A clinical reappraisal of electroconvulsive therapy (ECT) was conducted over the last decade after the observation that a proportion comprised between 20% and 40% of either unipolar or bipolar depressions did not respond satisfactorily to antidepressants (1, 2), despite the availability of more effective competitors for treatment of depression, like the selective serotonin reuptake inhibitors (SSRIs). An estimate of ECT efficacy can be obtained through the meta-analytic method that combines information from independent studies. Meta-analyses can use dichotomous or continuous data, with each type of data presenting advantages and disadvantages (3). Meta-analytic reviews that use continuous data produce reliable standardized weight mean differences, however, a decrease in a psychiatric scale does not necessarily means a clinical remission. A dichotomous measure, such as clinical response or no response, has a considerable advantage over derived statistical parameters because it employs raw data from each individual patients, allowing to know, for example, patients' proportions which achieve response with ECT in comparison other therapeutic tools. The objective of our study was to analyze the efficacy of ECT in depression by means a meta-analytic review of controlled clinical trials published between 1956 and 2003, using dichotomous data.
This study reviews, by means of a MEDLINE search procedure using the key words electroconvulsive therapy and depression, all peer-reviewed publications in English from January 1966 to February 2003. We also manually searched articles published prior to 1966 that might be relevant for our purpose. Among these studies, we selected those comparing ECT with simulated ECT, or placebo or antidepressants drugs. ECT trials conducted without comparison group were excluded from our analyses. Selected studies were classified and separated in nonrandomized controlled trials and randomized controlled trials.
Like Janicak et al, (4) we used the studies in which it was possible to individualize each patient's response to treatment, using author's own criterion of response or no response. Basically, the response criterion was defined either as a reduction of at least 50% from baseline to end point on the Hamilton Scale for Depression (HAM-D) or a HAM-D score of 10 or less at the end point or a clinical judgment of "recovered" or "marked improved" depending on which of these 3 outcome measures were used. The category of "moderately improved" was not considered as a response criterion in this review. The diagnostic categories were major depression, bipolar disorder depressed type, schizoaffective disorder depressive type, and other categorizations such as neurotic depression, reactive depression, endogenous depression, involutional depression, primary depression and secondary depression. Only those studies in which we could directly determine each patient's response to treatment were included in the meta-analysis. Therefore, some randomized controlled trials were excluded because they did not reveal the results of patients individually (5—10).
The software used for the meta-analysis was the EasyMA 2001 of the Department of Clinical Pharmacology, Cardiology Hospital, University of Lyon (France) (11). The selected summary statistic and statistical method to our dichotomous end point, response or not response to the treatments, was respectively the odds ratio and the random effect model. The odds ratio has statistical advantages relating to its sampling distribution and its suitability for modeling. The random effects model assumes a different underlying effect for each study and takes this into consideration as an additional source of variation. In the random model, the test of heterogeneity applied was the Cochran Q-test with the objective to examine statistically the degree of similarity in the studies' outcomes. If a test of heterogeneity between trials is statistically significant (P < 0.05) then it may not be appropriate to combine the results.
REAL ECT VERSUS PLACEBO EFFECT (SIMULATED ECT OR PLACEBO)
Initially, we did not distinguish the studies that compared the ECT with the simulated ECT and placebo itself (pill), constituting a comparison group denominated placebo effect. As shown in Figure 1, 11 randomized controlled trials were used in the comparison between real ECT and placebo effect (simulated ECT/placebo), involving a total number of 523 patients, with a mean number of 48 patients by trial. A discrepancy in sample sizes among studies was found, with the largest trial involving 109 patients and the smallest 8 patients.
Responsive Rate of ECT, ECT Simulated and Placebo in Randomized Controlled Trials
Data from these studies that compared the efficacy of real ECT and placebo effect (Table 1), at the end of the treatment's course, revealed a significant better response of the real ECT (association χ2 = 19.76, df = 1, P < 0.001). The Q Cochran test of heterogeneity revealed that the null hypothesis of homogeneity between these studies is valid (χ2 = 12.13, df = 10, P = 0.28). The probability, in terms of odds ratios (OR), of a positive response with ECT is approximately 5 more times greater than with simulated ECT or placebo (OR 4.77; 95% CI 2.39, 9.49).
Randomized Controlled Trials of ECT Versus Placebo Effect
Considering the hypothesis that simulated ECT had a greater placebo effect than placebo (pill), due to a possible larger effect of the preparation process to ECT, we separated the comparisons between, respectively, real ECT versus simulated ECT and real ECT versus placebo. Seven randomized controlled trials (Table 2) with a total number of 245 subjects were suitable for meta-analysis of ECT versus simulated ECT and show a significantly greater effect of ECT as compared with simulated ECT (association χ2 = 6.87, df = 1, P = 0.0087). The heterogeneity (Q Cochran) between these 7 studies was not statistically significant (χ2 = 7.35, df = 6, P = 0.29). The chance of response with ECT when compared with simulated ECT is approximately 3 (OR 2.83; CI 95% 1.30, 6.17) more times greater than with simulated ECT.
