Illness Risk Following Rapid Versus Gradual Discontinuation of Antidepressants

We analyzed information from systematic clinical assessments of adult patients with a recurrent major affective disorder or panic disorder based on diagnostic criteria updated to DSM-IV-TR. All patients included in the study had responded well to antidepressant treatment and were evaluated, treated, and followed at the Lucio Bini Mood Disorders Center affiliated with the University of Cagliari in Sardinia. Diagnostic and assessment methods have been reported previously (29, 30). We excluded patients who were even moderately clinically depressed, anxious, or hypomanic at the time of medication discontinuation as well as those whose rate of discontinuation was uncertain (22.3% of potential antidepressant-treated candidates). For the present analyses, consecutive patients were included who met the following criteria: diagnosed clinically with DSM-based recurrent major depressive disorder, bipolar I or II disorder, or panic disorder; received a tricyclic antidepressant (or the tricyclic-like tetracyclics maprotiline and mianserin), a modern antidepressant (serotonin reuptake inhibitors or bupropion, duloxetine, or venlafaxine), or more than one antidepressant, with or without a mood stabilizer, following standard clinical practices regarding drug selection and dosing in the study community; recovered from an antidepressant-treated index episode of a major depression or panic disorder, based on clinical euthymia and a score ≤7 on the Hamilton Depression Rating Scale (HAM-D; 31) sustained for at least 30 days (including patients with panic disorder evaluated with the same rating scale for consistency); discontinued medication electively for clinical or personal reasons over a known period of time, allowing categorization into groups based on rapid (1–7 days) or gradual (≥2 weeks) discontinuation (none of the patients were tapered off antidepressants in the 8- to 14-day range); remained clinically stable or euthymic for at least 1 week after discontinuing treatment; and remained under prospective observation for at least 1 year, during initial treatment and through a first new episode of major depression or panic disorder that met DSM-IV diagnostic criteria at clinical assessment. Follow-up was censored at 100 months, based on the likelihood that discontinuation pace would exert little effect after the first few months (21–25).

All patients underwent initial diagnostic assessments, treatment, and follow-up evaluations by the same mood disorders expert (L.T.), based on semistructured interviews that followed the mood disorder components of the Research Diagnostic Criteria and Structured Clinical Interview for DSM-IV research assessment procedures (30), as well as extensive follow-up clinical assessments and repeated assessments with standard mood disorder rating scales during systematic prospective follow-up every 2–4 months. Diagnoses were updated to meet DSM-IV-TR criteria in 2008 and 2009.

Patients provided written informed consent for analyses presented anonymously in aggregate form. The project database and data management comply with U.S. federal Health Insurance Portability and Accountability Act regulations pertaining to confidentiality of patient records, and research use of the database for this study was approved by the Ethical Committee of the District 8 Health Agency of Cagliari (Azienda Sanitaria Locale-8), Sardinia, as well as the Investigations Review Board of McLean Hospital, Belmont, Mass. Required data were entered into a database (by C.G., B.L., and L.T.) in coded form to protect patient identity.

Treatment was determined clinically and included use of standard antidepressants, alone or with mood stabilizers (lithium carbonate or anticonvulsants that have received regulatory approval), atypical antipsychotics, or sedative-anxiolytic benzodiazepines as adjunctive medications. Use of such adjunctive treatments typically indicates greater clinical severity or lack of adequate response to antidepressant monotherapy (32). Among study patients, adjunctive psychotropic medications were continued unchanged after discontinuation of antidepressants. We standardized antidepressant dosages (total mg/day) based on the relative potency (ratio of median manufacturer-recommended doses) of the various agents (32), modified by median daily doses of specific antidepressants based on experience with nearly 2,000 patients with major depression treated at the study center so as to reflect local practice, to provide total imipramine equivalents in milligrams per day, including the sum of individual antidepressants when more than one were given simultaneously. We also considered the manufacturer-reported approximate elimination half-life of each antidepressant as a predictor of time to new episodes, using the longest elimination half-life when more than one antidepressant was used, or of a major active metabolite if its half-life differed substantially from that of the parent agent (e.g., norfluoxetine from fluoxetine) (32). In cases involving more than one course of antidepressant treatment, only data from the most recent trial were considered.

We compared demographic and clinical factors, selected a priori, between patients who discontinued antidepressant treatment abruptly or rapidly (1–7 days) and those who discontinued gradually (≥2 weeks). Factors considered included sex, family history of any psychiatric disorder (including substance abuse or suicide attempt), education level, marital status, current employment status, history of a substance use disorder, age at illness onset, total duration of illness, number of prior nonmanic episodes per year, approximate average time between the end of one episode and the start of the next, duration of the index episode of depression or panic, total HAM-D score, types of antidepressants received, total estimated dosages (in imipramine equivalents) of antidepressants received, approximate half-lives of antidepressants received, use of adjunctive psychotropic agents, duration of index antidepressant treatment, age at antidepressant discontinuation, and whether the decision to discontinue was made by the patient or by the treating physician (L.T.). The primary outcome measure was time to first new episode of a DSM-IV major depressive or panic episode, dated from the last day of antidepressant treatment.

We compared characteristics of patients undergoing rapid versus gradual discontinuation of antidepressants, using analysis of variance methods for continuous measures and contingency tables for categorical measures. We used Kaplan-Meier survival analysis to estimate latency to a first new episode of depression or panic, with Cox modeling of covariates associated with time to relapse or recurrence, using the Mantel-Cox log-rank test with both methods. Differences between patients who discontinued rapidly and those who discontinued gradually were also compared as rate ratios. Statistical analyses were conducted with Statview, version 5 (SAS Institute, Cary, N.C.), and Stata, version 8 (StataCorp, College Station, Tex.). The primary hypothesis tested was that latency to first new illness would be shorter after abrupt or rapid discontinuation than after gradual discontinuation.

