Multi-modal hallucinations across diagnoses: What relationships do they have with voice-related distress?

Background: Research into hallucinatory experiences has focused primarily upon hallucinations within the auditory modality, to the relative neglect of other modalities. Furthermore, the exploration of auditory hallucinations (or ‘voices ’ ) has focused primarily upon the experiences of people with a diagnosis of psychosis. The presence of multi-modal hallucinations may have implications across diagnoses for levels of distress, formulation and the targeting of psychological interventions. Methods: This study presents a cross-sectional analysis of observational data from the PREFER survey ( N = 335). Linear regression was used to explore the relationships between voice-related distress and the presence, number, type and timing of multi-modal hallucinations. Results: Simple relationships were not found between distress and the presence of hallucinations in visual, tactile, olfactory or gustatory modalities, or in the number of modalities experienced. When considering the degree to which another modality hallucination was experienced simultaneously with voices, there was some evidence that the degree of co-occurrence with visual hallucinations was predictive of distress. Conclusions: The co-occurrence of voices with visual hallucinations may be associated with relatively greater distress, but not consistently, and the association between multimodal hallucinations and clinical impact appear complex and potentially variable from individual to individual. Further study of associated variables such as perceived voice power may further illuminate these relationships.


Introduction
Research into hallucinatory experiences has focused primarily upon hallucinations within the auditory modality, to the relative neglect of other modalities (Toh et al., 2019).Furthermore, the exploration of auditory hallucinations (or 'voices') has focused primarily upon the experiences of people with a diagnosis of psychosis (Waters et al., 2018).The narrowness of these approaches has maintained an assessment focus upon voices in the context of psychosis and shifted the focus away from hallucinatory experiences in other modalities (Montagnese et al., 2021).
As understandings of voice hearing experiences have broadened to be inclusive of diagnoses other than psychosis (Waters and Fernyhough, 2017) and the lifespan (Choi et al., 2021;Maijer et al., 2019), awareness has grown of the existence and experience of multi-modal hallucinations (MMH), whereby hallucinations are experienced in two or more modalities, concurrently or in isolation.In the context of psychosis, the lifetime prevalence of MMH has been estimated at 53 %, twice the amount of uni-modal hallucinations (UMH) (Lim et al., 2016).Across diagnoses, 72 % of patients seeking treatment for voices reported MMH (Badcock et al., 2021).Within this clinical sample, 44 % of patients reported hallucinations in two modalities, 38 % reported three, 11 % reported four, and 8 % reported five, thereby corroborating the suggestion that an inverse relationship exists between the number of modalities and the number of people reporting them (Montagnese et al., 2021).After voices, visual hallucinations have been most frequently reported, followed by tactile hallucinations, olfactory hallucinations and gustatory hallucinations (Badcock et al., 2021;Lim et al., 2016).
Beyond an emerging awareness of the prevalence of MMH and the modalities to which they are linked, a framework has been proposed to consider the features shared by different hallucinatory modalities regarding timing, source, and coherence (Montagnese et al., 2021): timing refers to the temporal scale across modalities and MMH are categorised as either 'simultaneous' (if hallucinations co-occur within different modalities) or 'serial' (if hallucinations occur in a serial manner over time); source looks at whether MMH are experienced from the same source or entity; and coherence concerns the degree of contextual coherence across MMH.
The presence of MMH has been proposed to be of potential clinical relevance, with conjectures that the presence of hallucinations in other modalities may make voice entities appear more real, veridical or powerful (Montagnese et al., 2021).This also ties in with the cognitive model of hallucination-related distress, in which accompanying symptoms may be interpreted as evidence of the power and threat associated with voices (Chadwick and Birchwood, 1994).In particular, experiences of voices manifesting in visual form, or being felt by the person physically, may increase the immediacy of voice-related threat by them being able to impress upon the person's reality in multiple ways.Montagnese et al. (2021) also suggest their framework may have clinical implications for levels of distress, formulation and the targeting of psychological interventions; for example, serial MMH may be perceived to have the power to affect the person in different ways and beliefs that MMH are interconnected could be targets for cognitive therapy.A particular possibility is that simultaneous MMH may be experienced as particularly distressing.However, there is currently no consensus on the implications of MMH for hallucination-related distress, when compared to UMH.Some studies have reported MMH to be associated with higher levels of psychopathology (Clark et al., 2017) and adverse mental health outcomes (e.g., Dudley et al., 2019;Laloyaux et al., 2019).However, in the limited examination of voice-related distress to date, no overall differences in distress have been observed between UMH and MMH groups (e.g., Badcock et al., 2021).
This study conducted an analysis of a large trans-diagnostic sample of patients experiencing voices in order to clarify some of the issues emerging in relation to MMH and distress.Specifically, the study aimed to explore whether voice-related distress was predicted by: (i) the presence of MMH as opposed to UMH (voices); (ii) the number of modalities in which hallucinations were experienced, (iii) the presence of specific hallucination modalities, and (iv) the degree to which MMH were experienced simultaneously as opposed to serially.

