Term Definition
Precision Medicine
A form of practice using specific therapies which are selected (‘personalised’) for patients based on their individual characteristics or the characteristics of a group to which they belong [1]. The aim of personalised therapy is to maximise patient outcomes whilst reducing adverse effects. An example of precision psychiatry, could be taken as selecting a therapy for depression e.g. cognitive behavioural therapy (CBT) versus an SSRI based on the likelihood of success for that treatment given a patient’s characteristics (clinical/biological).
Biomarkers
“A defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes or responses to an exposure or intervention.” [2] Biomarkers refer to substances which, found in the body, indicate information about a disease. These substances are usually component parts or bi-products of the disease process itself, and have traditionally been biological substrates such as proteins e.g. C-reactive protein as a biomarker for inflammation. As part of the disease process, biomarkers have value as they indicate the presence or prognosis of a disease. In psychiatric disease, due to a lack of traditional biomarkers for disease prognostication, increasing attention is being paid to computational parameters which can capture a behavioural process related to a particular disease.
Transdiagnostic Psychiatry Transdiagnostic psychiatry aims to look across diagnoses to discover new dimensions of disease based on biological and behavioural mechanisms [3].
Nosology Related to the classification of disease
Factor Analysis Modelling observed variables as a weighted combination of a smaller number of latent variables (e.g. modelling scores from 9 questionnaires as 3 factors).
Reinforcement Learning A framework for adaptive decision-making in the context of rewards and punishment.
Computational Model In neuroscientific terms, a computational model is a mathematical description that can be used to characterize complex cognitive processes, such as learning or decision-making. Parameters (see below) of the model can be estimated and quantify a specific part of the learning process, e.g. the weighting of new information compared to old. These parameters are estimated based on an individual’s behavioural responses during a cognitive task.
Parameter
Models are composed of parameters which represent a specific part of the learning process. When models are fitted to data from a task, parameters can be used to describe an individual’s performance mechanistically. For example, two people with major depressive disorder may show the same negative emotional bias on a cognitive task, but that behaviour may be caused by two different mechanisms - captured as differences in model parameters.
High dimensionality Data can be described as highly dimensional when there are more measurable features or variables than there are independent samples. In these scenarios, machine learning algorithms perform poorly. Reducing the number of variables is important as it improves the performance of the algorithm. [4]
Machine Learning Machine learning involves applying algorithms to data in order to make predictions or classifications based on input data, which either does (supervised) or does not (unsupervised) have known labels. A machine learning algorithm will produce an estimate about pattern or structure in the data.
Dynamic Causal Modelling (DCM) A method commonly used for the quantification of effective connectivity (e.g. the influence that one neural population exerts over another), DCM allows comparison between models of interconnected networks of neuronal populations in order to explain data gained from dynamic imaging during cognitive tasks. [5]
Overfitting Overfitting is a process in which models become extremely sensitive to noise when they are fitted to a training data set. The model inaccurately treats noise as signal of interest, so that it can better predict outcomes for the data that it is trained on. Highly dimensional data sets can lead to overfitting which in turn leads to poor predictions in new data (poor generalisability). [6]
Number needed to treat (NNT)
The number needed to treat describes the number of patients needing to receive a particular intervention so that one additional patient has a positive outcome. For example in a computational context, applying an algorithm which can predict remission in depression, the number needed to treat describes the number of patients the algorithm has to be applied to for the algorithm predict remission in an additional patient.