Matching the Right Antidepressants with the Right Patients: Using Artificial Intelligence to Personalize the Treatment of Depression

Proposal
2029

Title of Proposed Research

Lead Researcher

Gustavo Turecki

Affiliation

Douglas Mental Health University InstituteMcGill University

Funding Source

None

Potential Conflicts of Interest

Data Sharing Agreement Date

18 June 2019

Lay Summary

Major depression is a debilitating health condition that will affect more than one in ten people during the course of their lives. Over a dozen effective antidepressant drugs are now available in psychiatric practice. However, our abilities to predict which antidepressant will work best for a given patient remains limited. We aim to solve this problem by using the latest in artificial intelligence (AI) and deep learning technologies to pair patients with ideal treatments for depression. Deep learning is a process that can use clinical trial data to embed knowledge of drug treatments within an artificial neural network (ANN), computer technology trained for problem solving via recognition of complex patterns in data. Our research aims to train an ANN with clinical trial data in order to find patterns in patient characteristics that respond best to a specific antidepressant. We predict that clinicians can then use this ANN as a tool to help guide them in predicting which antidepressants will work best for future patients. In addition, our work will allow for identification of patient subgroups who respond better to specific medications. Improving our ability to select the most effective antidepressant for individual patients would mark a significant advancement in the treatment of depression. Current methods typically follow an educated “guess and check” approach: a patient and their physician will try one treatment, then another, with the hope that some drug or combination of drugs will eventually be effective. Given the multitude of antidepressants available, the probability of selecting the ideal therapeutic option on the first try is low. Indeed, only a third of patients report improvements in their depression following their first attempt at relief. The majority of patients must try two or more medications before identifying an antidepressant that works best for them; these trials can span months or years, and are often accompanied with sometimes unpredictable side effects. By conducting a secondary analysis of clinical trial data available from CSDR, we believe our ANN can find patterns that predict treatment responses based on individual patient characteristics, such as symptom profile or specific biomarkers. We will assess the utility of our ANN as a clinical decision aid for physicians using a combination of computer analyses, expert review, and validation through clinical trials. The ultimate aim of this research is to develop an evidence-based approach to personalizing pharmacological treatment of depression while minimizing side effects that lead to reduced quality of life and treatment adherence.