Randomized Controlled Trials of ECT Versus Simulated ECT
Only 3 randomized controlled trials (Table 3), involving a total number of 266 patients, made possible a comparison between ECT versus placebo. This comparison revealed a significant more favorable outcome in the ECT group (association χ2 = 13.68, df = 1, P < 0.001). The Q Cochran test revealed no significant statistical heterogeneity between these 3 studies (χ2 = 2.71, df = 2, P = 0.26) and the odds ratio suggested a treatment response chance of approximately 11 (OR 11.083; CI 95% 3.10, 39.65) in favor of ECT when compared with placebo. One trial compared real ECT with simulated ECT plus placebo (pill) (17).
Randomized Controlled Trials of ECT Versus Placebo
ECT VERSUS ANTIDEPRESSANTS
In a first analysis we compared the ECT with the group of antidepressants in an indistinctive manner, including tricyclic antidepressants (TCA), monoamine oxidase inhibitors (MAOIs), lithium, and SSRIs. This comparison involved 13 randomized controlled trials, with a total number of 892 patients and a mean number of 69 patients for trial; the biggest trial used 242 subjects and the smallest trial involved 8 patients. Of these, 3 studies included 2 comparison groups each of which was analyzed separately (18, 19, 23). Therefore, we had 9 trials comparing ECT versus TCA (531 patients), 5 trials ECT versus MAOIs (438 patients), 1 trial of ECT versus SSRIs (39 subjects), and 1 trial of ECT versus the combination lithium-TCA (30 patients; Figure 2).
Responsive Rate of ECT and Antidepressants in Nonrandomized Controlled Trials.
Overall, the comparison of ECT versus antidepressants in general demonstrated a significant superior effect of ECT (association χ2 = 51.88, df = 1, P < 0.001). The test of heterogeneity (Q Cochran) demonstrated the homogeneity between these studies (χ2 = 11.45, df = 12, P = 0.49). The chance of response with ECT was about 4 times greater than with the antidepressant drugs (OR 3.72; CI 95% 2.60, 5.32; Table 4).
Nonrandomized Controlled Trials of ECT Versus Antidepressants
Separated comparisons between, respectively, ECT and TCA and between ECT and MAOIs were also performed. The analysis of ECT trials versus TCA trials revealed a significant greater efficacy of ECT than TCAs (association χ2 = 22.81, df = 1, P < 0.001). The Q Cochran test of heterogeneity between the trials was not significant (χ2 = 6.04, df = 8, P = 0.64). Patients receiving real ECT had nearly 3 more chance of a positive response than a patient that used TCA (OR 2.99; 95% CI 1.91, 4.71; Table 5).
Randomized Controlled Trials of ECT Versus TCA
As compared with MAOIs, the ECT showed a significant greater efficacy (association χ2 = 56.55, df = 1, P < 0.001) and the likelihood of a positive response with ECT was approximately 6 greater than with MAOIs (OR 6.13; 95% CI 3.82, 9.83; Table 6). A statistical homogeneity was present between the studies of this subgroup [heterogeneity χ2 (Q Cochran) = 1.47, df = 4, P = 0.83].
Randomized Controlled Trials of ECT Versus MAOI
ECT VERSUS ANTIDEPRESSANTS IN NONRANDOMIZED CONTROLLED TRIALS
We also made a systematic review of the nonrandomized controlled trials that compared ECT versus antidepressants. Seven nonrandomized controlled trials were used in this meta-analysis, revealing a large total number of patients (n = 2275) and a large mean number of patients by trial (n = 325; Figure 3).
Responsive Rate of ECT and Antidepressants in Randomized Controlled Trials.
Two nonrandomized controlled trials compared ECT with imipramine (30, 31) and the other 5 studies compared ECT with various antidepressants (TCA and MAOIs) in adequate doses. These studies confirmed the superiority of ECT in comparison with antidepressants even in clinical settings (association χ2 = 26.77, df = 1, P < 0.001), with a similar common odds ratio (OR 2.84) and statistical homogeneity between the trials [heterogeneity χ2 (Q Cochran) = 7.77, df = 6, P = 0.26; Table 7]. The nonrandomized controlled trials results fell within the confidence interval of the randomized controlled trials (95% CI 1.91, 4.21).
Randomized Controlled Trials of ECT Versus Antidepressants
In the present meta-analysis, as compared with the recent review of The UK ECT Review Group, (37) we used a different statistical strategy identifying each patient's response to treatment. Depressive symptoms were assessed by a dichotomous end point. Furthermore, we adopted more stringent criteria than those used by Janicak et al. (4). In fact, we considered as responsive only those patients who presented "marked improvement" or "recovery," whereas we excluded from the analyses those who were reported to have "moderate improvement." Additionally, we conducted a large meta-analysis of observational studies that compared ECT versus antidepressants, which included 2272 patients.
