In treatment of most chronic disorders, one of the factors that limits therapeutic benefit is poor compliance (also referred to as adherence or concordance) with prescribed management, including drug treatment (1–3). It has long been known that compliance with tricyclic antidepressants is poor in hospital outpatients (4) and in primary care (5, 6). The possible reasons for lack of compliance include the patient’s health beliefs (e.g., that depression is not a condition needing drug treatment), lack of knowledge about antidepressants (that they are addictive or can be stopped on recovery), and aversion to side effects (7, 8), as well as the doctor’s prescribing behavior.
The introduction of the selective serotonin reuptake inhibitors (SSRIs), with a different side effect profile led to hopes that compliance might be improved compared to tricyclic antidepressants. However, improved compliance with SSRIs has not been easy to demonstrate, for several reasons. In naturalistic studies, patients with previously poor compliance are selected for treatment with the newer compound, and tricyclic antidepressant prescriptions in primary care are often at subtherapeutic doses (9); both of these study conditions favor compliance with the tricyclic antidepressants (10).
In double-blind clinical trials, the analysis of dropouts cannot easily be generalized to clinical practice and has given conflicting results. Some studies have shown an advantage for fluoxetine (11). One large meta-analysis of studies of dropout for all SSRIs failed to demonstrate this advantage for the entire class of medications (12), although data on side-effect-related dropouts did appear to show an advantage for the SSRIs (13). The rigorous study procedures of the double-blind trials included in these meta-analyses may have minimized real-life clinical differences in compliance by creating ceiling effects. Furthermore, meta-analyses have lumped together older and newer tricyclic antidepressants whose characteristics may vary considerably (10).
Studies that have examined compliance with antidepressants as a primary objective have usually been carried out in hospital outpatients (4). The majority of prescriptions for antidepressants are made in primary care, but most studies conducted in this setting have been small (2, 14, 15), and almost all have relied on self-reports of tablet consumption by patients or on tablet counts by doctors. Recently, however, electronic devices that record the opening of the tablet container (Medication Event Monitoring System, Aprex, Union City, Calif.) have been developed. In a previous study we successfully used the device to show improved compliance after a brief intervention with medication counseling (16).
Against this background we carried out a randomized comparison of a tricyclic antidepressant and an SSRI in primary care, using three measures of compliance as the primary outcomes and change in depression severity and health status as secondary outcomes. The hypothesis was that patients with major depressive disorder would show better compliance with their medication regimen if they were treated with therapeutic doses of fluoxetine rather than therapeutic doses of dothiepin over 12 weeks. Dothiepin is a sedative tricyclic antidepressant that is widely used as a first-line antidepressant in the United Kingdom (9).
We conducted a multicenter, randomized, parallel-group, open-label comparison of dothiepin and fluoxetine at accepted therapeutic doses in patients with major depressive disorder in 10 primary care settings. All practices were in urban and suburban settings. In one practice, two primary care physicians referred patients for the study. The minimum number of patients enrolled in the study per physician was eight and the maximum was 28. All 11 physicians referred equal numbers of patients to the two treatment conditions.
The potential for investigator bias was minimized in two main ways. First, the physicians who prescribed the medications were blinded to treatment allocation by the use of sealed envelope inserts. The physician entered the patient’s number and initials and the date in spaces on the outside of a sealed envelope provided by the study monitors. The information was automatically transcribed onto carbon inserts within the envelope, on which the treatment allocated had previously been printed. Thus, no patients could be covertly removed from the study after allocation to treatment. Second, the questionnaires completed by the patients were forwarded directly to the study monitors so that compliance details remained confidential and were not disclosed to the physicians.
All physicians were trained in the use of the physician-rated scales, diagnostic criteria, and other procedures used in the study protocol. After the physicians gave a written description of the study to the patients, patients’ written informed consent was obtained. The patients were given the following information about the purpose of the study: "to compare how two different antidepressants suit different people. The antidepressants are called fluoxetine and dothiepin. They are both commonly prescribed by doctors to help people get better from depression. They work equally well." The patients were not told that the primary objective was to measure compliance. The local ethics committee sanctioned this approach.
Patients aged 18–70 were included in the study if they met DSM-III-R criteria for major depressive disorder, unipolar type (296.2 and 296.3), except for the criterion of suicidal ideation, as applied by the recruiting physician, and had a score of 12 points or higher on the Hamilton Rating Scale for Depression. This is the threshold at which a therapeutic response to amitriptyline could be identified in a U.K. primary care population (17). No record was kept of the number of such patients who were seen but not randomly assigned to a study group.
