Design
We conducted a randomized controlled trial in which adult primary care patients with minor or mild-major depression were randomized to receive either usual care plus 3 months of antidepressant treatment (UCandAD) or usual care alone (UCnoAD). Because we were interested in the treatment effectiveness in everyday practice, we decided to conduct a pragmatic trial, implying that the interventions were provided by PCPs to typical primary care patients under normal practice circumstances. UCandAD was our treatment of interest, the experimental treatment. We did not want to compare antidepressant medication with placebo medication as such a treatment is not a feasible alternative to medication in daily practice. Instead, our control intervention, usual care, was based on the guideline on depression issued by the Dutch College of General Practitioners [
12]. We hypothesized that UCandAD was as effective as, i.e. equivalent to, UCnoAD. Therefore, we designed an equivalence trial. The investigation was carried out in accordance with the latest version of the Declaration of Helsinki [
14]. The Medical Ethical Committee of the VU University Medical Center approved the study design.
Participants
The study was conducted in 2002 and 2003. PCPs in the west and middle of The Netherlands were invited to participate in this trial. During a practice visit the PCPs were informed about the study's aim and procedures. Participating PCPs recruited, diagnosed, and treated consecutive eligible patients themselves. Patients were considered eligible if they had been diagnosed by their PCP as suffering from a current episode of minor or mild-major depression (i.e. three to six out of nine DSM-IV symptoms of depression, including at least one of the core symptoms, 'sadness' or 'loss of pleasure'). In accordance with the Dutch guideline on depression we defined minor depression as a depressive disorder with three to four DSM-IV depressive symptoms [
12]. Largely in accordance with the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV [
15]), we defined mild-major depression as a depressive disorder with five to six symptoms. The symptoms had to be present nearly every day for at least 2 weeks. Also, they had to cause clinically significant distress or impairment in social, occupational or other important areas of functioning.
Patients were excluded from the trial for the following reasons: age under 18 years, currently on antidepressant medication, currently receiving psychological therapy, experiencing psychotic features, addiction to alcohol or drugs, loss of a loved one or significant other in the past 6 months, pregnancy or breastfeeding, inability to complete questionnaires because of language difficulties, illiteracy or cognitive decline, or not having a telephone.
When the PCP considered a patient to be eligible for the study, a research assistant made an appointment for a baseline interview at the patient's home. At the start of the interview, the patient received a full explanation of the study and written informed consent was obtained.
Randomization
Patients were randomly assigned to one of the treatment conditions. Block randomization (block size 4) was used to ensure equal numbers of patients in the two conditions per PCP. Allocation schemes were generated by random number tables before the trial started. After the baseline interview with the patient, staff not in contact with the patient opened the appropriate opaque sealed envelope. PCPs and patients were not informed about the allocated treatment until the first treatment session. Due to the nature of the study, blinding of patients, research assistants or PCPs was not possible.
Interventions
Before the start of the trial, PCPs received a 3-h training session to improve recognition of patients with minor and mild-major depression, to make a standardized DSM-IV depression diagnosis, and to refresh their treatment skills.
All patients were scheduled for four 10–20 min consultations with the PCP at 2, 4, 7, and 11 weeks after inclusion. During these consultations, patients randomized to the usual care condition (UCnoAD) received patient education, information about depression and its prognosis, and advice on how to deal with depression (focus on the present, maintain social activities and daily routines, exercise, putting achievable goals, and restrain alcohol usage). This treatment may be considered as a standardized form of usual PCP care. Patients in this condition were not to receive any antidepressant medication.
Patients randomized to the antidepressant condition (UCandAD) received the selective serotonin reuptake inhibitor (SSRI) paroxetine beside the usual care treatment described above. The paroxetine dose was 20 mg/day for at least 3 months, along with education about the effects and side-effects of paroxetine. After 4 weeks the dose could be increased to 40 mg/day in case of poor clinical response. In case of intolerance to paroxetine, sertraline was advised. Paroxetine was chosen as first choice antidepressant because it was the most commonly prescribed antidepressant drug in The Netherlands at the time the study started [
13], and no clinically meaningful differences between antidepressants are found in primary care patients [
16].
In both conditions, prescription of benzodiazepines was allowed in patients with severe sleeping disturbances. After 3 months, treatment could end or continue in the way PCPs and patients preferred.
In equivalence trials, it is important to optimize the contrast between the two treatment conditions to assure a conservative approach. Therefore, we gave PCPs some instructions on how to shape both treatments in such a way that the two interventions would differ mostly with respect to the use of antidepressants. During the first 3 months, PCPs were asked to give patients in both conditions the same number of consultations and not to deviate from the medication protocol unless the PCP judged this to be imperative. To prevent contamination from co-interventions, PCPs were asked to refrain from referral to specialized mental health care (e.g. to a psychiatrist, psychologist, psychotherapist, or a social worker).
