Introduction
Bipolar disorder (BD) is characterized by recurrent episodes of depression and (hypo)mania with intervening periods of euthymia, but mood deviances between episodes can also present in the form of subsyndromal mood changes. BD is one of the most important causes of disability worldwide (Pini et al., Reference Pini, de Queiroz, Pagnin, Pezawas, Angst, Cassano and Wittchen2005; Harrison et al., Reference Harrison, Cipriani, Harmer, Nobre, Saunders, Goodwin and Geddes2016). Naturalistic follow-up studies suggest that the progressive development of BD is not prevented with the present treatment options (Angst and Sellaro, Reference Angst and Sellaro2000; Kessing et al., Reference Kessing, Hansen and Andersen2004). Given the limited resources available, a recent report by the World Health Organization stated that ‘the use of mobile and wireless technologies (mhealth) to support the achievement of health objectives has the potential to transform the face of health service delivery across the globe’ (WHO, 2011). During recent years, there has been a rapid increase in mhealth technologies for self-monitoring within mental health (Smith, Reference Smith2012; Donker et al., Reference Donker, Petrie, Proudfoot, Clarke, Birch and Christensen2013; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Munkholm, Frost, Bardram and Kessing2016b; Wang et al., Reference Wang, Varma and Prosperi2018). Within BD research, a number of studies have reported on electronic self-monitoring of depressive and manic symptoms using regular cell phones (Bopp et al., Reference Bopp, Miklowitz, Goodwin, Stevens, Rendell and Geddes2010), personal digital assistants (Depp et al., Reference Depp, Mausbach, Granholm, Cardenas, Ben-Zeev, Patterson, Lebowitz and Jeste2010; Mutschler et al., Reference Mutschler, von Zitzewitz, Rossler and Grosshans2012), computers (Whybrow et al., Reference Whybrow, Grof, Gyulai, Rasgon, Glenn and Bauer2003; Bauer et al., Reference Bauer, Grof, Gyulai, Rasgon, Glenn and Whybrow2004, Reference Bauer, Rasgon, Grof, Gyulai, Glenn and Whybrow2005, Reference Bauer, Grof, Rasgon, Glenn, Alda, Priebe, Ricken and Whybrow2006, Reference Bauer, Glenn, Grof, Pfennig, Rasgon, Marsh, Munoz, Sagduyu, Alda, Quiroz, Sasse and Whybrow2007, Reference Bauer, Wilson, Neuhaus, Sasse, Pfennig, Lewitzka, Grof, Glenn, Rasgon, Bschor and Whybrow2008; Chinman et al., Reference Chinman, Young, Schell, Hassell and Mintz2004; Adli et al., Reference Adli, Whybrow, Grof, Rasgon, Gyulai, Baethge, Glenn and Bauer2005; Lieberman et al., Reference Lieberman, Swayze and Goodwin2011) and smartphones (Depp et al., Reference Depp, Kim, de Dios, Wang and Ceglowski2012; Hidalgo-Mazzei et al., Reference Hidalgo-Mazzei, Mateu, Reinares, Undurraga, Bonnín Cdel, Sánchez-Moreno, Vieta and Colom2015; Bilderbeck et al., Reference Bilderbeck, Atkinson, McMahon, Voysey, Simon, Price, Rendell, Hinds, Geddes, Holmes, Miklowitz and Goodwin2016) as the electronic self-monitoring tools. However, none of the studies has included data on objective measures of illness activity, and the effect of smartphone-based monitoring has only been investigated in a single randomized controlled trial (RCT), the MONARCA I trial, conducted by the authors (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a). The MONARCA I trial investigated the effect of smartphone-based monitoring including a clinical feedback loop during 6 months compared with using a placebo smartphone in patients with BD. Overall, no differences between the intervention group and the control group were found, but importantly sub-analyses suggested that smartphone-based monitoring may sustain depressive symptoms and reduce manic symptoms (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a).
Social activity and communication (Weinstock and Miller, Reference Weinstock and Miller2008), as well as mobility (Kupfer et al., Reference Kupfer, Weiss, Foster, Detre and McPartland1974; Kuhs and Reschke, Reference Kuhs and Reschke1992; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Brage, Vinberg, Christensen, Knorr, Jensen and Kessing2012) represent central aspects reflecting illness activity in BD. It is likely that the ability of smartphone-based self-monitored measures may not be sufficient to detect prodromal depressive and manic symptoms compared with automatically generated objective smartphone data (referred to as objective smartphone data) on measures of illness activity such as phone usage, social activity and mobility which we have shown in previous studies correlate with the level of depressive and manic symptoms, discriminates between states and may represent a potential diagnostic marker (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Vinberg, Christensen, Bardram and Kessing2014a, Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2015b, Reference Faurholt-Jepsen, Vinberg, Frost, Debel, Margrethe Christensen, Bardram and Kessing2016c, Reference Faurholt-Jepsen, Busk, Þórarinsdóttir, Frost, Bardram, Vinberg and Kessing2018). It has never been tested in an RCT whether smartphone-based monitoring including a clinical feedback loop integrating subjective as well as objective smartphone data in patients with BD improves illness outcome.