Patients were excluded from the study for any of the following reasons: 1) history of treatment-resistant depression (nonresponse to a single antidepressant at therapeutic doses for at least 6 weeks), bipolar disorder, organic brain disease, substance use disorder, use of antidepressants within the last 6 months, or participation in any other study within 3 months; 2) a medical contraindication to either drug; 3) for women, pregnancy, lactation, or not using contraception while of childbearing potential; 4) administration of any other psychotropic medication, and 5) serious suicide risk.
Materials and Dosage Procedures
Patients received either fluoxetine capsules (20 mg) or dothiepin tablets (75 mg). Patients returned to the prescribing physician for follow-up visits at 1, 2, 4, 8, and 12 weeks after the medication was prescribed. A drug dispensing log maintained by the prescribing physician was used to record information about the material dispensed and returned at each visit. Patients randomly assigned to receive dothiepin were treated with 75 mg/day at night for 1 week proceeding to 150 mg/day (2 × 75 mg at night) for subsequent weeks. Patients allocated to fluoxetine were treated with 20 mg/day in the morning throughout. Both final doses are accepted as therapeutically active. During the study, concomitant therapy with benzodiazepines was allowed for the shortest clinically acceptable time to facilitate sleep. Patients were allowed to withdraw from the study at any time.
The primary outcome measures were all dichotomized as above or below 80% of full compliance. Although this threshold is arbitrary, it has been used in studies of other classes of medications such as antihypertensives (18). The measures were as follows:
1. Pill count: At each visit (weeks 1, 2, 4, 8, and 12), the physician recorded the number of pills remaining in the bottle. Patients were categorized as compliant if they returned less than 20%, i.e., they appeared to be taking 80% or more of prescribed medication. This method measures what treating doctors could know about compliance if they routinely carried out this procedure in clinical practice.
2. Patient-completed questionnaire: At each visit the patients completed a questionnaire giving their estimate of the number of days they took no medication or took a reduced dose and the reason for not taking it. Patients were regarded as compliant if they reported that they had taken 80% or more of their medication as prescribed. This method measures what patients are prepared to report about the amount of medication they remember taking.
3. Medication Event Monitoring System: The Medication Event Monitoring System consists of a white medicine bottle with a cap that contains a microchip that is hidden from the patient. The chip records the date and time of each bottle opening. For this study, the microchip was read automatically by computer to produce indices of compliance at weeks 2, 4, 8, and 12. Patients were categorized as compliant if 80% or more of openings were within 4 hours of the recommended dosage time of 8:00 a.m. or 8:00 p.m. This method reveals when the bottle was opened but not whether or how many tablets were taken.
The high level of detail of the information about compliance provided by the Medication Event Monitoring System was thought to justify two secondary analyses. The underlying principle of both of these analyses was that a period of total noncompliance would be more damaging to treatment response than the same number of missed days scattered over the full treatment period. Seven days of abstinence was adopted as the criterion for identifying the end of a period of therapy compliance, based on a priori assumptions relating to the long half-life of fluoxetine. The day of the last bottle opening recorded by the Medication Event Monitoring System before such a period was the end of the "therapy-compliant period." The duration of the therapy-compliant period was calculated for each patient and used in two analyses: 1) a survival analysis of the number of patients in each group still in the therapy-compliant period for each day of the study, and 2) calculation of a "compliance index." The compliance index was calculated by expressing the therapy-compliant period as a proportion of the total intended treatment period, i.e., 84 days (the persistency element of the index). The proportion of days during the therapy-compliant period with one or more Medication Event Monitoring System openings was then calculated (the precision element of the index), and these two components were multiplied to obtain the single compliance index. For example, patient 59 first opened the container on day 2 then had a gap starting on day 14 and lasting to day 26. The length of the therapy-compliant period was therefore 12 days (day 2–13 inclusive). The persistency element of the index was therefore 12/84=14.3% of the length of the study. The number of days on which there was a Medication Event Monitoring System opening during this time was 11, so the precision element was 11/12=91.7%. The compliance index for that patient was therefore 14.3% × 91.7%=13.1%.
In addition to comparing the two treatment groups on these derived variables, we also calculated the correlation between the compliance index and the reduction in scores on the Hamilton depression scale (19) between study entry and the last visit.
Efficacy and safety evaluations.