Design
This study presents a cross-sectional analysis of observational data from the PREFER (Patient preferences regarding psychological therapies for voice-hearing experiences) survey (Berry et al., 2022).

Participants and procedure
PREFER was a large national survey of therapy preferences in participants who had heard voices for at least six months.Between April 2020 and January 2022, a total of 342 participants were recruited from inpatient and outpatient youth and adult mental health services across 25 National Health Service (NHS) Trusts in England.A total of 335 participants completed the PREFER survey.
To meet the inclusion criteria participants had to: be at least 16 years old, receiving secondary mental health services with a designated lead practitioner, speak sufficient English to give informed consent and participate in the study, and have experienced voices for at least six months, either recently or in the past.There were no exclusions based on participant diagnosis (presence/absence or nature), medication use (presence/absence or type), or historical/current exposure to psychological therapy.
Participants completed a self-report questionnaire battery, which provided quantitative data on preferences and demographic and clinical characteristics.The battery was completed with the assistance of a researcher, either online using Qualtrics or on paper, as chosen.The sequence of the questionnaires was randomised in the online questionnaire.When COVID-19 restrictions were imposed, an adjustment was created to allow the questionnaire battery to be completed by phone and/or video-conferencing.Participants provided written consent or, after the COVID-19 modification, electronically via a digital consent form.

Primary clinical outcome
The Hamilton Program for Schizophrenia Voices Questionnaire (HPSVQ) is a 9-item self-report assessment of voice severity that has been validated with transdiagnostic voice hearers in the United Kingdom (Berry et al., 2021).We used the voice impact subscale to measure voice-related distress (Van Lieshout and Goldberg, 2007) which has five items rated 0-4.

Demographics
Demographics were age, ethnicity, gender, employment status, highest educational qualification, relationship status, duration of voices, frequency of voices (one of the four items from the HPSVQ phenomenology subscale rated 0-4), and self-reported diagnosis.