Following the MONARCA I study, adjustments to the smartphone-based monitoring system were made and a new integrated clinical feedback loop, based on real-time prediction models including both subjective measures and objective smartphone data, was established and implemented in the Monsenso system.
We therefore conducted the MONARCA II RCT and hypothesized that smartphone-based monitoring and mood prediction including a clinical feedback loop in patients with BD would reduce the level of depressive and manic symptoms more than standard treatment.
Methods
The trial is reported according to the CONsolidated Standards Of Reporting Trials (CONSORT) statement and Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) (Boutron et al., Reference Boutron, Moher, Altman, Schulz and Ravaud2008; Moher et al., Reference Moher, Hopewell, Schulz, Montori, Gøtzsche, Devereaux, Elbourne, Egger and Altman2010; Chan et al., Reference Chan, Tetzlaff, Altman, Dickersin and Moher2013).
Details concerning the design and methods of the trial have previously been published (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2014b).
Design, settings and participants
The trial was a randomized controlled single-blind parallel-group design with an unbalanced allocation ratio (2:1) of adult patients with BD with a 9-month follow-up period. All patients with a diagnosis of BD who previously (but not currently) had been treated at The Copenhagen Clinic for Affective Disorder Copenhagen, Denmark in the period 2004 to January 2016 but currently treated elsewhere (e.g. at community psychiatric centres, private psychiatrists and general practitioners) were invited to participate in the trial, corresponding to approximately 735 potential participants (Fig. 1). The Copenhagen Clinic for Affective Disorder is a specialized outpatient clinic covering a recruitment area of the Capital Region, Denmark, corresponding to 1.4 million people. Patients with BD were initially referred to the Copenhagen Clinic for Affective Disorder from secondary healthcare when a diagnosis of a single mania or BD was made for the first time or in the case of occurrence of treatment resistance (persistent depressive or manic symptoms or recurrence despite treatment with standard care). Treatment at the Copenhagen Clinic for Affective Disorder comprised combined psychopharmacological treatment as according to international guidelines and supporting therapy for a 2-year period. All the patients included in the present trial had finalized their course of treatment at the Copenhagen Clinic for Affective Disorder before they were invited to participate in the present trial.
Fig. 1. The MONARCA II – flow diagram.
Inclusion criteria: BD diagnosis according to the International Classification of Diseases (ICD-10) (World Health Organisation, 2011) using Schedules for Clinical Assessments in Neuropsychiatry (SCAN) (Wing et al., Reference Wing, Babor, Brugha, Burke, Cooper, Giel, Jablenski, Regier and Sartorius1990) and previous treatment at the Copenhagen Clinic for Affective Disorder, Denmark.
Exclusion criteria: Schizophrenia, schizotypal or delusional disorders; previous use of the MONARCA system; pregnancy and a lack of Danish language skills.
Randomization, allocation and implementation
In addition to standard treatment, patients were randomized with an unbalanced allocation of 2:1 to either (1) the intervention group: active use of the smartphone-based Monsenso system on either (i) a smartphone capable of collecting objective smartphone data on measures of illness activity (Android smartphones) or (ii) a smartphone not capable of collecting objective smartphone data (iPhone) or to (2) the control group: use of a smartphone for usual communication. A computer-generated list of randomization allocation numbers was obtained by an independent researcher (KM) using http://randomization.com. The unbalanced allocation ratio was chosen due to differences in the level of available data collected for the clinical feedback loop in the intervention group. The type of smartphone used during the trial (Android or iPhone) was chosen by the patients themselves before randomization and was estimated to be approximately 50% on each of these types of smartphones – Android (capable of collecting objective smartphone data) v. iPhone (not capable of collecting objective smartphone data). A non-stratified randomization with random block sizes was used. The allocation sequence was concealed in numbered opaque envelopes in a locked cabinet of unknown location to the blinded researchers.
