Psychological approaches to understanding and promoting recovery in psychosis and bipolar disorder: a mixed-methods approach

Participants

Participants were invited to take part in the study if they had experience of psychosis in the last year, were aged 16–65 years and were currently in contact with mental health services. Exclusion criteria included not being able to speak English, not able to give informed consent and having taken part in other research with the past 6 months. Participants were recruited from statutory care providers across the Greater Manchester West Mental Health NHS Foundation Trust. Recruitment was conducted until the research team felt that saturation of themes was achieved.

Design

The study utilised a semistructured interview approach, with an interview schedule developed by a clinical psychologist and service user researcher (GH and MK). A service user group facilitated generation of questions surrounding personal background, experience of symptoms, recovery and impacts of symptoms. These questions were piloted with three service users and amendments were made following their feedback. The final version included the following headings: information on initial contact with mental health services, background on personal experiences, current experiences, what they felt had changed over time/recovered, how they feel they have changed (over time/recovered, ways of coping, impacts and changes to their life).

Procedure

Recruitment for the study took place in early intervention services (EISs), assertive outreach (AO) teams and community mental health teams (CMHTs). Participants were given a minimum of 24 hours to read the information sheet before being asked to sign a consent form. A service user researcher conducted the majority of the interviews (75%). All interviews were audio-recorded and transcribed verbatim by the service user researcher and research assistant in order to help familiarise themselves with the data.

Analysis

Interpretative phenomenological analysis66 is an approach to qualitative research with a focus on how an individual makes sense of an event or phenomenon. IPA involves collecting information from participants, in this case using interviews, on a given experience and how the individual has interpreted and made sense of that experience. IPA was utilised to explore the data in this study, as it is well suited to the exploration of subjective experience. 66 A core concept of IPA is that the analyst should become immersed in the data66 to enable the researcher to gain an insider perspective. This was achieved by listening to the audio-recorded interviews and reading through the final transcripts a number of times. Two researchers analysed all the interviews independently and extracted pertinent themes. The third researcher acted as a mediator if there was any disagreement.

Theme 1: impacts on mental health

All people interviewed discussed alleviation of symptoms and/or negative emotions as key to their recovery. Participants discussed specific changes in symptom characteristics as well as changes in their emotional state.

Theme 2: self-change and adaptation

Experience of psychosis was shown to have great impact on one’s self. The themes illustrated the importance of overcoming psychosis and being able to regain self-identity.

Theme 3: social redefinition

Mental health problems were shown to have a direct impact on an individuals’ social role. Redefining and reconciling their social circumstances was frequently spoken about in all interviews.

Theme 4: individualised coping mechanisms

Developing an individualised coping mechanism was considered important to all people interviewed. By accessing support and treatment, people were able to assist their recovery. Furthermore, gaining insight and understanding was also shown to be important.

Development of the Q-sort

Q-methodology is undertaken in a number of steps. 69 First, a database of relevant codes (Q-concourse) is developed. This usually consists of a large database of diverse information from a variety of sources about the topic being examined. The Q-concourse was developed from two main sources: themes identified from the interviews conducted in phase 1 and also from existing literature. This Q-concourse codes are then grouped together to form a final set of descriptive themes. These are then translated into statements cards (this is called the Q-set). An initial set of 132 codes was condensed into 52 clear themes that were representative of the expansive concourse by a multidisciplinary team that included service users. These themes were then categorised by the research team and deemed to broadly cover eight areas of recovery: symptoms, emotional aspects, social functioning, the self, cognitions, services, coping mechanisms and occupational aspects. Participants (known as the P-set) are required to sort these statements in order of importance in relation to one another using a forced normal distribution table.

Participants

Forty participants were identified as being required to take part in the study, which was deemed to be a sufficient number for a study of this design based on previous research. 69,70 The inclusion criteria for the participants in the study were: aged between 18 and 65 years, experienced delusions and/or hallucinations for at least 1 year and in contact with a statutory mental health service in Greater Manchester in the North West of England.

Participants’ self-report demographics are outlined in Table 1. Participants’ diagnoses were self-report, with the majority (40%) reporting a diagnosis of schizophrenia, or experience of psychosis (20%).

Procedure

Participants were shown the Q-sort matrix (Figure 1) and given 52 shuffled statement cards (the Q-set). They were asked to read through them thoroughly to ensure they understood them before sorting the cards into three piles, one that represented things that were important to their recovery, one for things that were not important and ones that they were not sure about. They were then asked to consider the ‘most important’ pile and pick the two most important statements and place them on the far right of the grid (column 12), they were then asked to pick the three next most important and place it onto column 11, and so forth until they placed all the most important cards onto the grid. The same procedure was repeated with the least important pile and the not sure pile until the table was full. Participants were given the opportunity to move the statements around until they were happy with their final sort and were asked follow-up questions about the reasoning behind the rating of the two highest and two lowest items, if they thought any recovery items were missing, and for any further comments.

FIGURE 1.

Q-sort response matrix.

Data analysis

A Q-method software package (PQ method version 2.11; Kent State University, Kent, OH; URL: http://schmolck.userweb.mwn.de/qmethod/pqmanual.htm) was used to analyse the data. Principal component analysis (PCA) with varimax rotation was implemented to sort factors and to explain the maximum amount of variance. Q-method analysis factors groups of people together as opposed to items in traditional factor analysis.

Most important viewpoints of recovery

All 52 items were endorsed (placed in columns 9–12) by at least two of all 40 participants (5%). The number of endorsements and percentage of endorsements are shown in Table 2.

Interpretation of viewpoints

The Q-sort resulted in four viewpoints: collaborative support and understanding, emotional change though social and medical support, regaining functional and occupational goals, and self-focused recovery. The correlations between the factor scores indicated no overlap between factors.

Tables for each factor array, indicating where all statements were ranked for all four viewports are provided in Table 3.

Viewpoint 1: collaborative support and understanding

This viewpoint (n = 8) consisted of people who felt that positive and collaborative engagement with others and services was key to their recovery. This group of people was positively motivated to overcome their experiences and willingly accepted help from others to achieve this. This group did not focus on the negative aspects of having a mental health problem. Aspects such as ‘the amount of alcohol and drug use that worsens my experiences’ (−6), ‘the amount I think about harming myself as a result of my experiences’ (−5) and similar aspects were not considered important.

Viewpoint 2: emotional change through social and medical support

This viewpoint (n = 10) consisted of people who considered emotional stability through support and treatment as a key factor to the recovery process. The affective impacts of having psychosis were prioritised whereas the symptoms of psychosis were not considered as important. This group do not find the psychiatric characteristics of symptoms or occupational aspects of any importance.

Viewpoint 3: regaining functional and occupational goals

This viewpoint (n = 9) consisted of people who considered functional and occupational goals as key to their recovery. It was important to this group of people to regain life functioning and aspects that hindered this process were considered important to change in their recovery. This group did not find the internal cognitive aspects important.

Viewpoint 4: self-focused recovery

This viewpoint (n = 5) consisted of people who were internally focused in recovery. Aspects found important to this group involved the self and feelings of isolation. It seemed this group felt persecuted and unsupported by services and felt that they could only be dependent on themselves for recovery. Aspects that were not important to this group appeared to relate to support from others and psychiatric aspects of symptoms.

Additional feedback

Five participants left additional feedback outlining how they found completing the questionnaire and how useful the Q-sort was in helping understand their experiences. All feedback was positive about the procedure.

Development of the Subjective Experiences of Psychosis Scale

Potential items for the new scale [the Subjective Experiences of Psychosis Scale (SEPS)] came from three sources: (1) items identified from the existing literature on outcomes within mental health; (2) themes generated from phase 1; and (3) material generated from phase 2. A total of 146 items were identified from these methods which were reduced down to 52 items within three distinct categories: (1) impact of experiences of psychosis on mental health and well-being; (2) impact of support (such as treatment, support from friends and family, alcohol) on experiences of psychosis; and (3) severity of different dimensions of experiences of psychosis (e.g. frequency and pleasantness). Both positive and negative aspects of experiences of psychosis were identified in all three categories.

This initial draft scale was further scrutinised for content and face validity by a SURG resulting in the following changes: reducing the number of items to eliminate redundant or repeated items, introducing a scoring method that represented both the positive and negative aspects of psychosis for all items, including a scored example and rewording the instructions.

These modifications resulted in a 45-item self-report scale, consisting of three subscales assessing positive and negative aspects of experience of psychosis. Subscale 1 was ‘positive impacts of experiences’ and ‘negative impacts of experiences’, respectively, (31 items each), subscale 2 was ‘positive impact of support’ and ‘negative impact of support’ on experiences of psychosis (eight items each) and subscale 3 was ‘positive dimensions’ (two items) and ‘negative dimensions’ of experiences of psychosis (four items). Items were scored from 1 (not at all) to 5 (very much) with the addition of a ‘not applicable’ box for subscale 2.

