Early improvement in individual symptoms and response to antidepressants in patients with major depressive disorder
Interdisciplinary Center Psychopathology and Emotion regulation, University Center of Psychiatry, University Medical Center Groningen, the Netherlands
No specific funding will be used for the proposed project. All researchers are employed by and receive a salary from either the University Medical Center Groningen or the University of Groningen.
Major depressive disorder (MDD) affects around 7% of the population yearly and has been estimated to be the third leading cause of disease burden worldwide. Although effective treatments are available, many patients fail to respond to the first treatment that is tried. In clinical trials, only around half of all patients respond to 6 to 12 weeks of antidepressant treatment. Given these high failure rates, the ability to predict as early as possible whether a patient is (un)likely to respond would be of great value, as it would enable physicians to change treatment strategies faster. Early improvement, usually defined as an improvement in depressive symptoms of at least 20% by the end of two weeks of treatment, has consistently been found to be a strong predictor of later response. However, misclassification is still quite common, with perhaps a third of those who do not show early improvement going on to respond. Conversely, a substantial proportion of those who do show early improvement do not go on to respond. One possibility for improving the predictive power of early improvement is to examine individual symptoms, rather than the total score on a depression rating scale. Some items, for example, could reflect antidepressant side effects (e.g. gastrointestinal symptoms) and may not be very predictive. In the STAR*D study, of citalopram, remission was associated with early improvement in five particular symptoms: sad mood, negative self-view, feeling slowed down, low energy, and restlessness. In another study, of mirtazapine, remission was predicted by early improvements in late insomnia (i.e., waking up too early), general somatic symptoms, and illness insight. In the proposed project, we aim to examine the relationship between early improvement in individual symptoms and response to antidepressants in a very large patient sample. This large sample size will enable us to use more rigorous methods than previous studies, such as the use of cross-validation to confirm our findings. It will also allow us to examine a large set of predictors, including possible interactions among early-improving symptoms and between symptoms and demographic factors like age and gender. We will also examine the added value of individual symptoms over and above using the total symptom score alone and possible differences between different antidepressant classes. We will use penalized (lasso) regression, which is well-suited to analyzing data with a large number of (potentially highly correlated) predictors. Lasso regression selects a subset of predictors, yielding models that have higher predictive accuracy and are easier to interpret. In our primary analysis, we will predict response after 6 weeks of treatment. In secondary analyses, we will also predict remission at week 6 and response and remission at week 12. Our findings will be communicated through conference presentations and publication in a peer-reviewed medical journal.
GSK-MY-1043/BRL-029060/115: A multicenter, randomized, double-blind, placebo-controlled comparison of paroxetine and fluoxetine in the treatment of major depressive disorder.
GSK-29060/448: 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
GSK-29060/449: 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
GSK-29060/810: 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.
GSK-29060/874: Assessment of Paxil CR, 12.5 and 25 mg/day in treating elderly patients with major depression
LILLY-F1J-US-HMFA: Duloxetine Versus Placebo in the Long-Term Treatment of Patients With Late-Life Major Depression
LILLY-F1J-MC-HMAQ(A): Duloxetine Versus Placebo in the
Treatment of Major Depression
LILLY-F1J-MC-HMAQ(B): Duloxetine Versus Placebo in the
Treatment of Major Depression
LILLY-F1J-MC-HMAT(A): Duloxetine Versus Placebo and Paroxetine in the Acute Treatment of Major Depression
LILLY-F1J-MC-HMAT(B): Duloxetine Versus Placebo and Paroxetine in the Acute Treatment of Major Depression
LILLY-F1J-US-HMFS: Duloxetine Versus Placebo in Patients With Major Depressive Disorder (MDD): Assessment of Energy and Vitality in MDD
GSK-NKF100096: 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)
LILLY-F1J-AA-HMCV: Duloxetine Versus Paroxetine in the Acute Treatment of Major Depression
LILLY-F1J-MC-HMAY(A): Duloxetine Versus Placebo and Paroxetine in the Treatment of Major Depression
LILLY-F1J-MC-HMAY(B): Duloxetine Versus Placebo and Paroxetine in the Treatment of Major Depression
LILLY-F1J-MC-HMBH(A): Duloxetine Once-Daily Dosing
Versus Placebo in the Acute Treatment of Major Depression
LILLY-F1J-MC-HMBH(B): Duloxetine Once-Daily Dosing