Blinding
The MONARCA II trial was a single-blinded trial. Owing to the type of intervention, the patients, the patients' health care provider and the study nurse were aware of the allocated randomization group. The researchers conducting outcome assessments, data entry, data analysis, interpretation and writing of the present paper were kept blinded to allocation.
Intervention
The intervention group
The Monsenso system was available for smartphones capable of collecting objective smartphone data (Android) or smartphones not capable of collecting objective smartphone data (iPhone). Regardless of the type of smartphone used, the patients randomized to the intervention group had the Monsenso system installed on their own or a borrowed smartphone and instructed to use the system for daily self-monitoring (Fig. 2). Self-monitored measures included mood, sleep duration, activity level, etc. [details provided in the online Supplementary data e1 and in Faurholt-Jepsen et al. (Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2014b)].
Fig. 2. The Monsenso system for smartphones.
Objective smartphone data including data on phone usage, social activity and mobility were included in the clinical feedback loop (details provided in the online Supplementary data). All data were used in a real-time forecasting algorithm to predict future mood. The clinical feedback loop between patients and clinicians comprised a study nurse who examined the collected data on a web-page, including the mood prediction (details provided in the online Supplementary data). If mood prediction was exceeding a predefined threshold, a message would be sent both to the patients and the study nurse (details provided in the online Supplementary data).
The control group
The patients who were randomized to the control group were offered to borrow a smartphone or used their smartphones for usual communication.
All patients continued with standard treatment.
Outcomes and assessments
Outcome measures were defined a priori (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2014b). Primary outcome measures were differences in clinically rated depressive and manic symptoms measured using the Hamilton Depression Rating Scale 17-items (HDRS-17) and the Young Mania Rating Scale (YMRS), respectively (Hamilton, Reference Hamilton1967; Young et al., Reference Young, Biggs, Ziegler and Meyer1978) between the intervention group and the control group.
Sub-analyses in relation to the primary outcomes: Since the effect of smartphone-based monitoring in BD has been sparingly investigated and based on interesting findings from the MONARCA I trial, sub-analyses in relation to the primary outcome were defined a priori (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2014b). Differences in depressive and manic symptoms measured using HDRS-17 item and YMRS, respectively, between the intervention group and the control group in patients with (a) no mixed symptoms (defined as YMRS⩾7 or HDRS⩾7), (b) presence of depressive and manic symptoms (defined as HDRS-17>0 or YMRS>0) at any given visit and (c) with depressive or manic symptoms at baseline (defined as HDRS⩾7 or YMRS⩾7) were investigated. Also, differences in rates of relapse of depressive and manic episodes during the trial period between the intervention group and the control group were investigated.
Secondary outcomes were defined as differences in functioning according to the Functional Assessment Short Test (FAST) (Rosa et al., Reference Rosa, Sánchez-Moreno, Martínez-Aran, Salamero, Torrent, Reinares, Comes, Colom, Van Riel, Ayuso-Mateos, Kapczinski and Vieta2007) between the intervention group and the control group.
Tertiary outcomes were defined as differences between the intervention group and the control group in perceived stress according to Cohen's Perceived Stress Scale (PSS) (Cohen et al., Reference Cohen, Kamarck and Mermelstein1983), quality of life according to the WHO Quality of Life-BREF (WHOQoL-BREF) (WHO, 1998), self-rated depressive symptoms according to Beck's Depressive Inventory (BDI) (Bech et al., Reference Bech, Rasmussen, Olsen, Noerholm and Abildgaard2001), self-rated manic symptoms according to the Altman Self Rating scale for Mania (ASRM) (Altman et al., Reference Altman, Hedeker, Peterson and Davis1997), self-rated recovery according to the Recovery Assessment Scale (Corrigan et al., Reference Corrigan, Salzer, Ralph, Sangster and Keck2004), self-rated medicine adherence according to the Medicine Adherence Rating Scale (MARS) (Thompson et al., Reference Thompson, Kulkarni and Sergejew2000) and empowerment according to Roger's Empowerment Scale (Rogers et al., Reference Rogers, Chamberlin, Ellison and Crean1997).
All assessments were carried out rater-blinded by one researcher (MFJ) at baseline, 4 weeks, 3, 6 and 9 months.
Statistical methods
Statistical power
A priori, we defined that the minimal clinically relevant difference in the level of depressive and manic symptoms between the intervention group and the control group was three points on HDRS-17 item and YMRS, respectively. Further, the standard deviation (s.d.) on the HDRS-17 item and the YMRS in the overall population included was 4.7 points. The statistical power was 80% with α = 0.05 for a two-sample comparison of means when including a minimum of 39 patients in each of the intervention groups (39 × 2: n = 78) and 39 patients in the control group (N = 117).
