Artificial Intelligence (AI) and Mental Health Care
AI tools will facilitate individualized treatment planning and improve outcomes.
Posted Oct 10, 2019
This post is offered as a concise overview of important advances in artificial intelligence that will soon impact the way mental health care is practiced in day to day clinical settings. The result will be more individualized treatment incorporating both conventional and evidence-based complementary and alternative medicine (CAM) modalities, more effective and more cost-effective treatments of many common mental health problems, and improved outcomes.
The promise of AI for improving mental health care
- Rapidly extract useful information from very large data sets on patient medical information that were previously difficult and time-consuming to access.
- Rapidly access reliable resources containing up to date information on a broad range of Western medical and CAM modalities.
- Use advanced AI tools for guidance in identifying optimal treatment protocols addressing many common mental health problems and predicting treatment outcomes.
To have practical clinical utility in medicine and mental health care, an AI system must encompass machine-learning software capable of processing very large volumes of structured data, and natural language processing (NLP) software capable of mining unstructured data such as narrative text in electronic health records and medical imaging data. To assist health-care providers with clinical decision-making, the AI system must be ‘trained’ to a requisite level of expertise within a particular domain of medical knowledge. Following completion of training, it is vital to keep the supply of pertinent medical data current. Widespread data sharing between payers and providers is critical in order for this to succeed (Kayyali 2013). IBM’s Watson AI system is an example of a system that includes both machine learning and NLP capabilities and is already being widely used in the field of cancer research (Lohr 2016).
AI programs incorporating neural network algorithms are being used to characterize complex non-linear relationships between symptoms being treated, disparate treatment modalities, and clinical outcome measures that are difficult to ascertain using traditional software tools. Deep learning is a recent offshoot of neural network-based algorithms capable of investigating complex nonlinear relationships in data that are difficult or impossible to characterize using less advanced software. In recent years deep learning algorithms have been used to identify complex nonlinear relationships in data from functional brain imaging research that were previously impossible to analyze (Vieiraa et al 2017).
‘Big data’ makes possible analysis of very large volumes of complex medical data
‘Big data’ refers to very large, complex data sets for which existing data processing methods cannot provide useful analysis. Advances in big data analysis methods will soon permit the automation of literature research yielding high-quality information on a wide range of complementary and alternative medicine (CAM) modalities. Obtaining big data that is useful for decision making in medicine and mental health care is a nontrivial problem because payers and providers have different kinds of confidential data on the same patient often coded in different ways. In big data there is often a trade-off between accuracy at the micro level and insights about treatment benefits at the macro level. This problem is being addressed by combining big data sets on multiple domains such as clinical research data, quality improvement data, electronic health records, and administrative claims data and using multivariate analysis to identify patient subgroups that may more likely respond to different treatments in different settings (Matthews 2014).
Super learning: An AI system for predicting treatment outcomes in mental health care
An AI program called Super learning is being developed to assist clinicians in predicting outcomes to treatments of substance use disorders (Acion et al 2017). The program compares data generated from a variety of prediction algorithms such as deep learning neural networks and logistic regression. A database of 100,000 patients being treated for a substance use disorder analyzed using Super learning yielded outcomes predictions superior to all but one of the traditional non-AI algorithms being compared. The same software could be used to predict outcomes of psychiatric disorders in response to different conventional and CAM therapies. The results could then be used to modify treatment protocols on an ongoing basis to optimize outcomes.
Natural language processing and dynamic simulation modeling
Accessing pertinent clinical data in physicians’ notes calls for natural language processing software and requires overcoming hurdles of confidentiality. Natural language processing software is being used to extract key concepts and relationships in very large textual data sets contained in published biomedical literature, electronic health records, and web-based medical resources (Doan 2014). Studies using natural language processing have analyzed unstructured data from millions of patients, converting key information into structured content leading to improved surveillance of treatment response as well as potentially harmful medication side effects (Le Pendu 2013).
Dynamic simulation modeling (DSM) is an approach used to design and develop mathematical representations that simulate interventions and predict responses to them over time based on patient preferences and outcomes when limited or no data are available. DSM is being successfully used to estimate differences in effectiveness between healthcare interventions before they are implemented (Marshall 2015). Big data and DSM have reciprocal synergistic relationships. DSM will allow widespread applications of big data to decision making in large health care systems. By the same token, big data will provide updated research findings to ensure that the model is simulating outcomes based on the most current findings.
In the coming years progress in artificial intelligence (AI) will yield practical clinical tools that mental health providers will use to plan individualized treatment incorporating a broad range of conventional treatments and evidence-based complementary and alternative medicine (CAM) modalities. The result will be more effective and more cost-effective management of common psychiatric disorders such as major depressive disorder, ADHD, PTSD, bipolar disorder, anxiety disorders, schizophrenia, Alzheimer's disease, and others.