Study Data Provided

[{ "PostingID": 1617, "Title": "GSK-29060/785", "Description": "A double-blind, placebo-controlled, fixed-dosage study comparing the efficacy and tolerability of paroxetine CR and citalopram to placebo in the treatment of Major Depressive Disorder with anxiety" },{ "PostingID": 1623, "Title": "GSK-MY-1043/BRL-029060/115", "Description": "A multicenter, randomized, double-blind, placebo-controlled comparison of paroxetine and fluoxetine in the treatment of major depressive disorder." },{ "PostingID": 1633, "Title": "GSK-29060/448", "Description": "A Double-Blind, Placebo Controlled Trial to Evaluate the Clinical Effects of Immediate Release Paroxetine and Modified Release Paroxetine in the Treatment of Major Depression" },{ "PostingID": 1634, "Title": "GSK-29060/449", "Description": "A Double-Blind, Placebo Controlled Trial to Evaluate the Clinical Effects of Immediate Release Paroxetine and Modified Release Paroxetine in the Treatment of Major Depression" },{ "PostingID": 1638, "Title": "GSK-29060/810", "Description": "A double-blind, placebo-controlled, 3-arm, fixed-dose study of 12.5 mg/day and 25mg/day Paroxetine CR in the treatment of Major Depression." },{ "PostingID": 1639, "Title": "GSK-29060/874", "Description": "Assessment of Paxil CR, 12.5 and 25 mg/day in treating elderly patients with major depression" },{ "PostingID": 1649, "Title": "LILLY-F1J-US-HMFA", "Description": "Duloxetine Versus Placebo in the Long-Term Treatment of Patients With Late-Life Major Depression" },{ "PostingID": 1949, "Title": "LILLY-F1J-MC-HMAI", "Description": "A Double-Blind, Placebo- and Clomipramine-Controlled Study of Duloxetine in Patients with Major
Depression" },{ "PostingID": 1955, "Title": "LILLY-F1J-MC-HMDI", "Description": "Duloxetine Versus Placebo in the Prevention of Recurrence of Major Depressive Disorder" },{ "PostingID": 1982, "Title": "GSK-NKF100096", "Description": "A Randomised, Double-Blind, Double-Dummy, Parallel-Group, Placebo-Controlled, Forced Dose Titration Study Evaluating the Efficacy and Safety of GW679769 and Paroxetine in Subjects with Major depressive Disorder (MDD)" },{ "PostingID": 2003, "Title": "LILLY-F1J-EW-HMGD", "Description": "Comparison of Two Different Treatment Strategies in Patients With Major Depressive Disorder Not Exhibiting Improvement on Escitalopram Treatment: Early vs. Delayed Intervention Strategy" },{ "PostingID": 2014, "Title": "LILLY-H8I-MC-HQAC", "Description": "Validation of Daily Telephone Self-Assessment in the Study of Antidepressant Treatment Outcome" },{ "PostingID": 2063, "Title": "GSK-AK130939", "Description": "A Multi-Centre, Randomised, Double-Blind, Parallel-Group, Placebo- and Active-Controlled, Flexible Dose Study Evaluating the Efficacy, Safety and Tolerability of Extended-Release Bupropion Hydrochloride (150mg - 300mg once daily), Extended-Release Venlafaxine Hydrochloride (75mg - 150mg once daily) and Placebo in Subjects with Major Depressive Disorder." },{ "PostingID": 2064, "Title": "GSK-WXL101497", "Description": "A Multi-Centre, Randomised, Double-Blind, Parallel-Group, Placebo- and Active-Controlled, Flexible Dose Study Evaluating the Efficacy, Safety and Tolerability of Extended-Release Bupropion Hydrochloride (150mg - 300mg once daily), Extended-Release Venlafaxine Hydrochloride (75mg - 150mg once daily) and Placebo in Subjects with Major Depressive Disorder." },{ "PostingID": 2067, "Title": "GSK-WELL ZYB40021", "Description": "A 6 Month Multicenter, Randomized, Double-Blind, Pilot Study to Investigate the Tolerability and Efficacy of Bupropion SR Compared to Placebo for the Treatment of Mild Depressive Symptoms and