Telephone-delivered psychosocial interventions targeting key health priorities in adults with a psychotic disorder: systematic review

Background The mental and physical health of individuals with a psychotic illness are typically poor. Access to psychosocial interventions is important but currently limited. Telephone-delivered interventions may assist. In the current systematic review, we aim to summarise and critically analyse evidence for telephone-delivered psychosocial interventions targeting key health priorities in adults with a psychotic disorder, including (i) relapse, (ii) adherence to psychiatric medication and/or (iii) modifiable cardiovascular disease risk behaviours. Methods Ten peer-reviewed and four grey literature databases were searched for English-language studies examining psychosocial telephone-delivered interventions targeting relapse, medication adherence and/or health behaviours in adults with a psychotic disorder. Study heterogeneity precluded meta-analyses. Results Twenty trials [13 randomised controlled trials (RCTs)] were included, involving 2473 participants (relapse prevention = 867; medication adherence = 1273; and health behaviour = 333). Five of eight RCTs targeting relapse prevention and one of three targeting medication adherence reported at least 50% of outcomes in favour of the telephone-delivered intervention. The two health-behaviour RCTs found comparable levels of improvement across treatment conditions. Conclusions Although most interventions combined telephone and face-to-face delivery, there was evidence to support the benefit of entirely telephone-delivered interventions. Telephone interventions represent a potentially feasible and effective option for improving key health priorities among people with psychotic disorders. Further methodologically rigorous evaluations are warranted.

A more recent systematic review of telepsychiatry (telephone, internet or videoconferencing) in the assessment and treatment of people with a schizophrenia spectrum disorder included six studies (Kasckow et al., 2014). However, neither review included studies targeting people with bipolar disorder. Moreover, neither reviewed the evidence for multiple key health priorities in adults with a psychotic disorder (namely relapse prevention, medication adherence and health behaviours).

Aims of the current review
Given the poor physical and mental health of people with a psychotic disorder, limited access to healthcare and the potential promise of telephone-delivered interventions, we aim to provide an overview and critical analysis of the current state of evidence for telephone-delivered psychosocial interventions for relapse prevention, medication adherence, and modifiable CVD risk behaviours among people with a psychotic disorder (schizophrenia spectrum disorder or bipolar disorder). The focus of this review will be on person-delivered interventions using the spoken word (i.e. interventions delivered entirely by text, web and/or automated systems were excluded) and one or more psychological strategies (see published protocol for further details; Beck et al., 2015).

Protocol and registration
This systematic review is registered with PROSPERO (Registration Number CRD42015025402) and the protocol has been published (Beck et al., 2015).

