Description of the work carried out for the four reported research questions:
1) and 2) Genetic markers of non-response (failure of one treatment) and resistance (failure of two or more treatments) to antidepressant medications were studied in a sample of about 1200 patients with major depressive disorder (MDD). Common genetic variants throughout the genome and rare variants in coding regions were obtained using genome-wide genotyping and whole exome sequencing. The distribution of variants predicted to have a functional impact on protein function/levels was compared between response groups, as well as more complex scores reflecting the burden of variants in specific regions, but no gene or group of functionally related genes (pathway) clearly emerged as associated with non-response or resistance. Therefore, the predictive performance of a combination of genes/pathways was tested using machine learning (Figure 1). This approach was able to correctly classify 61-66% of patients with treatment resistance and it showed improvement when considering only patients with genetic profiles in extremes percentiles.
3) Five clinical risk factors were identified as particularly relevant in the risk of antidepressant resistance (duration of the depressive episode, number of previous depressive episodes, suicidal ideation, pessimism and symptoms reflecting lack of interest and poor involvement in activities). The addition of these variables to the genetic predictors increased the proportion of patients correctly identified as treatment-resistant up to 75%.
4) The developed predictive models could be used to guide treatment prescription if effective and well tolerated therapeutic options are available as alternative to standard care (pharmacotherapy). The most suitable alternative to standard care was identified as combined antidepressant pharmacotherapy and psychotherapy such as cognitive-behavioural treatment, since it has good evidence of higher efficacy compared to pharmacotherapy alone and it is not associated with increased side effects, but it has limited availability because of higher costs compared to pharmacotherapy. The cost-effectiveness of using the developed genetic/clinical predictive model of treatment resistance in guiding the prescription of combined pharmacotherapy and psychotherapy vs. pharmacotherapy was tested using the obtained sensitivity/specificity and costs-utilities from publicly available databases; the comparator groups were standard care (pharmacotherapy to all patients) and prescription of combined pharmacotherapy and psychotherapy vs. pharmacotherapy guided by the identified clinical predictors only. The use of clinical predictors only was cost-effective compared to using clinical predictors combined with genetic variables: one quality-adjusted life year (QALY, a year lived in perfect health) improvement costed 2341 GBP in the group guided by clinical predictors and 3937 GBP in the group guided by clinical/genetic predictors (Figure 2). One major contributor to this result was the cost of whole exome sequencing; if the cost of genotyping/sequencing is 100 GBP or less the genetic/clinical model would become cost-effective compared to the clinical one.
The results were disseminated during eight conferences and through two peer-reviewed publications (doi: 10.1038/s41398-020-0738-5 and 10.1097/YIC.0000000000000305) one online preprint (10.1101/2020.03.31.20048538) social networks and the Pharmacogenomics Research Network (PGRN) (
https://www.pgrn.org/featured-investigators/chiara-fabbri-phd(s’ouvre dans une nouvelle fenêtre)).
Replication and extension of the results of this project is planned as part of already funded studies using independent samples of patients with depression (UK Biobank (
https://www.ukbiobank.ac.uk(s’ouvre dans une nouvelle fenêtre)) and a new sample in phase of recruitment by the European Group for the Study of Resistant Depression (GSRD)).
