Among this sample of clinically depressed, sedentary, middle-aged, and older adults, 46% were fully remitted at the end of the original 4-month study treatment, and 66% were fully remitted 1 year after the end of treatment. In the original study, patients receiving exercise or sertraline achieved comparable benefits, which tended to be greater compared with placebo controls (10
). When a small subgroup of “early responders” (those 14 individuals who showed a ≥50% reduction in HAM-D score after only 1 week of treatment) were eliminated from the analysis, the contrast between active treatments and placebo controls went from 0.06 to 0.02. In contrast, initial treatment group assignment did not predict depressive symptoms at 1- year follow-up. However, self-reported exercise during the follow-up period was associated with lower depression scores and greater likelihood of improved depressive status at the time of follow-up, after adjusting for age, race, gender, prior history of MDD, and use of antidepressant medication at 1 year. We found a linear, inverse association between exercise level and depressive symptom severity between 0 minute and about 180 minutes per week of exercise, with a weaker association after 180 minutes, suggesting that the antidepressive benefit of moderate-to-vigorous exercise may diminish after an average of about 3 hours per week. The expected difference in HAM-D score between a person who reported 180 minutes of exercise and a person who reported 0 minute of exercise was 3.1 points, which is considered to be a clinically meaningful difference (26
To our knowledge, only four previous studies (27
) of exercise training reported follow-up data of depressive symptoms, each with important methodological limitation, including small sample sizes (28
), imprecise diagnosis of depression (29
), high drop-out rates (28
), or brief exercise interventions that failed to produce improvements in aerobic capacity (28
). Only one study examined the association between continued exercise participation during the follow-up period and longer-term outcomes (27
), and no study included a placebo control in the study design.
The observed association of posttreatment exercise and reduced depressive symptoms at the time of follow-up in the present study confirms the findings from our previous exercise trial in which patients who exercised over a 6-month follow-up period were 50% less likely to relapse (27
). Furthermore, the positive effects of posttreatment exercise did not vary by treatment group assignment, suggesting that exercise may be effective in maintaining treatment gains from an aerobic exercise intervention and may be an effective adjunctive treatment for patients who receive pharmacotherapy. However, it should be noted that, because of the naturalistic study design of the follow-up assessments, it cannot be determined if physical activity caused people to be less depressed or whether reduced depression caused people to be less sedentary and more likely to exercise.
Our exploratory analyses suggest that exercise may be particularly beneficial for depressed participants with elevated levels of anxiety. Younger participants and participants with greater perceived social support also may be more likely to sustain an exercise regimen over time.
The absence of an association between antidepressant use at the time of follow-up and MDD remission was unexpected, as other studies (7
) have reported that antidepressant use is associated with reduced risk for relapse. It is possible that participants in the present study who were depressed at the time of follow-up were more likely to be treatment resistant but were, nevertheless, motivated to use antidepressant medications in an effort to reduce their depressive symptoms.
Missing data are a study limitation, as 15% of our initial cohort were not available for follow-up, which may have biased these results. However, the statistical models that were estimated using imputed missing follow-up data yielded very similar results to models using completers-only data, mitigating this concern at least to some extent. Nevertheless, as with all nonrandomized studies, the general possibility of unmeasured confounders always exists.