The study sample of 398 patients included 261 women and 137 men. Their mean age at drug discontinuation was 41.9 years (SD=15.1), their intake for the index illness occurred a mean of 13.8 years (SD=11.6) from their first lifetime illness episode, and their prior mean (nonmania) recurrence rate was 0.37 episodes (SD=0.51) per year. During the index episode of major depression or panic, 62.6% of patients received tricyclic-like agents, and 37.4% received other modern agents, at a total mean dosage of 110 mg (SD=70) of imipramine equivalents per day, continued for an average of 8.48 months (SD=13.8) before discontinuation (Table 1), with a mean follow-up of 2.81 years (SD=3.63) after discontinuation. Decisions to discontinue treatment were clinical, not experimental; they were decided by the patient in 80.7% of cases and at the advice of the prescribing physician (L.T.) in 19.3% of cases. Gradual discontinuation was slightly more prevalent than rapid (53% compared with 47% of cases), and stopping abruptly or rapidly was almost always the patient's decision (94.1% of discontinuations). All patients included in these analyses were clinically well at the time of discontinuing antidepressant treatment and had no clinical evidence of even mild depression or anxiety disorder.

Compared with patients who discontinued antidepressant treatment gradually (N=210), those who discontinued abruptly or rapidly (N=188), in descending order of significance, 1) had a crude median time to first new episode of depression or panic that was shorter by 4.80 months; 2) were 4.4 years younger at illness onset and 4.6 years younger at antidepressant discontinuation; 3) had 0.15 fewer previous nonmanic recurrences per year; 4) were 12.0% less likely to have a family history of psychiatric illness; 5) had an index exposure to antidepressant treatment that was shorter by 2.96 months; and 6) were 16 times more likely to have made the decision themselves to discontinue treatment (Table 1). The discontinuation pace groups otherwise were similar, including in sex distribution, education level, marital status, employment status, lifetime history of suicidal behavior and substance abuse (including cigarette smoking), total years of illness, initial HAM-D score, months ill in the index episode before treatment started, duration of antidepressant treatment, and concomitant treatment with mood stabilizers, antipsychotics, or sedatives (Table 1). In addition, neither the type of antidepressant nor its estimated average total daily dose differed significantly between the two groups. Moreover, discontinuation pace did not differ by diagnosis or by type of antidepressant (modern versus tricyclic or tetracyclic).

Preliminary assessment of outcomes following rapid versus gradual discontinuation of antidepressants (Table 1) indicated that the crude median latency to first new illness with rapid discontinuation was 0.4 times the latency with gradual discontinuation (3.62 months [SD=8.78] compared with 8.42 months [SD=22.2]; t=20.8, df=1, 396, p<0.0001). Notably, this latency after rapid discontinuation (3.62 months) was one-fourth the estimated average previous interepisode interval in the same patients (14.9 months [SD=21.0]; paired t=5.26, df=1, 186, p<0.0001) and about one-fifth of the mean previous interval following gradual discontinuation (16.6 months [SD=30.5]; paired t=8.71, df=1, 208, p<0.0001). These differences suggest that gradual antidepressant discontinuation did not fully avoid or delay early illness episodes, or may indicate delay of previous episodes by prior treatment.

We then compared latency to first new depressive or panic episode among patients who discontinued antidepressant treatment rapidly versus gradually, using Kaplan-Meier survival analysis (Figure 1). The survival-computed median time to new episodes was 3.77 months (95% CI=1.99–5.55) after rapid discontinuation compared with 9.07 months (CI=6.78–11.4; Mantel-Cox χ²=20.8, df=1, p<0.0001) after gradual discontinuation—a 2.4-fold difference. Notably, the risk-by-time functions remained separated for more than 1 year (Figure 1) and for at least 100 months (not shown), indicating major reduction as well as delay of occurrence risk following gradual discontinuation of treatment.

With Cox multivariate modeling of survival functions, rapid drug discontinuation remained highly significantly associated with illness latency after adjusting for covariates. Covariates included factors likely to reflect illness severity—such as diagnosis, prior recurrence rate, concomitant treatment with a mood stabilizer or other drugs, duration of depression before the index treatment, and duration of the first new episode of illness—as well as aspects of treatment, including total daily antidepressant dose and duration of index treatment (not shown). Independent of antidepressant discontinuation pace (shorter latency after rapid discontinuation; hazard ratio=1.50 [95% CI=1.14–2.17]; χ²=38.3, df=1, p<0.0001), only the type of antidepressant (shorter latencies with modern antidepressants after both rapid and gradual drug discontinuation; hazard ratio=1.67 [95% CI=1.12–3.28]; χ²=15.5, df=1, p<0.0001) and the approximate drug elimination half-life (greater effect of discontinuation pace with drugs of relatively short half-life [latency ratios, 5.62 and 2.07]; hazard ratio=2.04 [95% CI=1.30–6.78]; χ²=11.4, df=1, p=0.0007) were associated significantly and independently with time to first new illness. Also, there was no overall effect of diagnosis, but the ratios of occurrence latency following gradual or rapid discontinuation were as follows, in descending order: bipolar I disorder (4.93/1.07 months=4.61); panic disorder (13.2/4.23=3.12); bipolar II disorder (5.43/2.17=2.50); major depressive disorder (7.60/3.17=2.40) (Table 2).