Participants

Participants were required to meet the following inclusion criteria:

1. an International Classification of Diseases, Tenth Edition (ICD-10) diagnosis of schizophrenia, schizotypal disorder and delusional disorder

2. recent experience of delusions or hallucinations verified by the PANSS55 and Psychotic Symptoms Rating Scale (PSYRATS)72

3. aged between 18 and 65 years.

Materials

Measurement of symptoms of psychosis was assessed using the PANSS55 and PSYRATS. 72 Depression, hopelessness, anxiety and self-esteem were assessed using the Calgary Depression Scale,73 Beck Hopelessness Scale (BHS74), Beck Anxiety Inventory (BAI75) and the Self-Esteem Rating Scale (SERS76), respectively. Functioning was assessed using the GAF scale. 77 In addition, we explored how the SEPS related to self-reported progress in recovery from psychosis and using the QPR. 32 All measures have been shown to have good reliability and validity with psychosis samples.

Procedure

Participants were recruited from mental health teams in the north-west of England via mental health professionals. The study was carried out over three stages. Stage 1 required participants to fill out the SEPS and self-report measures and participate in PANSS and PSYRATS interviews. Data from stage 1 were used to carry out a PCA and investigation of internal consistency and concurrent validity of the SEPS. In stage 2, which was completed 1–2 weeks later, participants completed the SEPS again to investigate test–retest reliability.

In stage 3, which involved exploration of the SEPS in relation to the other measures over time, participants were required to repeat the assessments carried out in stage 1. Participants were given the option to take part in any, or all, of the stages.

Statistical analysis

An initial PCA was carried out on the SEPS and the internal consistency of identified subscales was computed using Cronbach’s alpha. Intraclass correlations (ICCs) and Spearman’s rank-order correlation analysis were used to examine test–retest reliability, concurrent validity and sensitivity to change between the SEPS and the other measures.

Initial data scrutiny

From the initial pool of items for the SEPS examined in stage 1, items with very low endorsements (< 20% of sample) and either high or low inter item correlations (< 0.200 or > 0.900) were removed. In addition, items with low test–retest scores and which contributed to a decrease in internal consistency were examined further. Together with these considerations and initial PCA feedback from participants and from the SURG, a number of items were removed resulting in a final number of 41 items on which all subsequent analyses were carried.

Principal component analysis: in the first instance, the data set was checked for suitability of PCA by examining the correlation matrix to ensure no multicollinearity or singularity. A PCA was then performed on each of the three SEPS subscales using varimax rotation. From examination of the scree plots and Eigenvalues, all three analyses resulted in two-factor solutions, with loadings > 0.4 taken as significant. The subscales and factors defined in this analysis were used in all further psychometric analyses. The three subscales had 29, 6 and 6 items, respectively. Each subscale consisted of two factors positive and negative items. Total variance explained by these subscales was relatively low.

Subscale 1: impact of experiences

Two factors contributed to 40.13% of explained variance, with factor 1 (‘positive impacts of experiences of psychosis’) contributing to 21.54% of the variance and factor 2 (‘negative impacts of experiences of psychosis) contributing to 18.59% of the variance.

Subscale 2: impact of support

Two factors contributed to 54.29% of the total variance explained, with factor 1 (‘positive impact of support on experiences of psychosis) contributing to 29.49% of the variance and factor 2 (‘negative impact of support on experiences of psychosis) contributing to 24.80% of the variance.

Subscale 3: dimensions of experiences of psychosis

The two identified factors explained 60.12% of cumulative variance; factor 1 (‘negative dimensions of experiences of psychosis) contributed to 33.47% of the variance, and factor 2 (‘positive dimensions of experiences of psychosis’) contributed to 26.65% of the variance.

Reliability

Cronbach’s alpha statistics were calculated for each subscale and these ranged from acceptable to good (Table 5). ICCs, using a two-way random model and absolute agreement were also computed to provide further evidence for the reliability of the SEPS. All subscales within the SEPS showed moderate to good ICCs (see Table 5). Furthermore, the positive and negative aspects of each subscale were not significantly correlated with each other, although there was a moderate correlation between the positive and negative support items (ρ = 0.364; p < 0.01).

Test–retest means, standard deviations (SDs) and correlation coefficients of the subscale scores of the SEPS are also shown in Table 5. The subscales had good retest reliability overall although the positive support subscale was only moderate. The majority of individual items in the subscales also indicated good retest reliability with only five items with correlations below 0.2.

Validity

Means and SDs for all measures are shown in Table 6. The relationship between the SEPS and measures of symptoms of psychosis, affect, esteem, function and recovery can also be found in Table 6.

Relationship between the Subjective Experiences of Psychosis Scale and psychosis symptom measures

The positive impacts of experiences and positive impact of support were not correlated with any of the symptom measures. However, there was a modest, negative correlation between the positive dimensions items of subscale 3 and the PSYRATS delusions subscale.

Negative impacts of experiences showed small to moderate correlations with all PANSS subscales apart from the PANSS negative subscale. There were moderate correlations with PSYRATS auditory hallucinations and delusions subscales. There were no significant correlations between negative support and any symptom measures. A similar picture to subscale 1 was observed for the negative dimensions of experiences, with moderate correlations with all PANSS and PSYRATS subscales but no correlation between the PANSS negative subscale.

Relationship between the Subjective Experiences of Psychosis Scale and measures of affect, esteem, functioning and recovery

The positive impacts of experiences and positive impact of support were not correlated with any measures. However, positive dimensions showed small, positive correlations with progress on recovery on the intrapersonal and interpersonal subscales of the QPR and showed a moderate positive correlation with self-esteem on the SERS positive subscale.

Negative impacts of experiences correlated moderately with depression, intrapersonal aspects of recovery, hopelessness, anxiety and self-esteem (both positive and negative subscales of the SERS). Negative impact of support was not correlated with any affect or functioning measures other than a small, positive correlation with anxiety. A similar picture to the impact of experiences subscale was observed for negative dimensions, with small to moderate correlations with depression, intrapersonal aspects of recovery, hopelessness, anxiety and negative self-esteem. In addition, there was also a moderate relationship between negative dimensions of symptoms and lower functioning on the GAF scale (Table 7).

Sensitivity to change of the Subjective Experiences of Psychosis Scale

Sensitivity to change was derived by calculating the change in score (baseline score minus follow-up score) on each measure. The change scores for each section within the SEPS were then correlated with all measures, which can be found in Table 8.

Relationships between change on the Subjective Experiences of Psychosis Scale and psychosis symptom measures

Change in the positive subscales did not correlate with change in any symptom measures. Change in negative impacts of experiences showed moderate, positive correlations with the PANSS general subscale and total PANSS scores. Change in negative support did not correlate with change on any of the psychosis measures. Change in negative dimensions showed moderate correlations with delusion severity measured by the PSYRATS delusions scale and with the PANSS general subscale scores.

Relationship between change on the Subjective Experiences of Psychosis Scale and measures of affect, esteem, functioning and recovery

There was no relationship between change in the positive aspects of any subscale and any other measures. Negative aspects on the SEPS were related to esteem and progress in recovery and suggested that increases in negative impact, poor support and increase in self-reported symptoms was related to decreases in self-esteem and recovery.

Feedback and distress

Participants took between 5 and 30 minutes to complete the SEPS. Ninety-nine participants (99%) rated the degree of distress experienced from completing the SEPS. Seventy-five participants (75%) reported that they experienced no distress from completing the scale. Fourteen (14%) reported experiencing ‘a little’ distress, five (5%) reported ‘moderate’ distress and 5 (5%) rated ‘quite a lot’ of distress. When distress was disclosed, the researcher debriefed the participant, ensuring they had the opportunity to discuss any concerns and consent was sought to share this information with the participant’s clinical team if this was considered appropriate.

Stage 1

Elements identified as pertinent to the definition or conceptualisations of recovery from psychosis were identified through a literature search of journals, policy documents, recovery measures and websites. This list was reviewed by the authors and collated into an initial list of statements (n = 141). Owing to the complexities of including a large panel of service users as the experts to be consulted, the authors decided to use a smaller panel of service users (a local service user reference group with 10 members, all of whom have personal experience of psychosis and using mental health services) during stage 1 to further refine this initial statement list. This resulted in the addition of a further three items, rewording of several items to increase acceptability to service users (e.g. including the word ‘experiences’ alongside ‘symptoms’ and removing the word ‘illness’ where possible) and deletion of seven items which were felt by the service users to be duplications.

