Search

For many years, psychiatrists have tried to understand factors involved in response to medications or psychotherapies, in order to personalize their treatment choices. There is now a broad and growing interest in the idea that we can develop models to personalize treatment decisions using new statistical approaches from the field of machine learning and applying them to larger volumes of data. In this pursuit, there has been a paradigm shift away from experimental studies to confirm or refute specific hypotheses towards a focus on the overall explanatory power of a predictive model when tested on new, unseen datasets. In this paper, we review key studies using machine learning to predict treatment outcomes in psychiatry, ranging from medications and psychotherapies to digital interventions and neurobiological treatments. Next, we focus on some new sources of data that are being used for the development of predictive models based on machine learning, such as electronic health records, smartphone and social media data, and on the potential utility of data from genetics, electrophysiology, neuroimaging and cognitive testing. Finally, we discuss how far the field has come towards implementing prediction tools in real‐world clinical practice. Relatively few retrospective studies to‐date include appropriate external validation procedures, and there are even fewer prospective studies testing the clinical feasibility and effectiveness of predictive models. Applications of machine learning in psychiatry face some of the same ethical challenges posed by these techniques in other areas of medicine or computer science, which we discuss here. In short, machine learning is a nascent but important approach to improve the effectiveness of mental health care, and several prospective clinical studies suggest that it may be working already.

Background: E-mental health has become increasingly popular in interventions for individuals with eating disorders (EDs). It has the potential to offer low-threshold interventions and guide individuals to the needed care more promptly. Featback is such an Internet-based intervention and consists of psychoeducation and a fully automated monitoring and feedback system. Preliminary findings suggest Featback to be (cost-)effective in reducing ED symptomatology. Additionally, e-mail or chat support by a psychologist did not enhance theeffectiveness of Featback. Support by an expert patient (someone with a lived experience of an ED) might be more effective, since that person can effectively model healthy behavior and enhance self-efficacy in individuals struggling with an ED. The present study aims to replicate and build on earlier findings by further investigating the (cost-)effectiveness of Featback and the added value of expert-patient support.Methods: The study will be a randomized controlled trial with a two-by-two factorial design with repeatedmeasures. The four conditions will be (1) Featback, in which participants receive automated feedback on a short monitoring questionnaire weekly, (2) Featback with weekly e-mail or chat support from an expert patient, (3) weekly support from an expert patient, and (4) a waiting list. Participants who are 16 years or older and have at least mild self-reported ED symptoms receive a baseline measure. Subsequently, they are randomized to one of the four conditions for 8 weeks. Participants will be assessed again post-intervention and at 3, 6, 9, and 12 months follow-up. The primary outcome measure will be ED psychopathology. Secondary outcome measures are experienced social support, self-efficacy, symptoms of anxiety and depression, user satisfaction, intervention usage, and help-seeking attitudes and behaviors.Discussion: The current study is the first to investigate e-mental health in combination with expert-patient support for EDs and will add to the optimization of the delivery of Internet-based interventions and expert-patient support.Trial registration: Netherlands Trial Register, NTR7065. Registered on 7 June 2018.Keywords: Eating disorders, Internet, Internet-based, E-mental health, Intervention, Treatment, Prevention, Expert patient, Peer support, Cost-effectiveness

Of children with mental health problems who access specialist help, 50% show reliable improvement on self-report measures at case closure and 10% reliable deterioration. To contextualise these figures it is necessary to consider rates of improvement for those in the general population. This study examined rates of reliable improvement/deterioration for children in a school sample over time. N = 9074 children (mean age 12; 52% female; 79% white) from 118 secondary schools across England provided self-report mental health (SDQ), quality of life and demographic data (age, ethnicity and free school meals (FSM) at baseline and 1 year and self-report data on access to mental health support at 1 year). Multinomial logistic regressions and classification trees were used to analyse the data. Of 2270 (25%) scoring above threshold for mental health problems at outset, 27% reliably improved and 9% reliably deteriorated at 1-year follow up. Of 6804 (75%) scoring below threshold, 4% reliably improved and 12% reliably deteriorated. Greater emotional difficulties at outset were associated with greater rates of reliable improvement for both groups (above threshold group: OR = 1.89, p p p p p p