Statistical analysis
Descriptive summaries using counts, percentages, means, standard deviations (SD), medians, interquartile range (IQR), minima and maxima were conducted for the demographic variables, potential confounders (age, gender (male/female), education (left below 16 years/left post 16 years), ethnicity (Ethnic Minority/White British), voice frequency, duration of voices, and presence of self-reported schizophrenia diagnosis (no/ yes)) and the explanatory (independent) measures which were defined as: • Whether and how each multimodal hallucination (visual, olfactory, gustatory, and tactile) was experienced.For each hallucination, participants self-reported against the statement 'this happens to me', with one of the following four ordinal categories: 'No' (none/absent), 'Yes -not the same time as hearing voices' (serially), 'Yes -happens when and when not hearing voices' (mixed), or 'Yes -same time as hearing voices' (simultaneously).• Multimodal hallucinations experienced (absent v present; binary measure derived from above).• Total number of multimodal hallucinations experienced in addition to voices (0,1,2,3,4).
The association between the number of additional modalities and participant characteristics was tested using chi square tests (for categorical variables) and linear regression (for continuous variables treated as dependent variables).Demographics resulting in associations with p < 0.05 were deemed statistically significant.
A two-step process was used to explore the associations between the primary outcome, voice-related distress (dependent variable), and the potential confounders or explanatory measures (independent variables).In step one, bivariate linear regression models were used to examine whether there was a relationship between an independent variable and voice-related distress.Demographic variables deemed statistically significant were put into a candidate variable pool for entry into a multivariate model to describe voice-related distress.Potential confounders with p < 0.1 were also added to the candidate pool for further exploration.Joint statistical tests were used for categorical variables.Planned contrasts were used to compare distress levels across different pairs of modality categories.In step two, a single multivariate linear regression model was built to describe the variation in voice-related distress using explanatory measures with adjustment for confounders.
Starting with the null model with voice-related distress as the dependent variable, independent variables from the candidate pool were added one-by-one (starting with the most significant) and removed if p > 0.05.Parameter estimates from the final bivariate and multivariate models were reported as unstandardised effect sizes (ES) with corresponding standard errors, 95 % confidence intervals (95 % CIs) and p-values.
All analyses were exploratory therefore no adjustments were made for multiple testing nor for missing data due to the small (<5 %) level of missingness on the primary outcome.Analyses were conducted using STATA, version 16 (StataCorp, 2019).

Results
In total, 335 participants completed the PREFER survey and 311 (93-%) provided modality data.Table 1 describes selected participant characteristics according to the observed number of modalities experienced.Overall, participants were predominantly male (57 %); with a mean age of 41.9 (SD 17.8) years; from a mix of ethnic groups (4 % Asian/Asian British, 3 % Black/Black British, 3 % Mixed ethnicity, 84 % White British, and 5 % White Other); for 36 % of participants their dayto-day activities were limited a lot due to being disabled and 34 % experienced a little limitation in their activities due to disability; 43 % had a self-reported schizophrenia diagnosis; and 47 % had been hearing voices for 16 years or more.In total, 57 (18 %) reported hearing only voices, 81 (26 %) experienced one modality in addition to voices, 82 (26 %) experienced two additional modalities and 91 (29 %) experienced three or four.There was a statistically significant difference in the number of modalities experienced by ethnicity (Ethnic Minority vs White British; p = 0.06); there was a greater representation of participants from ethnic minority groups experiencing 3 or 4 modalities.Voicerelated distress data was missing for 2 % (n = 8) of participants which was below the threshold for addressing missingness.Table 2 provides a descriptive summary of voice-related distress by each modality according to four ordinal categories and as a binary measure.The table also shows how voice-related distress varies according to number of additional modalities.The relationships in the data are discussed in more detail in the sections below.

Effect of multimodal and unimodal hallucinations on voice-related distress
The relationships between voice-related distress and the absence or presence of visions, smells, tastes and body sensations and voice-related distress were explored (see Table 2 for data distributions).Effect sizes from bivariate linear regression models are displayed in Table 3 and show that across all modalities, there were no differences in voicerelated distress (p > 0.05) between MMH and UMH for any modality when presented in the binary form.
A relationship did exist when presented as an ordinal four-category measure rating the degree to which hallucinations were simultaneous, but only for visions, this was indicated by the joint test across all categories yielding p = 0.002.However, only the simultaneous (visions and voice hallucinations) category was statistically significant (ES -2.1: 95 % CI -3.7, − 0.5; p = 0.009).Evidence to support the hypothesis that MMH are associated with higher levels of voice-related distress compared to UMH was weak.The current evidence suggests lower levels of distress only for simaltaneous visions.

Effect of different numbers of additional modalities on voice-related distress
Table 2 indicates that an absence of other modalities in addition to voices (i.e., no additional modalities) was associated with the lowest level of distress (mean 9.0, SD 5.1).Meanwhile, having one additional modality was associated with the highest level of distress (mean 10.0, SD 4.8) (see Table 2).However, the bivariate model regressing voicerelated distress on the number of co-occurring modalities (see Table 3) indicated the relationship between the two was not statistically significant (p = 0.629).There was not enough evidence to support the suggestion that higher distress is associated with increased number of MMH.

