Longitudinal evaluation of an MRI-based predictive algorithm for early assessment of treatment response in depression (LEOPARD study)

Major depressive disorder (MDD) is a highly prevalent condition worldwide. It
is associated with increased morbidity and mortality. Symptoms include
depressed mood lasting more than 2 weeks, emotional distress, functional
impairment, health problems, and suicide. MDD is the leading cause of
disability10 resulting in a high socioeconomic burden.

Although MDD typically has a relatively good response to antidepressants (ADs),
only about one third of the patients show significant symptom relief in
response to the initial treatment4 and 50% have not found an efficacious AD
after 1 year. Clinical guidelines recommend 4-8 weeks of treatment before
considering an alternate medication in nonresponding patients14. The guidelines
recommend that, if the treatment is ineffective after 1-2 months, a new
medi-cation or treatment should be started, after reconsidering the diagnosis.
In summary, ineffec-tive pharmacotherapy may cause delay in adequate treatment,
persistence of depressive symp-toms and functional impairment, which could be
shortened by better prediction of therapeutic response. In general guidelines
recommend to use a Selective Serotonin Reuptake Inhibitor (SSRI) as a first
step treatment while for a second step treatment a second SSRI or a Serotonin
Norepinephrine Reuptake Inhibitor (SNRI) is often used.

This lengthy process can negatively impact patients* confidence in
pharmacotherapy and re-duces treatment adherence. Meanwhile, patients suffer
from MDD and might experience seri-ous adverse effects of different drugs
without effectively resolving symptoms. adverse effects include weight gain and
insomnia. Thus, a solution is urgently needed that allows faster de-termination
of AD non-response in MDD.

There is growing interest in the development of precision medicine algorithms
with the aim of tailoring treatment strategies to individual patients according
to unique biological signatures. This biomarker-based approach to precision
prescribing has the potential to improve therapeu-tic response, minimize
adverse reactions, and by stopping ineffective drugs as early as possi-ble
reduce time to symptomatic relief. Personalized medicine is already
revolutionizing can-cer treatment, in which treatments are tailored to a
tumor*s genomic profile.

The application of personalized medicine to psychiatry, however, is more
challenging. In con-trast to cancer, there is no biological or histological
test for definitive psychiatric diagnoses, because of the inaccessibility of
the human brain and the complexity of the link between biol-ogy and psychiatric
symptoms. For example, the diagnosis of MDD is based on a combination of
symptoms alone, by standard nosology, as reflected in diagnostic manuals, such
as the DSM or the International Classification of Diseases, which does not
incorporate any biologi-cal dimension, nor can guide any treatment selection.

The National Institute of Mental Health (NIMH)*s Research Domain Criteria
emphasize bi-omarker discovery as a clinical research priority by articulating
an approach to the integration of biological and clinical data. The emerging
field of psychoradiology, pioneered by Gong and colleagues17 aims to provide
biomarkers based on objective tests in support of the diag-nostic
classifications, as in other parts of medicine. Biomarkers derived from
neuroimaging data are potentially important contributors to the goal of guiding
treatment selection using clinical and biotyping data. Because of its
non-invasive nature, it has great potential to revolu-tionize clinical
psychiatry. Information on brain structure and function may be used to predict
non-response versus response to various treatments. Properties predictive of
treatment re-sponse presented in literature include pre-treatment brain
volumes, post-treatment chang-es in regional morphology, gray and white matter
patterns at baseline, presence of increase of subcortical white matter
hyperdensities (WMH), lowered DTI measures of fractional anisotropy (FA) and
mean diffusivity (MD)9, and baseline and regional changes in resting-state
functional connectivity (RSFC). Reviews on this topic are available by Fonseka
and colleagues and recently by us.

In DEPREDICT, we will develop a radiomics-based algorithm that allows early
(within 2 weeks after first administration) prediction and / or assessment of
the later (non-)response to AD in patients with MDD. Radiomics is the
high-throughput extraction of quantitative fea-tures that result in the
conversion of images into mineable data and the subsequent analysis of these
data for decision support. This is in contrast to the traditional practice of
treating medi-cal images as pictures intended solely for visual interpretation.
Radiomic data contain first-, second-, and higher-order statistics. These data
are combined with other patient data and are mined with sophisticated
bioinformatics tools to develop models that may potentially improve diagnostic,
prognostic, and predictive accuracy.

We will first develop the radiomics algorithm based on existing MRI datasets of
the brains of patients with MDD. Whereas existing literature predominantly
compares pre-treatment data between responders and non-responders
retrospectively, using a single outcome measure, DEPREDICT aims to employ
advanced radiomics analysis of MRI measurements of the brain as predictive
biomarker in a multivariate predictive solution. There are good indications
that this approach may offer improvement.

The first step for DEPREDICT towards clinical implementation is to demonstrate
reproducibility. Therefore, this study -LEOPARD- aims to determine the
predictive value of this algorithm by means of a longitudinal study. It will
test the algorithm's ability to predict week 8 response based on baseline and
week 8 clinical information, MRI scans, and activity and sleep information
obtained with a wrist-worn accelerometer. The results of LEOPARD will be
essential for the strategy of the development of the DEPREDICT algorithm
towards clinical implementation. A positive result of this study (being: strong
predictive properties of the algorithm) will enable the next step, a randomized
blind study. If the DEPREDICT algorithm turns out to be effective in this, it
will be a major step towards becoming a valuable objective tool to support
clinical decision making, with great health and economic impact.

To replicate the DEPREDICT algorithm (formed from external cohorts/datasets) as
a valid tool for prediction of AD treatment non-response, we aim to recruit 80
patients who are treat-ed with an SSRI or SNRI and are eligible and willing to
have an MRI-scan session before initiation of treatment and 2 weeks thereafter.

An 8-week non-randomized, longitudinal brain imaging study for assessment of
open label antidepressant (SSRI/SNRI) treatment response in the brain of adult
subjects suffering from MDD and in need of pharmacological treatment with AD
but free of psychotropic medications for at least 5 half-lives at baseline. The
effect of DEPREDICT algorithm is tested by predicting week 8 AD treatment
response from baseline and week 2 clinical data, MRI scans, and information
regarding activity and sleep collected using a wrist-worn accelerometer

Risks:
There are no risks associated with assessment of other MRI sequences at 3T,
clinical interviews, nor with the wrist-worn accelerometer.

Benefit:
Our hypothesis is that our DEPREDICT can sufficiently predict treatment outcome
at week 2. This would reduce the time to selection of an AD by 75%. If LEOPARD
confirms this, this algo-rithm could hold for major health and economic
benefits. It will not only reduce the lengthy *tri-al-and error* process of
finding the right drug, but also increase treatment adherence, and re-duce time
patients suffer from side effects of ineffective medication. In doing so, this
study will validate a novel tool in personalizing AD treatment, furthering its
way towards clinical deployment.

Although the patients participating in this study will not have direct benefits
themselves in participating, their fellow and future MDD patient-peers could
benefit, if our study is successful. The overall nature and extent of the added
risk associated with participation in the current study is to be classified as
negligible and the burden can be considered minimal.

Conclusion:
The overall nature and extent of the burden associated with participation in
the current study are to be classified as minimal and the risk negligible.
There is a group benefit associated with study participation.