The use of machine learning techniques in trauma-related disorders: a systematic review

Abstract

Establishing the diagnosis of trauma-related disorders such as Acute Stress Disorder (ASD) and Posttraumatic Stress Disorder (PTSD) have always been a challenge in clinical practice and in academic research, due to clinical and biological heterogeneity. Machine learning (ML) techniques can be applied to improve classification of disorders, to predict outcomes or to determine person-specific treatment selection. We aim to review the existing literature on the use of machine learning techniques in the assessment of subjects with ASD or PTSD. We systematically searched PubMed, Embase and Web of Science for articles published in any language up to May 2019. We found 806 abstracts and included 49 studies in our review. Most of the included studies used multiple levels of biological data to predict risk factors or to identify early symptoms related to PTSD. Other studies used ML classification techniques to distinguish individuals with ASD or PTSD from other psychiatric disorder or from trauma-exposed and healthy controls. We also found studies that attempted to define outcome profiles using clustering techniques and studies that assessed the relationship among symptoms using network analysis. Finally, we proposed a quality assessment in this review, evaluating methodological and technical features on machine learning studies. We concluded that etiologic and clinical heterogeneity of ASD/PTSD patients is suitable to machine learning techniques and a major challenge for the future is to use it in clinical practice for the benefit of patients in an individual level.

Introduction

Trauma-related disorders such as Posttraumatic stress disorder (PTSD) and acute stress disorder (ASD) are considered to be debilitating conditions, developed from exposure to traumatic events including war, mass violence, natural disasters, and accidents. The DSM-5 (American Psychiatric Association, 2013) lists 20 diagnostic criteria for PTSD divided into four clusters of symptoms: re-experience of the traumatic event; avoidance; persistent negative thoughts or feelings; trauma-related arousal and reactivity. The WHO World Mental Health Survey conducted across 24 countries found a lifetime prevalence of any traumatic event of 70.4% (Benjet et al., 2016), suggesting that constitutional and sociocultural factors are also involved in the development of the disorder, besides the magnitude of trauma (Yehuda, 2004). The prevalence of PTSD in a lifetime is 11% for women and 5.5% for men (Kessler et al., 1995). It is postulated that a dose–response relationship exists between exposure to traumatic events and the subsequent development of PTSD, indicating that prior trauma and/or multiple traumatic event exposures increase the risk of the disorder (Ozer et al., 2008; Kilpatrick et al., 2013).

Establishing the diagnosis of PTSD and ASD has always been a challenge in clinical practice, as well as in academic research. As indicated by its numerous risk factors, the etiologies of trauma disorders are multicausal and complex. In addition, the development of diagnostic criteria for classification systems (such as DSM-5) has been elaborated from research with chronic populations and in tertiary care settings; such phenotypic expressions may not reflect the instability and nonspecific nature of the phenomenology of the disorder in its development (McGorry et al., 2006). Evidence-based, trauma-focused therapies with the most support are cognitive- and exposure-based approaches, with prolonged exposure and cognitive processing therapy being the most investigated (Charney et al., 2018). Notwithstanding, establishing first-line psychotherapies may be difficult because of—among other aspects—the burden to patients and patient profiles (Nash and Watson, 2012). Some statistically significant results provided by evidence-based medicine may not represent a real benefit for an individual patient; subjects in clinical trials do not always reflect the multimorbidity profile of “real-life” patients (Greenhalgh et al., 2014). This may be particularly true in the field of PTSD, where clinical heterogeneity can be a very important factor, not always taken into account in research.

Machine learning, a field of computer science and a part of artificial intelligence, refers to the science and engineering by which machines (i.e., computer systems) can analyze and acquire information from data (Liu and Salinas, 2017). Machine learning can help develop sophisticated data models using advanced mathematical techniques and handling complex data sets with heterogeneous distribution. The ‘learning’ method is usually made by a supervised or an unsupervised approach (Bishop, 2006). In supervised learning, the user feeds the machine with input data and expected outcome: the machine learns a mapping from the input to the outcome target, through classification (where the output variable is a category, such as ‘disease’ or ‘no disease’) or regression (where the output variable is a numeric variable) methods. Common examples of supervised learning algorithms are Logistic Regression, Support Vector Machines and Neural Networks. Supervised learning is often used to estimate prediction and risk: the Framingham Risk Score for coronary heart disease may be one of the most famous uses of supervised learning in medicine (Deo, 2015; Kannel et al., 1975). Unsupervised learning does not depend on previous associations and output variables: the goal is to model the underlying data structure to learn more about the data. It can be performed by discovering groups of similar cases (clustering) or determining the distribution of available data (density estimation). Network analysis allows visualization of the connectivity among symptoms and clusters of symptoms providing knowledge about the strength and quantity of relationships (Sullivan et al., 2018), taking into account regression and clustering techniques. A revision of the relevant principles of machine learning and its limitations can be found elsewhere (Schultebraucks and Galatzer-Levy, 2019; Librenza-Garcia et al., 2017; Deo, 2015).

Machine learning techniques can be applied to improve classification of disorders, to predict risk factors and treatment outcomes, and to improve person-specific treatment selection (Hahn et al., 2017). Since PTSD and ASD are disorders that present clinical and biological heterogeneity, which may constitute a barrier to understanding the causative mechanisms and to developing optimal treatments and diagnostic tools, machine learning is a suitable approach to better achieve this understanding. The present study aims to systematically review data in which PTSD and ASD were assessed through machine learning techniques regarding classification, prognostic, and treatment selection studies. Furthermore, we proposed a method of quality measurement of these studies.