Personalized Depression Treatment
For a lot of people suffering from depression, traditional therapy and medication isn't effective. A customized treatment may be the solution.
Cue is a digital intervention platform that translates passively acquired normal sensor data from smartphones into customized micro-interventions to improve mental health. We analyzed the best-fitting personalized ML models to each person using Shapley values, in order to understand their features and predictors. The results revealed distinct characteristics that changed mood in a predictable manner over time.
Predictors of Mood
Depression is a leading cause of mental illness in the world.1 Yet, only half of those with the condition receive treatment. To improve outcomes, doctors must be able to recognize and treat patients with the highest likelihood of responding to particular treatments.
Personalized depression treatment can help. depression treatment modalities at the University of Illinois Chicago are developing new methods to predict which patients will gain the most from certain treatments. They make use of sensors on mobile phones as well as a voice assistant that incorporates artificial intelligence, and other digital tools. Two grants worth more than $10 million will be used to identify the biological and behavioral indicators of response.
The majority of research to date has focused on clinical and sociodemographic characteristics. These include demographic variables like age, sex and education, clinical characteristics such as the severity of symptoms and comorbidities and biological markers like neuroimaging and genetic variation.
While many of these factors can be predicted from data in medical records, few studies have utilized longitudinal data to explore the factors that influence mood in people. A few studies also take into consideration the fact that moods can differ significantly between individuals. Therefore, it is crucial to develop methods which allow for the identification and quantification of personal differences between mood predictors and treatment effects, for instance.
The team's new approach uses daily, in-person evaluations of mood and lifestyle variables using a smartphone app called AWARE, a cognitive evaluation with the BiAffect app and electroencephalography -- an imaging technique that monitors brain activity. The team can then develop algorithms to detect patterns of behaviour and emotions that are unique to each person.
The team also developed a machine-learning algorithm that can identify dynamic predictors of each person's depression mood. The algorithm blends these individual characteristics into a distinctive "digital phenotype" for each participant.
This digital phenotype was correlated with CAT DI scores, a psychometrically validated symptom severity scale. However the correlation was not strong (Pearson's r = 0.08, adjusted BH-adjusted P-value of 3.55 x 10-03) and varied widely across individuals.
Predictors of symptoms
Depression is one of the world's leading causes of disability1 but is often untreated and not diagnosed. In addition the absence of effective treatments and stigmatization associated with depressive disorders prevent many people from seeking help.
To aid in the development of a personalized treatment plan to improve treatment, identifying the patterns that can predict symptoms is essential. However, the current methods for predicting symptoms are based on the clinical interview, which has poor reliability and only detects a limited variety of characteristics associated with depression.2
Machine learning can enhance the accuracy of diagnosis and treatment for depression by combining continuous, digital behavioral phenotypes collected from smartphone sensors with a valid mental health tracker online (the Computerized Adaptive Testing Depression Inventory CAT-DI). Digital phenotypes permit continuous, high-resolution measurements and capture a wide range of distinctive behaviors and activity patterns that are difficult to capture with interviews.
The study involved University of California Los Angeles students with moderate to severe depression symptoms who were enrolled in the Screening and Treatment for Anxiety and Depression program29 that was developed as part of the UCLA Depression Grand Challenge. Participants were referred to online support or to clinical treatment according to the severity of their depression. Those with a CAT-DI score of 35 or 65 students were assigned online support via the help of a coach. Those with a score 75 patients were referred for psychotherapy in person.
At the beginning of the interview, participants were asked a series of questions about their personal characteristics and psychosocial traits. These included sex, age, education, work, and financial situation; whether they were divorced, partnered, or single; current suicidal thoughts, intentions or attempts; and the frequency at the frequency they consumed alcohol. Participants also scored their level of depression symptom severity on a 0-100 scale using the CAT-DI. CAT-DI assessments were conducted every week for those who received online support and once a week for those receiving in-person support.
Predictors of Treatment Reaction
A customized treatment for depression is currently a research priority and many studies aim at identifying predictors that help clinicians determine the most effective drugs for each person. Pharmacogenetics, for instance, uncovers genetic variations that affect how the body's metabolism reacts to drugs. This lets doctors choose the medications that are most likely to work for each patient, while minimizing time and effort spent on trial-and-error treatments and avoid any negative side negative effects.
Another approach that is promising is to build models of prediction using a variety of data sources, combining the clinical information with neural imaging data. These models can be used to determine which variables are most predictive of a particular outcome, like whether a drug will improve mood or symptoms. These models can be used to determine the response of a patient to treatment, allowing doctors to maximize the effectiveness.
A new generation of machines employs machine learning techniques like the supervised and classification algorithms, regularized logistic regression and tree-based methods to integrate the effects of multiple variables and increase the accuracy of predictions. These models have been demonstrated to be effective in predicting outcomes of treatment, such as response to antidepressants. These approaches are gaining popularity in psychiatry and it is likely that they will become the standard for the future of clinical practice.
In addition to the ML-based prediction models research into the mechanisms that cause depression continues. Recent research suggests that depression is related to the malfunctions of certain neural networks. This suggests that individualized depression treatment will be focused on therapies that target these neural circuits to restore normal function.
One method to achieve this is by using internet-based programs that offer a more personalized and customized experience for patients. One study found that an internet-based program helped improve symptoms and provided a better quality life for MDD patients. Additionally, a randomized controlled study of a personalised treatment for depression demonstrated steady improvement and decreased adverse effects in a large number of participants.
Predictors of Side Effects
In the treatment of depression a major challenge is predicting and identifying which antidepressant medication will have very little or no adverse negative effects. Many patients have a trial-and error approach, with a variety of medications being prescribed before settling on one that is safe and effective. Pharmacogenetics offers a fascinating new method for an efficient and targeted method of selecting antidepressant therapies.
There are a variety of variables that can be used to determine the antidepressant to be prescribed, such as gene variations, phenotypes of patients such as ethnicity or gender, and the presence of comorbidities. To identify the most reliable and valid predictors for a specific treatment, controlled trials that are randomized with larger numbers of participants will be required. This is because it may be more difficult to determine moderators or interactions in trials that only include one episode per participant instead of multiple episodes spread over a long period of time.
Furthermore the prediction of a patient's response will likely require information on comorbidities, symptom profiles and the patient's personal perception of the effectiveness and tolerability. Currently, only a few easily identifiable sociodemographic variables and clinical variables appear to be consistently associated with response to MDD. These include gender, age, race/ethnicity, BMI, SES and the presence of alexithymia.
The application of pharmacogenetics in depression treatment is still in its beginning stages and there are many hurdles to overcome. First, it is essential to be able to comprehend and understand the definition of the genetic factors that cause depression, as well as a clear definition of an accurate predictor of treatment response. Additionally, ethical issues, such as privacy and the responsible use of personal genetic information, must be carefully considered. The use of pharmacogenetics may eventually help reduce stigma around treatments for mental illness and improve the outcomes of treatment. As with any psychiatric approach, it is important to give careful consideration and implement the plan. The best method is to provide patients with an array of effective depression medication options and encourage them to speak freely with their doctors about their experiences and concerns.