Obesity, Followed by a 24-week Open-Label Extension" },{ "PostingID": 2122, "Title": "GSK-29060/01/001", "Description": "A Phase II, Placebo-Controlled, Double-Blind Study of Paroxetine in Depressed Outpatients" },{ "PostingID": 2123, "Title": "GSK-29060/02/001", "Description": "A Double-Blind, Placebo-Controlled Study of Paroxetine in Depressed Outpatients" },{ "PostingID": 2124, "Title": "GSK-29060/03/001", "Description": "A Double-Blind, Imipramine- and Placebo-Controlled Study of Paroxetine in Depressed Outpatients" },{ "PostingID": 2125, "Title": "GSK-29060/07/001", "Description": "A Double-Blind Comparison of Paroxetine, Amitriptyline, and Placebo in Inpatients with Major Depressive Disorder with Melancholia" },{ "PostingID": 2127, "Title": "GSK-29060/057", "Description": "A Double-blind comparative multicentre study of paroxetine plus supportive psychotherapy and psychotherapy alone in the prevention of recurrent suicidal behavior and episodes of intermittent brief depression" },{ "PostingID": 2128, "Title": "GSK-29060/106", "Description": "A double-blind comparative study of paroxetine and placebo in the treatment of episodes of intermittent brief depression (IBD)" },{ "PostingID": 2129, "Title": "GSK-29060/128", "Description": "A Multicenter, Randomized, Double-Blind, Placebo-Controlled Comparison of Paroxetine and Fluoxetine in the Treatment of Major Depressive Disorder" },{ "PostingID": 2130, "Title": "GSK-29060/251", "Description": "A Double-Blind, Randomized Trial of Paroxetine Versus Placebo In Patients With Depression Accompanied by Anxiety" },{ "PostingID": 2132, "Title": "GSK-26090/007", "Description": "A Double-Blind Placebo Controlled Study to Compare Paroxetine with Maprotiline in the Treatment of Depression" },{ "PostingID": 2133, "Title": "GSK-29060/276", "Description": "A double-blind study to investigate the efficacy, safety and tolerability of Paroxetine in the treatment of depression in comparison with placebo" },{ "PostingID": 2134, "Title": "GSK-29060/012_3", "Description": "A Study to Assess the Effectiveness and Tolerance of Paroxetine by Double-Blind Comparison with Placebo and Mianserin" },{ "PostingID": 2135, "Title": "GSK-29060/282", "Description": "A trial to assess the effectiveness and tolerance of paroxetine by double-blind comparison with placebo using a novel “shifted crossover” design" },{ "PostingID": 2136, "Title": "GSK-29060/327", "Description": "A double-blind, placebo-controlled, parallel group study of paroxetine in the treatment of dysthymia." },{ "PostingID": 2137, "Title": "GSK-29060/487", "Description": "A Double-Blind, Placebo Controlled Trial to Evaluate the Clinical Effects of Immediate Release Paroxetine and Controlled Release Paroxetine in the Treatment of Major Depression in Elderly Patients" },{ "PostingID": 2138, "Title": "GSK-29060/625", "Description": "A double-blind, placebo-controlled multi-centre study to evaluate the efficacy and tolerability of Paroxetine in the treatment of post-stroke depression." },{ "PostingID": 2577, "Title": "LILLY-F1J-MC-HMBO", "Description": "Duloxetine Versus Placebo in the Treatment of Fibromyalgia Patients With or Without Major Depressive Disorder" },{ "PostingID": 2579, "Title": "LILLY-F1J-MC-HMCA", "Description": "Duloxetine Versus Placebo in the Treatment of Fibromyalgia Patients With or Without Major Depressive Disorder" },{ "PostingID": 2581, "Title": "LILLY-F1J-MC-HMDD", "Description": "Duloxetine Versus Duloxetine Plus Non-Pharmacological Intervention in the Treatment of Depression" },{ "PostingID": 3017, "Title": "LILLY-F1D-MC-HGHZ", "Description": "The Combination of Olanzapine and Fluoxetine in