Effects of Actissist, a digital health intervention for early psychosis: A randomized clinical trial ☆

Schizophrenia affects 24 million people worldwide. Digital health interventions drawing on psychological principles have been developed, but their effectiveness remains unclear. This parallel, assessor-blinded, randomized clinical trial aimed to investigate whether a cognitive behaviour therapy-informed digital health intervention (Actissist app) confers added benefit on psychotic symptoms over and above remote symptom monitoring (ClinTouch app). Participants recruited from UK community health services were randomized 1:1 to receive either Actissist plus treatment as usual (TAU) or ClinTouch plus TAU. Eligible participants were adults with schizophrenia-spectrum psychosis within five years of first episode onset meeting a criterion level of positive symptoms severity. The primary outcome was Positive and Negative Syndrome Scale (PANSS) symptoms total score at 12 weeks post-randomization. Intention-to-treat analysis included 172 participants, with 149 participants (86.6 %) providing primary outcome data. Actissist plus TAU was not associated with greater reduction than an active control remote symptom monitoring app (ClinTouch) in PANSS total score at post-randomization. There were no significant effects between groups across secondary measures. There were no serious adverse reactions. Both groups improved on the primary psychotic symptoms measure at primary end-point and on secondary measures over time. The Actissist app is safe but not superior to digital symptom monitoring.


Introduction
Schizophrenia spectrum psychosis affects 24 million people worldwide, with significant associated costs (World Health Organisation, 2021a).Symptom onset is typically in early adulthood and up to 80 % relapse within five years (Alvarez-Jimenez et al., 2012;Robinson et al., 1999), resulting in costly unscheduled acute care and adverse effects on psychosocial development (Almond et al., 2004;Wiersma et al., 1998).Current delivery of psychosocial interventions for psychosis by scheduled appointment can result in relapse indicators being missed or treated too late.Developing efficient solutions to improve the speed and quality of recovery from psychosis, beyond conventional drug and psychosocial treatments, is a clinical priority.Health services worldwide have a clear digital agenda, with most World Health Organisation member states now having policies and strategies outlining how digital tools will support healthcare delivery (World Health Organization, 2021b).Furthermore, self-management of long-term conditions is a cornerstone of many health service policies.Smartphones offer an unprecedented opportunity to improve treatment quality, efficiency, cost and access, and to facilitate self-management, by making it possible to deliver near real-time, ecologically-valid therapy that extends the reach of standard healthcare delivery (Balaskas et al., 2021;Dao et al., 2021;Lattie et al., 2022;Nahum-Shani et al., 2018).
Cognitive-behavioral therapy for psychosis (CBTp) shows small-tomedium significant effects favoring CBTp for positive symptoms, hallucinations, delusions, and negative symptoms.However, timely access to such treatment is problematic (Haddock et al., 2014).We co-produced Actissist, a digital health intervention (DHI) for people with early psychosis, delivered via a smartphone app.Based on CBTp principles, Actissist uses structured, self-assessment question-answer exchanges that focus on cognitive appraisals, belief conviction, emotions and associated behaviours, followed by normalising messages and cognitive and behavioural strategies for managing distressing experiences (Berry et al., 2020).Actissist targets five key relapse risk indicators (perceived criticism; socialization; cannabis use; paranoia; and distressing voice-hearing) and is the end point of a systematic, co-produced development method (Berry et al., 2020;Bucci et al., 2018b).Our related papers provide full details of Actissist's development (Berry et al., 2020) and proof-of-concept testing (Bucci et al., 2018), and of user engagement with this DHI (Hassan et al., 2023;Eisner et al., 2023).Preliminary evidence from our proof-of-concept trial (Bucci et al., 2018a) showed Actissist was feasible, acceptable, safe, and brought benefit on negative symptoms, general psychotic symptoms and mood compared to usual care.Since the proof-of-concept trial, and in response to user feedback, we extended certain app features (e.g. more personalization, goal setting domain).In this study, we aimed to test the efficacy of Actissist in reducing psychotic symptoms.We hypothesised that participants receiving Actissist plus usual care would have lower total psychotic symptom scores post-randomization compared to those allocated to an active digital symptom monitoring app condition (ClinTouch (Palmier-Claus et al., 2012)).Our secondary hypothesis was that Actissist would improve secondary clinical and functional outcomes compared to ClinTouch.To our knowledge, this is the first rigorous test in psychosis of a CBT-informed DHI against an active control group that controls for the non-specifics of using a digital tool in daily life.

