The growing field of digital psychiatry: current evidence and the future of apps, social media, chatbots, and virtual reality

Mental health problems impact over one billion people world­wide annually1, with depression representing the leading cause of disability across the globe2. The World Health Organization’s Mental Health Gap Action Program (mhGAP) outlines evidence-based interventions to address this global crisis, yet acknowledges that barriers include lack of available services and funding3.

The extent of these barriers, even for high-income countries, is highlighted in a December 2020 report from the US government, which indicates that offering evidence-based mental health care in the US alone would require an additional 4 million trained professionals4. On a global scale, it is simply not feasible to propose that practices based entirely on in-person care will ever be able to meet this demand. Thus, even before the emergence of the COVID-19 pandemic, there was growing interest in the possible role of new technologies to extend care.

The rapid international growth in access to and capabilities of digital health technologies (DHTs) presents a feasible route towards augmenting traditional mental health care and bridging the gap between the need for treatment and the capacity to deliver it. In this paper, we consider DHTs to be innovations beyond electronic medical records or classical telepsychiatry, to instead focus on the recent developments in smartphone apps, virtual reality, social media, and chatbots.

While the integration of these DHTs into mental health care began somewhat slowly, restrictions driven by the COVID-19 pandemic have sparked a paradigm shift as assumptions, interest and utilization of digital health have undergone a fundamental transformation. Although there has been variability in the response of health care services to the unmet needs raised by the pandemic, a recent study encompassing 17 different countries reported an overall increased use of digital health in mental health care settings, as well as a renewed support for facilitating uptake during the pandemic5. This increased uptake in response to the pandemic is related not only to DHTs’ ability to connect people to care while social distancing regulations are in place, but also to recent innovations in these technologies that enable them to deliver scalable, affordable and accessible mental health care solutions6, 7.

In this state-of-the-art review, we explore the technologies, the available research evidence and the implementation issues most relevant to integrating digital psychiatry within mental health care. In the first section, we discuss technology mediums of smartphones, social media, virtual reality and chatbots as innovations in the digital psychiatry revolution. The second section critically discusses recent research informing the clinical evidence-based uses of DHTs, with a focus on smartphone studies, covering their use across multiple contexts, from the promotion of public mental health and well-being, to the management of long-term psychiatric conditions. The third section identifies the forefront challenges towards implementation, and discusses potential solutions for improving the use and facilitating evidence-based adoption of DHTs into mental health care across the world.

TOOLS AND TECHNOLOGIES

The aptly titled 2012 New York Times article The Therapist May See You Anytime, Anywhere 8 highlights that the use of smartphone devices in mental health care has been discussed and anticipated for nearly a decade. Smartphones have quickly become a driving force of digital health, due to special properties defining both the hardware and software of these devices.

From a hardware perspective, they are compact and wireless, with low purchasing and running costs, making them the first devices to provide ubiquitous connectivity/Internet access for a sizeable proportion of the global population. The sensors on these devices allow for new data capture and graphical/computing power for delivery of individualized interventions.

According to 2018 survey data, 76% of people in advanced economies and 45% in emerging economies owned a smartphone9, with recent data from the US showing that ownership rates may be as high as 70% even among people with severe mental illness10-12. While a digital divide still does exist, it is feasible to envisage that, in the near future, the majority of the world will have access to some form of a smartphone device.

From a software perspective, the relative ease of building new smartphone programs (termed applications or “apps”), combined with the centralized online platforms for finding, sharing and downloading these (i.e., the “app store”), creates an almost infinite potential for any new idea to quickly become an “app”, which can in turn be readily proliferated across any number of users, potentially reaching billions of people across the world. Further, smartphones can serve as a digital “hub” for integration of novel devices such as wearables and other sensors.

In the context of mental health, the clearest result of this focus on smartphones has been the massive influx of apps aiming to provide therapeutic interventions for virtually all known mental health problems13. Alongside app-based therapeutic interventions, smartphone devices also hold the potential for bolstering mental health care in a number of other ways, including: a) capturing longitudinal, dense and multimodal mental health data for use in diagnosis and monitoring; b) analyzing data, increasingly via machine learning paradigms, to generate clinically individual-level actionable insights and predictions; and c) offering a wide range of interventions often outside of the app itself, through facilitating connections to clinical care, peer support, personalized resources, emergency care, and even novel therapies. Below, we explain in more detail the evidence behind the multifaceted and large-scale applications of smartphones.

