The relationship between appraisals of auditory verbal hallucinations and real-time affect and social functioning

Abstract

In addition to being a hallmark symptom of schizophrenia-spectrum disorders, auditory verbal hallucinations (AVH) are present in a range of psychiatric disorders as well as among individuals who are otherwise healthy. People who experience AVH are heterogeneous, and research has aimed to better understand what characteristics distinguish, amongst those who experience AVH, those who experience significant disruption and distress from those who do not. The cognitive model of AVH suggests that appraisals of voices determine the extent to which voices cause distress and social dysfunction. Previous work has relied largely on comparisons of “clinical” and “non-clinical” voice hearers, and few studies have been able to provide insight into the moment-to-moment relationships between appraisals and outcomes. The current study examines longitudinal data provided through ecological momentary assessment and passive sensors of 465 individuals who experience cross-diagnostic AVH. Results demonstrated associations of AVH appraisals to negative affect and social functioning. Above and beyond within-individual averages, when a participant reported increased appraisals of their voices as powerful and difficult to control, they were more likely to feel increased negative affect and reduced feelings of safety. AVH power appraisals were also associated with next-day number and duration of phone calls placed, and AVH controllability appraisals were associated with increased time near speech and reduced next-day time away from primary location. These results suggest that appraisals are state-like characteristics linked with day-to-day and moment-to-moment changes in impactful affective and behavioral outcomes; intervention approaches should aim to address these domains in real-time.

1. Introduction

Though auditory verbal hallucinations (AVH) are most common amongst individuals with schizophrenia-spectrum (Ohayon, 2000) or other disorders (Van Os et al., 2009), they are also commonly reported in the general population (Johns et al., 2002; Scott et al., 2006; Tien, 1991) with lifetime prevalence rates estimated as high as 10% (Maijer et al., 2018). Outcomes associated with AVH vary. While in general, those experiencing AVH are more likely to report poor quality of life (Van Os et al., 2000), reduced functioning (Kelleher and Cannon, 2011), substance dependence (Johns et al., 2004), and victimization (De Vries et al., 2019), many experience little disruption or distress (Johns et al., 2014). A central aim of research on AVH has been to determine the factors that determine whether and how AVH result in dysfunction, and development of interventions to foster resilience.

The cognitive model of AVH (Beck and Rector, 2003) provides a framework to understand their emergence and maintenance. This model suggests that AVH are more likely to result in dysfunction when accompanied by particular appraisals (Mawson et al., 2010), for example, interpreting that they are powerful, important, or difficult to control. These appraisals likely lead to increased negative affect, and an increased desire to withdraw. While withdrawal can result in short-term relief (e.g. reduced obligations) and reinforce beliefs about avoidance (i.e., “I must stay home to manage stress”), inactivity can worsen negative affect over time (Upthegrove et al., 2016). Support for this model has grown over the last few decades, largely built on studies cross-sectionally comparing clinical vs. non-clinical populations (Johns et al., 2014). These studies have demonstrated that non-clinical voice hearers report higher perceived control over voices (Andrew et al., 2008), and are less likely to obey voices or engage in safety behaviors (Gaynor et al., 2013). Those with clinically significant AVH are more likely to appraise their voices as powerful, important, or malevolent (Lawrence et al., 2010; Sorrell et al., 2010). While this line of work has identified characteristics that may be associated with help-seeking among those with AVH, less remains known about whether these differences in appraisal are linked with more negative outcomes within individuals and moment-to-moment.

