelling Inconsistent group received placebos and observed side effects reported by a model who they believed had taken a different treatment to them. A No Social Modelling group received placebos but did not observe a model. A Natural History group did not receive placebos and did not observe a model. This group served as a control, allowing for a comparison with the three treatment groups to isolate the nocebo effect. Any difference in reported symptoms between the Natural History group and the treatment groups would indicate the presence of a nocebo effect overall. The key outcome of interest was the severity of side effects reported by participants.

Materials and measures

Placebo pills

No participants received active medication. The placebos comprised gelatine capsules filled with lactose. The placebo pills were contained in two distinct fake medication bottles designed to reinforce the cover story and the primary manipulation, see https://osf.io/m8tdh/.

Demographics

Participant age and gender were recorded.

Physical symptoms

Physical symptoms were assessed using a modified version of the General Assessment of Side Effects, GASE¹⁸. Symptoms communicated by the confederate (i.e., stomach discomfort) were added to the questionnaire and symptoms unlikely to occur within the time frame of the experiment were excluded (e.g., diarrhea, insomnia) resulting in a list of 10 symptoms. The original 4-point scale 0(Not Present) to 3(Severe) was modified to a 7-point scale to enhance the sensitivity to changes in symptoms. The full scale was decomposed into three scores: Target Symptoms (Mean severity of symptoms supposedly associated with the participants medication), Non-Target Symptoms (Mean severity of symptoms supposedly associated with the other medication) and General Symptoms (Mean severity of remaining 6 GASE items). The Target Symptoms for participants in the Natural History group (who did not receive medication) were yoked to the Target Symptoms of the previous participant in the No Social Modelling group.

Generalized state anxiety

General state anxiety was measured via the Spielberger State-Trait Anxiety Inventory-6¹⁹, (Cronbach’s alpha = 0.81).

Side effect specific anxiety

Participants rated their anxiety concerning experiencing side effects on the single item measure: “How anxious are you about experiencing adverse events (e.g., side effects) as a result of participating in this study?” on a visual analogue scale (VAS) ranging from 1(Not Anxious) to 100(Very Anxious).

Expectancy

Participants were asked to rate their expectancy of the experience of side effects on the single item measure: “How much do you expect to experience adverse events (e.g., side effects) as a result of participating in this study?” on a VAS ranging from 1(Not at all) to 100(Very much so).

Expectancy for cognitive enhancement

Participants were asked to rate their expectancy of the efficacy cognitive enhancement medication on the single item measure: “To what degree do you expect the cognitive enhancement medication to enhance your performance?” on a VAS ranging from 1(No enhancement) to 100(Significant enhancement).

Cognitive performance

The Rapid Visual Information Processing (RVIP) procedure was used to assess sustained attention²⁰. Numbers ranging from 1 to 9 were presented on a computer screen at a rate of 100/min for a total duration of five minutes. Participants were required to press the space bar as quickly as possible once either three consecutive even or three consecutive odd numbers appeared. They had 1.5s to make a correct response; all responses outside this time were considered false alarms. The sequence of numbers was semirandom such that the target sequences were separated by a minimum of 5 and maximum of 33 digits. Performance was assessed using proportion (%) of correct responses.

Self-reported cognitive performance

Participants were asked to rate their perceived performance on the RVIP: “How would you rate your performance on the cognitive task?” on a VAS ranging from 1(Very poor) to 100(Very good).

Self-reported influence on cognitive performance

Participants assigned to take the placebo were asked “How effective was the medication at enhancing your cognitive performance?” on a VAS ranging from 1(Not effective at all) to 100(Very effective).

Manipulation check

To ensure participants were aware of their assigned medication they were asked to recall the name of their assigned medication (I was assigned to…: ‘Vitatil’, ‘Monovigil’, ‘No Treatment Control’, ‘I took one of the medications but do not recall which one’ and ‘I do not recall if I received treatment’). To probe more generally about suspicion concerning the confederate they were asked to answer the general probe “Briefly describe (in 2–3 sentences) what you thought the purpose of the experiment was:”.