Statistical analyses
The statistical analyses were conducted using an intention-to-treat (ITT) analysis. Two-sample t tests and χ2 tests were used to assess the differences in means and proportions of background characteristics (Table 1). For each outcome, a two-level linear mixed-effect model, which accommodates both intra-individual and inter-individual variation of the specified outcome variable, including a fixed effect of visit number and a patients-specific random effect allowing for individual intercept and slope for each participant was defined (Table 2 and 3). Level 1 represented repeated measures of symptoms (e.g. HDRS, YMRS) and level 2 represented inter-individual variation. Differences between the intervention group and the control were investigated first in unadjusted models (except for differences in baseline values of the outcome variable) and second in models adjusted for age and sex. Model assumptions were checked. Potential interactions between randomization group and visit number on the outcome of interest were investigated. Differences in rates of relapse of episodes during the study were investigated using survival analysis [hazard rate (HR)] with reasons for censoring being death or date of drop out (Table 2). Analyses on probability of individual psychopharmacological prescriptions were investigated using logistic regression analyses [odds ratio (OR)] (Table 3). The threshold for statistical significance was p ⩽ 0.05 (two-tailed). Analyses were done using STATA 14 (StataCorp LP, College Station, TX, USA).
Table 1. Clinical and socio-demographic characteristics of study participants at baseline, N = 129
Data are mean (s.d.), median [IQR] or % (n) unless otherwise stated.
Table 2. Estimated differences in primary outcomes between the intervention group and the control group (control group serve as reference), N = 129
The use of bold indicate the primary main analyses.
a Adjusted for outcome variable at baseline.
b Adjusted for outcome variable at baseline, age and gender unless otherwise specified.
c The Hamilton Depression Rating Scale 17-items.
d The Hamilton Depression Rating Scale six-items.
e Defined as YMRS ⩾7.
f The Young Mania Rating Scale.
g Defined as HDRS 17-items ⩾7.
h Defined as HDRS-17 ⩾13.
i Defined as YMRS ⩾13.
Table 3. Estimated differences in secondary and tertiary outcomes between the intervention group and the control group (control group serve as reference), N = 129
FAST, The Functional Assessment Short Test; PSS, Cohen's Perceived Stress Scale; WHOQoL-bref, World Health Organization Quality of Life – short version; BDI, Beck's Depressive Inventory; ASRM, The Altman Self Rating scale for Mania; Recovery, The Recovery Assessment Scale; Empowerment, Roger's Empowerment Scale; Adherence to medication, The Medicine Adherence Rating Scale.
a Adjusted for outcome variable at baseline.
b Adjusted for outcome variable at baseline, age and gender unless otherwise specified.
c Adjusted for variable at baseline, age, gender, HDRS score and YMRS score.
Ethical considerations
Ethical permission was obtained from the Regional Ethics Committee in The Capital Region of Denmark and The Danish Data Protection Agency (H-2-2014-059). The law on handling of personal data was respected. The trial was registered at ClinicalTrials.gov as NCT02221336. Electronic data collected from smartphones were stored at a secure server at CIMT, Capital Region, Denmark (I-suite number RHP-2011-03). All patients were offered to borrow a smartphone free of charge during the trial period, and economic costs due to data traffic from the Monsenso system were refunded. No economic compensation for participating was provided. The trial complied with the Declaration of Helsinki of 1975, as revised in 2008.
Results
Background characteristics
Figure 1 presents the flow of patients in the MONARCA II trial. A total of 735 patients previously treated at the Copenhagen Clinic for Affective Disorder, Denmark from 2004 to January 2016, but currently treated at community psychiatric centres, private psychiatrists and general practitioners, were assessed for eligibility. Of these 544 patients were not included due to; unable to get in contact with the patient (n = 240), declined to participate (main reasons: did not have the time, did not want to participate in a research study or had moved too far away making transportation a problem) (n = 282 patients), or were excluded due to previous use of the MONARCA system (n = 22). The last patient visit was in January 2018. As 17 patients did not show up for baseline assessments and randomization, a total of 129 patients were randomized 2:1 to either the intervention group [active use of the smartphone-based Monsenso system on either (i) a smartphone collecting objective smartphone data or (ii) a smartphone not collecting objective smartphone data] or to the control group (use of a smartphone for usual communication) and left for ITT analyses. Of these, nine patients expressed that they wished to drop out during the trial (intervention group: six; control group: three) due to not having the time to participate (one patient wished to drop out due to self-monitoring). A total of seven patients (5.4%) had a discontinued intervention (intervention group: four; control group: three). Thus, there was only missing outcome data to a small extent and therefore not accounted for in the statistical analyses. Clinical and socio-demographic characteristics are presented in Table 1. Details regarding results on the adherence to the self-monitoring in the intervention group, technical aspects of the system and the clinical feedback loop are provided in the online Supplementary data (e2).