Although the study was not designed as an efficacy comparison of dothiepin and fluoxetine, recovery from illness was assessed by using the Hamilton depression scale. Patients completed the scale at study entry, and at 2, 4, and 12 weeks. The 36-item Short-Form Health Survey Questionnaire, with nine subscales (20), used for rating health-related quality of life, was completed by each patient at baseline and at 1, 2, 4, 8, and 12 weeks.
Power Calculation and Data Analysis
In primary care, 42%–45% of patients prescribed antidepressants are believed to continue their medication for 12 weeks, according to patient self-report at confidential interview (2). At 12 weeks, assuming at least 50% noncompliance in the dothiepin group and 25% noncompliance in the fluoxetine group, the number of subjects per group estimated to be required for a power of 80% was 65 (p<0.05). No comparable studies were available in the literature at the time of writing the protocol. The anticipated effect size was therefore estimated from the known dropout rates in randomized controlled trials amplified to meet our expectations on the basis of normal clinical practice and prescription of the full therapeutic dose to the dothiepin group.
All patients who were randomly assigned to a study group were included in the efficacy and safety analysis on an intention-to-treat basis. Patients who failed to attend the study visits or overtly discontinued medication were recorded as noncompliant on all measures. The primary dependent variable for this study was patient compliance. The proportions of patients categorized as compliant or noncompliant by using pill counts, questionnaire responses, and the Medication Event Monitoring System were compared between treatment groups by using Mantel-Haenszel analysis, stratified by center. The secondary dependent variables for this study were reduction of depression and improvement of health-related quality of life. Analysis of variance, with factors of center and treatment, was used to compare the treatment groups’ mean Hamilton depression scale scores at study entry and final visit. The change in quality of life between baseline and the patient’s last visit was compared between treatment groups. Pearson correlation coefficients were used to assess the association between compliance and recovery. The compliance index was compared between treatments by using analysis of variance adjusted for center and treatment. The length of time without a gap in medication taking of more than 7 days was compared between groups by using a log rank test.
A total of 152 patients were enrolled in the study, all of whom were included in the intent-to-treat population. Seventy-six patients were randomly assigned to each treatment. In the fluoxetine group, the mean age was 37 years (SD=12.9), and 74% (N=56) were women. In the dothiepin group, the mean age was 40 years (SD=11.3), and 67% (N=51) were women. The two treatment groups were well-balanced with respect to the baseline scores on the 36-item Short-Form Health Survey Questionnaire and the Hamilton depression scale, although the dothiepin group had a notably higher mean score on the health transition scale of the 36-item Short-Form Health Survey Questionnaire at baseline compared with the fluoxetine group (mean=40.5, SD=23.1, versus 31.6, SD=21.0). This scale measures perceived change in health status or personal well-being.
A similar number of patients in each treatment group withdrew from the study after being randomly assigned to a treatment group (37%, N=28, in the fluoxetine group and 39%, N=30, in the dothiepin group) (χ2=0.11, df=1, p=0.74). The main reason for withdrawal was an adverse event (14%, N=11, in the fluoxetine group and 20%, N=15, in the dothiepin group) (χ2=0.74, df=1, p=0.39). One patient in the fluoxetine group and two in the dothiepin group gave lack of efficacy as the reason for withdrawal. Some data from the Medication Event Monitoring System were recorded for 138 patients, but the system showed that all medicine bottle openings for one patient had taken place before the baseline visit. This patient was excluded, leaving 137 patients (90%) for the analyses of measures of compliance derived from the Medication Event Monitoring System.
Sixty-seven percent of the fluoxetine group (N=51 of 76) and 58% of the dothiepin group (N=44 of 76) experienced an adverse event of some kind (χ2=1.38, df=1, p=0.24). Headache was more common with fluoxetine (21%, N=16, versus 7%, N=5, in the dothiepin group) (χ2=6.69, df=1, p=0.01), but the difference between groups in the rate of nausea just failed to be significant (17%, N=13, versus 11%, N=8) (χ2=1.38, df=1, p=0.24). Dry mouth was more common with dothiepin than fluoxetine (20%, N=15, versus 3%, N=2) (χ2=11.19, df=1, p=0.001). The rate of adverse events that were evaluated as drug-related were similar in the fluoxetine and dothiepin groups (46%, N=35, and 43%, N=33, respectively) (χ2=0.11, df=1, p=0.74), with fluoxetine again being associated with headache (17%, N=13) and nausea (14%, N=11), and dothiepin with dry mouth (20%, N=15).
In the following section the odds ratios refer to the ratio of results for the fluoxetine group to the results for the dothiepin group.