Equivital sensor belt and module

Participant Heart Rate (HR) was measured using an Equivital Sensor Module fitted onto an Equivital Sensor Belt. HR was measured continuously for 15 min after administration of the treatment (or lack thereof). HR Variability (HRV) is a measure of the parasympathetic and sympathetic branches that modulate cardiac activity¹⁶. Frequency domain methodology was used to generate the High Frequency/Low Frequency (HF/LF) ratio using pyHRV²¹.

Electrodermal activity (EDA)

EDA was recorded as a measure of autonomic arousal in response to the treatment and social communication manipulations. This was achieved by using a PowerLab amplifier and a Galvanic Skin Response amplifier (ADInstruments) with two finger electrodes placed on the middle and index fingers of participants’ right hand. EDA was recorded over the same time period as HR. EDA can be decomposed into tonic activity (slower general changes in sympathetic arousal) and phasic activity (reactive fast changes), which each reflect important components of physiological arousal²². A convex optimization approach to electrodermal activity (cvxEDA) was applied to extract the two components²³, after which mean Skin Conductance Level ( SCL) was calculated as a measure of the tonic activity and the number of Skin Conductance Responses (nSCR) as a measure of the phasic activity.

Procedure

Refer to Fig. 1 for a flow chart of the study procedure. All participants were tested individually in a single 1 h session conducted by a single experimenter who was a 24-year-old white Australian female. On arrival, participants received a written information sheet, which was reinforced verbally. This outlined the cover story and contained the warning: “Vitatril has been associated with the experience of mild nausea and stomach discomfort. Monovigil has been associated with the experience of mild headaches and dizziness”. Side effect profiles were counterbalanced within each group. Participants who consented were then randomised and verbally informed of their treatment allocation by the researcher (i.e., No Treatment, ‘Vitatril’ or ‘Monovigil’). Next, participants were fitted with the Equivital Harness. Participants then completed demographics and baseline measures including: physical symptoms, state anxiety, expectancy for side effects and expectancy for cognitive enhancement. Participants were seated facing the door and set up with the electrodes to measure their EDA. The placebo capsules were then administered to treatment groups. All participants were seated and waited for 15 min, ostensibly providing time for the medication to take effect in the treatment groups. Exactly five minutes after the HR/EDA recording commenced, participants in the social modelling conditions saw a confederate enter the room. The confederate was a 26-year-old Asian-Australian male perceived as another participant of the study. Within view of the participant, the researcher asked the confederate, “So you received [‘Vitatril’/‘Monovigil’] around thirty minutes ago, how are you feeling now?”. Where the placebo medication either aligned (Social Modelling Consistent) or did not align (Social Modelling Inconsistent) with the supposed treatment the participant had just taken. The confederate then responded with either: “Not great, definitely feeling headachy and a bit dizzy” or “Not great, definitely feeling quite nauseous and my stomach is upset”. Importantly, the side effects the confederate reported aligned with the warnings participants had received concerning the treatment the confederate had supposedly taken. At the same timepoint as the social modelling conditions received this side effect modelling, participants in the Natural History and No Social Modelling groups overheard the experimenter have an irrelevant phone call to control for the effect of the conversation. After the 15-min wait period, all participants were taken to another room to complete the cognitive task to uphold the cover story. Upon completion, participants returned to the main room where they completed the physical symptoms questionnaire again and were also asked to report their perceived efficacy of the medication and complete a manipulation check. All participants then received a written debrief informing them about the true aims of the study.

Fig. 1.

Flowchart of study procedure.

Power and data analysis

We planned to recruit a total of 120 participants (30 participants per-group). This was based on a power analysis on the effect size for the social modelling manipulation (f = 0.31, derived from Faasse, 2015) with an alpha of 0.05 with 80% power. Of the 121 participants recruited, one participant was excluded from analysis as their sum-scored baseline physical symptoms exceeded the pre-registered threshold, resulting in a final sample of 120 participants.

