Figure 1.
(A) A simplified illustration of the different levels which may
underlie psychiatric diagnoses. Traditional diagnoses are signified by
blue and yellow branches, with a transdiagnostic approach denoted by the
green branch. (i) Underlying causes (aetiology), e.g.,
genetic expression or predisposition, contribute to (ii)
pathophysiological mechanisms, such as inflammation, aberrant synaptic
gain, perceptual differences, or can result in psychopathology in
response to the environment. For example, expression of a particular
gene may lead to differences in synaptic gain, affecting output at the
circuit level and ultimately influence behaviour through the interaction
with other social and psychological factors (environment) to produce
(iii) symptomatology. In this way, symptoms can be seen
as the product of single or multiple maladaptive processes which can
occur across multiple linked levels from biology through to the
environment. Notably, many mechanisms can give rise to
(v) the same symptom, as can the same mechanism give
rise to many symptoms. Traditionally, symptoms are then categorized into
a cluster known as a diagnosis (iv) , e.g., A orB . However, a transdiagnostic factor (vi) may
recapitulate common symptoms or behaviours across disorders, and is
related to a specific mechanism (‘Y’ denoted in green), which can
be captured in cognitive experiments and/or neuroimaging (see Fig 1B).
(B) Computational psychiatry techniques implementing a
precision approach. (i) Studies can include individuals
with a particular diagnosis or across related diagnoses.
(ii) Studies may involve factor analysis of
questionnaires that aim to encompass a broad range of transdiagnostic
behaviours and symptoms. (iii) The battery may involve
cognitive tasks and neuroimaging from which model parameters can be
estimated, thereby describing the behaviour/neurobiology of an
individual. (iv) Model parameters can be used to find
new subgroups within diagnoses, or alternatively (v)
they can be correlated against factor scores to determine if the factors
correspond to an underlying mechanism for disease across diagnoses.
(vi) Using these data, subjects can be grouped either
based on subgroups within existing disease categories (purple pathway)
or across diagnoses (orange pathway). (vii) Longitudinal
data from clinical trials can be used in conjunction with machine
learning techniques to determine if the parameters produced from the
models, or the factors associated with them, have any predictive value
in terms of treatment response or prognosis. Looking for cognitive
biomarkers within diagnoses is compatible with defining new dimensions
of disease as cognitive biomarkers in existing disease subgroups may be
common to multiple diagnoses associated with a single dimension.
(viii) Data gained from these trials can be used in
clinical practice in the form of a decision aid, where patients can
undergo clinical investigation in the form of a questionnaire battery or
undergo neuroimaging while performing a cognitive task to allow an
informed decision about which treatment may be better. This may improve
patient outcomes as it facilitates a quicker response to treatment and
circumvents the trialling of multiple different therapies.