Primary outcomes
Differences between the intervention group and the control in depressive and manic symptoms according to HDRS and YMRS, respectively, are presented in Table 2. There were no differences between the intervention group and the control group in level of depressive [adjusted model: coefficient B (B) = 0.61, 95% CI −0.77 to 2.00, p = 0.38] and manic symptoms (adjusted model: B = −0.25, −1.1 to 0.59, p = 0.56). Analyses on interactions between randomization group and time (visit number) were statistically non-significant, and therefore not reported.
Exploratory analyses in relation to primary outcomes
Sub-analyses are presented in Table 2. Excluding mixed symptoms, patients in the intervention group experienced statistically significantly lower levels of manic symptoms compared with the control group (adjusted model: B = −1.11, 95% CI −2.22 to −0.01, p = 0.050) (n = 88). When including patients with a HDRS score ⩾7 at baseline, there was no difference between the intervention group and the control group (adjusted model: B = 0.74, 95% CI −1.30 to 2.79, p = 0.48) (n = 64). However, patients in the intervention group presenting with an YMRS score ⩾7 at baseline experienced higher levels of manic symptoms compared with the control group (adjusted model: B = 4.21, 95% CI 0.45–7.97, p = 0.028) (n = 19).
In separate independent exploratory stratified analyses on (a) ⩾3 or <3 previous affective episodes, (b) BD type I or type II, (c) ⩾3 or <3 previous psychiatric hospitalization, (d) index affective episode depression or mania, there were no differences in the level of depressive and manic symptoms between the intervention group and the control group (all p > 0.05).
In survival analyses on differences in the rates of relapse in depressive and manic episodes based on patients without being in a current affective episode at baseline, there were no differences in the number of episodes, when not considering the polarity of episodes (depression or mania), between the intervention group and the control group (N = 120, HR 0.76, 95% CI 0.41–1.42, p = 0.39). However, there was a statistically significantly higher rate of depressive episodes between the intervention group and the control group (N = 97, HR 2.89, 95% CI 1.02–8.23, p = 0.047), but a lower rate of manic episodes in the intervention group (N = 116, HR 0.17, 95% CI 0.037–0.78, p = 0.022).
Secondary and tertiary outcomes
Analyses on secondary and tertiary outcomes are presented in Table 3. There were significantly lower perceived stress and higher quality of life in the intervention group compared with the control group (perceived stress, adjusted model: B = −2.08, 95% CI −4.15 to −0.01, p = 0.049; quality of life, adjusted model: B = 4.00, 95% CI 0.04–7.97, p = 0.048). There were no differences between the intervention group and the control group on other secondary and tertiary outcome measures.
In logistic regression models on probability of prescription of individual groups of psychopharmacological treatments, the intervention group had higher probability of being prescribed lithium than the control group (OR 2.78, 95% CI 1.33–6.20, p = 0.007). In survival analyses on differences in the rate of prescription of lithium among patients not in lithium treatment at baseline, there was no difference between the intervention group and the control group (N = 57, HR 0.30, 95% CI 0.032–2.93, p = 0.30). There was no difference in the rate of discontinuation of lithium among patients in lithium treatment at baseline between the intervention group and the control group (N = 66, HR 0.36, 95% CI 0.08–1.61, p = 0.18).
Notably, further adjustment of the analyses on the primary outcomes (HDRS and YMRS) for the use of lithium did not alter the estimates.
There were no differences in the probability of prescription of other psychopharmacological treatments (antipsychotics, antiepileptics and antidepressants) between the intervention group and the control group (all p values > 0.29). In addition, there were no differences in the rate of prescription or discontinuation of other psychopharmacological treatments (antipsychotics, antiepileptics and antidepressants) between the intervention group and the control group (all p values > 0.52). Further adjustment of the analyses on the primary outcomes (HDRS and YMRS) for the use of other psychopharmacological treatments (antipsychotics, antiepileptics and antidepressants) did not alter the estimates.