Primary data analysis

Statistical analysis was conducted using R 4.1.1¹⁷. The primary data analysis consisted of a one-way (Condition: Natural History, No Social Modelling, Social Modelling Consistent, Social Modelling Inconsistent) ANOVA, using the Target Symptom difference score (Active – Baseline) as the outcome. Orthogonal contrasts were used to determine:

1. Whether there was an overall nocebo effect—No Treatment (Natural History) vs. Treatment (No Social Modelling, Social Modelling Consistent, Social Modelling Inconsistent).

2. The effect of social modelling, above and beyond explicit instruction alone—No Social Modelling vs. Social Modelling (Social Modelling Consistent and Inconsistent).

3. The effect of generalisation—Social Modelling Consistent vs. Inconsistent.

Secondary analysis examined Non-Target Symptoms and General Symptoms as the outcome variable, in similar one-way ANOVAs. Self-reported and actual cognitive performance scores were compared between groups using one-way ANOVAs to assess the presence of a placebo effect but were not the focus of the study.

Analysis of physiological measures

HRV and EDA were extracted in two non-overlapping five-minute periods: (1) the first five minutes of the post-capsule waiting period (“Time 1”) and (2) the second five minutes (“Time 2”, immediately after the model communication in the social modelling groups). The data was analysed using a two-way mixed ANOVA comprising the within-subject factor Time (Time 1, Time 2) and the between-subject factor Condition. The same three orthogonal contrasts were used to test preregistered hypotheses. We report the Time main effect, each contrast’s main effect, and their respective Time × Contrast interactions. Retaining Time as a repeated-measures factor allowed us to both test overall group differences in physiological arousal and evaluate whether the social modelling manipulation altered arousal over time.

For brevity, only the planned orthogonal contrasts are reported in text. Partial eta squared were interpreted with 0.01, 0.06 and 0.14 as thresholds corresponding to small, medium, and large effect sizes²⁴. The results of the omnibus tests are available in Supplementary Materials 2 and 6. Exploratory analysis of symptoms is available in Supplementary Materials Figure S3.

Results

Baseline differences

There were no significant differences between groups in age, gender, or baseline measurements including physical symptoms, state anxiety, general anxiety, expectancy for symptoms, or expected enhancement (all ps > 0.05) indicating randomisation was successful (see Supplementary Materials Table S1).

Target symptoms

Figure 2 depicts the group means for the three pre-registered outcomes. Orthogonal contrasts revealed that there was a significant overall nocebo effect, where groups that received the placebo reported increased levels of their Target Symptom relative to the Natural History group, F(1,116) = 5.28, p = 0.023, _p² = 0.04. There was no significant difference in Target Symptoms between the No Social Modelling group and the social modelling groups combined, F(1,116) = 2.56, p = 0.112, _p² = 0.02, nor was there a significant difference in Target Symptoms between the Social Modelling Consistent and Inconsistent groups, F(1,116) = 0.03, p = 0.86, _p² < 0.001.

Fig. 2.

Pre-registered analyses. Note. From left to right, Mean baseline adjusted Target, Non-Target and General Symptom scores by group. Natural History (NH), No Social Modelling (No SM), Social Modelling Consistent (SM C) and Social Modelling Inconsistent (SM I). All error bars are ± 1 SEM.

Non-target symptoms

There was no significant overall nocebo effect, F(1,116) = 0.44, p = 0.50, _p² = 0.004, social modelling, F(1,116) = 1.99, p = 0.161, _p² = 0.02, or generalisation, F(1,116) = 0.30, p = 0.59, _p² = 0.002, on severity of Non-Target Symptoms.

General symptoms

There was no significant overall nocebo effect on severity of General Symptoms, F(1,116) < 0.001, p > 0.99, _p² < 0.001. However, there was a significant effect of social modelling, F(1,116) = 7.53, p = 0.007, _p² = 0.06, where the groups that did not receive social modelling reported a greater reduction in General Symptoms from baseline than those in the social modelling conditions. Similarly, a significant effect of generalisation was found such that participants in the Social Modelling Inconsistent group reported more General Symptoms than Social Modelling Consistent group, F(1,116) = 4.46, p = 0.037, _p² = 0.04.

