Repeated LIFE ratings were available for 236 participants (a total of 14 045 observations). Because of the large N, statistical power was high enough to detect very small effects, and emphasis was placed on evaluating the effect size. This sample was 64% women (n
= 152) and the mean age was 41.2 (SD = 10.95) years; 221 (93.64%) met criteria bipolar I disorder, 200 (84.75%) participants were taking a mood stabilizer. As with other bipolar samples (19
), 110 reported a lifetime history of substance misuse. Also parallel with other major longitudinal studies (8
), 91.5% of patients reported at least some depressive symptoms, and 69.5% reported at least some manic symptoms during the follow-up period.
With mania LIFE-II scores as an outcome variable, Phi (the estimated within-subject correlation among mania scores) was 0.72. As shown in , the intercept term was significant, suggesting that the average mania score was above one. There was a small but significant effect for week; for each week followed, mania scores increased by an average of 0.005 points. For the effect of current depression on mania, there was a significant positive effect that was qualified by a negative curvilinear effect, as shown in . In contrast to the strong inverse relationship assumed by the unidimensional model, depression accounted for very little of the variance in manic scores. Simple effects and curvilinear effects of depression scores at 1-week and 2-week lags were unrelated to mania (all B’s < 0.04, all t’s < 1.66) and so are not displayed in . Medication scores also were unrelated to mania (all B’s < 0.07, all t’s < 1.44). The final model did not include these non-significant effects.
Random effects regression parameters
Predicted mania scores as a function of current depression levels.
To examine the effects of mania on depression, we computed a parallel model. Phi was 0.80. Focusing on effects of current mania on depression, there was a positive association that was qualified by a negative curvilinear effect. For the effects of mania at 1-week lag, there was a simple effect but no significant curvilinear effect. Considering mania at a 2-week lag, there was a significant positive effect qualified by a negative curvilinear effect. As shown in , findings were not consistent with the hypothesized inverse relationship. As above, effects of mania scores were very small and accounted for very little of the variance in depression scores (All B’s < |0.11|). None of the medication effects were significant.
Predicted depression scores as a function of current, 1-week lag, and 2-week lag mania levels.
Parallel analyses were conducted within subgroups. Confining analyses only to persons with bipolar I disorder did not change the pattern of findings nor did separate parallel analyses by gender.