Existing methods – including retrospective assessments and comparisons of aggregated groups – are limited. First, retrospection is prone to biases, memory errors, or demand characteristics. Laboratory role-plays are also artificial and imperfect proxies for real-world behavior (Granholm et al., 2020; Holshausen et al., 2014). Second, retrospective measures do not provide insight into questions related to momentary changes. Emerging technologies provide new opportunities for ecological momentary assessment (EMA), which involves brief, targeted self-assessments in real-time and in real-place (Shiffman et al., 2008). A large and growing proportion of the U.S.– and a majority of individuals with serious mental illnesses (Young et al., 2020) – have access to smartphones. In addition to being equipped to deliver EMAs, smartphones can collect information about behavior through the use of passive sensors, including microphones, light sensors, GPS, and accelerometers (Wang et al., 2016), facilitating data collection without reliance on self-report. Previous studies have demonstrated that such multimodal packages are feasible to deploy in this population (Barnett et al., 2018; Ben-Zeev et al., 2017; Wang et al., 2017), and such tools have demonstrated associations with measures of social functioning both in non-clinical (Doryab et al., 2019) and clinical samples (Fulford et al., 2021). Studies using digital technologies can also be deployed remotely to participants (Buck et al., 2021), relying less exclusively on academic-affiliated institutions and medical centers for recruitment, where people with clinical levels of AVH might be overrepresented.

Several studies using a week of EMA have provided initial support for some components of the cognitive model of AVH, including demonstrating relationships between negative affect and AVH intensity (Kimhy et al., 2017), as well as links between social withdrawal (i.e. as a safety behavior) and subsequent reductions in AVH intensity (Delespaul et al., 2002). Fewer studies using EMA have targeted AVH appraisals and particularly (1) whether they are state or trait characteristics, or (2) whether changes in appraisals are associated with affective and behavioral disruptions. Further, because such systems can be downloaded and installed remotely, smartphone data collection provides the opportunity for sampling a broader representation of individuals who experience AVH, whereas most extant studies have recruited from clinical settings, thus overrepresenting help-seeking voice hearers, who may differ in meaningful ways from those who have not sought treatment (Lawrence et al., 2010).

Our group developed a multimodal data collection system designed for continuous assessment of the cognitive, affective, and behavioral indicators that may be associated with AVH. This system is an adaptation of a previous multimodal sensing system used in non-clinical samples as well as with individuals with schizophrenia-spectrum disorders (Ben-Zeev et al., 2017; Buck et al., 2019; Buck et al., 2019; Wang et al., 2016, 2014). Our team previously used this system to demonstrate differences in AVH appraisals and experience between individuals at differing levels of services (Ben-Zeev et al., 2020). Less is known about the momentary relationships between appraisals and outcomes predicted by the cognitive model, including negative affect and social functioning. The present study aims to determine whether momentary appraisals (i.e., controllability and power) are associated with momentary negative affect and passively sensed traces of social withdrawal in a cross-diagnostic sample of individuals experiencing AVH across the clinical and non-clinical spectrum. We hypothesized that increased appraisals of AVH as powerful and difficult to control would be associated with increased negative affect, decreased feelings of safety, and increased traces of social withdrawal (i.e. fewer interactions via phones and SMS, less time spent away from home or around others).

2. Methods

The present study involves a secondary analysis of a study whose primary study findings (Ben-Zeev et al., 2020) as well as additional details on study recruitment methodology (Buck et al., 2021) were described elsewhere.

2.1. Participants

465 individuals with AVH (n = 363 recruited online, n = 102 locally) completed data collection. The mean age of the full sample was 43.0 years (SD = 11.6) and 52.3 percent of participants were female (see Table 2). Participants were recruited in one of two ways: (1) remotely online or (2) in-person outreach to community mental health clinic at a university-affiliated public hospital and other community-based organizations (e.g. homeless shelters) in the Greater Seattle area (i.e. flyers, word of mouth). Participants were included if they were (1) 18 or older; (2) English-speaking; (3) able to use a smartphone and (4) reported experiencing AVH at least once weekly. Participants were excluded if they (1) did not live in the US, (2) had already participated in the study, or (3) were not available for 30 days of data collection. Individuals recruited online were required to own an Android smartphone with a data plan. In order to ensure the study sample comprised a broad representation of individuals with AVH, those recruited in-person were subject to an additional criterion of not owning a smartphone, thus countering overrepresentation of online participants who were required to have access. In-person participants were provided a smartphone by the research team for the duration of the study.

Table 2.