Stage 2

In this stage the finalised list of 137 statements from stage 1 was collated and formatted into an online questionnaire and an alternative paper version. A demographics sheet was added to collect data on age, sex, mental health trust, diagnosis and length of diagnosis. Participants were given the option to state ‘prefer not to say’ for all demographic items. Participants were also asked if they would like to provide a postal or e-mail address so they could be invited to take part in the final stage of the study, although this was optional to allow complete anonymity if preferred.

Participants were then asked to rate the importance of each item on the statement list, in relation to defining recovery from psychosis, on a 5-point Likert scale (1 – essential, 2 – important, 3 – do not know/depends, 4 – unimportant and 5 – should not be included). A total of 426 participants completed the stage 2 questionnaire, although 45 were not included in the final sample (26 were deemed to be ineligible owing to reporting no experience of psychosis, 14 people did not complete the questionnaire, one person added a note to say they had already completed the study before and four people posted the questionnaire after the deadline). Results from the remaining 381 eligible participants were entered into an anonymised database and analysed by obtaining group percentages.

In accordance with the methods used by Langlands et al. 103 the following criteria were used to determine items for inclusion, exclusion and rerating.

1. Items rated by 80% or more participants as essential or very important to defining or conceptualising recovery are included as standard.

2. Items rated as essential or important to defining or conceptualising recovery by 70–79% of respondents in stage 2 were rerated in stage 3.

3. Any statements that did not meet the above two conditions were excluded.

This resulted in the inclusion of 71 items, the exclusion of 30 items and 36 items to be rerated in stage 3.

Stage 3

In stage 3, participants were asked to rerate only those items that 70–79% of respondents had rated as essential or important during stage 2 (n = 36 items). Two hundred and six participants provided contact details to be invited to take part in stage 3. The majority of participants opted to be sent a postal paper version rather than complete the questionnaire online. A total of 154 postal questionnaires were distributed in stage 3 and 52 participants were sent the online questionnaire link. One hundred participants completed the final stage resulting in a further 23 statements being included and 13 statements being excluded.

Participants

Data for the present study are a secondary analysis of a large study of psychosocial and neuropsychiatric predictors of recovery from psychosis. 127 Participants were aged 16–65 years, with a schizophrenia spectrum diagnosis as well as sufficient level of English literacy to complete the measures and the capacity to provide informed consent. A total of 122 participants entered the study (88 male, 34 female), four of whom completed only half of the measures. The majority of participants were white British (84.4%). Diagnoses, confirmed using ICD-10 checklists were as follows: schizophrenia (n = 75), schizoaffective disorder (n = 10), substance-induced psychosis (n = 5), unspecified non-organic psychosis (n = 15), acute and transient psychotic disorder (n = 12), persistent delusional disorder (n = 4) and post-partum psychosis (n = 1).

The Process of Recovery from Psychosis Questionnaire32

The QPR is a 22-item questionnaire, which was collaboratively developed with service users. The questionnaire measures subjective recovery in two domains: intrapersonal functioning (17 items) and interpersonal functioning (five items). An example of an item on the intrapersonal scale is ‘I have been able to come to terms with things that have happened to me in the past and move on with my life’ and the interpersonal scale, ‘meeting people who have had similar experiences makes me feel better’. Participants rate their agreement with statements on a 5-point Likert scale rating from ‘strongly disagree’ to ‘strongly agree’. Respondents may score between 0 and 88. The subscales have good internal consistency and test–retest reliability over short periods. 32 Cronbach’s alpha coefficients for the intrapersonal and interpersonal scales for this sample were α = 0.94 and α = 0.66, respectively.

Dichotomous recovery question

Appended to the QPR, participants were asked: ‘Do you consider that you have recovered from your illness? (Yes/No)’.

Recovery Analogue Scale

Appended to the QPR: ‘Please place a cross on the line below to indicate the extent to which you think you have recovered from your illness’ (10-cm analogue scale with left anchor ‘completely recovered’ and right anchor ‘not recovered at all’) to gauge the extent to which participants believe they are recovered.

Beck Hopelessness Scale

The BHS74 is a 20-item self-report scale assessing negative expectations about the future. Participants respond ‘true’ or ‘false’ to the items and responses are summed to obtain a total hopelessness score (range = 0–20). Cronbach’s alpha value for this sample was α = 0.92.

The Brief Self-Esteem Rating Scale – Short Form

The SERS76 is a 20-item measure of explicit self-esteem, assessing both positive (10 items) and negative (10 items) beliefs about the self. Participants rate how often each of the statements reflect their feelings about the self, on a 7-point Likert scale of ‘never’ to ‘always’. Cronbach’s alpha values for the positive and negative scales for this sample were α = 0.92 and α = 0.91, respectively.

Hospital Anxiety and Depression Scale

The Hospital Anxiety and Depression Scale (HADS)128 is a 14-item self-report measure. Of these items, seven assess depression, while the remaining seven items assess anxiety, over a period of the preceding week. Cronbach’s alpha coefficients for the anxiety and depression scales for this sample were α = 0.86 and α = 0.83, respectively.

Procedure

Prior to commencement of the study, a dedicated service user panel was consulted on all aspects of design, including wording of information sheets, inclusion of questionnaires, length of assessments, semantic concerns and appropriate amount of payment. Participants were recruited from CMHTs, inpatient settings, early intervention (EI) teams and the voluntary sector to ensure heterogeneity in experience of psychosis and service provision in the Greater Manchester area. Individuals were approached via their care co-ordinator. Those who were interested in taking part were briefed about the study and provided with a participant information sheet. They were then given a minimum of 24 hours to decide whether or not to take part. On agreeing to take part, a research assistant met with the participant to take informed consent and administer questionnaires, over a single visit or two visits at the participant’s request. Participants were remunerated for their time.

Statistical analyses

Chi-squared tests were performed in order to examine the association between sociodemographic variables and self-selected group (‘recovered’ vs. ‘not recovered’). For these purposes, sociodemographic variables were transformed into dichotomous variables using the following criteria: not in education, employment or training (NEET) status: (1 = NEET; 2 = in education, employment or training); marital status (1 = married/cohabiting; 2 = not living with a partner); religiosity (1 = no religious beliefs, 2 = religious beliefs); level of education [1 = compulsory education (primary/secondary) only; 2 = further education post age 16 years]; mental health provider (1 = EISs, 2 = other community services). Data were analysed using Predictive Analytics SoftWare statistics 18 [Statistical Package for the Social Sciences (SPSS) Inc., Chicago, IL, USA]. All variables were scrutinised before inferential analysis and valid parametric analysis were conducted. Correlational analyses were performed to determine the association between recovery measures. Correlations were also calculated between each of the recovery measures and the psychological variables: positive and negative self-esteem, hopelessness, anxiety and depression. Regression analyses using forward entry were conducted with our recovery measures as dependent variables.

Participants

Participants were aged 16–65 years, with a schizophrenia spectrum diagnosis with a sufficient level of English literacy to complete the measures and capacity to provide informed consent. At study entry, there were 122 participants (88 male, 34 female), four of whom completed only some of the measures. The majority were white British (84.4%). Diagnoses, confirmed using ICD-10 checklists, were as follows: schizophrenia (n = 75), schizoaffective disorder (n = 10), substance-induced psychosis (n = 5), unspecified non-organic psychosis (n = 15), acute and transient psychotic disorder (n = 12), persistent delusional disorder (n = 4) and post-partum psychosis (n = 1). Participants were recruited from EISs (n = 40), other community-based mental health teams (n = 81) and an inpatient unit (n = 1).

Measures of recovery

Psychosocial measures

Neuropsychiatric measures

Procedure

Prior to commencement of the study, a dedicated service user panel was consulted on all aspects of design, including wording of information sheets, inclusion of questionnaires, length of assessments, semantic concerns and appropriate amount of payment. Participants were recruited from CMHTs, inpatient settings, EI teams and the voluntary sector to ensure heterogeneity in experience of psychosis and service provision in the Greater Manchester area. Individuals were approached via their care co-ordinator. Those who were interested in taking part were briefed about the study and provided with a participant information sheet. They were then given a minimum of 24 hours to decide whether or not to take part. When an individual agreed to take part, a research assistant met with the participant to take informed consent and administer questionnaires, over a single visit or two visits at the participant’s request. Participants were remunerated for their time.

