STUDY PROCEDURES

During the baseline visit, a detailed psychiatric and medical history was obtained, and the following standard rating scales were performed: Structured Clinical Interview for DSM-IV screening questions for current mania and depression, Young Mania Rating Scale13 (11-item structured interview version), Hamilton Rating Scale for Depression14 (31-item structured interview version), investigator- and patient-rated Clinical Global Impression scale,15 the Global Assessment Scale,10 and a brief adverse-effect rating scale. The rating scales were repeated during office visits at weeks 2, 4, 6, 8, 12, and 16. Because of a presumed delay in the therapeutic effects of 3 fatty acids, a priori criteria mandated that subjects remain in the study for 30 days or more to be included in the analysis. Identical gelatin capsules containing concentrated 3 fatty acid ethyl esters or placebo (olive oil ethyl esters) were obtained from the Fish Oil Test Materials Program, a joint research program of the National Institutes of Health and the National Marine Fisheries Service. Each capsule of 3 fatty acid concentrate contained 440 mg of eicosapentanoic acid (C20:5,3) and 240 mg of docosahexanoic acid (C22:6,3), which was vacuum deodorized and supplemented with tertiary-butylhydroquinone, 0.2 mg/g, and tocopherols, 2 mg/g, as antioxidants. The source of the 3 fatty acids was menhaden fish body oil concentrate.

Subjects were randomized by the Brigham and Women's Hospital Research Pharmacy to receive either 3 fatty acid treatment or placebo. The randomization was stratified according to sex, the presence or absence of concurrent lithium treatment, and the presence or absence of rapid cycling. Subjects received 7 capsules twice daily, for a total daily 3 fatty acid dosage of 6.2 g of eicosapentanoic acid and 3.4 g of docosahexanoic acid. Patients randomized to placebo also received 7 identical capsules twice daily. A relatively high dosage of eicosapentanoic acid and docosahexanoic acid was used, because similar doses have been safely and effectively administered in other disease states. Furthermore, because of the lack of data regarding the effective dosage of 3 fatty acids in mood disorders, a relatively high dosage was chosen to avoid a potentially ineffective low dose. Blood levels of 3 fatty acids were not monitored in this trial.

OUTCOME MEASURES

The main outcome measure chosen a priori was the duration of time to exit double-blind treatment because of symptoms of bipolar disorder of sufficient severity to warrant a change in medication. Specifically, patients ended their participation in the study and treatment was considered to have change in medication. Specifically, patients ended their participation in the study and treatment was considered to have failed if mood symptoms emerged, or continued beyond 30 days in patients who were not euthymic at baseline. Hence, duration of time in the study represented an overall measure of treatment efficacy. The two blinded principal investigators (A.L.S. and L.B.M.), in collaboration with each patient, were responsible for the decision whether to end a patient's participation in the study. Secondary outcome measures were the results of the Young Mania Rating Scale, Hamilton Rating Scale for Depression, Clinical Global Impression, and Global Assessment Scale ratings, before and after treatment.

STATISTICAL ANALYSIS

A power calculation was performed before the study to determine the appropriate sample size. Assuming a large effect size, we calculated that 60 patients (including dropouts) would be sufficient to demonstrate a difference between the 2 arms at 90% power with an .05.

The study was originally intended to include 60 randomized patients, each for 9 months of double-blind treatment. However, an unexpected cessation of production by the National Marine Fisheries Fish Oil Program led to a shortage of material. Simultaneously, a preplanned, blinded, interim analysis performed when 20 subjects had either failed treatment or completed 4 months suggested significant differences between the groups. The combination of these 2 factors led us to end accrual and then reanalyze the data after 30 patients had either failed treatment or completed at least 4 months of follow-up. A standard sequential design would prescribe looking for a P value of .02 or less to signal significance on the first interim analysis, and a P value of .04 or less to signal significance on the final analysis. Because of the 2 factors cited above, the results in this study fall between the interim and final analysis, and the P value designating significance could be taken conservatively as .015 or liberally as .042. A Kaplan-Meier "survival" analysis (Mantel-Cox log-rank statistic; df=1) was used to compare the duration of remission in the 2 groups. The rating scale scores on the last day of the study for each patient were used as the "final" data points (last observation carried forward). Categorical variables were analyzed by means of the Fisher exact test. Continuous variables were examined with the nonparametric Mann-Whitney test. Statistical significance for the primary outcome measure was set at <.01 (2 tailed).

Forty-four patients were randomized, but only 30 had evaluable data, based on the a priori criteria for inclusion. Four subjects dropped out before the 1 month point because of noncompliance with the study protocol (n=2), gastrointestinal tract side effects (n=1), or concern over the possibility of receiving placebo (n=1). The remaining 10 subjects had not yet reached the 4-month end point required for the main outcome measure when the trial was ended and therefore were not included in the analysis.