As clinicians, we all recognize that many of the depressed patients we see do not adequately respond to our first treatment. What to do next for these individuals remains unclear, as there are few data from well-controlled replicated trials to guide us. In the absence of clear guidance from empirical research, we use what evidence is available, as well as case reports, our own clinical experience, and recommendations from experts.

This lack of evidence-based guidance was the primary motivation behind the development of The Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study. This large study, funded at considerable cost by the National Institute of Mental Health and years in the making, has begun to yield results. Available clinical trials, many sponsored by the pharmaceutical industry and conducted in research centers (academic or private), generally focus on uncomplicated, nonchronic major depression; in this respect, they often do not reflect the clinical "real world." STAR*D was designed to be relevant to clinicians by including patients more typical of outpatient practice. The patients are drawn from psychiatric and primary care "real world" settings, and the exclusion criteria are much less restrictive than usual. Patients with chronic depression and other psychiatric and medical comorbidities are included, and outcomes focus on remission (a Hamilton Depression Rating Scale score of 7 or less) rather than just a decrease in symptom severity.

In this issue of the Journal, Fava and colleagues detail the Level 3 results of STAR*D. Their article follows an article published in the January issue of the Journal on the Level 1 results (1), and two articles published in the New England Journal of Medicine in March detailing the Level 2 results (2, 3). We can now begin to put STAR*D into context and examine what clinical guidance these results provide.

First, a brief overview of the study is worthwhile. In Level 1, over 4,000 patients with nonpsychotic major depression were entered into the study and given citalopram at doses of up to 60 mg. The primary outcome was remission, a relatively asymptomatic threshold. This first level approximates typical clinical practice in that the initial intervention was a selective serotonin reuptake inhibitor (SSRI). Level 2 included those patients who did not remit or could not tolerate citalopram. They were offered the option of switching to another treatment, or augmentation of the citalopram that they were taking. The switching options were sustained-release bupropion, sertraline or venlafaxine extended-release, and the augmentation options were bu-propion sustained-release, buspirone, or cognitive behavior therapy (the cognitive behavior therapy results have not yet been published). Again, this level approximates clinical practice in that patients were able to express a preference for switching or for continuing citalopram and adding an augmenter. Level 3 included patients who did not achieve remission or could not tolerate the treatment that they received at Level 2. These patients were randomly assigned to mirtazapine or nortriptyline. This is the study conducted by Fava and colleagues, presented in this issue of the Journal. It is worthwhile to read each study in the order of publication, as there are many details not covered in this editorial that may affect how one interprets the importance of the data.

So what were the results, and what have we learned that we did not already know? The overall remission rate in Level 1 (all patients taking open-label citalopram) was 27.5%. No surprise here, as clinical trials with antidepressants typically report remission rates of 30% or less (4). "What have we learned that we did not already know?"

At Level 2, remission rates did not differ significantly among the three antidepressant switch strategies: 21.3% with bupropion sustained-release, 18.1% with sertraline, and 24.4% with venlafaxine extended-release. Level 2 remission rates were also similar across drug augmentation strategies: 29.7% achieved remission with citalopram plus bupropion sustained-release, and 30.2% achieved remission with citalopram plus buspirone, but buspirone was not tolerated as well as bupropion. Here, we do have some surprises and, perhaps, new clinical guidance.

In clinical practice today, it is popular to target our second-tier approaches at specific neurotransmitters (5). That is, if an SSRI does not help, try a drug that targets dopamine, norepinephrine, or some combination of serotonin, norepinephrine, and dopamine. Furthermore, there is some suggestive evidence that the dual uptake inhibitors (those antidepressants that inhibit norepinephrine and serotonin) are associated with somewhat higher remission rates (6). However, STAR*D suggests that our clinical practice of switching to another class of antidepressant may not be any more effective than switching to another SSRI. While this may be counterintuitive, and we would like to see replication of the finding, it does give some pause to our current thinking.

STAR*D may also suggest that augmentation is somewhat better than switching. The study was not designed to compare these two approaches directly, but it does appear that, in general, we may be better off augmenting than switching. These results are consistent with our current practice and expert guidelines. If someone has partially responded to an antidepressant, augmentation makes sense. Likewise, if there has been no response or if a patient has not tolerated the initial antidepressant, then switching makes more sense. Last, it appears that buspirone, as an augmentation strategy, was not tolerated as well as bupropion. Augmenting with buspirone is an approach that, in practice, is not as widely used as other strategies.

In Level 3, 114 patients were randomly assigned to mirtazapine (up to 60 mg/day) and 121 to nor-triptyline (up to 200 mg/d) for up to 14 weeks of treatment. Remission rates did not differ between the treatment options: 12.3% with mirtazapine and 19.8% with nortriptyline. In addition, the treatments did not differ in tolerability. In truth, there was a numerical difference in remission rates between these two approaches, but with relatively few patients during this phase, the two arms of the study did not differ significantly. We are left wondering whether the difference was mere chance or if we really should be using tricyclic antidepressants, such as nortriptyline, more than we currently do.

It must be said that STAR*D is a laudable endeavor. The study is the largest randomized clinical trial in depression ever done, and it was very well done. The enrollment was efficient and included a large number of minorities, which is a rarity in clinical trials. The data from this study will continue to be published and, beyond this first wave, we will continue to see results that may affect our practice. Some will question whether or not the cost was worth what we have learned. Since we cannot know what would have come from this money spent elsewhere, there is no answer to this question. We are, however, left wanting guidance that this trial alone cannot give us.

Clinical trials tell us how groups of patients do on average. For an individual patient, a particular treatment may or may not work. These trials do not tell us which patient will respond to which treatment; they merely suggest what treatment is most likely to be helpful. Of course, we all want to know what treatment is right for a particular patient. For example, choosing an antibiotic for an individual patient requires knowledge of the causative organism and its sensitivity to various antibiotics, but we are not yet at a point in our understanding of depression where our technologies allow for such a detailed examination. We do not understand the pathophysiology and heterogeneity of the biology of depression sufficiently. The limits of our science also apply to our intervention research.

An issue that bedevils the design of such a large, long-term study is that treatments change, so establishing an evidence base for popular treatments is chasing a moving target. For example, there is considerable interest today in other approaches not tested in STAR*D, such as atypical antipsychotics (7), mood stabilizers (8), stimulants (5), and targeting specific residual symptoms (9). Each of these strategies is supported in clinical trials, case reports, and expert opinion. Unfortunately, it was not possible to test these approaches in STAR*D, and we are left wondering where these strategies belong in our clinical approach to depression.

STAR*D is valuable because it examines, in a systematic way, response and remission in a sequential approach for those patients who do not remit with standard treatment. Furthermore, it is valuable because it avoids some of the limitations in generalizability and external validity of most previous studies. It has also added some novel findings to the literature from which we derive our clinical guidance, but in the end, this first wave of data from STAR*D will not greatly affect the prescribing practices of most clinicians. The study has confirmed much of what we suspected, based on industry and federally sponsored clinical trials, case reports, and expert opinion. STAR*D does not answer all of our questions, but progress in science is incremental, and STAR*D has added incrementally to the evidence base for the use of antidepressants. However, while it is asking too much of any study, we are left wanting more—more attention to other current options, more certainty, and, in the end, more guidance.