Objective: Characteristics of difficult-to-treat depression (DTD), including infrequent symptom remission and poor durability of benefit, compel reconsideration of the outcome metrics historically used to gauge the effectiveness of antidepressant interventions. Methods: Self-report and clinician assessments of depression symptom severity were obtained regularly over a 2-year period in a difficult-to-treat depression registry sample receiving treatment as usual (TAU), with or without vagus nerve stimulation (VNS). Alternative outcome metrics for characterizing symptom change were compared in effect size and discriminating power in distinguishing the vagus nerve stimulation + treatment as usual and treatment as usual treatment groups. We expected metrics based on remission status to produce weaker between-group separation than those based on the classifications of partial response or response and metrics that integrate information over time to produce greater separation than those based on single endpoint assessment. Results: Metrics based on remission status had smaller effect size and poorer discrimination in separating the treatment groups than metrics based on partial response or response classifications. Metrics that integrated information over the 2-year observation period had stronger performance characteristics than those based on symptom scores at single endpoint assessment. For both the clinician-rated and self-report depression ratings, the metrics with the strongest performance characteristics were the median percentage change in symptom scores over the observation period and the proportion of the observation period in partial response or better. Conclusion: In difficult-to-treat depression, integrative symptom severity-based and time-based measures are sensitive and informative outcomes for assessing between-group treatment effects, while metrics based on remission status are not.