Introduction: Olanzapine is an established treatment for bipolar depression. Some authors also recommend olanzapine for melancholic depression. The growing observational OptiMA1 dataset permits the exploration and development of predictive models. Objectives: Exploration and testing of statistical prediction models as well as a priori bipolar and melancholic models of Olanzapine response. Methods: 375 participants with depression or anxiety disorders were recruited from outpatient mental health settings in Japan and New Zealand. Depressive symptom severity was measured using an online 30-item self-reported scale, included within a web-based system, Psynary, used by participants. Exploratory analysis was conducted using binary logistic regression models composed of: i. atheoretically selected, ii. bipolar, and iii. melancholic independent variables in three OpTIMA1-derived datasets with contrasting sample selection criteria and dependent variable definition of remission or response. Results: Binary logistic regression analysis produced bipolar models outperforming melancholic models in the first two analyses: remitters on olanzapine vs. remitters on other medication (n=159); remitters on olanzapine vs. non-remitters on olanzapine (n=56) respectively. No meaningful melancholic model could be computed for the third analysis: olanzapine responders vs. olanzapine nonresponders (n=51) with response defined as a 20% symptom reduction within 2 weeks. The second and third analysis produced strong predictive bipolar models with similar predictive power to the ideal statistical models. Conclusions: Measures of bipolarity accounted for around half of the variance in response to olanzapine. In contrast, melancholic features of depression explained little of the variance. These predictive models might be useful for predicting olanzapine response within the ecosystem used in this study. Disclosure: Dr Lee Andrew Kissane & Dr Richard Tranter are cofounders of Psynary. [ABSTRACT FROM AUTHOR]