Assessments
Sociodemographic and clinical information was collected at baseline, comprising the List of Threatening Experiences Questionnaire (LTE-Q [
17]), neuroticism (a subscale of the NEO-FFI [
18]), duration and history of depressive symptoms, and chronic somatic diseases. As a check of the diagnoses of the PCP, but without consequences for the inclusion in the study, standardized psychiatric diagnoses according to DSM-IV criteria were obtained with the Composite International Diagnostic Interview (CIDI [
19]). In addition, we collected both patients' and PCPs' treatment preferences before randomization.
Secondary outcome measure
The Short-Form 36 (SF-36 [
22,
23]) was completed by patients during the baseline interview and was sent to the patients' home at 6, 13, and 52 weeks follow-up. The SF-36 is a self-report quality of life measure, consisting of eight scales that are aggregated into two summary measures: the Physical (PCS) and Mental (MCS) Component Summary scores. Lower scores indicate worse health, higher scores indicate better health.
The Client Satisfaction Questionnaire (CSQ-8 [
24], self-report scale, eight items, item range 1–4, higher scores indicating higher satisfaction with the treatment received) was sent to the patients' home at 13 and 52 weeks follow-up.
Regarding the secondary outcomes we were only interested in possible differences between treatment groups, and thus not in demonstrating equivalence.
Statistical analysis
Using the sample size calculation of Jones et al for equivalence trials [
27], based on 90% power (1-β) to detect a clinically relevant difference in improvement of -5 points on the MADRS (α = 0.05, two-sided), 84 patients were required in each group.
To estimate the CIs for the mean difference between the groups, we used parameter estimates of repeated measures multilevel analysis in MLwiN [
28]. This method is characterized by an unrestricted repeated measurements design, allowing all observations to be used. Multilevel analysis was used for both the intention-to-treat analyses and the per-protocol analyses. To study change in MADRS scores, baseline measurements were used as a covariate as the improvement in depressive symptoms may be different for patients with higher versus lower MADRS baseline scores. In all analyses, all available time points were analyzed simultaneously.
Additionally, an adjusted model was analyzed. We adjusted for additional specialized help from mental health services during the experimental period, which was considered a co-intervention, to check the robustness of our outcomes.
In equivalence trials it is recommended to perform a per-protocol analysis in all cases; intention-to-treat analysis is no longer considered conservative [
27,
29]. When equivalence is demonstrated in both the intention-to-treat and the per-protocol analysis, the evidence is considered to be strongest. We performed both intention-to-treat and per-protocol analysis. If equivalence of the two treatments could not be established, we explored superiority of either treatment. As intention-to-treat analyses has been recognized as the most conservative strategy for superiority trials to analyse data, we focused at the intention-to-treat analysis. First statistical significance was explored (95% CI < 0 or 95% CI > 0), then clinical superiority was explored (95% CI < -5 or 95% CI > 5).
In the intention-to-treat analyses all patients were analyzed according to group assignment. Per-protocol analyses were performed for patients who received at least 70 defined daily doses (DDDs) of an antidepressant drug in the first 3 months when in the UCandAD group, and patients who received no more than 30 DDDs of an antidepressant drug in the first 3 months when in the UCnoAD group. Thus, protocol violation has a different meaning in the treatment groups; violators in the UCandAD group received a less extensive treatment (no antidepressants), whereas violators in the UCnoAD group received a more intensified treatment (additional antidepressants).
Differences in protocol violation between treatment groups were analyzed using the chi-square test (SPSS 11.0). To find predictors of protocol violation logistic regression was used for each treatment group separately. The influence of gender, age, the presence of life events, duration and history of depressive symptoms, chronic diseases, neuroticism, severity of depression diagnosis at baseline (i.e. minor versus mild-major depression, as assessed by the PCP), and use of mental health services during the first 3 months were explored.
In addition, some explorative analyses were performed. Differences in benzodiazepines prescription between treatment groups and differences in protocol violation in combination with patient's treatment preferences and assignment were analyzed using the chi-square test in SPSS 11.0 (SPSS Inc., Chicago, Il).
The secondary outcomes, the SF-36 and the CSQ-8, were explored for statistical significant differences in the intention-to-treat analyses. We used multilevel analysis in MLwiN [
28] for the SF-36. For the CSQ-8, total mean item scores were computed and the t statistic in SPSS 11.0 was used (two-sided, p < 0.05).