Statistical analyses

We hypothesised that recovery judgements, as measured by RecA, QPRIntra and QPRInter would form a single, latent dimension of recovery beliefs. We also hypothesised that the variables measuring negative emotion and related processes, namely anxiety, depression and negative self-esteem, would form a single latent dimension of negative emotion. We posited these measurement hypotheses as straightforward, robust and plausible ways of operationalising our multifaceted key constructs. If these hypotheses are supported, then the specification of the mediation model to test the primary hypothesis can be done in a structural equation modelling (SEM) framework using latent variables. Using latent variables allows for the separation of ‘true’ from ‘error’ (more accurately, ‘specific’) construct variance, giving the model greater statistical power and validity than using observed variables alone. 134

A series of factor analyses [exploratory factor analysis (EFA) and confirmatory factor analysis (CFA)] and SEM were specified to test our measurement modelling and mediation hypotheses. Using a SEM approach to estimating mediation models has been shown to reduce bias in the parameter estimates for the mediation effects and to give adequate power to detect mediation effects with samples as small as 100. 134 All models were fitted using Mplus 6.1 (Muthén & Muthén, Los Angeles, CA, USA) and estimated by maximum likelihood. Because of the sensitivity of mediation models to parametric assumptions and the modest sample size available in this study, we computed parameter standard errors (SEs) using bias-corrected bootstrapping based on 1000 bootstrap replications. 135 Latent variables in all factor models were statistically identified by setting the factor loading of one of the factor indicators to one. Alpha for hypothesis testing was set at 0.05 and all hypothesis tests were two-tailed.

To have confidence in hypothesis tests of parameter estimates, it is reasonable (though not strictly necessary) to require that an over-identified statistical model as a whole should provide a reasonable fit to the data. Model fit was assessed following the recommendation that a combination of the standardised root mean-square residual (SRMR) < 0.08 and the Cumulative Fit Index (CFI) > 0.95 provides an optimal trade-off between type I and type II errors in models estimated by maximum likelihood in smaller samples. 136

Measurement model of recovery

The unidimensional CFA model of QPRIntra, QPRInter and RecA proved to be a poor one. Although RecA and QPRIntra were well correlated (r = 0.58), RecA and QPRInter were not (r = 0.19). Although no formal measures of goodness of fit were available for this just identified unidimensional CFA model1, the estimated solution was improper, with negative residual variances for the interindicator and a non-positive-definite hessian matrix. The a priori measurement hypothesis for recovery was therefore rejected. We subsequently embarked on exploratory psychometric analysis of the QPR, to search for a coherent subset of the 22 QPR items that would reflect the desired notion of recovery beliefs. This use of exploratory, as opposed to a priori confirmatory, measurement modelling runs the risk of capitalising on chance and sampling variability. However, our goal here was only to find an internally reliable and interpretable set of recovery beliefs; the measurement model was not selected based on its providing a good fit with the other variables that would comprise the mediation model.

A two-factor EFA model of the 22-item QPR was estimated by maximum likelihood. Model fit statistics were somewhat mixed, the CFI indicating moderate but not close fit (CFI = 0.91), the SRMR indicating close fit (SRMR = 0.05), but the chi-squared test resoundingly significant [χ² (df 188) = 308.46; p < 0.01]. Neither varimax, oblimin nor geomin rotation could recover an approximate simple structure for the two factors, with many items having significant loadings on both factors. Using the oblique geomin rotation (suitable when item complexity is high, i.e. items load on more than one factor), only two out of the five items considered a priori comprising the interpersonal subscale had strong loadings on only the second (i.e. ‘interpersonal’) factor.

The hypothesised two-factor solution did not appear well supported in these data. However, a subset of the QPR items had strong loadings (> 0.7) on just one of the factors, and we decided to use these items as a unidimensional scale of recovery beliefs. These items were: (1) ‘I feel better about myself’, (2) ‘I feel able to take chances in life’, (6) ‘I feel that my life has purpose’, (12) ‘I can take charge of my life’ and (19) ‘I can actively engage with life’. Alpha reliability for these items was 0.88.

A second unidimensional CFA model was fitted by maximum likelihood, using these five QPR items and RecA as indicators. Model fit statistics were excellent [χ² (9) = 11.42, p = 0.25; CFI = 0.99; SRMR = 0.03]. All standardised factor loadings were highly significant and between 0.61 (RecA) and 0.79 (Questionnaire about the Process of Recovery item number 12). This factor was labelled Recov.

Measurement model of negative emotion

Anxiety, depression, and negative self-esteem were well correlated, between r = 0.66 and r = 0.72. A unidimensional CFA model of these three indicators was estimated. As stated above, overall goodness of fit statistics are not available for such a just identified model. However, with standardised factor loadings all highly significant (depression 0.78, negative self-esteem 0.83 and anxiety 0.86), the hypothesis of unidimensionality was clearly plausible. This factor was labelled Nemo.

Joint model of recovery and emotion

A joint CFA model of Recov and Nemo was fitted, to further test the measurement reliability of these constructs as operationalised above. Two correlated latent factors were specified, each factor indicated uniquely by the variables listed above, with no cross-loadings. Model fit statistics for this now over-identified model were good, indicating close fit between the model and the data [χ² (26) = 53.70, p < 0.01; CFI = 0.96; SRMR = 0.05].

The correlation between the recovery and emotion factors was very high, r = –0.83 (SE = 0.05), casting some doubt on the discriminant validity of these two related constructs. We, therefore, fitted two further models in an ad hoc test of the dimensionality of the recovery and negative emotion constructs. The first model was a unidimensional CFA model, the single latent factor indicated by all nine of the Recov and Nemo variables. The second model had two latent factors. The first factor was again a general factor, indicated by all nine variables. The second factor was indicated by the Nemo variables only and would, therefore, represent any correlations among these variables not accounted for by the general factor. The two factors were specified as uncorrelated, the association between the latent factors being represented by the cross-loadings on the Nemo variables. These two models were nested, meaning that the parameters of the first (unidimensional) model were a subset of those in the second model. The improvement in fit of the second model over the first model could therefore be formally tested using a likelihood ratio chi-squared test.

The likelihood ratio chi-squared test was significant [χ²_LR (3) = 41.90; p < 0.01], the second factor improving model fit significantly and supporting the hypothesis of separate recovery and negative emotion factors, despite their high correlation.

Latent mediation model of recovery on predictors, mediated by negative emotion

Finally, we specified a SEM mediation model to test our primary hypothesis. Recov was the dependent variable and Nemo the mediator. Both the dependent and mediator variables were regressed on the predictor variables Pos, Neg, Multidimensional Locus of Control Scale – internality subscale (MLCSi), BACS and IS. This was, therefore, a model of partial mediation, with direct effects of the predictors on the dependent variable as well as indirect effects via the mediator (Figure 5).

FIGURE 5.

Latent mediation model of recovery. (See Table 23a–c for parameter estimates.) χ² (df 61) = 109.92, p < 0.01; CFI = 0.93; SRMR = 0.06. QPR, Questionnaire about the Process of Recovery item number.

The substantive conclusions regarding our mediation hypotheses would be assessed using significance tests of the indirect effects (i.e. the product of the coefficients for the mediator regressed on the predictors and the outcome variable regressed on the mediator). Because indirect effects are based on products of parameters, confidence limits based on asymptotic (i.e. normal) theory are expected to be biased; we therefore used significance tests based on SEs computed using bias-corrected bootstrapping (1000 replications), as recommended. 135

Tables 23a–c show the results of the mediation model. The mediator, Nemo, had a very strong, negative and significant relationship with Recov, accounting for an R² of 68%. Of the predictor variables, only MLCSi had a significant direct effect on Recov, with a positive relationship. MLCSi also had a significant, positive, indirect effect on Rec, which was composed of a negative relationship between MLCSi and Nemo and a negative relationship between Nemo and Rec. The only other significant indirect effect was for Pos, a negative relationship. Pos was positively associated with Nemo, which was negatively related to Recov. The standardised effect sizes of these significant mediation effects were –0.311 for Pos and 0.211 for MLCSi, corresponding, by Cohen’s rule of thumb, to ‘medium’ sized effects. 137 The R² for Recov in the full model was very large, 72%. The majority of this was due to the R² of 50% uniquely associated with the total indirect effects. 138

A final model was fitted to check the robustness of these results to the possibly confounding effects of Sex (binary), Age (years), relationship status (single vs. not single) and employment status (NEET vs. not NEET). Both Recov and Nemo were regressed on these four extra predictors and the mediation model was rerun. Comparing the difference in model chi-squares between this and the original model using a likelihood ratio test found no significant improvement by including the four extra predictors [χ²_LR (28) = 32.995; p = 0.236], and so presented no evidence of confounding by these variables. Zero-order correlations for the variables are provided in Table 24.

Participants

Data for this study were collected throughout the recovery programme from all appropriate studies to create a comprehensive data set. Participants were recruited from EI teams, CMHTs, inpatient settings and voluntary services across the north-west of England. Participants were included in the data set if they were aged 16–65 years, had a diagnosis of schizophrenia spectrum disorder, sufficient understanding of the English language to allow them to complete the measures and the capacity to provide informed consent. A total of 110 participants were assessed at baseline and 6 months.