Smartphone sensor data and digital phenotyping

Until recently, a large portion of the understanding around the determinants of the onset, relapse or temporal variation in mental disorders was primarily based on data from large prospective studies. Although useful, the broad insights gained from such data may fail to capture individual differences or the more fine-grained temporal relationships between causes and consequences of mental ill-health. Across the entire field of health care, smartphones are providing a plethora of data enabling new insights into various conditions, through combining their increasingly detailed streams of longitudinal, multimodal and temporally dense data collection. To better clarify the nature and clinical utility of these data, the concepts of “active” and “passive” data are useful.

Active data typically refers to smartphone-based surveys – i.e., active symptom monitoring or ecological momentary assessment – which can be completed by the user either spontaneously or in response to a prompt, and then stored while crucially time-stamped (a digital record of the date and time when an item was completed) onto the collecting app. Active data capture offers a new means to characterize a patient’s clinical course.

While most clinical assessment scales have not been validated for deployment on mobile devices, strong correlations between traditional in-clinic metrics and their often-simplified mobile versions suggest adequate face validity14. The evolution of these assessments to focus on non-traditional metrics such as perception of self, functioning and social life (which research has shown to be particularly important to patients15) provides new opportunities for furthering the potential of active data collection. The use of smartphones for cognitive assessment16 and for remotely monitoring symptoms17 also appears feasible, with promising results even for severe mental illness such as schizophrenia18, 19.

While concern is often raised around using mental health apps for monitoring suicidal thoughts and urges, or even eliciting an increase in symptoms through reactivity to monitoring, research shows that actively collecting data on suicidal thoughts and urges does not elicit adverse effects20.

Passive data are obtained automatically through sensors, either on the smartphone or via a wearable device, ranging from simple device use metrics to accelerometry, global positioning system (GPS), and even now voice tone (via microphone) or facial expression (via camera) data. These automatically collected data offer a means to reduce patient burden typically related to active data collection, while also capturing novel digital markers of behavior.

Often referred to as “digital phenotyping”21 within the emerging framework of precision psychiatry, the multimodal nature of passive data obtained from consumer grade devices offers a means to understand the lived experiences of mental health in context22. For example, GPS data have recently offered insights into the relationship between reduced mobility and poorer mental health during the COVID-19 pandemic23. Passive data from smartphones have also been shown to correlate with outcomes such as social functioning and loneliness24, 25. An important trend emerging from passive data studies in various conditions is that variance, or measures of entropy or deviation from a personal mean, appear of more value than absolute measurements from any sensors23-25.

Recent reviews suggest that “most studies still only scratch the surface of advanced smartphone capabilities”26, and less than 2% of apps on the commercial marketplaces appear to leverage digital phenotyping potential27. Still, recent studies are employing digital phenotyping methods across diverse mental disorders28-30, and research interest in this field is expanding at a rapid pace.

The density and complexity of passive data31 is far greater than current clinical assessments, which continue to rely on static scales that ask a patient to recall symptoms over a defined period of time – e.g., a two week period in the case of the ubiquitous depression assessment by the Patient Health Questionnaire-9 (PHQ-9)32. However, the depth and diversity of passive data (which already typically combine measures such as step counts from wearables, text analytics from social media, metadata from electronic medical records, or green-space exposure from geolocation) require new techniques in data science, such as artificial intelligence and machine learning, to meaningfully combine and utilize such “big data” to inform mental health care33.

Advances in artificial intelligence and machine learning will likely represent a prominent bridge for translating new data into clinically relevant digital biomarkers34-36. Like all biomarkers, though, impact will be determined not only by statistical significance but also by clinical utility. A case in point refers to digital markers of self-harm and suicide, which, according to a recent review, possess high classification accuracy yet near zero accuracy for predicting future events37.

However, other approaches to digital phenotyping for different conditions/outcomes are beginning to show some promise. For instance, relapse risk in schizophrenia may be foreseen by “anomaly detection”, which involves the use of smartphone sensor data to monitor divergences of an individual’s behavioral patterns compared to his/her personal baseline. Preliminary studies in small samples have found reasonable sensitivity and specificity from applying this approach to date30.

