Treating Comorbid Depression During Care Transitions Using Relational Agents

Depression is a treatable disorder with effective pharmacologic and psychotherapy options for treatment. Evidence suggests that psychotherapy interventions such as cognitive behavior therapy (CBT) and self-management education (Lorig) can improve depression, physical symptoms and quality of life in patients with chronic medical conditions.While depression is treatable, less than 36% of patients afflicted with mood disorders get treatment. In order to meet the imminent need for post-discharge mental health support for patients with chronic illness, innovative approaches to delivering mental healthcare that are accessible on short notice, are urgently needed. The study investigators now aim to translate the RED-D intervention to an automated system to expand immediate access and scalability for this support service.

Cognitive behavioral therapy for depression (CBT) and self-management education have been studied for the treatment of comorbid depression in chronically ill patients. CBT is a psychosocial intervention that posits that depressive symptoms occur when individuals have a dysfunctional view of themselves, the world and the future. CBT teaches techniques to examine dysfunctional beliefs that perpetuate depression and to develop more accurate and helpful ones to alleviate that condition. It has largely been documented that CBT is among the most effective treatments for depression and shown to be more effective than other psychosocial treatments, as effective as pharmacotherapy and as effective as adding another medication for patients who do not respond to one antidepressant alone. CBT has also been successfully used to treat comorbid depression; the ENRICHD trial demonstrated that a collaborative care approach, hinged on CBT, is an effective treatment for depressive symptoms among post MI patients and superior to usual care.

Several studies have also demonstrated the efficacy of self-management education. For example, Lorig et al showed that self-management education enhances self-efficacy and reduces health service utilization among patients with chronic illness. This and similar programs include key processes of identifying and addressing problems, problem solving skills training, action plan design, and applying skills to medical and psychosocial aspects of chronic illness. However, self-management education alone shows limited efficacy for chronically ill patients with significant depression.

A few interventions have combined CBT for depression with self-management support. Culley et al showed that a brief CBT intervention integrating self-management strategies reduced both depressive and dyspnea symptoms in a trial with veterans diagnosed with either CHF or COPD in a primary care setting. This evidence suggests a blended and brief skills-based approach to treatment that addresses both physical symptoms and depression is more effective for depressed patients with chronically illness. However, conventional CBT has limitations. Traditionally, CBT is delivered in a face-to-face format requiring a patient to meet with a therapist in an office setting for weekly sessions for 1 weeks or more. Adherence can be a barrier for CBT due to factors related to stigma, transportation, or access to trained CBT clinicians. This has led to trials of alternative formats for delivering CBT including both telephone delivered CBT and computer-based CBT (cCBT), both of which are shown to increase adherence to treatment compared with face-to-face CBT.

Many cCBT models are in use today because of their numerous technological advantages including cost effectiveness, increased access and flexibility, and also reduced stigma for those seeking treatment for mental health conditions. cCBT is also shown to be effective for people over the age of 50, suggesting that cCBT is feasible with older patients. However, there are still several limitations to cCBT that can impact efficacy. For example, current cCBT systems largely rely on textbased interactions, creating a potential literacy barrier for some users. Similar to face-to-face CBT, nonadherence is also a problem for cCBT. Sustained adherence is essential to reach a successful therapeutic outcome with cCBT. Yet, despite their promise, evidence shows that outside of clinical trial conditions, many cCBT users do not complete the intervention, thus adherence is still a problem, although cCBT systems with embedded adherence support messages such as weekly reminders have demonstrated higher rates of program completion.

The study investigators propose to automate delivery of AHRQ's RED-D intervention and extend it with integrated self-management education for patients with CHF or COPD and comorbid depression following hospitalization. This intervention will be delivered using Relational Agent (RA) technology. Relational agents are animated counselors that provide patients with virtual consultations by simulating face-to-face conversation with an empathic provider. RAs provide an ideal medium for the CBT intervention because of: 1) their demonstrated ability to address non-adherence; 2) empathic communication features; 3) the ability to tailor the intervention to the individual patient; 4) the study investigators' past work demonstrating that hospitalized patients with depressive symptoms report high acceptance of the RA; and 5) the potential to address mental healthcare access barriers. Similar efforts using RA technology are being explored by the US Department of Defense to address the burden of PTSD among war veterans.

Traditional CBT has been effective for medically ill patients but limited in its impact on physical symptom experiences and therefore quality of life and healthcare utilization. Cully et al have shown a brief, face-to-face CBT intervention integrating self-management education is more effective for medically ill patients in primary care for reducing depression symptoms and improving quality of life. The relational agent CBT (RA-CBT) system will include six modules tailored for chronically ill patients with comorbid depression, including key components of face-to-face CBT. The final session will include a follow up exit interview to understand the patinet participant's experience with the RA-CBT program and to evaluate the effectiveness of the RA-CBT agent. The final session will incorporate a quantitative questionnaire and a voluntary qualitative interview. The RA-CBT program will include the following topics: cognitive restructuring (identify behaviors, emotions, and thoughts associated with chronic disease and depression, evaluate them and identify more helpful ones), problem solving, behavioral activation, adherence with medical appointments and medication management, and disease self-management. The disease self-management module will address relaxation, sleep hygiene and physical symptom management. Each module will contain a homework component. The RA-CBT protocol will be empathic, reflective and interactive, enhancing the possibility to develop engagement, increase adherence, and establish a trusting relationship and therapeutic alliance. Whereas current cCBT users cannot ask questions, the study team's RA-CBT intervention will make use of psychosocial technology that will allow the RA to interact with and respond to questions that arise.

Drs. Bickmore, Jack and Mitchell have conducted a series of studies using RA technology for health education and behavior change, including inpatient education (AHRQ grant 1R18HS017196-01 and AHRQ RFTO "Communication- Focused Technologies") and counseling for patients of all levels of health literacy. Dr. Bickmore and his team at Northeastern University have extensive experience designing and implementing RA psychosocial technology that simulates face-to-face conversation that is sensitive to the qualities of CBT that help establish an empathetic relationship, honing the use of social interaction with conversational technology in order to build trust and a therapeutic alliance between the agent and patient. There are many reasons why RAs provide an effective medium for health communication with patients, especially those with limited health literacy. The human-computer interface relies minimally on text (as compared to current cCBT models) and uses the universally understood format of face-to-face conversation, making it accessible to patients with limited reading ability. Most importantly, RAs enhance recall of critical information through the use of nonverbal conversational behaviors (gesture, tone) by providing redundant channels for conveying semantic content in order to enhance the likelihood of message comprehension. In addition, all cultures have nonverbal means for marking emphasis (for example, eyebrow raising and 'beat' or 'baton' hand gestures in American English) that may highlight salient parts of a message. Unlike a pre-recorded video lecture, the study team's RA-CBT intervention provides a flexible and dynamic communication medium that relies on synthetic speech tailored to the patient and the discourse context using language dynamically assembled using a dialogue engine (i.e. use of the patient's name, medical record information, and other personally relevant information) that responds to the context of the conversation (i.e. the questions asked, patient responses during earlier sessions, etc.) Patient contributions to the dialogue are made by selecting what they want to say from a multiple choice menu, dynamically updated for each turn of the conversation (no natural language understanding of unconstrained text or speech is performed). Alternative technologies do not offer this adaptive, face-to-face encounter that accommodates patients' needs. This interface has been used by over 2,500 patients including many with no prior computer experience.