Development and Improvement of prediction/decision making tools in treating depression in primary care
lead supervisor: assoc prof kate saunders, department of Psychiatry
Co-supervisor: Prof Terry Lyons, Mathematical Institute
Commercial partner: P1vital Products, Wallingford
The focus of the project will be the use of advanced mathematical techniques for
time series analysis of existing data sets held by P1vital products. The objective will
be improved understanding of the diagnosis and treatment of mood disorder in
primary care.
As an example, the PReDicT study (Kingslake et al, 2017) was designed to
determine whether early response to antidepressants can be used to guide changes
in treatment that leads to better outcomes for depressed patients. It was
implemented in five European counties (France, German Spain, Netherland and UK)
and recruited over 900 participants whose symptom profiles, cognitive function,
capacity to work and health economic costs were followed for 12 months using a
stable digital platform developed for commercial use in clinical trials and health care
by P1vital products. The primary analysis showed that predicting treatment response
changed clinical behavior very significantly and that this led to improved outcomes
for anxiety but not depression at 8 weeks. Elements of this approach have now been
implemented in the real world practice of two large primary care centres where data
is continuing to accumulate.
The datasets are rich in the variety of measures collected and the 90%
completeness of the data at 8 weeks follow up. The planned analysis has raised
many questions about trial methodology and ordinary practice. Critically, the
relationships between symptoms, functional outcome and the value of treatment
remain broadly unanswered in this trial and more broadly across the depression
field, which has been for too long dominated by measurement of subjective
symptoms and the assumption that depressive symptoms are somehow of greater
clinical significance than anxiety symptoms.
The project will determine:
1) the most meaningful individual and composite outcomes for depression in primary
care, integrating symptomatic with functional measures of treatment response.
Validation will be sought by the association with health economic measures in
particularly expensive patient sub-groups rarely considered in clinical trials in
psychiatry.
2) the potential for measures of anxiety rather than depression to permit prediction of
treatment response in a machine learning analysis.
3) determine whether novel analyses of the time course of response can be used to
predict treatment outcomes. Very simple analyses of symptom variability have
already suggested that this is the case.
Other datasets from smaller scale clinical trials with placebo arms will also be
available for confirmatory analysis.