Overall, while active and passive data have the potential to make smartphones crucial elements for the development and implementation of precision psychiatry38, the validity of the measures, how the data can be meaningfully represented, and the potential for ethical and effective uses in treatment delivery have all yet to be established.

Smartphone technologies for closed loop interventions

A rich legacy of Internet-delivered and computerized therapy research and experience39 is now in the process of being translated into new smartphone-based interventions, with promising results as well as challenges. These app-based interventions often utilize established aspects of cognitive and behavioral therapies to offer patients “on demand” access to evidence-based care tools. Examples abound of studies targeting mental health problems such as depression and anxiety40-43, and early psychosis and schizophrenia44, 45, that have been the subject of previous reviews46-51. The existing clinical evidence for digital health interventions across specific disorders is reviewed in more detail in the second section of this paper.

The potential for more personalized digital health interventions is bright. Known as a just-in-time-adaptive-intervention (JITAI), active and passive symptom data capture may aid in the development of personalized and real-time intervention strategies52, 53. For example, the smartphone may be able to infer low mood in the context of social isolation and offer a relevant intervention, whilst, in another circumstance, it may infer low mood in the context of poor sleep and recommend an alternative intervention. Although in its infancy, using JITAIs to offer “closed loop” mental health interventions is a promising area for future research.

Nevertheless, app marketplaces rarely reflect evidence from recent studies, or otherwise take advantage of the unique potential of app interventions54. For instance, just one percent of marketplace apps support use of sensors55, suggesting that concepts of digital phenotyping to support JITAI or behavioral interventions via apps are largely not incorporated into existing commercial technologies. Rather, even when considering more static interventions that do not take advantage of advanced smartphone features, the evidence base for widely proliferated apps remains poor54. For example, one review suggests that only ~2% of commercially available mental health apps are supported by original research evidence27. As we explore more details of app interventions in later sections, it is useful to consider that integration with sensors and digital phenotyping will likely soon transform this space.

Social media

The relationship between social media and mental health has received much attention from not only the academic literature, but also the traditional media and general public56. Frequently accessed via smartphone apps and connecting people from their own devices to global networks of friends, information, and health resources, social media can represent both a means to quantify mental health as well as a source of both positive and negative interactions.

Increasingly, research suggests that absolute screen time or exposure itself is not strongly associated with adverse mental health outcomes57. This is in marked contrast to the more popular conception that screen time and social media use is detrimental to mental health. In part, this view gained ascendence from the older literature, which was largely based on self-reported usage and cross-sectional analysis, thus offering limited evidence in this regard. Recent studies, however, based on objective screen use and social media engagement measurements, prospective cohorts, and new scales to assess problematic Internet use, are painting a more nuanced picture of social media and mental health58, 59. For example, during the COVID-19 pandemic, social media have been a source of social support for many who have been socially isolated and lonely.

While excessive use of social media and screen time is likely not beneficial for mental health (in the same manner that excessive use of any activity or behavior is often associated with deleterious outcomes), the quality of screen time and social media interactions appears to be more important than the quantity60. It is interesting that in recent years social media companies such as Facebook and Pinterest have undertaken new efforts to flag content that may be related to self-harm or suicide56. Nevertheless, it is currently difficult to determine the results of such interventions. The impact of social media on the developing brain also remains an unresolved61 yet frequently discussed topic, especially as the pandemic has forced increasing reliance on technology to connect people.

Patterns of social media use may represent a means to detect worsening of mental health symptoms. For example, changes in the content and style of social media posts may offer an early warning sign of relapse in schizophrenia62. Social media, combined with natural language processing methods, also offer a practical means to understand population-level mental health trends. For example, an analysis of 60 million Twitter posts in March-May 2020, as compared to one year prior, was able to detect pandemic-related increases in coping mechanisms63. These methods have also been employed in studies exploring psychosocial reactions to the COVID-19 pandemic64, 65, as well as the effects of psychiatric medications66.

While currently available work has largely focused on text-based natural language processing methods, the increasingly voice- and video-based nature of newer social media content has sparked interest in emotion recognition67. For example, early studies identified relationships between negative mood and posting pictures with darker colors68, although such relationships are now known to be more nuanced, thus highlighting inherent challenges in assessing mental health without a broader context.