RANDOMIZED AND NONRANDOMIZED CONTROLLED TRIALS
The results of our meta-analytic review of the randomized controlled trials revealed a significant superiority of the ECT in all comparisons: ECT versus placebo effect, ECT versus simulated ECT, ECT versus placebo, ECT versus antidepressants, ECT versus TCA and ECT versus MAOIs. Additionally, the meta-analysis of the observational studies that compared ECT versus antidepressants presented similar results, reinforcing the assumption that systematic reviews of observational studies and randomized controlled trials usually produce similar conclusions (38). The fundamental criticism toward observational studies is that they have inherent biases. On the other hand, the approach that includes only randomized controlled trials not always ensures relevancy in the reviews even though it minimizes biases (39).
Inside the group of randomized controlled trials we increased the relevancy of the meta-analysis combining all anti-depressants in a single group and the simulated ECT and the placebo itself (pill) in another group.
However, more specific comparisons between ECT versus other treatments taken singly (TCA, MAOIs, simulated ECT, placebo itself), even with minor relevancy, revealed significantly greater chance of treatment response with ECT as compared with MAOIs, placebo and, to a less extent, TCA. The larger effect of simulated ECT as compared with placebo (pill) suggested that the process of ECT preparation might have a greater impact on patient's outcome than placebo itself; however, definitive conclusions cannot be obtained because only 3 studies comprised a group of patients treated with placebo (pill).
Since 1985, few controlled trials compared the ECT with others treatments or placebo. The only study that compared ECT versus SSRIs in patients with treatment resistant depression showed a significant higher response rate of ECT (28). Nevertheless, more studies with larger samples are needed to confirm the superiority of ECT over SSRIs in general for the treatment of resistant depression. No studies, to our knowledge, compared ECT with other classes of antidepressants (SNRI, NASSA, NRI).
Some limitations inherent the meta-analytic method must be acknowledged: problems in the randomization processes used in controlled trials, publication biases, variation of standard treatments over time and heterogeneity of studies (39, 40). Second, ECT studies were in general heterogeneous for multiple aspects. For example, we did not take into account possible variations in the ECT techniques and procedures adopted in the various studies. In fact, most studies did not specify electrical parameters and type of equipment adopted. Furthermore, we did not discriminate between studies using unilateral ECT from those using bilateral ECT.
The reason why we did not distinguish between unilateral and bilateral ECT stays in some controversies regarding which of the 2 techniques is most efficient method of electrical induction of the ECT (41—43). Some studies pointed out that the efficacy of unilateral ECT is simply a question of adequate dosage levels; with high-dose unilateral ECT and bilateral ECT present equivalents response rate, with the advantage that unilateral ECT produces less anterograde and retrograde memory deficits (44—49).
Another consideration regarding the variation of standard treatments is the use of traditional ECT (constant-voltage modified sine-wave stimuli) or modern ECT (constant current brief-pulse). The physiological efficiency of the brief-pulse device is more optimal than the sine wave, but comparisons between the traditional sine-wave ECT and the bilateral suprathreshold modern ECT revealed that the clinical improvement were virtually identical for the 2 methods (50—52).
Another caveat for interpreting data on ECT, regards the diagnostic heterogeneity of samples used in the various studies, which included diagnoses such as neurotic depression (22, 31, 53), depression with psychotic symptoms (20, 22), melancholic depression (29), treatment-resistant major depression (28), and schizoaffective disorders, depressed type (33—36). There are contradictions in the analysis of the association of specific symptom profiles with ECT outcome. Early observational studies found that endogenous or melancholic depression were predictive of greater response to ECT than "neurotic depression"; however, subsequent trials did not reveal a difference in ECT response between patients with versus without melancholia (54). A combined analysis of randomized controlled trials of ECT versus simulated ECT showed that real ECT had a therapeutic advantage, specifically among patients with delusions and/or retardation (55). Nevertheless, in 2 randomized controlled trials, involving 143 patients, Sobin et al (56) investigated the utility of depression subtypes in predicting ECT response and concluded that ECT is a treatment option for patients with major depression regardless of the presence of psychotic features, retardation and/or agitation. More responsible of the variation of results among studies on ECT can be, respectively, the different instruments used to measure the reduction of the depressive symptoms, the difficulties to maintain some research team blind to this therapeutic method, the number of ECT sessions applied and the methodological weakness of some studies that did not specify the electrical parameters of the bilateral ECT and/or unilateral ECT. Moreover, the proportion of patients who previously failed adequate antidepressant medication trials could impact on rates of response to ECT (57—59).
In conclusion, by this meta-analysis we tried to analyze systematically available scientific information on ECT trials to provide a reliable estimate of ECT efficacy for depression. Data analyzed suggest that ECT is a valid therapeutic tool in the armamentarium for depression, including severe and resistant forms. Future studies are needed to clarify whether and when such an intervention can be a first choice treatment of some patients.