Criteria for selecting studies for this review
Methods were informed by Cochrane Guidelines for systematic reviews (Higgins and Green, 2011) and are extensively detailed in the review protocol (Beck et al., 2015). The population of interest was adults (⩾18 years) with a psychotic disorder (as defined by any criteria). We included studies with populations involving adults with non-psychotic disorders only if more than 50% of participants had a psychotic disorder, or if data limited to those with psychotic disorders were available. The intervention of interest was telephone support targeting: (i) relapse prevention, (ii) adherence to psychiatric medication and/or (iii) smoking and other CVD health risk behaviours [see (Beck et al., 2015) for definitions]. These domains were targeted as they represent an important avenue for improving the health and wellbeing of adults with psychosis since they are common challenges that have profound implications for the individual and are amenable to change following psychological intervention. Telephone support was defined as a person delivered intervention of at least 10 min using spoken word and one or more psychological strategies (see published protocol for further details; Beck et al., 2015). The telephone support could be a standalone intervention or delivered in combination with other treatment components. However, studies with multiple components were only included if the telephone was the predominant method of intervention delivery (defined as ⩾ 50% of the total number of participant contacts conducted by telephone). Interventions delivered in any setting (e.g. community, hospital, rehabilitation or residential treatment centre, etc.) were included. The telephone support could be compared with inactive (e.g. standard care, waiting list control) and/or active controls (e.g. pharmacological and/or psychological alone and/or in combination with usual care) whereby telephone was not the predominant method of intervention delivery (e.g. individual, group, internet). Studies had to provide data for at least one of the following: (a) relapse, (b) medication adherence, (c) health risk behaviours/CVD risk, (d) process variables (e.g. treatment engagement) or (e) feasibility [see (Beck et al., 2015) for definitions]. Process variables are included in Supplementary File 1. Qualitative studies were the only study design excluded.
Search methods for identification of studies Figure 1 summarises the procedure used to identify studies, (see online Supplementary Appendix 1 for the full MEDLINE search strategy). Abstract, title, keywords and subject headings specific to each of the identified databases were searched. All subject headings were exploded so that narrower terms were included. No limits were placed on publication year. Publications had to be available in English. Reference lists were hand searched to identify any additional publications. Publications were organised in reference manager Endnote. The first search was run in May 2015 and re-run just before final analyses (December 2016). Articles were identified and classified according to the following steps: Step 1: Identification and screening AKB performed the searches and reviewed the titles and abstracts of the identified 297 publications and used the inclusion criteria to exclude clearly ineligible articles. If eligibility was unclear, the full-text article was accessed.
Step 2: Eligibility and classification The full-text version of 76 publications was manually reviewed and 42 publications were excluded. The remaining 34 were classified as 'evaluation', 'review', 'discussion' or 'other' according to published definitions (Beck et al., 2015).
Step 3: Cross-checking The 76 publications from step two were cross-checked by ALB. The 22 studies independently classified as 'evaluation' were retained for further examination.

Data collection and analysis
Data extraction was performed by ALB and checked by AT, SB and KB. When multiple reports of the same study were identified (Simon et al., 2002(Simon et al., , 2005(Simon et al., , 2006 data were extracted separately and combined across data collection forms. Criteria for data extraction (detailed in the protocol; Beck et al., 2015) were adapted from the Cochrane Handbook for Systematic Reviews (Higgins and Green, 2011) and the Downs and Black Scale (Downs and Black, 1998).

Assessment of methodological quality and risk of bias
Methodological critique and assessment of risk of bias on individual studies were performed independently by ALB and AT, with final ratings made by consensus. As we included both randomised and non-randomised designs multiple tools were used.

Downs and Black scale
All studies were assessed against the Downs and Black Scale (Downs and Black, 1998). This scale is recommended by the Cochrane Guidelines for assessing the quality of non-randomised trials (Higgins and Green, 2011). Consistent with previous research (e.g. Baker et al., 2012) two items were not used. Scoring of the final item (power) was unclear so the following convention was used: 0 = no power calculation reported; 1 = power analysis reported, but insufficient power achieved and 2 = power analysis reported and sufficient power achieved. All other items were scored per published guidelines (Downs and Black, 1998) for a total maximum of 27, with higher scores reflecting greater methodological quality.

PEDro scale
Randomised controlled trials (RCTs) were assessed against the 11 item Physiotherapy Evidence Database (PEDro) scale (Maher Fig. 1. PRISMA flow diagram summarising systematic search identifying evaluations of telephone delivered psychosocial interventions for relapse prevention, medication adherence and health risk behaviours in adults with a psychotic disorder. et al., 2003), a widely implemented and validated tool for assessing the quality of randomised trials. As per above, the two items regarding blinding were not used (e.g. Spring et al., 2011, Baker et al., 2012. The remaining nine criteria were assigned a yes (1 point) or no (0 points) rating, and a quality score ranging from 0 to 8 points was calculated for each study.

Cochrane collaboration's risk of bias tool
Risk of bias (within and across all studies) was assessed using the Collaboration's Risk of Bias tool, as described in the Cochrane Handbook for Systematic Review of Interventions (Higgins and Green, 2011). Each item was judged as being high, low or unclear risk as per the criteria provided by Higgins and Green (Higgins and Green, 2011). Given the evidence that sequence generation and allocation concealment represent particularly important potential sources of bias, studies were deemed to be at the highest risk of bias if either item was scored as 'high' or 'unclear'.

