Mental health is a basic human right, impacting global general health and well-being. The estimated cost of mental health conditions is US$ 6 trillion by 2030 worldwide, combining lost economic productivity, direct costs of care, and prevention strategies. Investments in mental health improvement can lead to better public health and socioeconomic development [1]. The Mental Health Action Plan (2013–2030) from the World Health Organization (WHO) aims at improving mental health service delivery through prevention strategies, including the integration of mental health into routine health information systems [2]. Good mental health allows people to be productive, contribute to the community, and recover from daily stressors [3].
One of the most damaging and prevailing mental disorders is depression, affecting approximately 280 million people around the world [4]. The WHO projects that depression will be the main contributing factor to disease burden by 2030 [5]. Moreover, anxiety and bipolar disorder are leading causes of mental health disability worldwide [6], [7]. Bipolar disorder increases physical comorbidity, functional impairment, mortality, and suicide, reducing the life expectancy by 8–12 years [6], [8]. In this sense, untreated and undertreated mental illnesses are the main causes of suicide [9], which in turn is one of the main causes of death around the world [3]. The WHO estimated that 703,000 people committed suicide in 2019 in the world [10]. Depression, anxiety, and bipolar disorder pertain to the group of diseases related to anxiety, depression, mood, and trauma [11]. This group of diseases does not approach disorders related to brain injuries or brain development but those related to feelings and mood, which are affected by behaviors guided by dysfunctional thoughts according to Cognitive-Behavioral Therapy [12].
Thus, human behavior may reveal the manifestation of mental diseases [13]. Personal and wearable devices allow for collecting data regarding human behavior through passive sensors, which enhances the remote monitoring of individuals, presenting the potential for identifying real-time oscillations in psychological factors [14]. Collecting and analyzing individuals’ data through smartphones, wearable devices, and other digital platforms conceptualize the term “digital phenotyping” [15]. This term refers to a frequent and constant measurement of human phenotypes in situ based on data from smartphones and other personal digital devices [16]. Digital phenotypes can expand the ability to identify health conditions, including digital data in the health analysis in addition to traditional forms of disease expression. These digital data comprise people’s information collected at an individual level and in the environment where the individuals are [15]. This set of digital information related to individuals’ behaviors and lifestyles is also known as “digital biomarkers” [17].
The combination of multisource data to constitute the digital phenotypes might leverage the application of advanced data analysis to improve mental health diagnosis [7]. Machine learning (ML) is a strategic approach for analyzing these data in order to understand the relationship between data and individuals’ behaviors [13]. In this scenario, there is an opportunity for training ML models to learn from digital phenotyping data to identify mental health diseases. These ML models can contribute to more precise and fast diagnosis of mental health conditions based on massive data collection after patients’ consent, supporting mental health professionals, improving patients’ experience, and reducing costs [18].
Digital phenotyping approaches became popular because of the broad adoption of smartphones and other wearable devices such as smartwatches and fitness trackers [19]. The number of smartphone users has increased continuously in the last few years, tending to reach 6.4 billion users in 2029 [20]. These devices have embedded sensors that facilitate digital phenotyping applications, allowing continuous analysis of individuals’ behavior [13]. Therefore, this article provides a computer science view on data analytics for digital phenotyping in mental health through a systematic literature review. Our study approaches this research topic by reviewing the literature with a well-defined methodology, considering ten databases to investigate ten research questions about study design, such as volunteers’ profiles, explored data, available databases, data collection devices, and data analysis techniques.
According to our knowledge, this is the first study that systematically reviews the literature about data analytics for digital phenotyping in mental health in a comprehensible way, focusing on disorders regarding anxiety, depression, mood, and trauma, as well as related symptoms and consequences, such as emotions, stress, and suicide. The contributions of this literature review are: (i) provide a mapping of the application domains of this research topic, as well as specifications of the studies, such as techniques, explored data, and devices, among others; (ii) outline the employed data analytics techniques for analyzing digital phenotyping data; and (iii) identify challenges, opportunities, and future research directions towards more robust approaches for digital phenotyping in mental health.
The remainder of this article is organized as follows. Section 2 describes related studies that review the literature. Section 3 approaches the methodology used for performing this literature review. Section 4 shows the results from this study, addressing the answers to the research questions. Section 5 discusses the results and limitations of this study, presenting a list of lessons learned during the review process. Finally, Section 6 presents the conclusions from this review and directions for future research.