Placebo effect

There was no significant difference in self-reported cognitive performance between groups, F(3,116) = 0.02, p > 0.99, <0.001, nor was there a significant difference in actual performance, F(3,116) = 0.15, p = 0.93, =0.003. Within the three groups that received the placebo treatment, there was no significant difference in self-reported influence of treatment on cognitive performance, F(2,86) = 0.67, p = 0.51, =0.02. See Supplementary Materials Table S7 for group means.

HRV. Refer to Table 1 for group means for the physiological data. Due to technical issues with the Equivital Sensor Module, heart rate data was only available for 81 participants. There was no significant relationship between missing data and participant group, =1.63, p = 0.65, Cramer’s V = 0.12. Two-way mixed 2(Time) × 4(Condition) ANOVA was conducted to explore differences in HRV. The two-way mixed ANOVA revealed no significant effect of treatment (F(1, 77) = 0.08, p = 0.773, <0.001), social modelling (F(1, 77) = 1.78, p = 0.186, =0.02), nor generalisation (F(1, 77) = 0.33, p = 0.570, = 0.004) on HRV. The ANOVA found no significant effect of time F(1,77) = 0.17, p = 0.68, ² = 0.02, and no interaction between treatment, social modelling, and generalisation, and time, F(1, 77) = 0.09, p = 0.763, =0.001, F(1, 77) = 0.94, p = 0.336, =0.012. F(1, 77) = 0.00, p = 0.969, <0.001 respectively.

Table 1.

Physiological measures.

		Natural history	No social modelling	Social modelling consistent	Social modelling inconsistent
HRV	M	2.11	1.81	2.31	2.64
	SE	0.42	0.41	0.38	0.42
nSCR	M	19.65	28.25	29.59	28.60
	SE	3.08	3.30	2.79	3.22
μSCL	M	−2.83	2.51	3.27	2.31
	SE	1.37	1.47	1.24	1.43

Note. Means are averaged across the factor time.

EDA. Due to technical issues, EDA recordings for 92 participants were available. There was no significant relationship between missing data and participant group, =7.04, p = 0.07, Cramer’s V = 0.25. Two-way mixed 2(Time) × 4(Condition) ANOVA was conducted to explore differences in tonic (μSCL) and phasic (nSCR) activity separately. Averaged across time, contrasts revealed increased μSCL in groups that received treatment compared to those who did not, F(1,88) = 12.17, p < 0.001, =0.12. Averaged across the other variables in the model, there was no significant effect of social modelling, F(1,88) = 0.02, p = 0.87, <0.001, generalisation, F(1,88) = 0.26, p = 0.61, =0.002, nor time, F(1,88) < 0.01 p = 0.96, <0.001. Averaged across time, contrasts revealed increased nSCR in groups that received treatment compared to those who did not, F(1,88) = 6.59, p = 0.01, =0.07. There was no significant effect of social modelling, F(1,88) = 0.05, p = 0.83, <0.001, generalisation, F(1,88) = 0.05, p = 0.82, <0.001, nor time, F(1,88) = 3.09 p = 0.08, =0.03. There was no significant relationship between tonic, B = 0.01, t(89) = 0.44, p = 0.67, or phasic SC, B = 0.09, t(89) = 1.00, p = 0.32, and Target Symptom severity. Table 1 contains group means for the Physiological Measures.

Manipulation check

Most participants correctly recalled the ‘treatment’ they had taken (93%). The overall social modelling manipulation was successful, with only one participant in the sample reporting suspicion concerning the other participant. A small proportion of participants expressed suspicion that the treatment was a placebo (N = 23). However, raising suspicion concerning the placebo treatment did not vary between groups, χ²(3, N = 120) = 7.05, p = 0.070, Cramer’s V = 0.24, nor was it associated with a change in target symptom reporting, controlling for group, F(1, 112) = 0.44, p = 0.51 _p² = 0.004.

Exploratory analyses

Counterbalancing effect

Contrary to hypotheses, social modelling did not increase Target Symptoms above the effect of instruction alone. To investigate, we conducted exploratory analyses to investigate the influence of counterbalancing side effects. The counterbalancing was intended to increase sensitivity to detect genuine nocebo effects²⁵. However, it is possible that if one side effect profile was less responsive to the nocebo manipulation, then this could obscure evidence of social modelling. Further, knowledge regarding the types of symptoms that are most receptive to social modelling is important for future research³.

