Why Deep Learning Technology Strengthens Our Abilities to Treat Depression
Aifred health uses cutting-edge artificial intelligence to overcome challenges in the treatment of common mental health conditions.
Major depression is a serious mental illness that globally affects between 11% and 18% of people over the course of their lives. At any one time, 300 million people around the world struggle with depression, making this common condition a significant source of suffering for patients and their families, and results in tragedies such as suicide.
Current clinical practice typically follows an educated “guess and check” approach when treating depression: following a clinical assessment, a physician is then challenged in determining what combination of medications, neurostimulation techniques, lifestyle and behavioral modifications, and psychotherapies will work best for a patient. Given the many available treatment strategies, the probability of selecting the ideal therapeutic option on the first try is low. Many patients must try a number of treatments — which can take months or years of assessments — before finding the treatment combination that works best for them. Clinicians want to streamline this process so they can provide the best care, so we seized this opportunity to innovate with aifred health.
Today, best clinical practice consists of referring to consensus-based treatment guidelines that are updated too slowly to take advantage of cutting-edge research; these often lack guidance for personalizing treatments. While advancements in psychiatric research uncover better understanding of how best to treat depression, clinicians are hard-pressed to incorporate this new knowledge into daily practice. Data from clinical trials and anonymized patient records hold important insights, but a clinician cannot parse all of this information in order to make a decision about treatment for the patient in front of her.
How can we use vast new banks of knowledge to determine which combination of treatment strategies will best serve patients on a case-by-case basis?
Advancements in artificial intelligence (AI) provide solutions: our project aims to be the first step towards personalizing mental health services by using AI, specifically deep learning, to produce a treatment decision aid for depression.
Deep learning and big data open new possibilities in mental health
Introducing AI-innovations within healthcare is a hot topic in the technology sector. The recent issue of The Economist magazine is exemplary: a collection of articles explain how tech giants of the likes of Google, IBM, Microsoft and Apple are all vying to develop AI-powered tools that enable clinicians and patients to benefit from medical data. Deep learning is one domain in AI that holds great potential for innovation.
In deep learning, artificial neural networks are trained to solve problems by recognizing patterns in large datasets. Because of this ability to learn the context and relationships between variables in data, deep learning can provide insights about best clinical practices when it is trained on high-quality patient-level data. Medical innovations made possible by deep learning have proven their utility in a growing number of applications in medicine — in cancer imaging, for example.
The time is ripe to expand beyond diagnosis and apply this technology in the realm of treatment selection. Deep learning can leverage many clinical predictive features to find “hidden” patterns in medical data from clinical trials and electronic hospital records. Once identified, clinicians can use predictions based on these patterns to help select treatments. Importantly, it is possible to train deep learning algorithms with complex and incomplete datasets, which are common in mental health research and practice.
Mental health practitioners will be able to use aifred health to match a new patient’s profile — consisting of biomarkers, clinical factors, and sociodemographic information — with treatment regimens that our deep learning tool predicts to have the highest probability for success (possible treatment categories include psychotherapy, medication, behavioral and lifestyle interventions, and neuromodulation). Our innovation will be one of the first to personalize treatment for depression across a range of different therapies. In addition, our system will also predict personalized side effect profiles, which, along with the treatment recommendations, should enrich shared decision-making between the patient and the clinician.
Developing our innovation aims at using existing data resources and knowledge more effectively, rather than the development of new treatments. Given the recent abandonment of research into neuropsychiatric disorders by several large pharmaceutical firms, using existing treatment resources to their fullest potential is a necessary strategy to improve outcomes in mental health. What is exciting is that a deep learning approach can identify patient subtypes that do not respond well to any existing therapies, knowledge which could encourage new research into therapeutics by providing a more targeted research question and a better defined population of study.
Aifred health: An opportunity for significant impact in mental health services
Employing AI as a guide in the treatment of depression is a cutting-edge innovation in psychiatry and marks a significant deviation from the status quo. Large studies, such as the STAR*D study, indicate that fewer than one in three depressed patient will receive an effective treatment following their first clinical assessment, and roughly 30% of patients will not respond even after four different courses of treatment. If we are able to improve from a 1⁄3 initial treatment success rate to even a modest 2⁄3 success rate — or from a 70% overall success rate to an 80 or 90% rate after several treatments — our innovation will significantly reduce population-level morbidity from depression and mortality from suicide.
Our technology keeps pace with new knowledge and provides tangible means to introduce research findings into clinical practice: we can input new research findings into our model as patient-level data become available. Our algorithm is thus in a continual state of learning and gets more accurate with use since it will train on data collected from patients and thus will remain up-to-date. Aifred provides a cost-effective way to personalize treatments to the needs of patients, improving the rate of recovery for patients with depression. Patients and clinicians tell us that being able to choose more effective treatments more quickly will provide significant benefits to patients who often must endure months of treatment trials before finding relief. Once deployed in the healthcare system, we believe our technology will dramatically improve patient care and increase efficient access to treatments while reducing healthcare costs by reducing the time it takes for patients to improve. In addition, when employed in a care network our software will track how often certain therapies (such as psychotherapy) were recommended but not chosen because of poor availability; this in turn could help administrators and government re-apportion funds to pay for effective treatments that are not consistently available. Our innovation thus contributes towards the sustainability of public healthcare.