Measures

Procedure

Recruitment took place across EI teams, CMHTs, inpatient settings and voluntary services across the Greater Manchester area to ensure heterogeneity of service provision and experience of psychosis. Potential participants were approached by the care team and offered information about the study. Interested participants were given a minimum of 24 hours to read the participant information sheet and decide whether or not to take part. Those who agreed to take part met with a researcher to complete a consent form and the baseline study measures. The researcher then attempted to contact all participants again 6 months later to repeat the set of measures in a follow-up assessment. Participants were recompensed for their time.

Statistical analysis

All models were fitted in Mplus version 7 (Muthén & Muthén, Los Angeles, CA, USA) and estimated by maximum likelihood. SEs were estimated using the Huber–White Sandwich estimator, robust to non-normality and heteroscedasticity in the outcome variables. Model log-likelihoods and the likelihood ratio tests were computed using Satorra–Bentler adjustments for non-normality. Nested models were compared using Satorra–Bentler-corrected likelihood ratio chi-squared tests.

Sample characteristics

Demographic characteristics of the sample are shown in Table 25. The average age of participants was 37.3 (SD 11.62) years. Diagnoses at referral were schizophrenia (n = 50), schizoaffective disorder (n = 13), persistent delusional disorder (n = 7), unspecified non-organic psychosis (n = 4), and acute and transient psychotic disorder (n = 2). The remaining 30 participants had not been given a diagnosis but were experiencing psychosis. Participants were recruited from EISs (n = 27), other community-based mental health teams (n = 45) and an inpatient service (n = 1). Data on service type at referral were missing for 37 participants.

Model variables

Variables are sufficed 1 to indicate time 1 (baseline) assessments and 2 to indicate time 2 (6-month follow-up) assessments.

Core variables

• Recovery 1 (Rec1) and recovery 2 (Rec2). This variable consisted of the 15-item total QPR score at time 1 or time 2, respectively.

• Negative emotion (Nemo1 and Nemo2). A composite variable for negative emotion was constructed by taking the mean of scores from the Calgary Depression Scale and the SERS-N. The SERS-N is scored from 10 to 70 whereas the Calgary Depression Scale is scored from 0 to 27. To avoid the composite measure being dominated by the higher scores of the SERS-N, the raw SERS-N scores were divided by seven before taking the composite mean, which gave both contributing scales similar means and SDs.

Test variables

• Symptoms (PANSS1 and PANSS2). A composite variable representing the overall mean of the seven positive, seven negative and 16 general PANSS scale items was created.

• Hopelessness (Hopeless1 and Hopeless2). Total score from BHS was utilised.

• Positive self-esteem (SERS-P1 and SERS-P2). Total score from the positive subscale of the SERS.

• Functioning (Func1 and Func2). Functioning score utilised the PSP scale if available and the functioning subscale of the GAF scale if not.

Exogenous covariates

All of these variables were measured at time 1:

• Age.

• Education or employment. A binary variable indicating whether or not the participant was engaged in paid employment (full or part time), study or voluntary work (coded as 1) versus being unemployed, retired or in receipt of disability living allowance (coded as 0).

• Marital status. A binary variable indicating whether or not the participant was married or living with a common-law spouse (coded as 1) versus being single, divorced or widowed.

• Religious beliefs. A binary variable indicating whether or not the participant was a theist, that is believed in the existence of a deity (coded as 1).

• Early intervention. A binary variable indicating whether or not the participant was recruited from an EIS.

Model of recovery and negative emotion

Both Rec and Nemo were highly correlated within time point (r = –0.66 in both time points). This correlation was reduced to r = –0.45 at time point 2, after controlling for the previous time point.

Table 26 shows that both recovery and negative emotion were significant predictors of recovery at time 2, but that only negative emotion was a significant predictor of negative emotion at time 2.

The R² for recovery at time 2 was 31.8% and for negative emotion 58.3%. The large R² for negative emotion at time 2 was mainly accounted for by its relationship with negative emotion in the previous time point.

Further development and testing of the model

The core model of recovery and negative emotion (as illustrated in Figure 6) was compared with test Models 2.1–2.4 using Satorra–Bentler corrected likelihood ratio chi-squared tests for nested models.

FIGURE 6.

Core model (Model 1): recovery and negative emotion at time 2 predicted by recovery and negative emotions at time 1.

In each test model, the core model was added to by including extra predictors of the outcome variables (recovery and negative emotion at time 2). In Model 2.1, the extra predictors were the overall PANSS symptom scores at times 1 and 2. In Model 2.2 the extra predictors were hopelessness time 1 and hopelessness time 2 scores. Positive self-esteem was the extra predictor in Model 2.3 and functioning was included in Model 2.4.

Table 27 shows the results of the likelihood ratio tests comparing each of Models 2.1–2.4 with the core model.

All models improved on the fit of the core model, with the largest improvements seen in the prediction of recovery scores at time 2 owing to hopelessness and positive self-esteem.

A further model was then fitted, which combined the predictors that were tested separately in Models 2.1–2.4 into a single Model 2.5.

The parameter estimates for Model 2.5 are shown in Table 28.

Recovery at time 2 was predicted significantly by symptoms at time 1 and hopelessness and positive self-esteem at time 2. After accounting for these influences, recovery at time 1 was no longer a significant predictor of recovery at time 2.

Negative emotion at time 1 was a significant predictor of negative emotion at time 2, along with symptoms, hopelessness and positive self-esteem at time 2.

Checking for endogeneity

There was a possibility that regressing closely related constructs on one another within each data collection time point would be stretching assumptions of exogeneity with regard to these constructs. To test for this we ran a model, Model 2.6, which regressed recovery and negative emotion at time 2 on the other test variables from time 1 only, not including the other time 2 variables as predictors.

From Table 29, we see that symptoms and positive self-esteem at time 1 are significant predictors of recovery beliefs at time 2, each with broadly equal magnitude. These predictors accounted for 44% of the variance in recovery at time 2.

By far the strongest predictor of negative emotion at time 2 is the time 1 score on this variable. No other time 1 variables were significant predictors of negative emotion at time 2. The R² for this model was 61%.

The fact that recovery and negative emotion have different sets of predictors is evidence in support of the fact that these are distinct constructs.

Checking for confounding

In the previous analyses, no attempt was made to control for the effects of potential demographic and other confounding factors. Such variables available in this study were age, sex, marital status, employment status, religious beliefs and whether or not the participant was drawn from an EI or other service.

Fitting the same series of models above, but this time conditioning the outcome variables (i.e. Rec2 and Nemo2) on these covariates, the pattern of model improvement was identical to that seen in Table 29(Table 30). In these models, the only variable that had any significant effects was sex.

We decided to fit an additional model with sex as a predictor, but this time exploiting the fact that we could plausibly assume that sex was a truly exogenous variable and so include it as a predictor of both the time 1 and time 2 outcomes.

The results for this model, Model 3, are shown in Table 31.

The pattern of significant results in Model 3 is identical to that in Model 2.6 with the notable exception that although sex is not a significant predictor of recovery beliefs at time 1, it is a significant and substantial predictor of recovery beliefs at time 2, with men having an average recovery score 4 points less than women. This is despite the fact that sex was not a significant predictor of negative emotion at either time point.

Participants

This study was reviewed and approved by the NHS North West 10 Research Ethics Committee Greater Manchester North (REC reference 10/H1011/36) and was sponsored by the University of Bangor.

Sixty-eight service users meeting criteria for a diagnosis on the schizophrenia spectrum disorders (ICD-10 schizophrenia, delusional disorder, schizoaffective disorder and psychotic disorder not otherwise specified) were recruited into the present study. Service users were recruited from CMHTs, hospital inpatient wards and day hospitals across several trusts in the North West, including Greater Manchester West Mental Health NHS Foundation Trust, Cheshire and Wirral NHS, Pennine Care, 5 Boroughs Partnership, and Manchester Mental Health and Social Care Trust. Inclusion criteria were: aged between 16 (required for informed consent) and 65 years; being in current contact with mental health services; ability to provide informed consent; and sufficient competency in English to complete the assessments.

Procedure

Participants completed baseline clinical and questionnaire assessments and were then instructed to complete the ESM diaries. Some service user participants were later followed up at 3 and 6 months to assess changes in and stability of recovery; these data are not reported here.

Baseline measures

Experience sampling methodology

The ESM is a pseudorandom time sampling assessment that has previously demonstrated feasibility, validity and reliability in people diagnosed with psychosis. 121,123,164 All service user participants received an electronic digital watch to wear, which was programmed to bleep 10 times a day on a pseudorandom schedule between 07.30 and 22.00 over the course of 6 days. Participants were asked to complete a short diary around their recovery on hearing a bleep.

The items included in the ESM diary measured psychological experiences on a 7-point Likert scale. The items were adapted from our previous studies121,122,165 and integrated recovery-related variables including positive and negative affect, self-esteem, activity-related stress, self-efficacy, rumination, recovery status and savouring ability.