To investigate this possibility, we conducted a post-hoc two-sample t-test comparing the Target Symptom severity for participants randomised to headache/dizziness as the target profile versus participants with nausea/stomach discomfort as the target profile, collapsed across the four experimental groups. This revealed participants’ whose target was headache/dizziness experienced significantly more of their Target Symptoms (M = 1.17, SD = 1.71) than the nausea/stomach discomfort participants, (M = −0.02, SD = 1.01), t(118) = 4.61, p < 0.001, d = 0.84. This suggested that headache/dizziness was responsive, but nausea/stomach discomfort was not. Therefore, further exploratory analysis was conducted on the sub-group of participants who were warned about headache/dizziness (n = 60). Despite a reduction in power, this revealed a significant overall nocebo effect, F(1,56) = 6.44, p = 0.014, _p² = 0.10, and social modelling effect, F(1,56) = 4.61, p = 0.036, _p² = 0.08, but no effect of generalisation, F(1,56) = 0.43, p = 0.52, _p² = 0.007, on the severity of headache/dizziness experienced. Similar analyses were conducted with participants warned about nausea/stomach discomfort, which found no significant effect on any contrasts (ps > 0.63), see Supplementary Materials Tables S4 and S5. When the full sample was included using headache/dizziness as the outcome, the pattern of results was replicated, with a significant overall nocebo effect found on severity of headache reported, F(1,116) = 6.85, p = 0.010, _p² = 0.06, and significant social modelling effect F(1,116) = 5.17, p = 0.025, _p² = 0.04, but no effect of generalisation F(1,116) = 0.008, p = 0.93, _p² < 0.001. Results are presented in Fig. 3.

Fig. 3.

Exploratory analyses. Note. From left to right, mean Target Symptoms of headache/dizziness warned participants only; mean Target Symptoms of nausea/stomach discomfort warned participants only. Mean headache/dizziness for all participants. All error bars are ± 1 SEM.

Discussion

The present study investigated whether socially acquired nocebo effects generalise to similar treatments. Planned analyses revealed an overall nocebo effect, with increased Target Symptom severity in groups that received treatment relative to control. Unexpectedly, there was no significant effect of social modelling above and beyond explicit instruction on Target Symptom severity. Because very few prior studies counter-balance symptom warnings, we conducted a post-hoc, symptom-specific exploration to avoid overlooking differential symptom responsiveness. This exploratory analysis revealed an interesting difference: social modelling increased the nocebo effect when headaches/dizziness were the target profile, but not when nausea/stomach discomfort were, with a large effect of type of target symptom. Most importantly, across both pre-registered and exploratory analyses, the nocebo effect in the Social Modelling Inconsistent group was always as large as, if not larger than, the Social Modelling Consistent group. Taken together, this provides preliminary evidence that while socially induced nocebo side effects may not always be present, when they are present, they do tend to generalise.

Previous research has established the propensity for directly-conditioned nocebo effects to generalise broadly^11,12,26 and socially acquired nocebo effects to generalise across VR contexts⁶. The present research extends this with preliminary evidence demonstrating generalisation of socially acquired nocebo effects across treatments. The possibility of this type of generalisation of socially acquired nocebo effects is concerning considering the wealth of treatment-related information communicated between individuals face-to-face, online, and via social media; something which has only been exacerbated by the COVID-19 pandemic^27,28. It suggests that observing another person experience a negative outcome to a specific treatment can cause the observer to experience nocebo effects not just to that treatment, but to other similar treatments. Interestingly, traditional theories of generalisation from the associative learning literature would predict a weakening of the nocebo effect for different treatments⁹. In the current study, as well as in Saunders et al.⁶, the socially acquired nocebo effect was equally large whether participants received the same or a different treatment to the model. In fact, group means across all analyses trended towards an exacerbation in the Social Modelling Inconsistent group. This indicates the absence of any statistically significant reduction of the nocebo effect due to generalisation was not due to a lack of statistical power. As such, when socially acquired nocebo effects are present, generalisation across treatments and contexts appears to be as strong as the original effect and hence even more concerning.

