Predicting changes in depressive symptomatology following oral esketamine treatment in treatment-resistant depression: A machine-learning approach.

Abstract

Oral esketamine is a potentially effective and well-tolerated treatment for treatment-resistant depression (TRD). However, around 50% of TRD individuals treated with oral esketamine do not achieve response, and this variation in response may be accounted for by interindividual differences between patients. Efforts to develop effective and personalized depression treatment strategies are crucial, as early improvement is associated with higher response and remission rates. One strategy increasingly used to identify patient characteristics that might predict antidepressant response is the use of machine learning approaches. This study aimed to assess the predictive value of sociodemographic and clinical characteristics in reducing depressive symptomatology in a TRD population treated with oral esketamine. Clinical characteristics included depressive symptomatology and treatment resistance. Data from an open-label trial with a sample of 131 TRD patients who received individually adjusted dosages of oral esketamine, ranging from 0.5 mg/kg to 3 mg/kg, twice a week for six weeks were analyzed. The predictive performances of a linear regression, elastic net learner, and random forest models were compared to a featureless learner. The results showed that none of the models were able to predict change in depressive symptomatology above chance. This suggests that, within the scope of the selected features, oral esketamine may have similar effectiveness across the TRD population, regardless of levels of treatment-resistance. Future attempts to predict the treatment outcomes of esketamine should consider including a wider range of features and utilizing other analysis methods that counter small sample sizes and accounts for time-dependent interactions within systems.

Topics & conditions

Detected automatically by Psychedelic Commons from the title and abstract, not labels from the source. They may be incomplete.

Keywords