Measures of self-esteem (four items) and single-item measures of visual hallucinations, auditory hallucinations and current paranoia were taken from previous ESM studies. 121,122,165 From these studies we also took single-item measures of appraisals of symptoms: whether or not visual hallucinations were present, whether or not auditory hallucinations were pleasant and whether or not persecution was believed to be deserved. Two additional measures were piloted by a team of service user researchers before inclusion in the study.

Momentary recovery was generated as the mean of the three ESM recovery items: (1) ‘I felt limited by psychological problems’; (2) ‘I have worries about psychiatric problems’; and (3) ‘I have felt mentally well’. Items (1) and (2) were first reverse coded so the overall recovery score would be on a positive scale, that is higher values indicate better recovery. Momentary recovery was negatively skewed but residuals of the models seemed Normally distributed enough for the assumption to hold.

Three items derived from the BHS166 were included to measure hopelessness: ‘my future seems darks to me’, ‘I feel optimistic about the future’ and ‘the future seems vague and uncertain to me’.

Analysis

All analyses were performed in Stata version 11.0 (StataCorp LP, College Station, TX, USA) using the xtmixed command, and Mplus version 7.1 (Muthén & Muthén, Los Angeles, CA, USA). To test the factor structure of the predictor scales, we used a two-level CFA with participant and beep as the levels, postulating a one-factor solution for each scale to test that the items did load onto a single factor.

Fluctuations in scores were calculated as the absolute difference between ESM measures at two consecutive time points.

As the ESM measures have a three-level hierarchical structure (beeps nested within days nested within participant), we use multilevel modelling to account for the clustering in outcomes within participants. We fitted three-level random intercept models containing in the error structure: a random intercept for each participant, a random intercept for each day within a participant and a beep-specific error term.

In order to test the hypothesis that the ‘Right Now’ variables are correlated with recovery at time 1 we considered predictors and outcomes at concurrent entry points; the results will demonstrate if there is an association between the predictors and outcomes which, if significant, might suggest that they occur simultaneously. To test the hypothesis that the Right Now variables at time 0 predict recovery at time 1 lagged values of the predictors were used in the multilevel models instead of concurrent predictor values.

Participant characteristics

A total of 80 service users took part in the study (Table 32 shows summary table of participant characteristics). Demographic information was missing for 10 participants. The majority of participants were referred based on having experience of psychosis but no diagnosis specified (n = 35, 58%), while 36.7% (n = 22) had a diagnosis of schizophrenia and 3 participants (5%) had a diagnosis of schizoaffective disorder.

Summary data for completed measures can be seen in Table 33.

Is there a relationship between baseline recovery and fluctuations in recovery?

We compared fluctuations in momentary recovery between groups defined according to whether their baseline ratings were ‘yes’ (recovered) or ‘no’ (Table 34). Baseline recovery was entered into the model as a fixed effect and allowing different residual variance of the two baseline recovery groups. The ‘recovered’ group at baseline have significantly higher mean recovery scores recorded in everyday life than the ‘not recovered’ group, but there was more variation in recovery in the ‘not recovered’ baseline group.

For the baseline QPR recovery (Table 35), the total baseline score (of the 15 items) was entered as a fixed effect. This score was then split into tertiles and entered into the random part of the model to allow the residual variance to be different for each group, as with the binary baseline recovery. Group 1 included total baseline scores of 30–49, group 2 scores of 50–56 and group 3 scores of 57–73. The results imply that for a unit increase in baseline QPR score, mean momentary recovery scores increase by 0.09. However, showing a similar pattern to the binary baseline recovery variable, there is more variation in momentary recovery judgements for those with lower baseline QPR scores than for those with higher baseline QPR scores.

Which experience sampling methodology variables correlate with current momentary recovery judgements?

Table 36 shows the relationship between momentary recovery judgements and individual ESM measures taken at the same time point. There are significant correlations between the concurrent variables and momentary recovery judgements. As expected, positive measures (self-esteem) are positively correlated with recovery, whereas the symptom measures are negatively correlated with recovery. For example, for a unit increase in self-esteem, momentary recovery increases by 0.38, whereas for a unit increase in hopelessness momentary recovery decreases by 0.35.

All of the current variables were entered into one model (Table 37), excluding the pleasantness of hallucinations and deservedness ratings for which there were not sufficient data (including these variables would cause the model to use only 145 observations whereas, excluding them the model uses 2273 observations). In this model, self-esteem, hallucinations (visual and auditory), and paranoid are all significantly associated with recovery judgements.

Do variables at the previous time point predict momentary recovery judgements?

Lagged variables for predictor ESM variables were generated, restricting lag to be within day. Each line in Table 38 corresponds to a separate model with the lagged variables as the predictors, controlling for baseline recovery group (yes/no binary judgement) as this was found to be a significant predictor of recovery scores in a previous analysis. It can be seen that self-esteem, hopelessness and the symptom measures (but not participants’ appraisals of their symptoms, as indicated by the pleasantness of their hallucinations or their beliefs about whether or not they deserved to be persecuted) predicted the extent to which they believed they were recovered.

Table 39 shows the results when all variables were entered into a single model, controlling for baseline recovery group. Again, because of the small number of observations for the symptom appraisal variables, these were excluded, enabling an analysis with 169 observations instead of 172. As can be seen, in this analysis all of the variables with the exception of visual hallucinations are significant predictors. Substantially similar results were obtained by using baseline QPR scores instead of the binary recovery classification.

Sensitivity analysis

The results above include data from all participants regardless of the total number of observations completed per person. In line with common practice of only including data from participants who have completed at least one-third of the potential observation, a sensitivity analysis was conducted to identify any changes to the results. Analysis 3 (in which lagged variables were used to predict recovery judgements) was repeated including only participants who had completed at least 20 observations overall: 54 out of the 68 service users. Results were very similar to the full sample, with only a slight decrease in magnitude of the effect for each of the predictor variables. The only variable to significantly change was visual hallucinations, which now had an effect size of –0.05 (SE 0.03; p-value 0.056).

Participants

Participants were included in the study if they were aged 18–65 years, had sufficient English language skills, were in contact with mental health services, met ICD-10 criteria187 for non-affective psychosis and had been considered by their clinical team to be symptomatically stable for at least 1 month.

Recruitment

Participants were recruited from community secondary mental health services in three NHS trusts in the north-west of England. Of 119 service users identified by mental health professionals, 90 gave informed consent to take part. Participants were interviewed in their home or a clinical service base, receiving £10 as a token of appreciation.

Measures

Assessments included a sociodemographic questionnaire, the PANSS55 and the GAF77 scale. Inter-rater reliability was high for the PANSS [ICC coefficient 0.96, 95% confidence interval (CI) 0.93 to 0.98] and GAF scale (ICC coefficient 0.95, 95% CI 0.85 to 0.99).

Participants were provided with an information sheet (with counterbalanced presentation) describing hypothetical treatment options: (1) TAU, (2) self-help manual with telephone CBT and peer support (weekly 30-minute sessions for up to 6 months) in addition to TAU [self-help and telephone support (SHT)] or (3) as SHT plus a fortnightly 2-hour group support session for up to 6 months [self-help and group support (SHG)] facilitated by a psychological therapist and a peer supporter. TAU in secondary care mental health services typically consists of the provision of support by a care co-ordinator and monitoring of psychiatric medication. Telephone CBT has been shown to be equivalent to face-to-face CBT in other populations185 and provides increased flexibility about therapy contact which may be particularly helpful with this client group; it was offered as an option here as a way of maximising the flexibility for delivery of CBT. Participants were asked about their hypothetical preferences for participation in a future study, preferred treatment, willingness to be randomised and reasons for their choices. A qualitative interview method was chosen to ensure participants’ understanding and allow open-ended responses and clarification of reasons for preferences.

Data analysis

The data were analysed using SPSS version 17. Missing items were substituted with the mean when < 10% of the scale items were missing. Comparisons between the groups opting for each of the treatment options (TAU, SHT or SHG) were performed using chi-squared tests for categorical data, t-tests/analysis of variance (ANOVA) tests for parametric continuous data, and Mann–Whitney U-tests for non-parametric continuous data. Owing to the use of multiple comparisons, a stringent significance level of p < 0.01 was used.

