Computational Psychiatry

Psychiatric disorders impose a heavy burden on patients and their families. These disorders also cost the Canadian economy $51 billion each year (Lim 2008, Chronic Dis Can 28(3): 92-98). Economic costs of psychiatric illnesses are similarly gigantic in other countries.

Treating psychiatric illnesses is challenging in several ways. Psychiatric diagnosis is often not straight-forward, and this can make treatment less effective. For example, the early stages of major depressive disorder and bipolar disorder are often indistinguishable. Differentiating these two diseases early is highly desirable as the treatments are very different.

Psychiatric prognosis is also difficult. While effective psychiatric drug treatments are available, individual patients react differently to a given drug, making it difficult to predict the best treatment for an individual. Psychiatrists may try various drugs before finding an effective combination. This trial-and-error process is especially problematic as many drugs have unpleasant side-effects and most drugs take several weeks to become therapeutic. Psychiatric patients often become disillusioned and discontinue drug treatment that would have helped them.

We are working on technologies for improving diagnosis for psychiatric patients and for providing more accurate prognostic predictions of how patients will respond to treatment. The goal is to provide tools to assist medical doctors in making more accurate diagnoses and more effective treatment decisions. This should improve patients' response to treatment, improve adherence to pharmacotherapy, and improve patient outcomes.

We apply advanced computational methods from machine learning, data science, and data mining to the problems of psychiatric diagnosis and prognosis. These approaches can utilize a variety of input data sources, ranging from written questionnaires and structured interviews to advanced neuroimaging and metabolomics. This computational approach is particularly well-suited to discovering subtle but clinically useful patterns in large, diverse datasets - the sort of pattern that a human researcher looking at the data with traditional analytical tools would not find. The goal of the computational approach is to discover patterns in the data - essentially complex biomarkers - that distinguish among patient types (diagnosis) or predict an individual patient's response to treatment (prognosis). Such discoveries would allow for improved personalized psychiatric treatment that is customized to the individual.

We are a multidisciplinary group based in Alberta, Canada, including clinicians and scientists who work in the fields of mental health, neuroscience, population health, and machine learning. We have an active collaboration with the IBM Centers for Advanced Studies (CAS), CAS Alberta.