Summary measures
A study was considered to have a positive outcome if more than 50% of the reported outcome measures (primary and secondary) demonstrated a between-group difference in favour of the telephone group at the treatment end. Positive maintenance outcome(s) were identified when this effect was evident at short and/or medium and/or long-term follow-up (1-6; 7-12 and >12 months after intervention completion, respectively).

Synthesis of results
Comparability of study design and outcome measures across studies was assessed by a consultant statistician to determine the possibility of conducting meta-analyses on RCTs to examine effects on relapse, medication adherence and smoking and other health behaviours and CVD risk. A narrative synthesis of the findings was conducted, structured around intervention type, outcome, population and methodological quality. As Clinical Guidelines recommend an improved focus on personally meaningful recovery (e.g. quality of life, functioning) relative to traditional clinical outcomes (e.g. symptoms and relapse) in mental health care, to help inform clinical practice, the assessment, reporting and/ or change in these additional outcomes is also central to the structure of the review.

Participant Characteristics
Across all studies, the total number of participants was 2473, with 867 in relapse prevention, 1273 in medication adherence and 333 in smoking and/or other health risk behaviour studies (see online Supplementary Table S1). The average age was 40.7 years (41.9 in relapse prevention, 39.5 in medication adherence and 42.2 in smoking and/or other CVD risk behaviours). Overall, the percentage of males across the studies was 50.1%. However, there was a higher percentage of males in studies of schizophrenia samples (64.5%) compared with studies of bipolar (37.7%) and mixed samples (44.2%). No study used a first episode sample.

Study characteristics
The 22 papers comprised a total of 20 trials, with Simon et al. (Simon et al. 2002(Simon et al. , 2005(Simon et al. , 2006 reporting on the same study. There were 16 controlled (Table 1) and four single-arm (Table 2) studies. Nine trials recruited people with bipolar disorder, six with schizophrenia spectrum disorder, four with schizophrenia and one a range of diagnoses (see online Supplementary  Table S1). For the RCTs the telephone was the sole method of intervention delivery in one relapse prevention (Beebe, 2001) and three medication adherence trials (Salzer et al., 2004;Cook et al., 2008;Beebe et al., 2016). For the studies without a comparison condition, the intervention was delivered entirely by telephone for two relapse prevention (Miklowitz et al., 2012;Boardman et al., 2014) and one healthy lifestyle (Baker et al., 2014) study.

Outcomes assessed
Outcome measures utilised in each study are reported in Tables 1 and 2. There was considerable heterogeneity. In studies of relapse prevention, the primary outcome was typically relapse, which was variously defined according to number of days until psychiatric hospitalisation, number of days until DSM criteria (IV or IV-TR) were met for a mood episode [(hypo)mania, depression, mixed)] and/or severity of symptoms. All 10 studies included one or more measures of psychiatric symptomatology, but only three included measures of quality of life and/or functioning (Castle et al., 2007;Javadpour et al., 2013;Wenze et al., 2015) and only one utilised an index of personally meaningful recovery as a primary outcome (Haddock et al., 2017). In studies of medication adherence, the primary outcome was typically medication compliance, as per self-report or clinician administered assessment. Studies typically included one or more measures to assess the impact on symptoms, service utilisation and attitudes (including self-efficacy and insight), but only two assessed the impact on quality of life and/or functioning (Salzer et al., 2004;Montes et al., 2010). In studies of CVD/health risk behaviours, primary outcomes typically included an index of smoking Heffner et al., 2015) or CVD risk (Kilbourne et al., 2012;Baker et al., 2015). One study (Baker et al., 2014) focused on sedentary activity and intake of fruit and vegetables. Functioning and/or quality of life were assessed in three of the four studies (Kilbourne et al., 2012;Baker et al., 2014;Baker et al., 2015).

