Abstract
Response rates to available treatments for psychological and chronic pain disorders are poor, and there is a substantial burden of suffering and disability for patients, who often cycle through several rounds of ineffective treatment. As individuals presenting to the clinic with symptoms of these disorders are likely to be heterogeneous, there is considerable interest in the possibility that different constellations of signs could be used to identify subgroups of patients that might preferentially benefit from particular kinds of treatment. To this end, there has been a recent focus on the application of machine learning methods to attempt to identify sets of predictor variables (demographic, genetic, etc.) that could be used to target individuals towards treatments that are more likely to work for them in the first instance. Importantly, the training of such models generally relies on datasets where groups of individual predictor variables are labelled with a binary outcome category - usually 'responder' or 'non-responder' (to a particular treatment). However, as previously highlighted in other areas of medicine, there is a basic statistical problem in classifying individuals as 'responding' to a particular treatment on the basis of data from conventional randomized controlled trials. Specifically, insufficient information on the partition of variance components in individual symptom changes mean that it is inappropriate to consider data from the active treatment arm alone in this way. This may be particularly problematic in the case of psychiatric and chronic pain symptom data, where both within-subject variability and measurement error are likely to be high. Here, we outline some possible solutions to this problem in terms of dataset design and machine learning methodology, and conclude that it is important to carefully consider the kind of inferences that particular training data are able to afford, especially in arenas where the potential clinical benefit is so large.
Highlights
The proportion of patients who respond to available treatments for psychological and chronic pain disorders is often low
The success of causal inference modelling approaches to response prediction may depend upon availability of different kinds of data to that derived from traditional randomised controlled trial (RCT) – involving semi-continuous measurement of the relevant clinical outcome, and gathered from more representative sources than some previous RCT datasets
The issues discussed above underline the importance of focusing on where data comes from when considering strategies for personalised medicine
Summary
Big data, machine learning, psychiatry, chronic pain, individual differences, response heterogeneity, clinical trial design. We have made it more explicit that the underlying RCTs (designed to determine treatment effectiveness at the group level) are not single arm, machine learning algorithms concerned with predicting individual differences in treatment response are usually only trained on the active treatment arm data (i.e. without reference to symptom changes in the control arm data). We have updated the manuscript to touch upon how prediction of differential treatment responses in future patients (i.e., outof-sample data) can be framed as a causal inference problem (with respect to current datasets) - with brief discussion of some pertinent issues, including representativeness of training data samples, and the requirement of explicit assumptions about the causal consequences of any confounding variables. Some changes to grammar have been made throughout the manuscript in order to increase clarity
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have