Participant characteristics

		Total (N = 465)
Age		40.67 (11.37)
Gender	Female	243 (52.3%)
	Male	205 (44.1%)
	Other	4 (0.9%)
	Transgender Woman	5 (1.1%)
	Transgender Man	8 (1.7%)
Race	American Indian or Alaskan Native	7 (1.5%)
	Asian	7 (1.5%)
	Black or African American	98 (21.3%)
	More than one race	59 (12.8%)
	White	290 (62.9%)
	Missing / Declined	4 (0.9%)
Ethnicity	Hispanic / Latino	66 (14.3%)
	Not Hispanic / Latino	396 (85.7%)
	Missing / Declined	3 (0.6%)
Diagnoses	Bipolar Disorder	185 (39.8%)
	Major Depressive Disorder	310 (66.7%)
	PTSD	201 (43.2%)
	Schizoaffective Disorder	119 (25.6%)
	Schizophrenia	136 (29.2%)
	Substance Use Disorder	151 (32.5%)
Living situation	Assisted/supported living	33 (7.1%)
	Homeless	62 (13.3%)
	Independent/Living on my own	211 (45.4%)
	Living with family	156 (33.5%)
	Substance treatment institution	3 (0.6%)

2.2. Measures.

The data collection app prompted participants to complete a 12-item self-report assessment four times daily (i.e. every three hours, +/− 5 to 8 minutes, randomized) between the hours of 9AM and 9PM (Ben-Zeev et al., 2020). Items examined in this study are reported in Table 1. In initial analyses, the three items assessing negative affect were found to be highly intercorrelated (average r = .68), and thus were combined into a mean composite score. (Ben-Zeev et al., 2020) The application used passive sensors on the device to collect on-going information such as geospatial activity, physical activity, speech frequency and duration, phone calls, and SMS messages. For the present study, we examined sensors that could provide an estimate of behaviors that together represent social avoidance or isolation, including (1) time spent on phone calls, (2) SMS messages sent and received, (3) time spent away from one’s primary location, (4) time spent near speech.

Table 1.

EMA items examined.

Presence of AVH:	Are you experiencing voices right now?	Yes / No
AVH appraisals	How much power do the voices have? (AVH power)	1 = not at all 2 = a little 3 = moderately 4 = extremely
	How much power do you have over the voices? (AVH controllability)
Feelings of safety:	How safe do you feel right now?	1 = not at all 2 = a little 3 = moderately 4 = extremely
* Negative affect:	How distressed do you feel right now?	1 = not at all 2 = a little 3 = moderately 4 = extremely
	How anxious do you feel right now?
	How sad do you feel right now?

^*These three items were combined into a mean composite negative affect score

2.3. Procedures

All study procedures were approved by the Institutional Review Boards of the University of Washington and Dartmouth College. Online remote recruitment was conducted with Google Ads. The study ad was presented to users based on the extent to which their search terms match pre-selected clinical (e.g. hearing voices, schizophrenia and bipolar), non-clinical (e.g. talking to ghosts, am I crazy, stress relief), and related (i.e. generated by the Google Ads “broad match” algorithm) keywords. A complete description of our online recruitment process can be found elsewhere (Buck et al., 2021). The study application was an updated version of a system used and described in previous studies (Ben-Zeev et al., 2017, 2016); it ran only on Android devices.

All participants were instructed to carry the device and respond to prompts for 30 days. During that time, if a participant provided incomplete data, the research team would contact participants directly to answer questions and provide technical assistance. All participants were offered $125 (or, for in-person participants, the study device) for participation.