Social media can also be used as a therapeutic tool. Novel research using carefully curated and monitored social networks as interventions has shown promise in youth with diverse mental health needs69-71. For example, the PRIME app72 is designed to help people with schizophrenia through the promotion of functional recovery and the mitigation of negative symptoms (e.g., amotivation) through a supportive and personalized network. The Moderated Online Social Therapy (MOST) platform is another example of an innovation that offers personalized therapy combined with social connections among other features71, 73.

It is noteworthy that social media are not without risk. Disinformation74 and stigma on social media are forces that cannot be ignored. Stigma on social media is common75, although efforts are also underway to challenge and reverse this trend76. Using social media for mental health work also remains a catalyst for ethical tensions, and a recent review offers a practical taxonomy of these tensions as well as guidance for navigating through these ongoing challenges77.

Chatbots

Conversational agents, such as Apple’s Siri or Amazon's Alexa, have become common in the digital marketplace. Termed “chatbots”, the use of these conversational style interfaces offers an intelligent, automated system for detecting and responding to immediate mental health needs. Chatbots have the look and feeling of interacting with a human, despite being run by an automated software program. Thus, chatbots or “robot therapists” have become a galvanizing force for those seeking to automate therapy where software programs listen and respond to people's mental health needs. While the words “robot therapist” conjure images of a physical robot, most are actually text based, although animated video and even physical robot versions have been researched78, 79.

One ongoing challenge in chatbot work is seeking to offer emotional support from inherently inanimate computer code. There is some evidence that people can develop therapeutic relationships with digital technologies (referred to as “digital therapeutic alliance”80). As therapeutic alliance with an in-person therapist is related to more positive outcomes in mental health treatment81, harnessing the digital therapeutic alliance through human-style interactions with a chatbot might promote change without the need of human support82. Research has found that some people feel more comfortable conversing anonymously with a chatbot83, and that this may open up the possibility to improve detection of distress and in turn provide momentary interventions to those who feel less comfortable with face-to-face contact84.

Chatbot interfaces have become a key feature of many commercially available mental health apps. However, their evidence base is not well established85. Across two recent systematic reviews, 24 studies investigating chatbots for health care were identified85, 86. Of the 11 trials targeting mental health problems, most were for depression, with a smaller number targeting anxiety, schizophrenia, post-traumatic stress disorder (PTSD), and autism spectrum disorder. Only two randomized controlled trials were included, and, while some mental health benefits from chatbot interventions were indicated, the types of benefits observed were not consistent across studies, which were further limited by small sample sizes, short duration, and a lack of follow-up data.

While the development and implementation of more complex interactive systems is inevitable, current chatbots are limited in their ability to deliver appropriate contextual responses to complex language inputs, presenting important safety concerns. One study of commercial chatbots such as Siri found that they often failed to recognize serious mental health concerns and provide appropriate responses such as referral to a support service87. For example, chatbots were found to not recognize when suicidal ideation was being discussed, and these devices also seemed to ignore domestic violence problems. Further, surveys of consumer attitudes reveal concerns about the privacy of chatbots as well as their potential to replace human care. Nevertheless, satisfaction ratings in the limited number of pilot and feasibility studies tend to be high, and rates of adverse events low88. Given the evidence and governance in place at this time, chatbots are best used only as a supportive tool in the context of a broader treatment plan.

Virtual reality

Virtual reality involves an immersion in an interactive, computer-simulated environment via a headset. The ability to create and control exposure to real-world environments presents important opportunities for mental health assessment and treatment89-91. Standard psychological assessments are limited by a lack of real-world validity and overreliance on subjective ratings92. Virtual reality allows precise, real-time data capture of responses to stimuli within controlled virtual environments, and hence provides critical insight into the way in which clinically relevant phenomena develop in real world89, 93.

Controlled exposure to anxiety-inducing stimuli within a virtual environment offers a safe, convenient and accessible medium to deliver exposure-based behavioral treatments. The benefit of virtual reality treatment lies in the repeated exposure to feared stimuli, enabling the individual to adapt to triggers and develop healthy responses in a safe and controlled therapeutic platform94. For example, randomized controlled trials have shown that learning to engage in virtual social interactions can reduce paranoia in people experiencing psychosis95, 96.