Data from pill counts showed that 76% of fluoxetine patients (N=58) and 64% of dothiepin patients (N=49) were compliant with treatment at or above the 80% level over the 12-week study (odds ratio=1.76, 95% confidence interval [CI]=0.9–3.5; Mantel Haenszel χ2=2.62, df=1, p=0.11).
Of the 146 patients who completed the confidential questionnaire 79% of the fluoxetine patients (N=59) and 80% of the dothiepin patients (N=57) were compliant at or above the 80% level (odds ratio=0.91, 95% CI=0.4–1.9; Mantel Haenszel χ2=0.06, df=1, p=0.80). Forty-five percent of the fluoxetine patients (N=34) and 32% of dothiepin patients (N=23) were compliant at the 100% level (odds ratio=1.69, 95% CI=0.9–3.2; Mantel Haenszel χ2=2.47, df=1, p=0.12).
The results from the Medication Event Monitoring System showed that 43% of the fluoxetine group (N=30) and 36% of the dothiepin group (N=25) opened the container within 4 hours of the recommended dose time on 80% or more occasions (odds ratio=1.40, 95% CI=0.7–2.8; Mantel Haenszel χ2=0.91, df=1, p=0.34).
F1 shows the results of a survival analysis of the percentage of patients who remained in the therapy-compliant period (before a gap in openings of 7 or more days) over the study period. The fluoxetine group had a longer therapy-compliant period (log rank test, χ2=4.32, df=1, p=0.04).
The compliance index (t1) gave adjusted mean compliance rates of 70.0% for the 66 patients in the fluoxetine group and 55.8% for the 66 in the dothiepin group for whom calculations could be made (95% CI for the difference=3.1%–25.4%; F=6.43, df=1, 130, p=0.01), a significant advantage to fluoxetine.
The fluoxetine group (N=72) had a nonsignificantly greater mean decrease in the Hamilton depression scale score compared with the dothiepin group (N=73) (difference=–2.34 points, 95% CI=–4.8–0.1, F=3.47, df=1, 134, p=0.07).
Patients’ responses to the 36-item Short-Form Health Survey Questionnaire showed greater improvement on fluoxetine only on the health transition scale (adjusted mean change=29.13 for the fluoxetine group [N=69] compared with 18.22 for the dothiepin group [N=68]; F=4.28, df=1, 126, p=0.04). This difference appeared to be due to the poorer score of the fluoxetine group before treatment; the final scores of the two groups were identical.
Association Between Compliance and Recovery
Using the simple 80% criterion on the Medication Event Monitoring System, we found no association between compliance and response on the Hamilton depression scale. Seventy-one of the 105 patients who were compliant (68%) and 22 of the 30 patients who were noncompliant (73%) responded to the medication. However, using the compliance index, we found an association between reduction in Hamilton depression scale scores and compliance; for the fluoxetine group the correlation was –0.42 (p<0.001), and for the dothiepin group it was –0.56 (p<0.001).
This randomized, controlled comparison of compliance with an SSRI and a tricyclic antidepressant is the first we are aware of to have been carried out in primary care, where the majority of prescriptions are written. Using similar methods to assess compliance, we previously showed that a brief psychological intervention (medication counseling) delivered by a trained nurse can improve compliance with a tricyclic antidepressant (16).
Strengths of the study reported here include its use of a mechanical medication event monitoring system and the recording of improvement rates in depressive symptoms at the same time. The study did, however, have an open-label design. This had the advantage of permitting patients to receive treatment from primary care physicians with a minimum of interference in routine clinical procedures, thus giving a more realistic indication of compliance in the clinical setting than a double-blind study. Furthermore, we did not consider it ethical to keep the physician and the patient unaware of the treatment for three reasons: 1) the objective of the study was partially withheld from the patients, 2) both treatments were already known to be effective, and 3) one medication was considerably more toxic in overdose than the other. This design complements the analyses of previous double-blind trials of medication efficacy. However, although precautions were taken, our study design could have allowed bias as a result of the treatment expectations of the physicians or patients.
By avoiding the double-blind design, we were able to recruit a study group that was reasonably representative of the population treated in primary care. Although we do not know the proportion of eligible patients who refused to participate, the study group appears to have been representative because overall compliance in this study was no greater than in naturalistic studies in the same setting. For example, Maddox et al. (2) used a prototype of the questionnaire used in this study and found that only 40% of a group treated in primary care with a range of antidepressants (equal numbers treated with tricyclic antidepressants and SSRIs) were still taking the medication after 12 weeks. A similar proportion of the original groups in this study were still attending study visits and taking medication at 12 weeks, suggesting that the study group is representative of the treated population.