2.4. Data analytic plan

Our approach examined how appraisal variability across and within individuals relate to both concurrent emotion and subsequent passively sensed behavior using linear mixed effects models. We used the R package glmmTMB (Brooks et al., 2017) with standard Gaussian family distributions for outcome variables based on EMA ratings. In these models predicting emotion (negative affect, safety) from appraisals (power, control) we used 3 variables to characterize different temporal components of the model. We fit models with random intercepts and fixed effects for EMA predictors with a heterogenous unstructured covariance structure. All available data were used for each model; our multilevel approach is robust to differences between individuals in observations. For EMA variables, we examined real-time relationships between appraisals broken down across separate time intervals. First, we included each participant’s mean appraisal across the 30-day period. This provides a between-subjects estimate reflecting the association of AVH appraisal and each outcome averaged over the study period (i.e. whether participants who tend to report the appraisal also have high average levels of the outcome). Second, we included the person-centered daily appraisal rating. This parameter examines whether individual day-to-day elevations in appraisals are associated with momentary emotion ratings separate from the influence of the individual’s overall average (i.e. whether when individuals report a particular appraisal more or less than is typical for them, whether they also experience a change in the outcome). Lastly, we included the day-centered momentary appraisal rating. This parameter examines how within-day variability of appraisals are associated with momentary emotion ratings (i.e. whether there exists a relationship, within the course of a day, between an individuals’ appraisals and affect, above and beyond their daily average).

The behavioral outcome variables derived from daily aggregate data via passive smartphone data collection (minutes of daily phone call time, number of SMS messages, minutes of exposure to conversation, and hours of time spent away from primary location) were positively skewed and overdispersed. The phone call duration, and number of phone calls and SMS messages variables also showed zero-inflation; thus, we used hurdle models within the glmmTMB package with a truncated negative binomial model. Thus, the phone call duration, and number of phone calls and SMS messages variables were examined with hurdle models, a two-component mixture model in which the first component examines the probability of zero versus non-zero values (e.g., the effect of appraisals on the likelihood of a phone call being placed during the day) and the second component, the probability of non-zero values (e.g., the effect of the appraisal on the total amount of time on a phone call). The measures of daily audio conversation minutes and minutes spent away from primary location were less strongly skewed when examined with Cullen and Frey graphs; each was square-root transformed and examined using standard Gaussian family distributions. Models include each participant’s mean appraisal across the 30-day period and an individual mean-centered daily average score, representing the participant’s daily level of each appraisal.

For EMA outcomes (i.e. negative affect and safety), we examined concurrent relationships between each appraisal and affect. We hypothesized that appraising one’s voices as powerful would be related to increased negative affect and reduced levels of safety and appraising them as controllable would be related to reduced negative affect and increased levels of safety. Further, we expected these relationships to be present on the aggregate (i.e. between subjects) level, as well as the daily average and within-day level. For social functioning approximated through passive sensors, we examined prospective relationships, thus ensuring that daily appraisal ratings preceded the measured behavior variables. We hypothesized that increased appraisals of power and reduced appraisals of control would be associated with reduced social engagement as evidenced by reduced phone call time, SMS, time near speech, and time spent away from primary location.

3. Results.

Of all EMA surveys completed, participants reported that they were experiencing AVH 32.7% of all entries, and reported them at least once on 50.2% of the days.

3.1. Association of appraisals and negative affect.

3.1.1. Power of voices.

The relationship between participants’ mean score on the appraisal of the power of voices was related to negative affect (Z = 12.04, p < .001) and safety (Z = −4.00, p <.001), suggesting that individuals who tended to appraise their voices as powerful experienced elevated negative affect and reduced feelings of safety. In addition, person-centered daily appraisals of AVH power were associated with negative affect (Z = 26.99, p < .001) and safety (Z = −4.94, p <.001). This suggests that on days when participants appraised their voices as more powerful than is typical for them, they experienced increased distress and a decreased sense of safety. Further, day-centered momentary appraisals of power were also associated with both negative affect (Z = 19.88, p <.001) and safety ratings (Z = −3.70, p <.001). This suggests that even beyond a person’s typical average appraisals, and changes in their appraisal for a full day, moment-to-moment fluctuations in appraisals are related to negative affect and feelings of safety (see Table 3).

Table 3.

