Mediaboss Marketing

Standards and guidelines of machine learning in psychiatry were followed when this study was conducted and reported [20].

This study included 291 inpatients in a tertiary hospital who were diagnosed as major depressive disorders. Patient eligibility was determined based on the criteria of the Diagnostic and Statistical Manual of the American Psychiatric Association, Fourth Edition (DSM-IV). Blood samples were collected before antidepressant treatment.

All patients met the following criteria: Han Chinese, 1865 years old, baseline 17-item Hamilton Depression Rating Scale (HAMD-17) [21] scores>17 points, and their depressive symptoms lasted at least 2 weeks. All patients had just been diagnosed or had recently relapsed and had not been on medication for at least two weeks prior to enrollment. All diagnoses were made independently by two psychiatrists with professional tenure or higher, and confirmed by a third psychiatrist. Participants had never been diagnosed with other DSM-IV Axis I diagnosis (including substance use disorder, schizophrenia, affective disorder, bipolar disorder, generalized anxiety disorder, panic disorder, obsessive-compulsive disorder). They had never been diagnosed with personality disorder or mental retardation. Patients with a history of organic brain syndrome, endocrine, and primary organic diseases, or other medical conditions that would hinder psychiatric evaluation were excluded from the study. Other exclusion criteria included blood, heart, liver, and kidney disorders; electroconvulsive therapy in the past 6 months; or an episode of mania in the previous 12 months. Pregnant and nursing females were also excluded from participation.

All study subjects in the study endorsed written consent that was approved by the Zhongda Hospital Ethics Committee (2016ZDSYLL100-P01) under the Declaration of Helsinki.

Response was defined as 50% reduction in HAMD-17 scores from baseline to two weeks [22]. Accordingly, the two-week treatment participants were divided into two groups, responders and non-responders.

Two retrospective self-report questionnaires, the Childhood Trauma Questionnaire (28-item short-form, CTQ-SF) and the Life Events Scale (LES), were used to evaluate recent stress exposures and childhood adversities, respectively. The evaluation of LES and CTQ scales was completed by the same nurse using consistent, scripted language. LES is a self-assessed questionnaire composed of 48 items, reflecting both positive and negative life events experienced within the past year. The LES is divided into positive and negative life events (NLES). The CTQ-SF was dichotomized for use in the gene-environment interaction analyses.

Twelve considered demographic and clinical features are age, gender, years of education, marital status, family history, first occurrence or not, age of onset, number of occurrences, illness duration, HAMD-17, NLES and CTQ-SF baseline scores (Supplemental Material Table1).

Primers were previously designed by us to encompass 100bp upstream and 100bp downstream of TPH2 SNPs that showed a significant association with the antidepressant response, as well as with GC sequence content of CpGs>20% after methylation [11, 12]. Out of the total 24 TPH2 SNPs, only 11 SNPs (rs7305115, rs2129575, rs11179002, rs11178998, rs7954758, rs1386494, rs1487278, rs17110563, rs34115267, rs10784941, rs17110489) met the DNA methylation status criteria of the sequences to be detected (Supplemental Material Table2). Methylation levels of 38 TPH2 CpGs were calculated and presented as the ratio of the number of methylated cytosines to the total number of cytosines.

In the data set comprising 291 observations of 51 variables (12 demographic and clinical features, 38 CpGs methylation levels and 1 response variable), 6% entries were missing (see Fig.1). Of the CpGs methylation levels, 3 CpGs (TPH2-7-99, TPH2-7-142, TPH2-7-170) were excluded because they had more than 45% missing values. Due to the randomness of experimental/technological errors and interrelatedness of the variables, missing completely at random (MCAR)/missing at random (MAR) was assumed for the DNA methylation data and the mean imputation can deal with the missing data [23, 24]. The values of other features with missing values were imputed with mode and mean in the case of categorical and numerical features, respectively.

Missingness pattern in the DNA methylation data set

Normalization (Linear transformation) was used to improve the numerical stability of the model and reduce training time [25]. To avoid overfitting when harnessing maximum amount of data, cross-validation (CV) using entire sample was used to report prediction performance. The CV was 5-fold and the averaged prediction metrics including the area under the receiver operating curve (AUC), F-Measure, G-Mean, accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were reported. Hyperparameter tuning was based on AUC with random search using the caret default tuning settings. A packaging method (Recursive Feature Elimination with random forest, RFE-RF) [26] with 5-fold CV was employed to select the features that contributed the most to the prediction of the early antidepressant response in MDD patients. The variable importance was also estimated using random forest. For better replicability, the 5-fold CV procedure was repeated 10 times.

ML methods were implemented via their interface with the open-source R package caret in a standardized and reproducible way. Five different supervised ML algorithms were used in this study, including logistic regression, classification and regression trees (CART), support vector machine with radial basis function kernel (SVM-RBF), a boosting method (logitboost) and random forests (RF) to develop predictive models. All analyses were implemented in R statistical software (version 4.0.4). We utilized the caret package which implements rpart, caTools, e1071 and RandomForest packages for CART, logitboost, SVM-RBF and RF, respectively.

Read this article:
Early antidepressant treatment response prediction in major ... - BMC Psychiatry