Evidence for machine learning guided early prediction of acute outcomes in the treatment of depressed children and adolescents with antidepressants

Abstract

Background: The treatment of depression in children and adolescents is a substantial public health challenge. This study examined artificial intelligence tools for the prediction of early outcomes in depressed children and adolescents treated with fluoxetine, duloxetine, or placebo.

Methods: The study samples included training datasets (N = 271) from patients with major depressive disorder (MDD) treated with fluoxetine and testing datasets from patients with MDD treated with duloxetine (N = 255) or placebo (N = 265). Treatment trajectories were generated using probabilistic graphical models (PGMs). Unsupervised machine learning identified specific depressive symptom profiles and related thresholds of improvement during acute treatment.

Results: Variation in six depressive symptoms (difficulty having fun, social withdrawal, excessive fatigue, irritability, low self-esteem, and depressed feelings) assessed with the Children's Depression Rating Scale-Revised at 4-6 weeks predicted treatment outcomes with fluoxetine at 10-12 weeks with an average accuracy of 73% in the training dataset. The same six symptoms predicted 10-12 week outcomes at 4-6 weeks in (a) duloxetine testing datasets with an average accuracy of 76% and (b) placebo-treated patients with accuracies of 67%. In placebo-treated patients, the accuracies of predicting response and remission were similar to antidepressants. Accuracies for predicting nonresponse to placebo treatment were significantly lower than antidepressants.

Conclusions: PGMs provided clinically meaningful predictions in samples of depressed children and adolescents treated with fluoxetine or duloxetine. Future work should augment PGMs with biological data for refined predictions to guide the selection of pharmacological and psychotherapeutic treatment in children and adolescents with depression.

Keywords: Depression; adolescents; decision support tools; machine learning.

Figure 1

(A) Envisioned use of proposed probabilistic graph‐based tool to derive prognoses of treatment outcomes in children and adolescents treated with fluoxetine or duloxetine. (B) Trajectories of Children’s Depression Rating Scale‐Revised (CDRS‐R) total score in patients treated with fluoxetine. (C) The machine learning workflow

Figure 2

(A) Construction of the probabilistic graphical model (PGM) – a hidden Markov model comprising hidden states, observation states, and transitions. The hidden states were defined using ranges of total depression severity, wherein the ranges for depression severity at baseline were inferred using unsupervised machine learning. Observation states at the treatment’s intermediate (4–6 weeks) or endpoint (10–12 weeks) record active depression [i.e., Children’s Depression Rating Scale‐Revised (CDRS‐R) total depression severity score ≥ 40]. (B) Compact representation of CDRS‐R total score variations derived using PGM. (C) Symptom clusters of patients in A2 strata based on depression severity of CDRS‐R total scores less ≤ 55

Figure 3

(A–D) Solid blue line was the variation in mean symptom severity, the shaded region around the solid blue line was 95% confidence interval of the variation around the mean symptom severity and the box plots visualize the variation in symptom severity at each time point. (B–D) We observe variations in symptom severity on the symptom dynamic paths originating in A1 stratum at baseline. (A) Variation in the severity of irritability in [Children’s Depression Rating Scale‐Revised (CDRS‐R) item 4] in patients originating from A1 stratum at baseline. Although the band of the shaded region was narrow around the mean indicating reduction in symptom severity in response to therapy, the height of the boxplot and the extent of whiskers indicate high degrees of variability in scores at each time point. (B and D) Visualizing variation in prognostic (irritability) and nonprognostic symptoms (impaired schoolwork) in symptom dynamic paths of patients originating in A1 stratum and treated with either fluoxetine (Figure B) or placebo (Figure C and D). (E) Accuracy of predictions derived using short‐term improvement in prognostic symptoms and CDRS‐R total depression severity