Predicting EMA Affect and Safety from Appraisals of Power and Control

	Negative Affect			Safety				Negative Affect			Safety
Predictors	Est	SE	Z	Est	SE	Z	Predictors	Est	SE	Z	Est	SE	Z
(Intercept)	0.86	0.06	14.75***	2.15	0.06	34.80***	(Intercept)	1.47	0.06	24.37***	1.73	0.05	31.57***
Average Power	0.48	0.04	12.04***	−0.17	0.04	−4.00***	Average Control	0.00	0.05	0.03	0.19	0.04	4.59***
Daily Power	0.23	0.01	26.99***	−0.06	0.01	−4.94***	Daily Control	−0.04	0.01	−4.66***	0.2	0.01	17.56***
Momentary Power	0.18	0.01	19.88***	−0.04	0.01	3.70***	Momentary Control	−0.06	0.01	5.94***	0.13	0.01	10.85***
Random Effects							Random Effects
σ2	0.37			0.65			σ2	0.39			0.63
τ00	0.35 id			0.37 id			τ00	0.48 id			0.36 id
ICC	0.49			0.36			ICC	0.55			0.36
N	452 id			451 id			N	453 id			453 id
Observations	13369			13395			Observations	13369			13404
Marginal R2 / Conditional R2	0.177 / 0.579			0.017 / 0.373			Marginal R2 / Conditional R2	0.002 / 0.549			0.033 / 0.386
AIC	25871.02			33323.31			AIC	26852.61			33019

^*p<0.05

^**p<0.01

^***p<0.001

3.1.2. Control over voices.

When examining appraisals of control over voices, we found similar relationships as with power, only with negative values that reflect changes in the scaling of the variable. The only exception is that the individual’s overall average control rating was not significantly related to negative affect (Z = 0.03, p=.975), whereas it was related to feelings of safety (Z = 4.59, p < .001). Person-centered daily control was significantly related to both negative affect (Z = −4.66, p < .001) and safety (Z = 17.56, p < .001) as was day-centered momentary control (negative affect, Z = −5.94, p < .001; safety, Z = 10.85, p < .001). This suggests that averaged across the study period, feeling control over voices is associated with feelings of safety, and further, that on days when a participant felt more control over their voices than is typical, they felt reduced negative affect and greater feelings of safety, and last, that above and beyond the average for a day, moment-to-moment increases in control appraisals were associated with reduced negative affect and increased feelings of safety (see Table 3).

3.2. Association of appraisals and social functioning.

Models involving behavioral sensing measures can be found in Tables 4 and 5. Individual mean values for appraisals were not significantly related to any next-day behavioral sensing values. Above the influence of these individual means, AVH power appraisals were significantly related to the duration of phone calls the subsequent day (count model component, Z = −2.38, p = 0.017), indicating that one day after participants reported feeling that their voices were more powerful than usual, they spent less time on phone calls. Above the influence of the overall individual mean rating of control, person-centered daily appraisals of AVH controllability were significantly positively related to the next day’s text messages (zero-inflated model component, Z = 2.24, p = 0.025), minutes around speech (Z = 2.58, p = 0.01), and counter to expectations, negatively related to minutes away from primary location (Z = −2.58, p = 0.01). This suggests that on days when participants reported higher levels of control over their voices than usual, they are more likely the next day to have exchanged text messages, spent more time around conversation, and spent fewer minutes away from home (see Tables 4 and 5).

Table 4.

Models examining prospective relationships of appraisals to phone call duration and exchange of SMS. The count model component examines continuous non-zero values of each outcome; zero-inflated component reports the likelihood of a non-zero value of the outcome.