Methodological quality and risk of bias in included studies
Studies are presented in descending order of methodological quality in Table 1 for controlled trials and Table 2 for single-arm studies. No clear pattern emerged between methodological rigour and whether or not the outcomes were in favour of the telephone condition. Across all trials, there was considerable variation in methodological quality scores on the Downs and Black scale, (total scores ranged from 9 to 25 out of 27). At least half of included studies scored 0 for the following items: adverse events; characteristics of those lost to follow-up; representativeness of the sample; attempts to have blinded outcomes assessors and adequate power (six studies reported power calculations, one had sufficient power). For the 12 RCTs Pedro scores ranged from two to eight out of eight. At least half of included studies scored 0 for 'blinding of outcomes assessors', and 'measures of at least one key outcome variable from at least 85% of original participants'.         TS completed significantly more telephone sessions ( p < 0.05).

---
Notes: CO verification in the in-person condition. Missing = smoking imputation for all cessation outcomes and return to baseline smoking for the reduction outcomes.
-Treatment satisfaction 100% phone and 90% in-person participants found the intervention helpful and would recommend the treatment.     Cochrane risk of bias assessments is presented in online Supplementary Fig. S1a and S1b, with overall risk of bias scores in Tables 1 and 2. To summarise, eight studies reported adequate random sequence generation, four reported allocation concealment procedures, four stated that assessors were blinded to intervention status, nine were unlikely to be subjected to attrition bias, and 11 may have been affected by reporting bias. Regarding the overall risk of bias, all non-RCTs were automatically rated as 'high' for overall risk of bias. Eight RCTs were rated as having a high overall risk of bias (Beebe, 2001;Salzer et al., 2004;Castle et al., 2007;Price, 2007;Kilbourne et al., 2012;Javadpour et al., 2013;Wenze et al., 2015;Beebe et al., 2016), with all rated as unclear regarding one or both of two key items (sequence generation and allocation concealment). The remaining five RCTs were rated as having a low overall risk of bias, although only two (Simon et al., 2006;Baker et al., 2015) had adequately blinded outcomes assessors and a pre-published protocol.

Synthesis of results
Results of individual studies are presented in Table 1 (controlled trials) and 2 (single arm studies). Heterogeneity of form of intervention delivery (telephone only or in combination), control group (active or inactive control) and outcome measures precluded a meta-analysis on (within outcomes or collapsed across groups). A narrative synthesis is presented below.

Relapse prevention
Of the 10 trials assessing relapse prevention, there were eight RCTs (Beebe, 2001;Simon et al., 2006;Castle et al., 2007;Price, 2007;Castle et al., 2010;Javadpour et al., 2013;Komatsu et al., 2013;Wenze et al., 2015), one partially randomised preference trial (Haddock et al., 2017) and one open trial (Miklowitz et al., 2012). Numbers in the RCT component of the preference trial were low (only three participants chose to be randomised), therefore this study has been categorised as an observational for the purpose of this review. Five RCTs reported at least 50% of outcomes significantly in favour of the telephone intervention, over time periods of up to 18 months (Javadpour et al., 2013); four relative to an active comparison condition (Castle et al., 2007;Castle et al., 2010;Komatsu et al., 2013;Wenze et al., 2015) and one relative to TAU (Javadpour et al., 2013). For the remaining RCTs, Beebe and colleagues (the only study in which the telephone was the sole delivery method) did not detect significant differences between active treatment conditions on the three indicators of relapse used (Beebe, 2001); Simon and colleagues demonstrated significant effects in favour of the telephone condition in two of the eight outcomes, but otherwise equivalent performance to TAU (Simon et al., 2002;2005, ) while Price found that the difference seen in hospital admissions and treatment compliance for the telephone condition (relative to TAU) did not reach statistical significance (Price, 2007). For the two non-RCTs, Haddock et al., did not detect significant differences between active treatment conditions and/or TAU for eight of the nine outcomes assessed, with the remaining outcome (Recovery from Negative Impacts of Psychosis) in favour of TAU [although the authors urge caution when interpreting this finding due to multiple comparisons (Haddock et al., 2017)] and Miklowitz et al. found significant improvement in knowledge of mood management strategies, but was unable to calculate the statistical significance of observed improvements in mania and depression (Miklowitz et al., 2012).

