Many people who receive treatment for major depressive disorder (MDD) do not respond to pharmacological treatment, meaning several drugs may have to be trialled to find a successful therapy.
In a study presented at the 56th Annual Meeting of the American College of Neuropsychopharmacology in December 2017, researchers from Harvard University examined data from electronic health records on 49,322 patients with MDD who had a successful treatment, characterised as a treatment with one of nine common antidepressants that was repeated at least twice in 12 months with no change in treatment
The team used machine learning to categorise patients into subtypes based on characteristics such as age, presence of sexual dysfunction and obesity, and then generated a model to link these subtypes to the likelihood of an antidepressant’s efficacy. They found that the model was able to predict treatment efficacy with a high level of accuracy (AUC [area under the curve]: 59–71%).
The team said the predictive model could help to reduce the trial-and-error approach to finding an effective antidepressant in MDD.
 Doshi-Velez, F. From electronic health records to treatment recommendations for depression. Presented at the 56th Annual Meeting of the American College of Neuropsychopharmacology; 4–7 December 2017; Palm Springs, California. Abstract available here: https://www.nature.com/articles/npp2017263 (accessed February 2018)