Childhood adversity is associated with a range of mental disorders, functional impairment and higher health care costs in adulthood. In this study we evaluated if childhood adversity was predictive of adverse clinical and functional outcomes and health care costs in a sample of patients at ultra-high risk (UHR) for developing a psychosis. Structural Equation Modeling was used to examine the effect of childhood adversity on depression, anxiety, transition to psychosis and overall functioning at 4-year follow-up. In addition, we evaluated economic costs of childhood adversity in terms of health care use and productivity loss. Data pertain to 105 UHR participants of the Dutch Early Detection and Intervention Evaluation (EDIE-NL). Physical abuse was associated with higher depression rates (b=0.381, p=0.012) and lower social functional outcome (b=−0.219, p=0.017) at 4-year follow-up. In addition, emotional neglect was negatively associated with social functioning (b=−0.313, p=0.018). We did not find evidence that childhood adversity was associated with transition to psychosis, but the experience of childhood adversity was associated with excess health care costs at follow-up. The data indicate long-term negative effects of childhood adversity on depression, social functioning and health care costs at follow-up in a sample of UHR patients.

In biomedical research, many different types of patient data can be collected, such as various types of omics data and medical imaging modalities. Applying multi-view learning to these different sources of information can increase the accuracy of medical classification models compared with single-view procedures. However, collecting biomedical data can be expensive and/or burdening for patients, so that it is important to reduce the amount of required data collection. It is therefore necessary to develop multi-view learning methods which can accurately identify those views that are most important for prediction. In recent years, several biomedical studies have used an approach known as multi-view stacking (MVS), where a model is trained on each view separately and the resulting predictions are combined through stacking. In these studies, MVS has been shown to increase classification accuracy. However, the MVS framework can also be used for selecting a subset of important views. To study the view selection potential of MVS, we develop a special case called stacked penalized logistic regression (StaPLR). Compared with existing view-selection methods, StaPLR can make use of faster optimization algorithms and is easily parallelized. We show that nonnegativity constraints on the parameters of the function which combines the views play an important role in preventing unimportant views from entering the model. We investigate the performance of StaPLR through simulations, and consider two real data examples. We compare the performance of StaPLR with an existing view selection method called the group lasso and observe that, in terms of view selection, StaPLR is often more conservative and has a consistently lower false positive rate., Comment: 26 pages, 9 figures. Accepted manuscript

IntroductionShort-term psychodynamic psychotherapy (STPP) is an empirically supported treatment that is often used to treat depression. However, it is largely unclear if certain subgroups of depressed patients can benefit specifically from this treatment method. We describe the protocol for a systematic review and meta-analysis of individual participant data (IPD) aimed at identifying predictors and moderators of STPP for depression efficacy.Method and analysisWe will conduct a systematic literature search in multiple bibliographic databases (PubMed, PsycINFO, Embase.com, Web of Science and Cochrane’s Central Register of Controlled Trials), ‘grey literature’ databases (GLIN and UMI ProQuest) and a prospective trial register (http://www.controlled-trials.com). We will include studies reporting (a) outcomes on standardised measures of (b) depressed (c) adult patients (d) receiving STPP. We will next invite the authors of these studies to share the participant-level data of their trials and combine these data to conduct IPD meta-analyses. The primary outcome for this study is post-treatment efficacy as assessed by a continuous depression measure. Potential predictors and moderators include all sociodemographic variables, clinical variables and psychological patient characteristics that are measured before the start of treatment and are assessed consistently across studies. One-stage IPD meta-analyses will be conducted using mixed-effects models.Ethics and disseminationInstitutional review board approval is not required for this study. We intend to submit reports of the outcomes of this study for publication to international peer-reviewed journals in the fields of psychiatry or clinical psychology. We also intend to present the outcomes at international scientific conferences aimed at psychotherapy researchers and clinicians. The findings of this study can have important clinical implications, as they can inform expectations of STPP efficacy for individual patients, and help to make an informed choice concerning the best treatment option for a given patient.PROSPERO registration numberCRD42017056029.