Medication adherence
Of the six trials reporting on medication adherence as the primary outcome three were RCTs (Salzer et al., 2004;Montes et al., 2010;Beebe et al., 2016), one non-randomised (Cook et al., 2008) and two single-group pre-post designs (Boardman et al., 2014;McKenzie and Chang, 2015). For the RCTs, the larger study (Montes et al., 2010) was the only to report at least 50% of outcomes in favour of the telephone condition. Although Salzer (Salzer et al., 2004) demonstrated effect sizes in the direction of the telephone for eight of the ten outcomes evaluated (using an intervention delivered entirely over the telephone). However, the two medication adherence outcomes (subjective response to medication and self-reported treatment adherence) did not significantly differ between groups (Salzer et al., 2004). Similarly, in their entirely telephone-delivered intervention Beebe (Beebe et al., 2016) did not detect a between-group difference for medication adherence. Conversely, in their non-randomised trial of an intervention delivered entirely by telephone, Cook reported improved adherence (both pharmacy based and self-report measures) in favour of the telephone condition (Cook et al., 2008). Both open trials reported improved self-reported medication adherence post-treatment (Boardman et al., 2014;McKenzie and Chang, 2015).

Smoking or CVD risk behaviours
There were four studies reporting smoking or CVD risk behaviour outcomes (Kilbourne et al., 2012;Baker et al., 2014;Baker et al., 2015;Heffner et al., 2015), with two RCTs (Kilbourne et al., 2012;Baker et al., 2015) one of those a pilot trial (Kilbourne et al., 2012). Both RCTs utilised an active comparison condition and neither demonstrated at least 50% of outcomes in favour of the telephone condition. Baker et al. (2015), demonstrated significant improvements in CVD risk and smoking at 12 months following either a largely telephone-delivered intervention or a multicomponent face-to-face intervention. Significant improvements in global functioning were also seen in both conditions . Neither condition demonstrated significant improvements in health behaviours other than smoking . Cardiometabolic risk (BMI and blood pressure) and health-related quality of life also remained stable for both conditions in the pilot RCT by Kilbourne et al. (2012) and between-group differences for functioning and depression symptoms approached significance, in favour of the telephone condition (Kilbourne et al., 2012). Further, for individuals at greater risk (BMI⩾30 or systolic BP>140), post hoc analyses demonstrated superior improvement in functioning and depressive symptoms for the telephone condition (Kilbourne et al., 2012). For the single-arm studies, results from Heffner et al. (2015) suggest largely equivalent performance of the phone and face-to-face delivery for a smoking cessation intervention, although between groups comparisons were not performed. Finally, in a singlegroup pre-post design Baker et al. (2014) demonstrated clinically important change across a range of health behaviours following an intervention delivered entirely by telephone.

Discussion
This review aimed to capture all relevant studies of interventions delivered on at least 50% of session occasions by telephone to improve relapse prevention, medication adherence or reduce smoking and/or other CVD risk behaviour. We sought to comment on the feasibility and efficacy of telephone-delivered psychosocial interventions in people with a psychotic disorder. A total of 20 trials were reviewed in full, with 13 RCTs. Overall, the literature is split relatively evenly across schizophrenia or schizoaffective disorder and bipolar disorder. Studies typically included one or more 'traditional' clinical outcomes (e.g. symptomatology, relapse, medication compliance), with considerably fewer assessing the quality of life or functioning. Little is known about the process variables that may influence treatment outcome and only one study conducted economic analysis.
Although the modest body of literature and diversity of methods precludes definitive comments on efficacy, positive effects were found. Five of eight RCTs evaluating relapse prevention and one of three RCTs evaluating medication adherence reported at least 50% of outcomes in favour of the telephone-delivered the intervention, for time periods up to 18 months. As for smoking and other CVD risk behaviour studies, comparable levels of improvement were seen across treatment conditions. Of note, the comparison condition for one of the studies  was an intensive, multi-component face-to-face delivered intervention with longer session duration. Accordingly, the equivalent level of improvement seen is important and points to the potential efficiency of telephone-delivered interventions for promoting clinically meaningful change.
The results in each domain of relapse prevention, medication adherence and smoking and CVD risk behaviour interventions are encouraging. Although most interventions combined telephone and face-to-face delivery, there were indications that entirely telephone-delivered interventions might be effective (e.g. Baker et al., 2014, Boardman et al., 2014, with evidence of at least equivalent (Beebe, 2001;Beebe et al., 2016) if not superior performance (Salzer et al., 2004;Cook et al., 2008) relative to standard care. In addition, in the relapse prevention preference trial conducted by Haddock et al., (2017), strong preferences were nominated by study participants for either telephone or telephone plus group delivery, with a significantly greater number of telephone sessions attended in the telephone only condition and few group sessions attended, on average. Thus, this review suggests that telephone-delivered interventions may be popular among service users, well attended, and at least as effective, if not superior to treatment as usual. Clearly, further methodologically rigorous research is warranted.