Effect of the presence of specific hallucination modalities on voicerelated distress
The only additional hallucination modality to significantly increase voice-related distress was visions when described as an ordinal measure (see Table 2 for descriptive summaries and Table 3 for unstandardized effect sizes).Overall, experiencing visions had a statistically significant (p = 0.002) main effect (see Table 3) and therefore association with voice-related distress.Table 2 shows that for the visions modality, those who experienced them when and when not hearing voices (mixed) had the highest levels of voice-related distress (mean 10.4,SD 4.7).This was followed by those who did not experience visions (none; mean 9.7, SD 4.9), experienced visions serially to hearing voices (mean 7.7, SD 4.8), and then at the same time as hearing voices (simultaneously; mean 7.6, SD 5.5).Exploratory contrasts revealed the following significant differences for visions: simultaneous vs none (ES -2.1: 95 % CI -3.7, − 0.5; p = 0.009), mixed vs serial (ES 2.6: 95 % CI 0.2, 5.1; p = 0.035), and simultaneous vs mixed (ES -2.7: 95 % CI -4.2, − 1.2; p < 0.001).Conversely, there was little difference (ES -0.1: 95 % CI -2.7, 2.5; p = 0.941) between simultaneous and serial vision experiences.Caution should be taken as only 18 participants had serial experiences.

Effect of simultaneous vs serial multimodal hallucinations on voicerelated distress
Descriptive summaries of voice-related distress over the different hallucination modalities (Table 2) and the resulting effect sizes from the bivariate regressions (Table 3) implied small and non-significant differences in voice-related distress for simultaneous MMH compared to serial MMH.Planned contrasts (results not shown) indicated there was not enough evidence (p > 0.05) to support the suggestion that simultaneous MMH are associated with higher levels of voice-related distress compared to serial MMH.It should be noted that in general the numbers of participants experiencing modalities serially to hearing voices were very small: visions (n = 18; 6 %), smells (n = 17; 6 %), tastes (n = 8, 3 %), and body (n = 16, 5 %).

Multivariate model to describe voice-related distress
The multivariate model was built starting by adding visions to the null model.Confounders were then added one by one in the order of statistical significance and non-significant variables were removed.The final model (shown in the righthand panel of Table 3) indicates that increasing levels of voice-related distress are associated with increasing voice frequency, as follows: hearing voices less than once a day (ES 4.3: 95 % CI 2.5,6.1;p < 0.001), hearing voices once or twice a day (ES 6.5: 95%CI 4.8, 8.2; p < 0.001), hearing voices several times a day (ES 7.0: 95 % CI 5.5, 8.5; p < 0.001), hearing voices constantly (8.6: 95%CI 7.2, 10.1; p < 0.001), and being female (ES 1.4: 95 % CI 0.5, 2.2; p = 0.001).Leaving education after 16 years of age was associated with a reduction in distress levels (ES -1.0: 95 % CI -1.9, − 0.1; p = 0.026).This model explains 36 % of the variation in voice-related distress but the majority is explained by frequency of voices as a highly correlating measure (r = 0.55; p < 0.01) (see also effect sizes in Table 3) which on its own explained 30 % of the variance in voice-related distress.

Discussion
This study examined the relationship between the presence of MMH and voice-related distress in a transdiagnostic group of people with voice hearing experiences.We did not find a simple relationship between

Table 1
Summary of selected PREFER participant characteristics by number of additional modalities in addition to voice hearing.
(continued on next page)

Table 2
Descriptive summary of voice-related distress by hallucination modality and number of modalities.
Note: N = count; HPSVQ = Hamilton Program for Schizophrenia Voices Questionnaire; SD = Standard Deviation; IQR = Interquartile Range.N = 8 participants excluded from this table due to missing HPSVQ data.

Table 3
Bivariate and multivariate regression model results exploring the associations between voice-related distress and potential confounders, explanatory variables, hallucination modalities and number of modalities.
(continued on next page)