Treatment Resistant Depression without Psychotic Features" },{ "PostingID": 3018, "Title": "LILLY-F1D-MC-HGIE", "Description": "Olanzapine Plus Fluoxetine Combination Therapy in Treatment-Resistant Depression: A Dose Ranging Study" },{ "PostingID": 3019, "Title": "LILLY-H6P-MC-HDAO", "Description": "The Study of Olanzapine plus Fluoxetine in Combination for Treatment-Resistant Depression Without Psychotic Features" },{ "PostingID": 3021, "Title": "LILLY-H6P-MC-HDAY", "Description": "A Study to Assess the Long-Term Efficacy and Safety of Olanzapine and Fluoxetine Combination Versus Fluoxetine Only in the Relapse Prevention of Stabilized Patients with Treatment-Resistant Depression" },{ "PostingID": 3090, "Title": "LILLY-F1J-US-HMGR", "Description": "A Phase 4, 8-week, Double-blind, Randomized, Placebo-controlled Study Evaluating the Efficacy of Duloxetine 60 mg Once Daily in Outpatients with Major Depressive Disorder and Associated Painful Physical Symptoms" },{ "PostingID": 3091, "Title": "LILLY-F1J-MC-HMDG", "Description": "Switching to Duloxetine From Other Antidepressants: A Regional Multicentre Trial Comparing Two Switching Techniques" },{ "PostingID": 3092, "Title": "LILLY-F1J-US-HMGU", "Description": "Duloxetine Versus Placebo in the Acute Treatment of Patients With Major Depressive Disorder and Associated Painful Physical Symptoms" },{ "PostingID": 3093, "Title": "LILLY-F1J-CR-HMGM", "Description": "A Phase 4, 8-week, Double-blind, Randomized Study Comparing Switching to Duloxetine or Escitalopram in Patients With Major Depressive Disorder and Residual Apathy in the Absence of Depressed Mood" },{ "PostingID": 3161, "Title": "GSK-NKD20006", "Description": "An 8-Week, Randomized, Double-Blind, Placebo-Controlled, Multicenter, Fixed-Dose Study Comparing the Efficacy and Safety of GW597599B or Paroxetine to Placebo in Moderately to Severely Depressed Patients with Major Depressive Disorder" },{ "PostingID": 3563, "Title": "GSK-HTP109035", "Description": "A Randomised, Double-Blind, Parallel-Group, Placebo-Controlled Study Evaluating the Efficacy and Safety of GSK163090 in Subjects with Major Depressive Disorder" },{ "PostingID": 3705, "Title": "GSK-AK1113351", "Description": "Study AK1113351, a fixed dose study of 323U66 SR in the treatment of Major Depressive Disorder (MDD) - a multi-center, placebo-controlled, randomized, double-blind, parallel-comparison study" },{ "PostingID": 4238, "Title": "GSK-PCR112810", "Description": "A Randomized, Double-blind, Placebo Controlled Trial to Evaluate the Clinical Effects of Controlled Release Paroxetine in the Treatment of Major Depressive Disorder" },{ "PostingID": 4424, "Title": "GSK-PKI113009", "Description": "A six week randomized, double-blind, multi-center, placebo-controlled, exploratory, adaptive design study to explore the antidepressant properties of the p38 MAP kinase inhibitor GW856553 compared to placebo in adult subjects with Major Depressive Disorder" },{ "PostingID": 4738, "Title": "GSK-PKI108574", "Description": "A randomized, double blind, placebo controlled study to explore the antidepressant properties of P38a Kinase inhibitor GW856553X 15mg compared to PBO in Subjects with Major Depressive Disorder exhibiting symptoms of loss of energy and interest and psychomotor retardation, for a six week treatment period" },{ "PostingID": 4919, "Title": "GSK-29060/274", "Description": "A study to assess the effectiveness and tolerance of paroxetine by double-blind comparison with placebo" },{ "PostingID": 4946, "Title": "LILLY-F1J-MC-HMBV", "Description": "Duloxetine Versus Placebo in the Treatment of Elderly Patients With Major Depressive Disorder