	Duration of phone calls (min)			No. of SMS messages					Duration of phone calls (min)			No. of SMS messages
Predictors	IRR	SE	Z	IRR	SE	Z	Predictors	IRR	SE	Z	IRR	SE	Z
Count Model							Count Model
(Intercept)	10.13	0.91	25.88***	16.16	2.08	21.63***	(Intercept)	11.28	0.9	30.42***	14.88	1.71	23.49***
Average Power	1.09	0.07	1.42	0.87	0.08	−1.61	Average Control	1.00	0.06	0.01	0.92	0.08	−0.94
Previous Day Power	0.95	0.02	−2.38*	1.01	0.02	0.47	Previous Day Control	0.98	0.02	−1.05	1.01	0.02	0.24
Zero-Inflated Model							Zero-Inflated Model
(Intercept)	0.19	0.04	−7.65***	0.06	0.02	−9.22***	(Intercept)	0.21	0.04	−8.09***	0.07	0.02	−9.94***
Average Power	0.94	0.14	−0.41	1.14	0.22	0.67	Average Control	0.85	0.12	−1.15	1.07	0.21	0.34
Previous Day Power	0.98	0.05	−0.37	1.05	0.07	0.77	Previous Day Control	1.01	0.06	0.1	1.16	0.08	2.24*
Random Effects							Random Effects
σ2	0			0			σ2	0			0
τ00 id	0.57			1.52			τ00 id	0.57			1.52
ICC	1			1			ICC	1			1
N id	441			441			N id	442			442
Observations	5732			5732			Observations	5720			5720
Marginal R2 / Conditional R2	0.009 / 1.000			0.007 / 1.000			Marginal R2 / Conditional R2	0.000 / 1.000			0.002 / 1.000
AIC	37750.82			39782.1			AIC	37650.85			39684.09

^*p<0.05

^**p<0.01

^***p<0.001

IRR = Incidence rate ratio

Table 5.

Models examining prospective relationships of AVH appraisals to time near speech and time not in primary location.

	Duration of speech (mins/day)a			Hours spent away from primary locationa				Duration of speech (mins/day)a			Hours spent away from primary locationa
Predictors	Est	SE	Z	Est	SE	Z	Predictors	Est	SE	Z	Est	SE	Z
(Intercept)	13.90	0.59	23.42***	17.31	0.70	24.85***	(Intercept)	14.28	0.53	26.94***	17.8	0.62	28.63***
Average Power	0.69	0.41	1.70	0.34	0.48	0.71	Average Control	0.44	0.40	1.10	−0.08	0.47	−0.17
Previous Day Power	−0.16	0.09	−1.88	−0.05	0.15	−0.32	Previous Day Control	0.23	0.09	2.58**	−0.38	0.15	−2.58**
Random Effects							Random Effects
σ2	21.46			56.34			σ2	21.4			56.52
τ00 id	35.89			44.87			τ00 id	35.93			44.77
ICC	0.63			0.44			ICC	0.63			0.44
N id	441			439			N id	442			440
Observations	5732			5522			Observations	5720			5511
Marginal R2 / Conditional R2	0.004 / 0.628			0.001 / 0.444			Marginal R2 / Conditional R2	0.002 / 0.627			0.001 / 0.442
AIC	35117.021			38899			AIC	35031.99			38837.86

^*p<0.05

^**p<0.01

^***p<0.001

^afamily=Gaussian, outcomes are square-root transformed

We repeated these analyses controlling for the previous day’s value of the behavioral outcome to examine whether changes in appraisals were related specifically to a next day increase in the respective outcome (rather than an increase relative to average). We found that above and beyond the participant mean appraisal and previous day behavioral value, previous daily appraisals of control were no longer associated with the presence of text messages and time around conversation. However, power of voices continued to show a significant negative relationship with duration of phone calls (Z = −2.43, p = 0.015) and control over voices a significant negative relationship with minutes away from primary location (Z = −2.36, p = 0.019). This suggests that on days participants reported feeling that their voices were powerful, they were likely to experience a decrease in time spent on the phone the subsequent day, and on days they felt more control over their voices, they were likely to experience a decrease in time spent away from their primary location (for full models, see Supplement).

4. Discussion

Our study provides granular insights into the cognitive model of AVH. In general, tendencies to appraise one’s voices as powerful and difficult to control were associated with increased negative affect and reduced feelings of safety. These associations were evident both on the daily (within-subject) and momentary (i.e. within-day) level, suggesting that appraisals of voices are at least partially state characteristics. Our results also build on previous work by demonstrating a relationship of these appraisals to passively sensed social behaviors. AVH power appraisals were negatively associated with next-day number and duration of phone calls, and AVH controllability appraisals were associated with increased time near speech. Importantly, however, in change-specific (i.e. controlling for the previous level of the outcome) analyses, only the relationship of power appraisals to phone calls maintained significance. Counter to expectations, AVH controllability was associated with reduced next-day time away from home, a finding which also maintained significance when controlling for the outcome’s previous value.