The two intervention groups were lumped together for the reported statistical analyses. Analyses according to each of the three randomization groups did not affect the estimates reported in Tables 2 and 3. There were no statistically significant differences in any of the outcome measures or in any of the statistical analyses (including sub-analyses) between the two groups receiving the active smartphone-based intervention (all p value > 0.30). In addition, there were no statistically significant differences in any of the baseline characteristics between any of the three groups (including the two groups receiving the active smartphone-based intervention) (p value > 0.78).
Discussion
The present paper presents results from the first RCT investigating the effect of smartphone-based patient-reported and objective monitoring including a mood prediction system used with a clinical feedback in patients with a psychiatric disorder. In this single-blinded trial, patients with BD were randomized to a smartphone-based monitoring system including a feedback loop or to the use of a smartphone for usual communication for 9 months. Overall, there was no effect on clinically rated depressive and manic symptoms. However, patients in the intervention group reported significantly higher quality of life and lower perceived stress compared with the control group. Since the effect of smartphone-based monitoring in BD has been sparingly investigated and based on interesting findings from the previous MONARCA I trial, sub-analyses in relation to the primary outcome were defined a priori and employed (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2014b).
Exploratory analyses suggested that smartphone-based monitoring may reduce the risk of relapse of manic episodes but increase the risk of relapse of depressive episodes. The exploratory sub-analyses were based on small number of patients, and may be chance findings, but interestingly some of the a priori defined analyses resulted in findings in line with findings from our previous trial (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a).
Currently, approximately 1/3 of the world's adult population owns and uses a smartphone, and it is estimated that by 2018, this proportion will increase to 50% (FierceWireless, 2011). During recent years, there have been a rapidly increasing international interest in and use of mhealth technologies within mental health (Smith, Reference Smith2012; Anthes, Reference Anthes2016; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Munkholm, Frost, Bardram and Kessing2016b; Insel, Reference Insel2017; Wang et al., Reference Wang, Varma and Prosperi2018). However, the scientific evidence base is small as there are yet few RCTs investigating the effect of smartphone-based monitoring in patients with a psychiatric disorder [e.g. psychosis (Krzystanek et al., Reference Krzystanek, Borkowski, Skałacka and Krysta2018), anxiety (Stolz et al., Reference Stolz, Schulz, Krieger, Vincent, Urech, Moser, Westermann and Berger2018), depression (Ben-Zeev et al., Reference Ben-Zeev, Brian, Jonathan, Razzano, Pashka, Carpenter-Song, Drake and Scherer2018; Furukawa et al., Reference Furukawa, Horikoshi, Fujita, Tsujino, Jinnin, Kako, Ogawa, Sato, Kitagawa, Shinagawa, Ikeda, Imai, Tajika, Ogawa, Akechi, Yamada, Shimodera, Watanabe, Inagaki and Hasegawa2018) and BD (Depp et al., Reference Depp, Ceglowski, Wang, Yaghouti, Mausbach, Thompson and Granholm2015; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a)], and these are limited by methodological issues such as small samples, short follow-up periods and including unblinded patient-reported outcome measures. Within BD, RCTs investigating the effect of mhealth interventions are strikingly few, with most trials investigating the effect of web-based interventions with varying contents (Lieberman et al., Reference Lieberman, Kelly, Douglas and Goodwin2010; Todd et al., Reference Todd, Jones, Hart and Lobban2014; Barnes et al., Reference Barnes, Hadzi-Pavlovic, Wilhelm and Mitchell2015; Lauder et al., Reference Lauder, Chester, Castle, Dodd, Gliddon, Berk, Chamberlain, Klein, Gilbert, Austin and Berk2015), and only two trials used smartphone as the monitoring tool (Depp et al., Reference Depp, Ceglowski, Wang, Yaghouti, Mausbach, Thompson and Granholm2015; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a). Of these, only one investigated the effect of smartphone-based monitoring and clinical feedback compared with standard treatment (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a), whereas the other investigated the effect of smartphone-based delivered interactive intervention linking patient-reported mood states with personalized self-management strategies without feedback compared with paper-and-pencil mood monitoring (Depp et al., Reference Depp, Ceglowski, Wang, Yaghouti, Mausbach, Thompson and Granholm2015). The MONARCA I RCT was the first to investigate the effect of smartphone-based monitoring including a feedback loop between patients and clinicians compared with standard treatment on the level of depressive and manic symptoms in patients with BD (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Ritz, Christensen, Jacoby, Mikkelsen, Knorr, Bardram, Vinberg and Kessing2015a). Findings from sub-analyses suggested that patients using the MONARCA system had more sustained depressive symptoms than the control group but had fewer manic symptoms. Interestingly, the findings from the present MONARCA II trial were in line with these findings. It may be that identifying and treating manic prodromes may be easier than for depressive prodromes (Perry et al., Reference Perry, Tarrier, Morriss, McCarthy and Limb1999; Simon et al., Reference Simon, Ludman, Bauer, Unützer and Operskalski2006), and that continuous monitoring of depressive symptoms may have a detrimental effect. Although the idea of electronic self-monitoring and mood prediction seems appealing, studies using rigorous methodology carefully investigating beneficial as well as possible harmful effects of electronic monitoring are needed.