A recent meta-review of 11 meta-analyses, covering predominantly anxiety disorders and PTSD, found that effect sizes for virtual reality exposure treatments were overall moderate to large, and were typically maintained at follow-up97. A smaller number of trials have been conducted for other psychiatric disorders, with emerging evidence that virtual reality treatment may be effective for depression, schizophrenia97 and eating disorders98. However, in the studies that have compared virtual reality to traditional treatment, there was little evidence for superior efficacy. Further, the quality of evidence is overall low to moderate, due to the predominance of studies with small sample sizes, the relatively limited number of randomized controlled trials, and issues around publication bias.

Fewer studies have explored virtual reality treatments beyond exposure therapy, with the exception of skills training, which has also demonstrated positive results99. Pilot studies have also shown that virtual reality applications can guide people to learn therapeutic skills such as mindfulness100-102, relaxation103 and self-compassion104, 105. Using virtual reality as a vehicle to deliver experiences that help people develop skills to manage mental health difficulties may increase treatment engagement and efficacy.

Virtual worlds offer a compelling solution to increased demand for technology platforms that can deliver personal clinical care remotely106. Virtual worlds enable users to meet within virtual environments, represented as personalized avatars, and interact with other users in real time. Whilst few studies have been conducted in mental health, there have been promising early results especially in psychosis107. Delivering therapy via virtual worlds has the clear potential of offering highly accessible care within personally tailored, engaging therapeutic environments that provide a safe and comfortable medium for social interactions.

Whilst commercial growth in virtual reality is occurring rapidly, with an estimated growth of $54 billion over the next 7 years108, the technology remains unfamiliar and inaccessible to many users, presenting a barrier to implementation89. As costs decrease and virtual reality becomes more mainstream (partly due to the increased capacity to deliver it via smartphones), there is a need for further research and subsequent provision of evidence-based treatments and protocols, with adequate training for relevant workforces to enable their implementation.

EVIDENCE FOR DIGITAL PSYCHIATRY WITHIN SPECIFIC CONTEXTS

The research base on the efficacy and acceptability of the various types of DHTs is rapidly expanding. In this section, we explore recent and notable findings from empirical studies of the DHTs described above, with a focus on smartphones, across four specific contexts of mental health care: self-management of depression and anxiety; clinical management of major mood disorders; remote monitoring and interventions for psychosis, eating disorders and substance use disorders; and child and adolescent mental health.

Self-management of depression and anxiety

Depression and anxiety disorders are among the most common types of mental health conditions in the world109, and many more individuals experience subthreshold albeit disabling symptoms. Due to the high demand for self-management strategies for depression and anxiety, smartphone apps claiming to help with these problems are widely available on app marketplaces110, 111.

A recent large-scale meta-analysis of 66 randomized controlled trials explored the efficacy of smartphone apps for mental health problems including depression and anxiety across clinical and non-clinical populations112. For depressive symptoms, this meta-analysis found that smartphone apps outperformed control conditions, with larger effect sizes found when waitlist or educational resources (health tips, information) were used compared to attention/placebo controls (e.g., gaming apps)112. Smartphone apps also outperformed control conditions for generalized anxiety and social anxiety symptoms112. App interventions for anxiety did not differ significantly from face-to-face or other computer-based interventions in terms of outcomes, although only a small number of studies were used in these comparisons. For both depression and anxiety, studies which provided professional support alongside the smartphone app (e.g., through supportive phone calls or personalized therapist feedback) produced larger effect sizes compared to studies which did not.

A common criticism of smartphone apps for depression and anxiety is that they lack an underlying evidence-based framework111, 113. A review of 293 commercially-available apps for anx­iety and/or depression found that just over half (55.3%) included a reference to an evidence-based framework in their app store descriptions111. When a reference was included, a range of therapeutic frameworks were mentioned, including cognitive behavioral therapy techniques (30.0%), mindfulness (15.7%), positive psychology (9.2%), dialectical behavior therapy (3.4%), acceptance and commitment therapy (1.7%), and other techniques (6.8%). However, of the 162 apps that claimed to use a theoretical framework, only 6.2% had published evidence supporting their efficacy111.