These findings corroborate components of the cognitive model (Birchwood and Chadwick, 1997). Our study builds on previous cross-sectional and prospective cohort studies of individuals with schizophrenia-spectrum disorders (Mawson et al., 2010; Tsang et al., 2021) by examining AVH cross-diagnostically, and on a day-to-day and moment-to-moment time scale. Our results suggested links of appraisals of power and control with momentary affect and feelings of safety. This suggests that appraisals could be targeted in interventions that seek to improve individuals’ ability to cope with voices in real-time. Importantly, these relationships are likely bidirectional. For example, when individuals hear voices with negative content, feeling that their voices are powerful and uncontrollable leaves them more vulnerable to such negative messages. At the same time, individuals who distressed may feel less equipped to cope. Strong relationships between these appraisals and affective outcomes argues for the potential for intervening on momentary appraisals to improve outcomes.

Our study also builds on previous work examining impacts of AVH appraisals by assessing their links with passively sensed social behavior. Previous studies have demonstrated that appraisals of AVH power and uncontrollability occur more often among individuals with clinical (vs. non-clinical) voice hearing experiences (Ben-Zeev et al., 2020; Johns et al., 2014) and are linked with complying with command hallucinations (Bucci et al., 2013) and engaging in safety behaviors (Hacker et al., 2008). Current results demonstrated that when participants reported feeling that their voices were more powerful than usual, they spent less time on phone calls the following day. When participants reported feeling more able control their voices, they sent more text messages and spent more minutes around speech the following day. One finding that emerged counter to expectations involved the relationship between increased appraised control over voices and next day reduced time away from primary location. This may contradict this aspect of the cognitive model of AVH, or could speak to the need for a more complex operationalization of daily engagement in passively sensed data. For example, it could be the case that participants may feel increased distress in anticipation of required activities (e.g. work, social engagements) and that this distress could lead to a decreased sense of control over voices.

This study has several limitations. First, little empirical work exists to guide hypothesis generation regarding specific time intervals within which these relationships ought to persist. While our team used intuitive, interpretable and potentially clinically meaningful a priori time intervals, future work could test differential time spans at which appraisals could impact affect and behavior. Second, interpretation of passive sensor data presents challenges. Two or more individuals both with powerful voices may experience social disruption but manifest it behaviorally in different ways. Related issues have been noted in reviews highlighting the need for studies examining population heterogeneity in sensing research (Qirtas et al., 2022); this variation could possibly account for the surprising finding linking AVH controllability and less time spent away from one’s primary location. This is particularly notable given other studies that have noted a link between dispositional loneliness and fewer passively sensed locations visited (Fulford et al., 2021). Finally, our momentary within-subjects analyses provide insight into the associations between AVH appraisals and outcomes; they do not speak to the ways in which these variables respond to interventions over time. In addition to the variables examined directly in this study, additional trait or social factors might also increase the negative impact of AVH (e.g. living alone, having a smaller social network).

This study extends work examining the cognitive model of AVH across diagnoses and beyond just those who have engaged in help-seeking. Appraisals of voices as powerful and difficult to control appear to fluctuate over time and are linked with subsequent distress and interference. Taken together, these results support interventions that respond to momentary changes. A growing selection of digital interventions is emerging for individuals with a range of mental health conditions. A key advantage of those interventions is their ability to provide ongoing momentary support in the individual’s environment at the moment they’re most vulnerable. This study demonstrates that AVH appraisals are a clear target for these kinds of interventions.

Figure 1.

Scatter plot displaying relationship of AVH appraisals and affective outcomes (power with negative affect, control with safety), broken down by between subjects averages (panels A and D), daily change (B and D) and within-day change (C and F).

Figure 2.

Example plot of one participant’s data during the study period, demonstrating daily average EMA rating of AVH power, AVH controllability, negative affect, and safety.