The lack of difference between the two groups in the primary outcome variables suggests that we overestimated the effect size in our power calculation. A 25% difference in compliance was not confirmed by any of the primary or secondary measures we used. The largest differences between groups were of the order of 12% for pill count, 13% for complete compliance reported in response to the questionnaire, and 14% for the compliance index. These differences in the absolute risk of noncompliance suggest a number needed to treat of around seven to achieve one extra completed course of antidepressants by using an SSRI compared with a tricyclic antidepressant. The differences are somewhat larger than those reported for clinical trials in secondary care, which suggest an 8% advantage of fluoxetine over tricyclic antidepressants in the proportion of dropouts owing to adverse events (13). (We found a 6% difference between the fluoxetine and dothiepin groups in dropout owing to adverse events.)
If a real difference in compliance between fluoxetine and dothiepin exists, it may be due to side effects, half-life, or dosing schedule. In this study both compounds were given on a once-a-day schedule, but dothiepin was given at night and fluoxetine in the morning. The difference in half-life is unlikely to have advantaged fluoxetine, as it was not translated into a difference in dose schedule. The difference in side effect profile therefore seems the most likely explanation.
Associations between compliance and recovery are complicated because the reasons for noncompliance are associated both with treatment failure (adverse events, perceived lack of effectiveness) and treatment success (mistaken discontinuation after recovery). Possibly because of these complicated associations, two studies have failed to find a relationship between measures of compliance and outcome of depression. Demyttenaere et al. (21) used the Medication Event Monitoring System in a double-blind comparison of compliance with fluoxetine and amitriptyline in 66 depressed outpatients. Their measure of compliance was completion of the course of treatment. Only in those who completed the course was the Medication Event Monitoring System used to gauge the proportion of days the container was opened (called "adherence" by the authors). Interpretation of the results of this study is difficult because the measures do not allow an intention-to-treat analysis. However, Demyttenaere et al. found no evidence of significant differences in adherence between treatments and no evidence of an association between compliance and the outcome of depression. Amsterdam et al. (22) failed to find an association between plasma levels of fluoxetine, norfluoxetine, or their ratio and recovery from depression. The lack of association may have been due to the long half-life of fluoxetine, which would have obscured changes caused by 1–3 days of missed doses. This possibility suggests that the compliance index may be a more clinically meaningful measure of compliance than plasma levels. Nevertheless, compliance and recovery may both be related to a common underlying variable such as personality.
In conclusion, none of the primary outcome variables in this study showed a difference between groups, but two measures derived from the Medication Event Monitoring System—survival analysis and the compliance index—both showed an advantage to fluoxetine. Two possible interpretations of this result are that 1) the primary outcome measures were adequate, but no real difference between treatments exists or 2) the primary outcome measures were insensitive, possibly because of artificial dichotomization, and the secondary outcome measures provided a better reflection of a difference between treatments. Because there is reason to believe that compliance is continuously distributed rather than naturally dichotomous, we believe the latter explanation to be more likely. In addition, the measures derived from the Medication Event Monitoring System data—survival analysis and the compliance index—are appropriate for use in intention-to-treat designs, which are more acceptable for randomized controlled trials. However, the question of whether to accept the primary or the secondary outcome measures in this trial can only be answered definitively by a replication study using the compliance index and survival analysis applied to Medication Event Monitoring System data as an a priori hypothesis.
Received Feb. 23, 1999; revision received Sept. 8, 1999; accepted Oct. 14, 1999From the University of Southampton; and Eli Lilly and Company, Basingstoke, England. Address reprint requests to Dr. Thompson, Department of Mental Health, University of Southampton, Royal South Hants Hospital, Brinton’s Terrace, Southampton SO 14 0YG, England; firstname.lastname@example.org (e-mail). This study was carried out by Eli Lilly and Company in accordance with Good Clinical Research Practice guidelines. Professor Thompson was the study consultant throughout. The authors thank the primary care physicians and patients who took part in the study.
Percentage of Primary Care Patients With Major Depressive Disorder Treated With Fluoxetine (N=66) or Dothiepin (N=66) Who Remained Compliant With Treatment Before a Gap in Treatment Over 12 Weeksa
aNoncompliance was defined as a gap in treatment of 7 or more days and was based on measures of compliance derived from the Medication Event Monitoring System. Of the 137 patients with data from the Medication Event Monitoring System, five had only one opening of the tablet container and were excluded from the analysis, leaving 66 patients per group.