Research design
We conducted a parallel-group randomized clinical trial (ISRCTN76986679) to test the efficacy of the Actissist DHI compared with routine digital symptom monitoring (ClinTouch) plus TAU.Assessments were done at 0 weeks, 12 (primary end-point) and 24-weeks post-randomization.Study procedures were conducted between March 2018 (first patient recruited 22/03/18) and June 2020.This was a single-centre study with patients recruited from two National Health Service (NHS) Trusts.The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2013.All procedures involving human subjects/patients were approved by West of Scotland Research Ethics Committee (REF: 17/WS/0221).All participants gave written informed consent.The trial was prospectively registered (ISRCTN76986679) and followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline (eMethods1, Supplementary Material).The treatment exposure window was 12 weeks.

Participants
Eligible participants were aged over 16 years, met ICD-10 criteria for a schizophrenia-spectrum diagnosis (ICD codes F20-F29, confirmed by the treating clinician) or EIS entry criteria (operationally defined using the Positive and Negative Syndrome Scale, PANSS (Kay et al., 1987), and/or the psychosis transition criteria of the Comprehensive Assessment of At-Risk Mental States, CAARMS), in contact with mental health services, within five years from onset of first psychotic episode, met a criterion level of positive symptoms severity (a score of >3 ("moderate" or greater severity) on any PANSS positive item and a score of >3 ("moderate" or greater severity) on any PANSS negative/general items), English speaking, able to provide informed consent (determined by their care co-ordinator or responsible consultant), and were not currently participating in another trial.

Randomization and masking
Following baseline assessment, participants were randomized using a computer based random number generator (www.sealedenvelope.com).Participants were randomized on a 1:1 ratio, stratified by service (EIS vs Community Mental Health Team, CMHT), using permuted blocks of size 4 and 6.The trial statistician was masked until the statistical analysis plan had been signed off, all data had been collected, and the database was locked.Group allocation was revealed only to the participant, responsible clinician, baseline research assessor, and trial manager.Research assessors were masked to group allocation when conducting follow-ups; breaks in masking were recorded.Strategies to protect masking included considering room use and diary arrangements and reminding participants not to disclose group allocation.

Interventions
The Actissist and ClinTouch apps are described below, with example screenshots of both apps (eFig.1) and details of the onboarding procedure (eMethods 2) provided in Supplementary Material.All participants were offered a phone, and data network costs at £10 per month (£30 total) were provided.Both apps were available for a 12 week treatment exposure window.

Actissist
The Actissist DHI is a smartphone app comprising a series of question-answer exchanges, based on CBTp principles, and access to a menu of multi-media tools that act in a stand-alone fashion to complement and support the feedback from the intervention domains.The Actissist intervention is theoretically underpinned by the cognitive maintenance model of psychosis which hypothesises that cognitive appraisals contribute to the development of unhelpful beliefs and influence the interpretation of psychotic experiences (Morrison, 2001;Garety et al., 2001).Interpretations of symptoms and beliefs about their consequences are proposed to be the primary drivers of distress.The Actissist intervention incorporates key theoretical elements of CBT for psychosis (e.g., normalising messages, evaluation of appraisals through worksheets capturing evidence for and against beliefs/evaluation of pros and cons of beliefs, behaviour change strategies), as identified by a previous Delphi study and competence framework (Morrison and Barratt, 2010;Roth and Pilling, 2013).
The app is divided in two parts, although presented as a single app.The core section of the app targets five key relapse risk indicators: perceived criticism; socialization; cannabis use; paranoia; and distressing voice-hearing.Participants choose a domain(s) that is bothering them, which then triggers a series of self-assessment questions structured as question-answer exchanges that focus on cognitive appraisals, belief conviction, emotions and associated behaviours.According to the appraisal selected, the user is then presented with a normalizing message and cognitive or behavioral strategy suggesting ways of coping with distressing experiences.To minimise boredom and repetition, the app presents a variety of messages and images for each appraisal type.Users can also access multimedia content (e.g.patient recovery videos, relaxation and mindfulness exercises, factsheets and web content), diary functions and customise the app aesthetics using their own images.Users are also invited to set up to three personal goals which were entered using free text by users on app setup.
At 3 pseudo-randomised time points per day, 6 days a week between 10.00 and 22.00, an auditory alert followed by a visual prompt is emitted from the app inviting participants to access the app.The notifications persist on the handset (i.e.no time out) until such point as they are accepted, dismissed, "snoozed" (up to 15 min), or another notification is received.The notifications serve merely as a reminder; the app also allows self-initiated use at any point.If a user accepts a notification or initiates use, they are invited to select an intervention domain(s)hearing voices, suspiciousness, socialisation, cannabis use, perceived criticism -and then complete a series of self-assessment questions structured as question-answer exchanges that focus on cognitive appraisals, belief conviction, emotions and associated behaviours.The app collects responses from the user and wirelessly uploads user responses to a server.We permitted the app to undergo three software version updates to ensure the technology was not obsolete by study end.Participants were given the option to use their own phone or a loaned study handset.

