Can data-driven Models be useful for understanding Mental Health?
How data-driven Models can shed light on Psychiatric Disorders that emerge during Development.
Different machine learning (ML) models are applied to high-dimensional data in order to improve psychiatric treatment by a) classifying disorders for a better or more automatic diagnosis and b) predicting treatment outcomes and disease progression longitudinally to enable a better treatment selection (Huys et al., 2016).
Based on the interaction between neural mechanisms and symptoms, a ML analyses used on data from neuroimaging can help more reliably differentiate patients from controls (Huys et al., 2016). However, one potential challenge for this binary distinction might be the gradually scaled difference in disorder severity between patients, which is often difficult to quantify and tends to blur the boundaries of a diagnosis.
But diagnostic classifications are also important to address the problem of co-morbidity, whereby an individual’s symptoms are classified under several psychiatric disorders (Borsboom, Cramer, Schmittmann, Epskamp, & Waldorp, 2011). Probabilistic techniques generally help estimate the certainty of these classifications (Sabuncu & Konukoglu, 2014). Multi-class approaches meanwhile can be used to distinguish between diagnostic groups (Wolfers, Buitelaar, Beckmann, Franke, & Marquand, 2015), although they often assess comorbidity incorrectly by treating different diagnoses as mutually exclusive (Huys et al., 2016). Moreover, these diagnostic classification models tend to be limited to the unambiguous cases on which they are trained and not always applicable to the ambiguity of clinical practice.
Based on these limitation, data-driven computational models have shifted from diagnosis to the prediction of treatment response. In case of depression, for example, patients usually require several trials before responding to an antidepressant (Rush et al., 2006). Data-driven models with a combination of quantitative electroencephalography markers, which are proven predictors of pharmacological response (Iosifescu, 2011; Olbrich & Arns, 2013), were more reliable in determining the treatment outcome than any individual predictor (Khodayari-Rostamabad, Reilly, Hasey, de Bruin, & MacCrimmon, 2013). Furthermore, the identification of patients that are likely to be unresponsive to antidepressant treatment was significantly improved when ML techniques such as the multivariate analysis of magnetic resonance imaging data were used (Korgaonkar et al., 2015).
Most relevant to the clinical practice of psychiatry is, along with the prediction of treatment responsiveness, also the selection of the best suited treatment option for a patient. For this purpose multiple regression models have been used to look for relevant factors and interactions between treatments, based on trial data in which patients were randomly assigned different treatment options (Huys et al., 2016). Interestingly, it turned out that demographic factors like marriage and employment as well as a record of previously unsuccessful antidepressant trials accurately predicted that cognitive-behavioural therapy would be a better treatment choice, whereas a comorbidity with personality disorders meant antidepressants would be more appropriate (DeRubeis et al., 2014).
As a consequence of the better treatment selection achieved with these models, a significant and measurable reduction on the Hamilton Rating Scale for Depression was attained, considerably beyond that following traditional treatment selections (National Collaborating Centre for Mental Health, 2010). However, a main point of criticism from psychiatrists has been the concern that complicated and highly theoretical computational models are neither very intuitive for clinicians to understand nor sufficient to replace the human component and multifaceted approach that professional psychiatrists with years of practical experience can provide their patients with (Huys et al., 2011). Nevertheless, it is worth overcoming these challenges and encouraging cross-disciplinary collaboration by reinforcing clearer communication and bridging the gap between the theoretical and practical considerations of treating these disorders.