Contrary to hypotheses, the primary pre-registered analysis on symptom severity did not find an additive effect of social modelling on side effects above explicit instruction. Importantly however, planned analysis of the secondary outcome general symptoms andexploratory analysis on the subset of participants for whom headaches/dizziness were the Target Symptoms, found significant medium sized effects of social modelling above and beyond instruction. The results of exploratory analyses are consistent with existing studies demonstrating socially acquired nocebo effects^15,29–31. Notably, these previous studies all included headache as a primary symptom of interest. As such, it may be the case that headache and dizziness are more susceptible to nocebo effects in general and/or in the context of ‘cognitive enhancers’. This can be compared to the concept of cue preparedness’ in learning, where for example rats more readily learn light-shock and taste-nausea contingencies than light-nausea or taste-shock contingencies³². Understanding if some symptoms are more receptive to social modelling than others could be useful in the development of recommendations regarding the contexts in which interventions might be most appropriate. Furthermore, given the nature of the exploratory analysis there is a need for pre-registered replication of this result.

Exploratory analyses revealed heightened physiological arousal due to treatment as measured by skin conductance, but no effect when measured using HRV. Physiological arousal was not, however, heightened by social modelling. Previous research has not reached consensus on this topic, with one study finding no effect of social learning on SCR³³ while another has³⁴. However, the present analysis of physiological measures is limited due to the lack of pre-treatment baseline measurements. Consequently, it is not clear whether the significantly higher level of arousal among the placebo-treated participants was due to the act of treatment administration itself.

The present study included natural history and treatment-only groups which allowed for a direct assessment of the nocebo effect and the additive effect of social learning above instruction. While this addressed some of the methodological issues in previous studies^15,31, several limitations must be noted. First, the counterbalancing of side effect profiles was intended to increase experimental control and distinguish treatments but appeared to reduce sensitivity to detect socially acquired nocebo effects because nausea and stomach discomfort appeared non-responsive to the manipulations within the measured timeframe. Because the symptom‑specific findings were derived from unplanned exploratory analyses, they should be viewed as hypothesis‑generating and warrant replication in future preregistered studies. Second, the study was conducted on a healthy student sample utilising an acute treatment. While this allowed for a large sample size, this translated to an overrepresentation of young, educated females in the sample and as such it is important to replicate this research in clinical samples involving longer treatment timeframes. Third, the modelling procedure took place live (i.e., face-to-face), which may differ from virtual platforms, such as social media. Finally, the manipulation check was brief to avoid arousing suspicion in the student cohort prior to the study’s conclusion. Thus, participant memory of the specific side effects associated with each “medication” was not assessed.

In conclusion, the current study provides mixed evidence for socially acquired nocebo effects, and also evidence that socially acquired nocebo effects can generalise from an observed treatment to other treatments, depending on the symptoms measured. The abundance of social information communicated between patients, face-to-face or via mainstream and social media, is therefore a concern due to the potential spread of nocebo effects and the burden they cause. Future studies should seek to examine these processes in clinical settings as well as to identify interventions to inhibit these effects.

Author contributions

All authors (C.S., W.T., K.B., N.M., & B.C.) were involved with study conception and design. C.S. and W.T. ran the experiment sessions with participants. C.S. conducted statistical analyses under the supervision of B.C., N.M. & K.B. C.S. wrote the first draft of the main manuscript text. All authors reviewed the manuscript (C.S., W.T., K.B., N.M., & B.C.).

Funding

Australian Government Research Training Program Stipend Scholarship, Australian Research Council Discovery Projects (DP200101748 and DP230102411).

Data availability

The study was preregistered at aspredicted. org (#109,597) before data collection, including a detailed analysis plan. Deidentified data and analytic code used to conduct the analyses presented in this study are available in a public archive: https://osf.io/m8tdh/.

Declarations

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-14118-5.