ClinTouch
ClinTouch (Palmier-Claus et al., 2012) is a symptom-monitoring app that triggers, collects and wirelessly uploads symptom data to a server and served as our active control condition to control for the non-specifics of using a smartphone.The notification schedule, aesthetics and interface mirrored those of the Actissist app, but ClinTouch does not facilitate self-initiated access; data entries must be in response to a notification.ClinTouch notifications prompt users to respond to a core set of items using a touchscreen slider to rate the severity of 12 individual mood, anxiety and psychotic symptoms on a 1-7 scale (validated against the PANSS (Kay et al., 1987)), with some items seeking further detail dependent on initial rating.Notifications time out after 30 min; the notification is no longer visible and ClinTouch items are not accessible.Users can view summary symptom data and see visualisations showing changes over time.ClinTouch has been shown to reduce positive symptoms in an early psychosis group over 12 weeks with significant between-group differences compared to no treatment (Lewis et al., 2020).We compared the Actissist DHI to an active control symptom-monitoring app condition, which matched the overall look-and-feel and functionality of the Actissist DHI, thus accounting for the non-specifics of smartphone use.

Treatment as usual (TAU)
Both groups continued to receive their usual treatment alongside the assigned study app.TAU was delivered according to UK national and local service guidelines and typically involved antipsychotic treatment, contact with mental health workers, risk monitoring, and outpatient psychiatric appointments.

Assessments
Assessments were completed at baseline, 12-and 24-weeks postrandomization by masked research assistants who underwent a rigorous training process and received weekly supervision by the chief investigator.Reliability assessment of symptom and functioning measures occurred in two phases.First, all researchers were calibrated to interviewer-administered measures by completing a structured training programme.Feedback on ratings was anchored to a set of training recordings and videos, with reference ratings provided by an expert rater until consistency was achieved.Raters met for reliability assessment for the study period.All assessors met departmental reliability standards after pre-trial training.Participants received £20 (approximately US $28) for each assessment and travel expenses were reimbursed.