Medicine: Duloxetine hydrochloride, Condition: Major Depressive Disorder, Phase: 3, Clinical Study ID: F1J-MC-HMBV , Sponsor: Lilly." },{ "PostingID": 4947, "Title": "LILLY-F1J-US-HMCB", "Description": "Duloxetine Once-Daily Dosing Versus Placebo in Patients With Major Depression and Pain

Medicine: Duloxetine hydrochloride, Condition: Major Depressive Disorder, Phase: 3, Clinical Study ID: F1J-US-HMCB , Sponsor: Lilly." }]

Statistical Analysis Plan

Our main outcomes measured will be the number of accurately predicted side effects and the accuracy of predicted treatment efficacy when the model, after having been trained on the training data, is used to analyze the validation data. Treatment efficacy will be measured by changes in normalized versions of standardized rating scales for depression as well as rating scales for patient function when available. Model Evaluation and Bias Control We will evaluate our ANN using the following standard tests: model sensitivity, model specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and Receiver Operating Curve (ROC) and the Area Under the Curve (AUC) analysis. Samples within a data set are classified into categories based on passing a set threshold; for example, an antidepressant is classified as an effective treatment for a given patient only if it passes a high predetermined threshold (i.e meeting the usual criteria for response or remission). As we increase the threshold, we can increase the segregation between categories and thus the specificity and selectivity of the ANN. Sensitivity indicates the ANN's predictive power for correctly identifying a positive association; specificity determines the ANN's ability to correctly identify a negative association. The ROC measures how responsive the ANN is to changes in thresholds as we optimize this value. ROC is determined from the sensitivity and specificity. The area underneath this ROC curve (AUC) is the standard in comparing different iterations of a classifier and can indicate superiority over the other trained algorithms if the current algorithm has the largest AUC. The NPV is related to the specificity as it is responsible for correctly predicting negative results to patients without a certain condition. The PPV (otherwise known as the precision score in the machine learning community) is responsible for determining the number of correct results among all results to illustrate the algorithm understands the nuances of a certain illness. To help illustrate how this works, imagine a rare illness. For this rare illness, an algorithm can predict the negative case (i.e not ill) for all patients and retain a very good accuracy but these metrics assist in decoding the true performance so as to be able to still predict this rare illness when necessary.Our ANN must acquire a high accuracy in associating patient characteristics with effective antidepressant treatments and potential side-effects. To do so, we set appropriate thresholds, or “minimum standards”, that must be met in order to classify one factor over the other. For instance, we must avoid having the ANN learn to mistakenly recommend excessive pharmacological treatments for relatively mild cases of depression. Here we set a high threshold to ensure the ANN will recommend effective antidepressant treatments with far greater precision. We face a contrary challenge in terms of predicting side-effects. Here we want a low threshold to ensure the ANN will include reporting of rare side-effects that are important nonetheless. To promote the accuracy of our ANN in identifying key patient characteristics with respect to treatment responses for antidepressants, we will use the k-fold validation assessment. We further assess the predictive accuracy of the ANN by measuring the variance in prediction error between its predictions made from the training data set compared to those made with the validation data set. The ideal scenario here is to keep the validation error as close to the training error as possible while still jointly lowering the training error. Large divergence of the validation error from that of the training will result in overfitting and a less generalizable model.Deep learning techniques used to train ANNs typically employ a frequentist statistical approach in order to establish patterns and associations in data. Our training strategy will focus on the frequentist approach in order to identify replicable links between patient characteristics (inputs) and effective treatments/side-effects (outputs). This frequentist method can introduce a bias: ANNs will readily learn to make strong associations between inputs and outputs abundant in the data while downplaying those between rare inputs or outputs. (For example, the data may abound with patient classifications based on gender but lack information on their socioeconomic status; the ANN will thus succeed in linking effective antidepressants and gender but not socioeconomic status. This may happen despite the fact that socioeconomic status may be as strong a predictor of treatment success as gender). We will identify this potential bias by use of precision and recall metrics and correct any imbalance by use of a “regularizing term”. A regularising term is an additional restriction on the model aimed at penalising the model more for incorrect classification of classes that are not abundantA potential issue requiring vigilance with deep learning of ANNs is the problem of “overfitting”. Long-term training of an ANN may cause it to memorize and thus fixate on certain patterns from the training data. Overfitting of the ANN is problematic because overfit patterns are: 1) not representative of actual “learning”; and 2) may not generalize to new data sets. To counter overfitting, we will employ a common “dropout strategy”. Here we will selectively remove variables from the training data--specifically, the input features of patient characteristics--during different training rounds. For example, a patient with characteristics ABC can be used in three training rounds -- one for A (removed BC), B (-AC), etc. -- to train the ANN to recognize these three features as being linked to an effective antidepressant. Using the dropout method, the ANN can learn more distinct and similarly representative patterns between patient characteristics and effective antidepressants rather than fixating on a select few. We can further avoid the problem of overfitting by employing a “snapshot learning” procedure. This common procedure in the field of deep learning enables us to iteratively push several ANNs towards learning together more general patterns in the data instead of memorizing data points.Statistical power calculations and levels of significanceWe will follow a “10X standard” to determine the amount of data our ANN requires to understand and identify patterns in the clinical trial data. This rule-of-thumb in deep learning recommends to use roughly 10 times more data points than trainable parameters (i.e. the weights and biases of connections between nodes at each layer of the ANN) embedded within the ANN. Recall that our project employs an unsupervised learning procedure to first train our ANN to understand the training data set. This initial training enables the ANN to learn how to accurately predict treatment efficacies for antidepressants using fewer data points concerning labelled treatment responses. As such, following the 10X standard should provide us with more than enough data for accurate prediction. Analysis of subgroups Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) can identify general patterns (i.e., “natural clusters”) in the training data. Once the ANN can associate categories to individual patients, both PCA and t-SNE techniques will sort the large number of features for each patient to simplify identification of patterns. These techniques simplify the visualization and identification by the investigators of the patterns learned by the ANN during training. Handling of missing dataDeep learning techniques can train ANNs using incomplete data sets using the “dropout method” described previously. This means that while the model is being trained, certain variables are removed from the input set to give the model the ability to form several representations of the same class using different indicating factors (to illustrate, let us imagine that responders to drug X are generally female, over 50, do not have diabetes, and have low cholesterol; the model would learn to associate response to this drug to different combinations of 3 of those 4 predictors. As such, even if some patient records are missing one of the three factors, patients from those records could still be correctly identified as belonging to the class of patients who respond to drug X).Overall, the dropout method enables us to group together large and incomplete data sets with heterogeneous patient features and antidepressants.

Publication Citation

Perlman, K., Mehltretter, J., Benrimoh, D. et al. Development of a differential treatment selection model for depression on consolidated and transformed clinical trial datasets. Transl Psychiatry 14, 263 (2024).
DOI: https://doi.org/10.1038/s41398-024-02970-4