Given the limited access to appropriate treatment facilities, smartphone-based monitoring represents a flexible and adjustable system that could potentially be of great support for both patients and health care providers (Musiat et al., Reference Musiat, Goldstone and Tarrier2014). Continuous availability and support with tailored content of intervention may engage patients who would not seek help through traditional routes. Continuous monitoring could possibly increase the patients' illness insight and awareness of sub-syndromal symptoms, which have been associated with poor prognostic factors including impaired functioning and high risk of relapse (Judd et al., Reference Judd, Schettler, Akiskal, Maser, Coryell, Solomon, Endicott and Keller2003; Strejilevich et al., Reference Strejilevich, Martino, Murru, Teitelbaum, Fassi, Marengo, Igoa and Colom2013; Patel et al., Reference Patel, Lloyd, Jackson, Ball, Shetty, Broadbent, Geddes, Stewart, McGuire and Taylor2015; O'Donnell et al., Reference O'Donnell, Ellis, de Loo, Stange, Axelson, Kowatch, Schneck and Miklowitz2018), and alertness of early warning signs of upcoming depressive or manic episodes. This will provide new opportunities for early intervention in outpatient settings which has been shown to reduce the risk of rehospitalization (Kessing et al., Reference Kessing, Hansen, Hvenegaard, Christensen, Dam, Gluud and Wetterslev2013). In this way, it is possible that outpatient treatment can be optimized, and that the frequency of necessary health care provider/clinician and other clinical staff visits can be reduced and be more flexible according to the needs of the patients.
Limitations
The overall findings from the present trial are negative. It could be that there is an effect of smartphone-based monitoring on other outcomes (e.g. readmissions, mood instability, illness insight, rumination, activity) which we did not include, or which the clinical rating scales used in the present trial (the HDRS and the YMRS) did not capture comprehensively enough. However, the HDRS and the YMRS are currently the international golden standard to evaluate the severity of depressive or manic symptoms in BD and will therefore give the opportunity to compare results across trials and include in potential future meta-analyses. Ongoing studies investigating the effect of smartphone-based monitoring on rate of readmission, mood instability and rumination in patients with BD will hopefully provide more insight (Hidalgo-Mazzei et al., Reference Hidalgo-Mazzei, Mateu, Reinares, Undurraga, Bonnín Cdel, Sánchez-Moreno, Vieta and Colom2015; Saunders et al., Reference Saunders, Cipriani, Rendell, Attenburrow, Nelissen, Bilderbeck, Vasudevan, Churchill, Goodwin, Nobre, Harmer, Harrison and Geddes2016; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Martiny, Tuxen, Rosenberg, Busk, Winther, Bardram and Kessing2017; Mühlbauer et al., Reference Mühlbauer, Bauer, Ebner-Priemer, Ritter, Hill, Beier, Kleindienst and Severus2018).
There may be different effects of the different components of the monitoring system. The MONARCA II trial was designed to investigate the effect of the total Monsenso system. Thus, it was not possible to address the possible positive or negative effects of the individual components of the intervention. Future trials including groups with different levels of smartphone-based monitoring could be able to investigate the effects of individual components. Since the objective of the present trial was to investigate the effect of smartphone-based monitoring compared with standard treatment, we did not conduct within groups investigating differences in outcome measures from baseline to the end of the trial in the individual groups.
The feedback loop was based on prediction models based on subjective measures and on smartphones capable of collecting these data also on objective smartphone data. Previous studies by the authors have shown that, even in patients with rather severe levels of depressive and manic symptoms, the self-monitored data as well as the objective smartphone data reflect clinically rated depressive and manic symptoms (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Christensen, Bardram and Kessing2015b, Reference Faurholt-Jepsen, Busk, Frost, Vinberg, Christensen, Winther, Bardram and Kessing2016a, Reference Faurholt-Jepsen, Vinberg, Frost, Debel, Margrethe Christensen, Bardram and Kessing2016c). The prediction models in the feedback loop were adjusted during the trial (a learning model) allowing for optimal titration of the intervention and to achieve the best prediction possible (Insel, Reference Insel2017). However, during the trial there were periods with technical problems regarding the prediction models, and thus feedback based on these data was functioning less optimal.
