Abstract
All medications currently used to treat schizophrenia, which exert their therapeutic effects by inhibiting dopaminergic neurotransmission, have their greatest efficacy against the positive symptoms of schizophrenia but have limited impact on negative symptoms and cognitive deficits, core symptoms that robustly predict outcome. Recent research, which has implicated glutamatergic neuronal dysfunction in a subgroup of subjects with schizophrenia, has given rise to the development of several experimental glutamatergic medications. While Phase III clinical trials have not shown significant group effectiveness of these drugs, some subjects were reported to exhibit substantial reductions of symptoms. Identifying such a subgroup prior to drug testing would permit more targeted design of Phase III clinical trials and could lead to more personalized prescription of drugs to treat schizophrenia, especially its core symptoms. Using data from two failed Phase III clinical trials (N = 163 and N = 235) of the experimental glutamatergic drug pomaglumetad methionil (an mGluR2/3 agonist) and applying a gradient-boosted machine learning algorithm, we identified novel, pre-treatment EEG biomarkers that predicted responders with accuracy rates over ninety percent. These constellations of EEG markers predicted pomaglumetad responders prior to treatment in comparison to standard-of-care antipsychotic treatment, indicating that they are specific to pomaglumetad and do not represent a marker for response to antipsychotic treatment generically. The effects were seen with positive and negative symptoms as well as cognitive deficits. The method described could be applied to identify likely responders to other mechanistically novel psychotropic medications in schizophrenia and other neuropsychiatric disorders.