Limitations
Firstly, this review identified a modest sample of heterogeneous studies. Differences in outcome assessment, intervention and comparator conditions precluded meta-analysis. Accordingly, it is difficult to draw strong conclusions about the impact of telephone-delivered interventions on the outcomes of interest. There was also considerable variation in methodological quality. Most studies were uncontrolled and less than half of the RCTs identified were deemed to be at low risk of bias. In addition to poor reporting around randomisation and allocation concealment, many studies did not report using blinded outcomes assessors. Adequately powered RCTs were also rare. Many had small sample sizes, and all but one of those reporting power calculations were underpowered to detect significant differences. The crosscultural generalisability of our findings is also restricted as we limited our search to English language publications.

Implications for practice
Despite psychological interventions being recommended (Galletly et al., 2016;National Institute for Health andCare Excellence, 2014a, 2014b) for the treatment of schizophrenia and other psychotic disorders, of those likely to benefit, only 10% or less have access (Gulliver et al., 2010;Haddock et al., 2014;Schizophrenia Commission, 2015). Our findings lend further support to the potential role of phone delivered interventions in improving access. Importantly, the treatment protocols included in the current review were delivered by a variety of health professionals and ranged from brief time-limited 'check-in's' (e.g. Price, 2007) to full psychological interventions (e.g. Baker et al., 2014). Accordingly, telephone delivery may help to overcome barriers related to accessibility of support services and availability of trained clinicians (Gulliver et al., 2010;Haddock et al., 2014;Schizophrenia Commission, 2015), while maintaining the verbal contact and social connectedness of face-to-face delivery. Moreover, contrary to reservations from service providers, especially with regards to severe mental illness [SMI (Perle et al., 2013)], evidence from the current, and other (Kasckow et al., 2014) reviews suggest that telephone interventions are acceptable and well attended by adults with SMI.

Implications for research
To better establish the effectiveness of telephone interventions for people with a psychotic disorder, high quality, adequately powered studies are an important priority. The latter might best be conducted within existing practice settings to better evaluate the real-world impact of telephone-delivered interventions. To better understand the comparative clinical and cost effectiveness of telephone-delivered interventions, more head to head trials are needed. This would also help inform what, if any modifications are needed to ensure that telephone-delivered interventions meet the needs and preferences of service users. With the increasing focus on peer workers in mental health services, future research may also benefit from examining the acceptability and effectiveness of using peer workers to deliver telephone interventions. While it is challenging in studies of psychological interventions to use a double-blind design, the use of blinded outcomes measurement [e.g. a prospective, randomised, open, blinded endpoint (PROBE) design] has been argued to be a sufficient alternative (Hansson et al., 1992). Greater attention to non-symptom indicators of wellbeing (e.g. quality of life and functioning) and process variables (e.g. therapeutic alliance) is also warranted. To allow comparison between studies, greater uniformity in outcome measures would be beneficial. Accordingly, agreement upon and adherence to standard definitions of common outcome variables is an important priority for future research.
Supplementary material. The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718001125