The selective adoption of self-management apps for depression and anxiety has also been explored. A consumer data-driven review highlighted that the proliferation of depression and anxiety apps on the marketplace is in contrast with the relatively small number of apps which are regularly downloaded and used. The review reported that just three apps (Headspace, Youper and Wysa) accounted for about 90% of app downloads for depression. Similarly, three apps (Headspace, Calm and Youper) accounted for approximately 90% of downloads and daily active users of anxiety apps114. Moreover, most apps for depression (63%) and anxiety (56%) had no active users for the one-month period under review114. While commercial app companies do not publish engagement data, it is clear that downloads do not automatically translate into active use. For example, the popular (and free) COVID Coach app designed to address stress during the pandemic reported over 140,000 downloads, but only 1.56% of individuals who have downloaded the app recorded at least two weeks of use115.

There are several areas in which improvements can be made for apps dedicated to depression and anxiety. They include: ensuring substantive involvement of relevant health care professionals in the development of the apps110; embedding apps within local health care settings116; more robust testing of apps, specifically more well-designed randomized controlled trials to assess their efficacy114; understanding engagement techniques to ensure optimal use114; and using validated treatment techniques/interventions within the apps116. Further evaluation of anxiety and depression apps is clearly warranted114, including the need for additional research into the efficacy of app-delivered interventions compared with face-to-face “care as usual”116. Further research is also needed to understand the long-term engagement, as well as to examine any possible deleterious effects related to app usage111.

The evidence to date suggests that smartphone apps could provide an accessible, scalable and low-cost mechanism to deliver effective self-management interventions for symptoms of depression and anxiety, particularly to non-clinical populations and those who cannot access face-to-face services110, 116. However, the promise of apps to increase low-cost access to evidence-based treatment for depression and anxiety has not yet been fully realized. Efficacy trial data are needed to support many anxiety and depression apps available on the marketplace. Most of such apps have no clear evidence of efficacy47, 51, 117.

Clinical management of major mood disorders

Despite the growing evidence base described above on the use of DHTs for self-management of depression and anxiety, much of the existing research has been conducted in general population samples or people with mild-to-moderate symptoms. Thus, the current applicability of such research in the actual clinical management of severe mood disorders, such as bipolar disorder and major depressive disorder, remains unclear.

A recent systematic review and meta-analysis concerning the efficacy of digital interventions in bipolar disorder found positive effects on both depressive and manic symptoms118, but only four of the ten included studies were randomized controlled trials45, 119-121. As to unipolar depressive disorder, while an increasing number of randomized controlled trials of apps with psychotherapy-related content have been published26, several of them have shown no evidence that delivering psychological interventions via smartphone confers a significant advantage beyond control conditions122-124. However, randomized controlled trials which have used app-based interventions alongside human coaching to bolster their usage in community patients with depression have produced more robust evidence42, suggesting that human engagement in supporting app-based interventions is critical. New roles such as digital navigators to support app use in mental health care may provide one solution to offer human support without overburdening the clinician125-128 (see below).

The fact that bipolar disorder and major depressive disorder are characterized by episodic fluctuations in mood and behavior may suggest that smartphone-based interventions which provide fine-grained monitoring and real-time treatment (including JITAIs) may improve outcomes, either by fostering early identification of deterioration and/or by providing means for flexible and timely treatment interventions. Preliminary evidence in patients with major depression indicates that smartphones do indeed represent an available platform for real-time monitoring of patient-reported symptoms, such as changes in mood and activity, through ecological momentary assessments129-131, and that this can feasibly be supported through collection of sensor-based data such as the number of incoming and outgoing calls and text messages, or location information which may reflect changes in behavior and psychomotor activity. Similarly, in bipolar disorder, several recent studies have shown that smartphone-based active and passive data reflect digital markers of symptoms132-134, and classifications of affective states135, 136 and affective traits28, 137 have been published. Collectively, these studies suggest that such digital data could provide important real-time information reflecting the psychopathological status of patients with major mood disorders.

An important consideration is that patient-reported symptoms collected in clinical encounters have an inherent risk of recall bias45. On the other hand, establishing the extent to which patient-reported mood ratings collected via smartphones are consistent with clinical symptom ratings in patients with severe mood disorders is imperative for determining the role of such technologies in the clinical landscape. Studies examining this issue have largely indicated that smartphone-based mood assessments represent promising alternatives or adjuvants to traditional clinical measures, while acknowledging the methodological limitations in the existing evidence base, including that the overwhelming majority of trials and observational studies to date have enrolled small samples

留言 (0)

沒有登入
gif