Outcomes
Measures were selected that map directly onto hypothesised mechanisms of change.The primary outcome was change in psychotic symptoms measured by PANSS total score at 12-weeks postrandomization.PANSS, the gold-standard measure of psychotic symptom severity, showed promising effects in our pilot trial.Secondary outcomes, measured at 12 and 24 weeks post-randomization were: symptom frequency, intensity and distress using the Psychotic Symptoms Rating Scales (PSYRATS; Haddock et al., 1999) delusions (interviewer-administered rated over the past week, range 0-24; higher scores indicate more severe delusions) and PSYRATS (Haddock et al., 1999) hallucinations (interviewer-administered rated over the past week, range 0-44; higher scores indicate more severe hallucinations) subscales; mood using the Calgary Depression Scale for Schizophrenia (Addington et al., 1990) (CDSS; interviewer-administered rated over the past week, range 0-36; higher scores indicate more severe depression); social functioning using the Personal and Social Performance Scale (Morosini et al., 2000) (PSP; interviewer-administered, past week); perceived criticism and perceived warmth using the Perceived Criticism and Perceived Warmth Scale (Hooley and Teasdale, 1989) (PCPW; higher scores indicate more severe perceived criticism / warmth; ratings completed for a significant other); recovery using the Questionnaire about the Process of Recovery (Neil et al., 2009) (QPR, past week, higher scores indicate better recovery); well-being using the Warwick-Edinburgh Mental Wellbeing Scale (Tennant et al., 2007) (WEMWEBS; past 2 weeks, range 14-70; higher scores indicate higher levels of mental well-being); internalised stigma using the 29-item Internalised Stigma of Mental Illness Inventory (Ritsher et al., 2003) (ISMI; higher scores indicate more severe internalised stigma); and empowerment using the 28-item Empowerment Scale (Rogers et al., 1997) (ERS).Substance use was measured using the Alcohol Use Disorders Inventory (Babor et al., 2001) (Group, 2002).Cannabis use frequency was measured using the Time Line Follow Back for drugs and alcohol (TLFB, past 3 months; Sobell and Sobell, 1992).The Quantitative feedback questionnaire (QFQ) is a 27-item questionnaire and was used to assess the acceptability of the apps (Bucci et al., 2018a).Health economic outcomes including EQ-5D-5 L and Client Service Receipt Inventory (CSRI) were recorded and will be reported separately.App usage is reported elsewhere (Hassan et al., 2023).

Statistical analysis
A statistical analysis plan was approved by the Trial Steering and Data Monitoring and Ethics Committee (TS/DMEC) prior to analysis.With 1:1 allocation, 63 participants per group had 80 % power to detect an effect size of 0.4 (6 points on the PANSS with an SD of 15), based on a two-group t-test with a 0.05 two-sided significance level, assuming a baseline-endpoint correlation of 0.6.Allowing 20 % attrition, we aimed to recruit 85 patients per group (N = 170).The SD, correlation and attrition rate are based on figures from our pilot trial (Bucci et al., 2018a).
All primary analyses were completed using Stata® version 16.1 (StataCorp, 2015) and done by intention-to-treat with no interim analyses.Primary and secondary outcomes, except cannabis free days, were analysed with linear mixed effects models.Proportion of cannabis free days was analysed using a mixed effects generalised linear model, with family binomial and an identity link.In all models, outcome measurement (at the two follow-up time points) was the dependant variable.Models included fixed effects for timepoint, treatment, timepoint-by-treatment interactions, the baseline measure of the respective outcome, and service (two categories: i) EIS; ii) CMHT).Observations were clustered by participant with an unstructured correlation matrix for residuals.The model was fitted using restricted maximum likelihood (Rabe-Hesketh and Skrondal, 2008).Alcohol, smoking and substance use data from the ASSIST was analysed using descriptive statistics only.Cohen d effect sizes at follow-up were calculated as the adjusted mean difference divided by the sample SD of the outcome at baseline and are shown in a forest plot.
For each outcome, participants with at least one completed follow up measure were included in the analysis.Missing values in baseline covariates were handled using mean imputation; the missing value was imputed with the mean of the covariate for all participants in the trial (White and Thompson, 2005).Details of sensitivity analysis conducted to assess whether outcomes were affected by the COVID-19 pandemic, whether the treatment effect differed by app version, and differing missing data assumptions are given in the statistical analysis plan.The independent TSC/DMEC reviewed adverse events occurring during the trial.
Fifty (30 %) of the sample used their own phone, and 119 (70 %) used a study phone.Four participants in the ClinTouch group did not receive an app onboarding session (uncontactable following randomization).For those onboarded on Actissist (n = 87) and ClinTouch (n = 81), the mean overall response rate to scheduled app alerts was 34.0 % (5670/16,689) and 33.6 % (4757/14,281), respectively.Actissist participants submitted a mean of 77.1 total responses during the study compared to 60.2 responses (scheduled only) among ClinTouch participants.Overall, 47.1 % of Actissist and 45.7 % of ClinTouch users completed at least a third of scheduled alerts while active in the study.Further details of app usage are reported elsewhere (Hassan et al., 2023).
Descriptive statistics, between-group mean differences, and associated P values, 95 % CIs and standardized effect sizes are shown in Table 2 for outcomes at 12 and 24 weeks, with cannabis-related secondary outcomes shown in eTable 1 (Supplementary Material).Fig. 1 shows standardized effect sizes for continuous variables as a forest plot.Actissist was not associated with greater reductions in the primary outcome (PANSS total score) than ClinTouch at 12 weeks (Cohen d, 0.1; 95 % CI, − 0.2 to 0.3; P =0.62) or 24 weeks (Cohen d, − 0.2; 95 % CI, − 0.4 to 0.1; P=.26; Fig. 2).Similarly, Actissist was not associated with greater reductions in any of the secondary outcomes at 12 or 24 weeks (Tables 1, 2, Fig. 1).
Overall, participants reported both apps to be acceptable (eTable 2, Supplementary Material).Participants said the apps were quick and easy to use, they would recommend them to others and answering questions made them feel better.
Ninety-five adverse events were reported (eTable 3, Supplementary Material), of which 19 were serious, occurring in 8 participants in the Actissist group and 8 in the ClinTouch group (3 occurred prerandomization).No serious events were related to the study hardware, software or study assessments, and no deaths occurred.Nine non-serious Adverse Reactions occurred; 3 related to the study software (e.g., participant reported the app made them focus negatively on their mental health difficulty, resulting in distress), 2 related to both the study hardware and software (e.g., participant reported the alert notifications were "too loud" resulting in distress; participant reported that the   software updates and reinstallation required on the study phone caused distress), and 4 related to study assessments (e.g., some assessment questions resulted in the participant feeling distressed; procedural issues, including receiving payment and arranging a research assessment visit in a clinic rather than the participant's home, resulted in distress).