The patients presented with a rather low level of depressive and manic symptoms making it harder to demonstrate a potential effect on depressive and manic symptoms (Table 1). Including patients presenting with more severe symptoms could have resulted in other findings, and will be investigated further by the authors (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Frost, Martiny, Tuxen, Rosenberg, Busk, Winther, Bardram and Kessing2017). A previous study by the authors found that patients were able to validly evaluate the level of depressive and manic symptoms during more severe episodes than in the present study and further, that patients tolerated and adhered to the self-monitoring during these states (Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Frost, Debel, Margrethe Christensen, Bardram and Kessing2016c).
Due to the type of intervention, it was not possible to blind the patients, the clinicians or the study nurse to randomization group. However, the researchers collecting clinically evaluated outcome data (HDRS, YMRS and FAST) were blinded to randomization group. The reported tertiary outcomes (e.g. perceived stress and quality of life) were due to the design of the trial collected unblinded to randomization and therefore prone to detection bias. Thus, given that patients in the intervention group knew that they were being monitored, this might have given an additional subject sensation of control over their symptoms and support (less perceived stress and higher quality of life).
In any non-pharmacological treatment trial, it is always a challenge to define a control group. The trial was designed to eliminate the effect of simply using a smartphone by including a control group of patients using a smartphone for usual communication.
Generalizability
An RCT represents a study design with high interval validity, with a possible cost of lower external validity and lower generalizability of the results. The present trial used a single-blinded design, with data on outcome measures collected by a researcher not aware of randomization group and therefore data were not affected by bias. The trial included patients treated in community centres, private psychiatrists or general practice at the time of the trial. Further, the trial had a pragmatic design with few exclusion criteria. The follow-up period was quite long. Thus, the results from the present trial reflect the use of smartphone-based monitoring during clinical settings. In addition, the Monsenso system is easy to use, and user-friendly with a high usability (Bardram et al., Reference Bardram, Frost, Szanto and Margu2012, Reference Bardram, Frost, Szánto, Faurholt-Jepsen, Vinberg and Kessing2013; Faurholt-Jepsen et al., Reference Faurholt-Jepsen, Vinberg, Christensen, Frost, Bardram and Kessing2013, Reference Faurholt-Jepsen, Frost, Vinberg, Christensen, Bardram and Kessing2014). The trial had a high retention rate, which could be due to the pragmatic design and few requirements from the participants during the trial. Also, patients expressed that they wished to contribute to the lack of evidence regarding smartphone-based monitoring in BD. Overall, the findings of the present trial are found to be generalizable to patients with BD in general.
Conclusion
Smartphone-based monitoring and real-time mood prediction, including a clinical feedback loop, did not reduce the severity of depressive and manic symptoms in the present trial. Exploratory sub-group analyses showed that the intervention group had increased risk of relapse of depressive episodes compared with the control group. In exploratory analysis of patients without mixed symptoms, the intervention group had statistically significantly fewer manic symptoms and lower risk of manic episodes. Findings from sub-analyses were based on small numbers, could represent chance findings and should be interpreted with caution. Patient-evaluated unblinded data showed that the intervention group experienced higher quality of life and lower levels of perceived stress compared with the control group. Potential negative effects of smartphone-based monitoring and mood prediction should be investigated carefully before implemented as a standard clinical tool.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291719000710.
Author ORCIDs
Maria Faurholt-Jepsen, 0000-0002-0462-6444.
Acknowledgements
The authors would like to thank the patients for participating in the trial and nurse Bente Støyer for running the integrated clinical feedback loop. The trial received financial support from The Mental Health Services in the Capital Region of Denmark including the Centre for Telemedicine in the Capital Region of Denmark and EU 7th Frame Programme, IT University of Copenhagen and the Lundbeck Foundation. The research team has no financial connection to any of the contributors.
Author contributions
MFJ and LVK conceived and conducted the trial, analysed data and authored the first draft of the manuscript. MV, MF and JB have been revising and optimising the manuscript. All authors contributed to and approved the final version of the manuscript.
Financial support
The funding organizations had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; and decision to submit the manuscript for publication.
Conflict of interest
MFJ has no competing interests. MF and JEB are co-founders and shareholders in Monsenso ApS. EMC has been a consultant for Eli Lilly, Astra Zeneca, Servier, Bristol-Myers Squibb, Lundbeck and Medilink. MV has been a consultant for Lundbeck within the past 3 years. LVK has been a consultant for Sunovion.
