Learning machines in Internet-delivered psychological treatment

Abstract

A learning machine, in the form of a gating network that governs a finite number of different machine learning methods, is described at the conceptual level with examples of concrete prediction subtasks. A historical data set with data from over 5000 patients in Internet-based psychological treatment will be used to equip healthcare staff with decision support for questions pertaining to ongoing and future cases in clinical care for depression, social anxiety, and panic disorder. The organizational knowledge graph is used to inform the weight adjustment of the gating network and for routing subtasks to the different methods employed locally for prediction. The result is an operational model for assisting therapists in their clinical work, about to be subjected to validation in a clinical trial.

Predicting treatment failure in regular care Internet-Delivered Cognitive Behavior Therapy for depression and anxiety using only weekly symptom measures.

“…More research is needed to understand what input, and which modeling, produce the most accurate and most helpful type of prediction tools for adaptive treatment strategies. Using machine-learning methods could increase accuracy and help maximize data utilization as so much data is routinely collected in ICBT that can be hard to use in traditional statistical models (Boman et al, 2019). Though this study is within the context of time-limited treatments, identifying unhelpful treatments as early as possible is still desirable even if treatment time is not limited, since the time of suffering for a patient can be reduced.…”

“…Future research could compare different models in terms of misclassification overlap and disparity, perhaps identifying subgroups of at-risk patients that are easy or difficult to detect. One potential way forward is ensemble methods used in machine learning (e.g., Boman et al, 2019), though that is beyond the scope of the current study.…”

“…Our and their findings indicate that clinicians know a lot more about whether the patient is actively engaged in treatment or not than we can capture with any of these rough proxies, at least so far. One possible future avenue for this is natural language processing or other text/language based machine learning methods, where the contents of patient and therapist messages and well as written work sheets could be mined for compliance/adherence/onboarding indicators (Boman et al, 2019;Nadkarni, Ohno-Machado, & Chapman, 2011). The findings by Kraepelien et al (2019), where messages and work sheets were rated by experts, suggest that such indicators can be gleaned from those sources and that they do predict outcomes.…”

“…One strategy to improve outcome predictions is by analyzing data in more complex ways; such as machine learning (Boman et al, 2019; Lenhard et al, 2018), growth mixture modeling (Lutz et al, 2017) or latent profile analysis (Saunders, Cape, Fearon, & Pilling, 2016). Now that data is often collected digitally and computational power increases, the idea of more complex analyses is achievable.…”

Toward Learning Machines at a Mother and Baby Unit

“…Its purpose is usually not to replace, but to augment, humans in its logical or physical vicinity. With repeated training, testing, and use, a correctly programmed learning machine will increase its usefulness over time and over task (Boman et al, 2019). The same data point can over time contribute to many rounds of perception and reasoning in the machine.…”