Content analysis was used to explore reasons reported by participants for their choices, which involved categorising participants’ reasons into mutually exclusive themes188 following discussion with a researcher with lived experience of mental health challenges. The agreement between two raters (KS and SW) for each of the categories was as follows: willingness to participate, 84.4%; unwillingness to participate, 69.2%; preference for TAU, 70.6%; preference for SHT, 91.2%; preference for SHG, 73.8%; willingness to be randomised, 76.7%; and unwillingness to be randomised, 84.4%; ranging from acceptable to good. 189

Sample

Ninety participants were included. The majority of participants were white (n = 77), male (n = 68), unemployed (n = 63) and met criteria for a diagnosis of schizophrenia (n = 59). The remainder met criteria for other ICD-10 defined non-affective disorders, for example schizoaffective disorder or delusional disorder. The mean age of participants was 37.22 years (SD 11.34 years; range 19–63 years). Means for the clinical measures were 28.6 (SD 7.32) for the PANSS general scale, 14.19 (SD 5.57) for the positive scale and 12.05 (SD 3.58) for the negative scale, and 47.98 (SD 12.90) for the GAF scale. The median number of years since first experience of psychosis was 8 years (range 1–34 years) and the median number of hospital admissions of participants was two (range 0 to ≥ 11 admissions).

Key findings

Participants’ willingness to participate in the proposed PRPT was high and SHT was the most popular option. Most participants reported that they would prefer not to agree to randomisation. The results are reported in further detail below, along with the results of the content analysis of participants’ reasons for their preferences.

Willingness to participate in the proposed PRPT

The majority of participants (90%, n = 81) stated that they would be willing to consider participating in a PRPT evaluating psychological therapy for psychosis. Results of the content analysis are reported in Table 40, which illustrates the diversity of participants’ views about participating in a psychological treatment trial.

Preferences for treatment

Most of the participants who were willing to participate in the PRPT had a preference for therapy as opposed to TAU only (Table 41). The telephone support option (SHT) was favoured more than the telephone plus group support option (SHG). Results of the content analysis highlighting the reasons for this are displayed in Table 42.

Willingness to be randomised

Of the 81 participants who reported that they would consider taking part in the PRPT, the majority (57%, n = 46) said they would be unwilling to be randomised to treatment group. Results of the content analysis highlighting the reasons for this are shown in Table 43.

Characteristics associated with preferences

None of the assessed sociodemographic and clinical variables was found to be significantly associated with preferences for therapy or willingness to be randomised. Comparisons of those who were willing and unwilling to participate could not be performed owing to the low number who were unwilling (n = 9).

Key findings

Hypothetical preferences indicated that willingness to participate in a PRPT of a cognitive–behavioural psychological treatment for psychosis was high in comparison with other psychiatric research studies. 183 Both personal and altruistic benefits were evident, as has been found in other research. 181 Reasons for potential participation included a desire to gain information or understanding, to have a new experience or activity, or to share experiences with others, suggesting that there may be gaps in service provision in these areas. Reasons for non-participation in the future trial largely reflected a negative view of the treatment offered, or satisfaction with current treatment, consistent with a cost–benefit analysis.

The majority of the sample had a preference for therapy in addition to TAU, consistent with previous research regarding psychological therapy for other mental health problems. 190 The telephone support condition (SHT) was more popular than the telephone plus group support option (SHG), although considerable numbers of participants opted for each. A substantial number of participants opted for TAU alone, in contrast with previous findings. 178 Many of those who opted for TAU in the current study reported perceived lack of need or desire for the additional therapy. Reasons for preferring SHT indicated a desire to receive therapy or gain information and the lack of desire for group support, often due to a specific dislike of group situations. In contrast, reasons for preferring SHG often reflected a positive view of the social or group support, or the higher level of support available. Notably, participants’ reasons for their treatment preferences showed wide individual variation.

Most participants reported unwillingness to be randomised, (in line with King et al. ’s175 review) and cited a desire to receive the preferred treatment or for allocation to be based on necessity, or therapy being viewed as unhelpful. Reasons for willingness to be randomly allocated included altruistic reasons, a personal positive belief in chance and the absence of a strong preference. Participants seemed able to understand randomisation despite previous research191 indicating that this may not be the case. None of the sociodemographic and clinical variables measured was significantly associated with preferences, supporting the need for an individualised approach.

Limitations

The sample may be biased owing to recruitment via mental health services, although it is likely that findings are generalisable to future similar trials. Further, the reported high rate of willingness to participate in the proposed PRPT may reflect participants’ general acceptance of research. Bespoke, non-validated measures were used to determine preferences, which is not ideal. In addition, hypothetical rather than actual preferences were assessed, although the two have been found to be relatively consistent. 192 It is possible that cognitive functioning (which was not assessed) and/or the presence of symptoms may have impacted on participants’ ability to make decisions. 193 However, participants were able to give coherent reasons for their preferences, and clinical severity was not related to preferences.

Implications for research and clinical practice

The findings indicate that many service users with experience of psychosis welcome choice about psychological treatment and highlight the importance of individual differences. Future research may involve the more detailed investigation of the PRPT design (including feasibility and statistical considerations and differences in outcomes compared with RCTs), in-depth qualitative research, development of standardised measures of preferences and interventions to improve the understanding of research. Finally, it would be of interest to determine whether or not the participants’ stated hypothetical preferences and reasons are reflected in real decisions.

This research adds to the growing literature regarding psychological treatment for individuals with experience of psychosis, for which the ethics of treatment choice may be particularly pertinent owing to the experience of marginalisation. The results indicate that preference trials are feasible in terms of overall popularity and proposed treatment options, including TAU. Randomisation was often viewed negatively; therefore, preference designs may enhance recruitment and prevent ‘resentful demoralisation’. 170 Preferences were not associated with any of the variables measured, suggesting that it is difficult to predict preferences on the basis of individual characteristics and that all service users should be supported to state their choices.

Pilot study context and design

This self-help therapy and recovery proof of concept study constituted phase 2 of project 4 in the recovery programme of research. This followed from phase 1 work developing the self-help recovery guide for psychosis and exploring people’s hypothetical preferences for different levels of therapeutic support. The proof of concept study used a randomised design whereby participants were allocated to either receive TAU or extra therapeutic support in the form of the recovery guide plus telephone CBT provided in 30-minute sessions over 6 months, in addition to monthly peer support sessions. Participants met with a researcher at the start of the project and at 6 months and 12 months to assess their experiences, emotions, functioning and recovery using a range of measures.

Recruitment

The proof of concept study aimed to recruit 30 participants from three NHS trusts: Pennine Care, Greater Manchester West and Manchester Mental Health and Social Care Trust. Participants met the following inclusion criteria:

• aged 18–65 years

• in contact with mental health services

• met ICD-10 criteria for non-affective psychosis (schizophrenia, schizophreniform disorder, schizoaffective disorder, delusional disorder)

• at least 1 month of stabilisation if the person had experienced a symptom exacerbation in the last 6 months

• able to provide written informed consent

• able to read the self-help manual/assessments

• able to use the telephone.

Retention and therapy uptake

Of the 29 people in the study at the start, 27 were followed up at 6 months and the full 29 were followed up at 12 months, demonstrating an excellent retention rate for the project as a whole.

As a result of the random allocation, 15 people continued with their TAU with no extra support provided by the project and 14 people were offered therapy and peer support sessions and the recovery guide.

Therapy session uptake

Of the whole therapy sample:

• mean number of therapy sessions conducted: 14

• mean number of therapy sessions offered: 17

• mean per cent of therapy sessions conducted, of those offered: 79%

• number of withdrawals from therapy: 5 (out of 14).

Among those who did not withdraw:

• mean number of therapy sessions conducted: 17

• mean per cent of therapy sessions offered: 20

• mean per cent of therapy sessions conducted, of those offered: 87%.

Sections of the guide used

The session notes of those participants in therapy were reviewed in order to discern which guide sections were utilised. A summary of this information is provided in Figure 8, showing that most participants were able to utilise chapters 1 and 2 of the self-help guide, and a good proportion also used additional chapter 3 sections (chapter 1, What is Recovery from Psychosis, chapter 2, Making Sense of Yourself and chapter 3, Your Current Situation).

FIGURE 8.

Sections of the guide used.

Quantitative analyses

An additional aim of the pilot study was to assess the potential impact of the guide and therapy on people’s functioning, recovery and experiences of psychosis.

The primary outcome was a measure of recovery; the QPR. 32 Analysis of the data showed that there were no significant differences between the therapy and TAU groups on the QPR measure at baseline, 6 months or 12 months. Other measures showed variations in their initial starting scores and change scores. Allocation group (i.e. TAU vs. therapy) significantly predicted 12-month follow-up scores while controlling for baseline scores on measures of negative self-esteem, depression, negative symptoms of psychosis and general symptoms of psychosis. Those in the therapy condition had higher mean scores on these measures at 12 months, which suggested the possibility that the additional therapy (consisting of a combination of the recovery guide, telephone-delivered CBT and peer support) actually hindered recovery. When we only included those people in the therapy group that had more than five therapy sessions in total, only the effects for negative self-esteem and general symptoms of psychosis remained significant.