Main findings
Using a CBT-informed DHI alongside usual care did not bring about significant improvements in psychotic symptoms relative to a digital symptom monitoring app used by our active control group at each follow-up time point; both groups improved over time on the primary psychotic symptoms outcome.We found no significant differences between groups in all secondary outcome measures at both time-points.There was no evidence of either the Actissist or ClinTouch apps being harmful, with a similar number of serious adverse events unrelated to the apps or trial procedures, in both groups, and participants found the apps acceptable to use.
The absence of an effect between groups in this study is important.Unlike our feasibility study (Bucci et al., 2018a), we did not observe a treatment effect in this powered study.It is possible that we saw an inflated effect size in our feasibility study, which is common in pilot studies (Torous et al., 2021).We did not incentivise app use in this study as we did in our feasibility study, which may have also contributed to observed differences between our studies.The evidence that DHIs are feasible and acceptable, not only in psychosis but across mental health conditions, is strong (Torous et al., 2021).Furthermore, efficacy studies show that DHIs bring clinical benefit when conducted under rigorous research conditions.However, to our knowledge, this is the first RCT in psychosis reporting results of a DHI compared against a highly relevant control condition.Participants in the control group used an app with the same look-and-feel and alert schedule as the Actissist app.The lack of an effect between our digital CBT and digital symptom monitoring groups draws into question for the first time the added benefit of a CBT-informed DHI over and above routine digital symptom monitoring, which suggests caution in interpreting efficacy when a DHI is compared against TAU in a psychosis sample.There was no evidence of harm of both apps, and both groups improved over time.Therefore, offering patients the choice of a digital health app that facilitates symptom monitoring in day-to-day life, or therapeutic strategies, is both safe, acceptable and beneficial and will allow for more personalised tailoring of treatment approaches.Further work to integrate digital symptom monitoring with intervention strategies, combining both active and passive monitoring and informed by experts-by-experience, is warranted and will allow for more personalised tailoring of healthcare for people with psychosis; for example, integrating the ClinTouch and Actissist apps so that CBT-informed strategies are provided specifically in response to escalations identified by symptom monitoring, along with the target domains a user identified as distressing.
The heterogeneity of interventions, engagement definitions and study designs make it difficult to directly compare studies of DHIs.However, the level of engagement for both apps in the trial was substantial when compared to the average Android app and reflect similar patterns in smartphone interventions observed in trials of smartphonedelivered mental health interventions (Linardon and Fuller-Tyszkiewicz, 2020) and previous research studies collecting self-reported health data longitudinally (Ben-Zeev et al., 2016;Reade et al., 2017).Furthermore, we retained a substantial proportion of participants for the duration of the study and at follow-up in both trial arms.The high uptake and engagement with the apps suggest that our co-production efforts when designing and developing both apps was effective and that people are interested in using DHIs to self-manage their mental health.
We did not perform gender or ethnicity-specific analyses because the numbers did not permit this; future studies could increase the sample size to investigate such differences because females and people from black and other minority ethnic groups are under-represented in psychosis research (Friskney et al., 2023).