Meta-analyses comparing the accuracy of clinical versus actuarial prediction have shown actuarial methods to outperform clinical methods, on average. However, actuarial methods are still not widely used in clinical practice, and there has been a call for the development of actuarial prediction methods for clinical practice. We argue that rule-based methods may be more useful than the linear main effect models usually employed in prediction studies, from a data and decision analytic as well as a practical perspective. In addition, decision rules derived with rule-based methods can be represented as fast and frugal trees, which, unlike main effects models, can be used in a sequential fashion, reducing the number of cues that have to be evaluated before making a prediction. We illustrate the usability of rule-based methods by applying RuleFit, an algorithm for deriving decision rules for classification and regression problems, to a dataset on prediction of the course of depressive and anxiety disorders from Penninx et al. (2011). The RuleFit algorithm provided a model consisting of 2 simple decision rules, requiring evaluation of only 2 to 4 cues. Predictive accuracy of the 2-rule model was very similar to that of a logistic regression model incorporating 20 predictor variables, originally applied to the dataset. In addition, the 2-rule model required, on average, evaluation of only 3 cues. Therefore, the RuleFit algorithm appears to be a promising method for creating decision tools that are less time consuming and easier to apply in psychological practice, and with accuracy comparable to traditional actuarial methods. (PsycINFO Database Record (c) 2015 APA, all rights reserved)

Objective: To examine how various predictors and subgroups of respondents contribute to the prediction of health care and productivity costs in a cohort of employees. Methods: We selected 1548 employed people from a cohort study with and without depressive and anxiety symptoms or disorders. Prediction rules, using the RuleFit program, were applied to identify predictors and subgroups of respondents, and to predict estimations of subsequent 1-year health care and productivity costs. RESULTS:: Symptom severity and diagnosis of depression and anxiety were the most important predictors of health care costs. Depressive symptom severity was the most important predictor for productivity costs. Several demographic, social, and work predictors did not predict economic costs. Conclusions: Our data suggest that from a business perspective it can be beneficial to offer interventions aimed at prevention of depression and anxiety. Copyright © 2014 by American College of Occupational and Environmental Medicine.

Purpose: Factor mixture analysis (FMA) and item response mixture models in the general population have shown that the psychosis phenotype has four classes. This study attempted to replicate this finding in help-seeking people accessing mental health services for symptoms of non-psychotic mental disorders. Methods: All patients (18-35 years old) referred for non-psychotic mental health problems to the secondary mental healthcare service in The Hague between February 2008 to February 2010 (N = 3,694), were included. Patients completed the Prodromal Questionnaire (PQ). Hybrid latent class analysis was applied to explore the number, size and symptom profiles of the classes. Results: The FMA resulted in four classes. Class 1 (N = 1,039, 28.1 %) scored high on conceptual disorganization, inattention and mood disorder. Patients in Class 2 (N = 619, 16.8 %) endorsed almost all PQ-items, were more often screened as being psychotic or at high risk of developing psychosis, without care takers noticing. In Class 3 (N = 1,747, 47.3 %) perplexity, paranoia and negative symptoms were more prevalent. Patients were more often at high risk of developing psychosis. Class 4 (N = 286, 7.7 %) represented the 'normative' group with low probabilities for all items. Discussion: The results support the hypothesis that a representation in four classes of psychotic-like experiences can also be applied in a help-seeking population. © 2013 Springer-Verlag Berlin Heidelberg.

Minimizing the respondent burden and maximizing the classification accuracy of tests is essential for efficacious screening for common mental health disorders. In previous studies, curtailment of tests has been shown to reduce average test length considerably, without loss of accuracy. In the current study, we simulate Deterministic (DC) and Stochastic (SC) Curtailment for three self-report questionnaires for common mental health disorders, to study the potential gains in efficiency that can be obtained in screening for these disorders. The curtailment algorithms were applied in an existing dataset of item scores of 502 help-seeking participants. Results indicate that DC reduces test length by up to 37%, and SC reduces test length by up to 46%, with only very slight decreases in diagnostic accuracy. Compared to an item response theory based adaptive test with similar test length, SC provided better diagnostic accuracy. Consequently, curtailment may be useful in improving the efficiency of mental health self-report questionnaires. © 2013 Elsevier Ireland Ltd.