The possibility that this intervention was delaying recovery was considered by the study team and the independent DMC. However, it was decided that the small sample size of the study made it difficult to make any firm conclusions on the basis of these analyses. Given that the preference design of the next phase meant that people were able to choose which arm of the trial they were allocated to and the qualitative feedback (see Qualitative feedback) was largely positive, it was decided that the next phase was merited.

Qualitative feedback

Feedback interviews were conducted by a service user researcher with four participants receiving telephone therapy. The interviews explored participants’ views on the guide (Table 44), telephone therapy (Table 45), reasons for taking part, peer support (Table 46), CBT model and suggestions for changes. A selection of comments on therapy, the guide and peer support is provided below (roughly proportionate to the total amount of comments in each category). Many suggestions were incorporated in revisions for the current trial, including reducing the amount of text, reducing the number of rating scales to complete, adding more blank copies of formulation diagrams and tables, offering longer and more telephone sessions and offering group sessions.

Participants

Recruitment was undertaken in three North West England NHS Trusts: Greater Manchester West Mental Health NHS Foundation Trust, Manchester Mental Health and Social Care Trust and Five Boroughs Partnership NHS Foundation Trust.

Participants met the following inclusion criteria:

• ICD-10 diagnosis of a schizophrenia spectrum disorder

• aged 18–65 years

• in contact with mental health services

• able to use the telephone

• able to read English

• able to provide written informed consent

• at least 1 month of stabilisation if the person has experienced a symptom exacerbation in the last 6 months.

Design

The STAR-T trial had three arms: TAU, low-support treatment (self-help recovery guide plus weekly CBT sessions delivered over the phone) and high-support treatment (the same as the low-support treatment, plus group therapy sessions). Participants were able to either choose their preferred treatment arm or they could choose to be randomised to treatment.

Procedure

Informed, written consent was obtained from all participants prior to the initiation of any assessment. All participants were given a payment of £10 for each assessment set as a token of appreciation.

Assessments were carried out at three time points: baseline, 9 months after baseline (post therapy) and a final follow-up at 15 months after baseline. Following the completion of the assessment procedures, participants were contacted by the trial manager, who conducted a semistructured interview to ascertain participants’ preferences for allocation. Participants were provided with an information sheet listing the four options (TAU, low-support treatment, high-support treatment or randomisation) and a short description of each. Based on the information gathered during this interview, participants were either allocated to their preferred therapy option or randomised to receive one of the therapy options. Randomisation was conducted remotely by an independent clinical trials unit and was performed using random permuted blocks.

Measures

Trial feasibility outcome: we will consider whether or not the trial design permitted the planned comparisons between randomised and non-randomised participants.

Primary outcome measure: QPR. 92

Secondary outcome measure: SEPS. 71

Other measures: included those of general symptom severity, delusion and hallucination severity, depression and anxiety. General symptom severity was assessed using the PANSS55,154,194 and delusion and hallucination severity was assessed using the PSYRATS,72 with PANSS scores. 195

Sociodemographic information was gathered relating to ethnicity, living arrangements, employment or education, marital status and religious beliefs.

Intervention

There were three intervention arms in the current study: TAU, low support and high support. TAU consisted of participants receiving their usual care only, alongside assessment visits at baseline, 9 months and 15 months. The low-support and high-support interventions were both provided in addition to usual care and assessments over 9 months. Low support consisted of the provision of the recovery guide, alongside weekly recovery-focused cognitive–behavioural therapy (RfCBT) sessions, which lasted approximately 45 minutes and were delivered over the telephone.

The recovery guide is a supported self-help resource that includes two core chapters (relating to the meaning of recovery, goals, resources and understanding the development of problems) and two chapters with numerous optional sections to facilitate work on overcoming obstacles (such as anxiety, paranoia, low self-esteem) or reaching wider recovery goals (such as finding employment, enjoying oneself and relaxing, developing social networks). The recovery guide was developed over approximately 1 year. It was produced in collaboration with both service users and staff from community-based mental health teams including EISs and CMHTs. Focus groups were held with both sets of participants in order to elicit what relevant questions such a guide should address and what kind of information would be useful in addressing these topics. In addition, several service users read drafts of the individual sections of the guide and provided extensive feedback in both written form and via additional focus groups. Participants were given the opportunity to work on the guide independently and also within telephone therapy sessions. Participants in phase 1 of the trial also received up to five sessions of peer support, provided by a researcher with lived experience of mental health challenges and delivered over the telephone.

The high-support arm consisted of all of the same elements of the low-support arm, with the addition of group sessions. Group sessions were delivered over 6 months of the 9-month therapy window and were offered roughly every 2 weeks, lasting 2 hours. Participants in the high-support arm attended a local group alongside two facilitators: a therapist and a researcher with lived experience of mental health challenges. The sessions followed the ethos of the recovery guide, exploring issues relating to recovery, cognitive–behavioural principles, normalising experiences and also provided a venue for participants to share ideas and experiences.

Therapy manuals for the purpose of replication and the fidelity scale used to assess therapist adherence to the intervention are available from the first author.

Analysis

The aim of the analysis was to evaluate the effects of the interventions and to assess whether or not they were unaffected by the mechanism of treatment selection (that is, by patient preference or random allocation). In the absence of significant numbers of participants accepting randomisation (see Preferences) this aspect of the analysis plan had to be abandoned. The outcome data were analysed using appropriate regression (analysis of covariance) models in order to assess the effects of treatment receipt, regardless of the mechanism of treatment selection, using several baseline covariates to allow for any possible confounding of the effect of treatment receipt on outcome. All analyses were carried out using Stata.

Preferences

Figure 9 shows the distribution of chosen treatment choices within the trial. Three participants were randomised to treatment arm while the rest were allocated to their therapy of preference.

FIGURE 9.

Treatment allocation on the basis of stated preferences.

A total of 89 subjects chose (n = 86) or were randomised (n = 3) to treatment. Of those who were randomised, one participant was allocated to each group. One subject who chose TAU but changed to low support was assigned to their original choice (i.e. TAU) for the provision of summary statistics and all the analyses.

In order to evaluate the predictors of patient treatment choice, a multinomial regression model was fitted with a trinary outcome (TAU, low support or high support) for the 86 subjects who chose their therapy. The model included the following treated as quantitative variables: age, education, baseline scores for PANSS positive and PSP, and sex as a categorical variable. The relative risk ratio was significant for PANSS positive at baseline for the high support vs. TAU outcomes (p = 0.046) and sex was marginally significant for this comparison (p = 0.063).

Primary outcomes analyses

In all outcome analyses the baseline score was entered as a covariate along with baseline PANSS positive and PSP totals. Consistent with the choice model above, age, education and sex were also included in each linear regression model.

Table 48 shows the estimated treatment effects from fitting models to the three primary outcomes at 9 and 15 months as choice or randomisation to therapy.

Considering all three treatment groups at 9 months, there was no evidence of an overall treatment effect for the QPR total primary outcome (p = 0.58). For QPR total, the adjusted mean difference comparing high support to low support was –0.86 units meaning that on average the high-support group were less recovered on QPR but this was not significant (p = 0.82). Comparing high support and low support with TAU, the adjusted mean difference was –2.70 units, that is, on average the ‘extra’ support group were less recovered on QPR, but this was not significant (p = 0.30).

At 15 months there was no overall treatment effect for QPR total (p = 0.82). The adjusted mean difference comparing high support with low support was –0.29 units meaning that on average the high-support group were less recovered than low support but this was not significant (p = 0.94). The estimated mean difference between ‘extra’ support and TAU was 1.75, but non-significant (p = 0.53).

Missing primary outcome data and sensitivity analyses

The extent of missing follow-up QPR total data is shown in Table 49. For each treatment allocation, a separate logistic regression model was fitted using the same covariates as in the choice and standard outcomes model to predict who provided QPR total outcome data 9 months after baseline completion. Each model was used to generate an expected probability of providing outcome data and the reciprocal of this estimated probability was used as an inverse probability weight for use in sensitivity analyses of the 9-month primary QPR outcome. The same procedure was undertaken for the intrapersonal and interpersonal QPR subscales. These steps were repeated for the coprimary data at 15 months.

This method assumes that for subjects with a particular combination of baseline covariates the value of the outcome is independent of whether observed or not (‘missing at random’ assumption). The results of fitting the regression models with inverse probability weights showed no evidence of significant difference between these results and the basic approach.

Secondary outcomes analyses

The same basic analyses were performed on a pre-specified selection of secondary outcomes: four out of the six SEPS subscales and PANSS positive and negative subscales; the results are shown in Table 50. Inverse probability weight sensitivity analyses were not performed. The only significant result was for the SEPS negative experience outcome at 15 months for which there was an estimated difference in adjusted means comparing high support with TAU of 16.85 units (95% CI 1.36 to 32.35; p = 0.03) (Table 51).