Versus Placebo in the Acute Treatment of Major Depression
LILLY-F1J-MC-HMBU: Duloxetine Versus Venlafaxine Extended Release in the Treatment of Major Depressive Disorder
LILLY-F1J-MC-HMCQ: Duloxetine Versus Venlafaxine Extended Release in the Treatment of Major
Depressive Disorder
LILLY-F1J-US-HMCR: Duloxetine Versus Escitalopram and Placebo in the Treatment of Patients With Major Depression
GSK-29060/01/001: A Phase II, Placebo-Controlled, Double-Blind Study of Paroxetine in Depressed Outpatients
GSK-29060/02/001: A Double-Blind, Placebo-Controlled Study of Paroxetine in Depressed Outpatients
GSK-29060/03/001: A Double-Blind, Imipramine- and Placebo-Controlled Study of Paroxetine in Depressed Outpatients
GSK-29060/07/001: A Double-Blind Comparison of Paroxetine, Amitriptyline, and Placebo in Inpatients with Major Depressive Disorder with Melancholia
GSK-29060/009: A Multicenter, Double-blind, Placebo-controlled Fixed-dose Evaluation of Four Doses of Paroxetine
GSK-29060/128: A Multicenter, Randomized, Double-Blind, Placebo-Controlled Comparison of Paroxetine and Fluoxetine in the Treatment of Major Depressive Disorder
GSK-29060/276: A double-blind study to investigate the efficacy, safety and tolerability of Paroxetine in the treatment of depression in comparison with placebo
GSK-29060/012_3: A Study to Assess the Effectiveness and Tolerance of Paroxetine by Double-Blind Comparison with Placebo and Mianserin
GSK-29060/487: 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
GSK-NKD20006: 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
LILLY-F1J-MC-HMBV: 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.
Missing data
We will take a complete cases approach, as we are interested in predicting response and remission in participants who have actually taken an antidepressant for the specified period of time. Therefore, we will select only participants who have valid baseline, week 2 and week 6 (±1) HDRS scores for our main analyses (and valid week 12 data for our secondary analyses of week 12 outcome).
Training and validation sample
The data will be randomly divided into an 80% training sample and a 20% validation sample (stratified by treatment group). Model discovery will be done in the training sample, while the predictive performance of the models will be assessed in the validation sample.
Predictors
Improvement in individual symptoms will be derived from the HDRS items at baseline and week 2. The answer choices for these items range from 0 - 2 for 7 items and 0 - 4 for 10 items. Early improvement will be dichotomized into “no improvement” and “improvement”. “No improvement” is indicated by worsening of the item score (e.g. from 1 at baseline to 2 at week 2) or no change in the item score. “Improvement” is indicated by an improvement in the item score of >=1. We will use cross-tables to check whether any variables are very highly correlated and if so, we will remove one of the items. Baseline scores on the HDRS items will also be included in the model in order to investigate the added value of improvement of individual items over and above the baseline item scores.
For the total HDRS-17 score, early improvement will also be dichotomized into no/minor improvement (<20% improvement) or improvement (>=20% improvement). The baseline HDRS-17 score will be standardized and included in the model as a covariate.
With regard to the demographic factors, gender is already a dichotomous variable and age will be standardized.
Lasso regression
We will apply lasso regression to the following models:
Primary analysis (in the antidepressant-treated group only)
1. A model containing variables for early improvement at week 2 in all 17 HDRS items, age, gender, and all two-way interactions between these variables; baseline HDRS item scores; and additionally total HDRS score at baseline and early improvement (>=20% improvement in score) in total HDRS score at week 2.
Exploratory analysis (in all participants, including those treated with placebo)
2. As model 1 above, but including treatment group (placebo, SSRI, SNRI) and all two- and three-way interactions with treatment group.
The tuning parameter (lambda) resulting in minimal prediction error (based on deviance) will be selected with the help of 10-fold cross-validation. We will use the GLMMLasso package for multilevel data and subsequently refit the model with mixed-effects logistic regression using only the variables selected by lasso regression.
Model performance
Prediction accuracy will be assessed by applying the mixed-effects logistic regression model to the independent validation sample. The area under the curve (AUC) for the receiver-operating characteristic (ROC) curve in predicting response/remission at 6 or 12 weeks will be used to assess prediction accuracy. We will also determine sensitivity, specificity, and accuracy (percentage of correct predictions).
Secondary analyses
Secondary analyses will examine 12-week outcomes within the subgroup of trials with a double-blind treatment duration of at least 12 weeks.