A new model analyzes sleep and circadian rhythm data from wearable devices to predict depressive, manic, and hypomanic episodes in mood disorder patients.