Limitations
Without a TAU only group, we were unable to determine whether our findings are different to what would have been observed in a no monitoring control condition.However, we chose ClinTouch as our active control as we have demonstrated previously digital symptom monitoring is superior to TAU in an early psychosis group.Although we have not demonstrated Actissist is superior to ClinTouch in this trial, as both groups demonstrated sustained effects over time, Actissist is by    b The reported treatment effect is adjusted for the baseline measure of the respective outcome and the service (CMHT or EIS).
S. Bucci et al. extension likely superior to TAU.It is possible that our trial attracted more digitally confident/ literate individuals with psychosis with positive attitudes to using digital tools in healthcare; further research is needed exploring engagement and benefits with digital tools in psychosis outside a research trial context.While certain aspects of the trial were designed to reflect real-world conditions (e.g.no financial incentives for app use), others (e.g.provision of phones and data plans) may limit generalisability.

Conclusions
To our knowledge, this is the first powered RCT of a CBT-informed DHI in psychosis compared against an active control condition that controls for the non-specifics of using a smartphone/app.Using the CBTinformed Actissist app did not confer added benefit over digital symptom monitoring; both apps were safe and are associated with clinical improvements over time.Early psychosis patients should be given the choice of using digital symptom monitoring or intervention tools as part of routine care delivery.Further work is needed to integrate digital monitoring and intervention to provide harmonised assessment and justin-time (adaptive) intervention.

Fig. 1 .
Fig. 1.Cohen's D for Primary and Secondary outcomes.Outcomes have been rescaled so for all outcomes negative effects favour Actissist and positive effects favour ClinTouch.Markers represent point estimates, with horizontal lines representing 95 % CIs.Abbreviations: PANSS, Positive and Negative Syndrome Scale; PSYRATS, Psychotic Symptoms Rating Scales; CDSS, Calgary Depression Scale for Schizophrenia; PSP, Personal and Social Performance Scale; PCPW, Perceived Criticism/Perceived Warmth Scale; QPR, Questionnaire about the Process of Recovery; WEMWEBS, Warwick-Edinburgh Mental Wellbeing Scale; ISMI, Internalised Stigma of Mental Illness Inventory; ERS, Empowerment Rating Scale; AUDIT, Alcohol Use Disorders Inventory.

Table 1
Baseline characteristics of the sample.
Abbreviations: CMHT, Community Mental Health Team; EIS, Early Intervention for Psychosis Service; PAYG, Pay As You Go.

Table 2
Primary and secondary outcomes.

Table 2
(continued ) Abbreviations: PANSS, Positive and Negative Syndrome Scale; PSYRATS, Psychotic Symptoms Rating Scales; CDSS, Calgary Depression Scale for Schizophrenia; PSP, Personal and Social Performance Scale; PCPW, Perceived Criticism/Perceived Warmth Scale; QPR, Questionnaire about the Process of Recovery; WEMWEBS, Warwick-Edinburgh Mental Wellbeing Scale; ISMI, Internalised Stigma of Mental Illness Inventory; ERS, Empowerment Rating Scale; AUDIT, Alcohol Use Disorders Inventory.a Unless outcome is marked * lower scores indicate better outcomes and negative treatment effects favour Actissist.For outcomes marked with * higher scores indicate better outcomes and positive treatment effects favour Actissist.