In this report we applied a descriptive approach and focussed on remission rates and its stability for a series of depressed subjects for a period of 3-years.
Cumulative remission rates
From that perspective the outcome looks quite promising: 67 % of the LOCF sample and most patients of the OC (88 %) sample recovered at some time point during three years (cumulative remission rate). These cumulative rates are in a comparable range to other naturalistic long-term follow-ups. Holma found 88.5 % after 5 years, O’Leary 88 % after 3 years and Ramana 80 % after 2 years [4, 6, 7]. Also the landmark study by Keller about the naturalistic 5-year course of 431 subjects with major depression found cumulative recovery rates (defined as 8 consecutive weeks with no or minimal symptoms) of 70 % after 1 year, 81 % within 2 years, 87 % within 4 years and 88 % within 5 years [8].
Absolute response and remission rates
In contrast to cumulative rates, absolute remission rates at a certain follow-up time tend to be considerably lower. With respect to remission rates after one, two and three years, the LOCF analysis revealed 55 %, 56 % and 58 % of remitters in the present study. With respect to observed cases after one, two and three years, almost 62 % completer after one, (59 %) after two and (69 %) after three years met criteria for remission (OC).
The naturalistic Vantaa sample comprised 163 outpatients (OC sample) and applied DSM-IV criteria over a continuous period of two months [4]. After five years 50 % of the observed cases were in full remission in this follow-up study.
The MADRS remission rates of observed cases (MADRS < 9) of the naturalistic SLICE study on a primary care population (n = 1031) after 1- (70.7 %) and 2-years (75.3 %) were higher than in the present report [22]. But it needs to be considered, that a primary outpatient sample usually includes less severe and less refractory patients in comparison to a tertiary care inpatient population as in the report at hand [23].
The PROSPECT study reported HAMD-17 remission rates of the observed cases of an intervention group receiving algorithm-based interventions of 40.1 % after one and of 49.7 % after two years although the mean HAMD baseline severity (18.1 vs. 24), was only in a moderate range [24].
Thus, all in all the rates of the present report are in a similar range to other naturalistic data with a tendency towards the upper range, despite its tertiary referral infrastructure. An important limitation in that context is the high drop-out rate of 82 % leading to a selection of patients with favourable outcomes (also see limitations).
Although “real world” patients and patients included in randomized controlled trials are not easily comparable [25, 26], a look into long-term data of a recent randomized controlled trial, might still be informative. For example in the Co-Med trial remission rates after 7-months were lower with 48 % (LOCF). But here the observational period was shorter and the remission criterion stricter (patients had to be in a remitted state on 2 consecutive visits) [27]. Another example is the PREVENT trial, here the remission rates of this double blind randomized controlled long term trial comparing fluoxetine against venlafaxine (N = 268) were higher at year one (67–68 %) and (71 %–77 %) at year two (OC) [28].
Stability of remission and illness course
The stability of remission for LOCF and OC analysis is illustrated in flow chart diagrams (see Figs. 2 and 3). Due to the high drop-out rate, the OC analysis seems to be the most reliable one in that respect (see limitations).
The highly fluctuating course of major depressive disorder is reflected by 52 % of patients (OC) showing a fluctuation from remission to non-remission and vices versa throughout the three years. This result is in line to the previously published corresponding relapse rates which have been retrospectively assessed at each follow-up. Of the 458 patients 155 (33.6 %) experienced at least one severe relapse during the 3-year follow-up period. The highest rate was found in the first month and the first year (25.3 %) after discharge from inpatient treatment declining to 16.1 % two years thereafter [10].
A finer grained picture can be obtained by only looking at the switch rates from remission to non-remission. These rates vary between 11 % and 25 % with 45 % loosing remission at some time point during the three years. Only 36 % of all patients (and 55 % of initially remitted patients) stayed in remission during the whole time (OC). The recent naturalistic PREDICT-NL study followed 174 primary care outpatients with major depressive disorder (out of 1338 attendants) for a period of three years. In line with our results the authors found a rate of 40 % with a fluctuating course and rate of 43 % staying in remission right from the start. The benign rates can again be well explained by the milder and less complex cases of a primary care outpatient population. However, still 17 % of patients in the PREDICT-NL study had a chronic course and stayed in an episode for the whole 3 years [26]. In the report at hand, 12 % of all discharged patients stayed in non-remission from discharge up to 3-years and thus had a chronic course (OC, Table 3). This level of chronicity is also in accordance with earlier reports. Keller reported that 12 % patients of the CDS study did not reach recovery after 5 years [9]. Likewise Jules Angst reported in his 21 year follow-up of 406 initially hospitalized patients that 13 % of all patients developed a chronic course [27]. Spijker found a slightly higher rate of 20 % of patients with MDD who had not fully recovered after a period of two years in the NEMESIS study [29].
Treatment
After three years 70.4 % of the completer sample received at least one antidepressant. A recent systematic review including 14 large observational naturalistic/epidemiologic surveys reported adherence rates ranging from 30–97 % (median 67 %) [30]. The comparably high psychotherapy rate as well as the high rate of patients being in specific mental health care (84 % year one, 81 % year two and 78 % year three), together with a higher chance of adherent patients staying in the completer sample might have led to high adherence rates in the completer sample. In addition it should be kept in mind that, the German health care system provides healthcare insurance for all community members, allowing a free choice of doctor. Therefore, the encouraging adherence rates can partly be traced back to specialized mental health care advanced by the German insurance policy.
Limitations
The most important limitation pertains to the high dropout rate. Attrition rates in long-term studies of similar time spans, range from 18 % [31], to 72 % [32, 33]. The Texas algorithm project showed similar attrition rate after one year of 47 % [34]. Thus, the three-year attrition rate of the present study of about 82 % of the patients entering the follow-up appears to be within a high range.
Amongst others, one reason for the high attrition rate in the study at hand may have been due to the way participants during the follow-up had to be contacted in accordance with the study protocol. Participants were only allowed to be contacted via letters and not via telephone or email. Thus, over the three years twenty mailing waves have been performed. In addition, no telephone interviews were intended for the yearly visits but face to face interviews had to be performed instead. Although these interviews might result on the one hand in a higher data quality, they may led to higher attrition rates on the other hand. For future studies contact via Email and especially telephone calls, which have shown to increase retention time in trials are clearly preferable [35].
The drop out analysis revealed that patients staying up to three years in the follow up exhibited variables that are associated with better longterm outcome (female gender, less comocrbid personality disorders, living more often with partner, less often discharged against medical advice, longer inpatient treatment time, lower Hamilton baseline score, (Table 1)).
These variables are largely in line with the variables associated with lower attrition in the STAR*D Study and the Texas algorithm project [34]. Patients of the drop out sample showed significantly lower remission rates in step one and two of STAR*D [34, 36]. Thus the outcome of the OC sample of the report at hand are if anything optimistic and it seems likely that we have overestimated positive outcome[37, 38].
Nevertheless, this sample of 3-year long-term data of initially hospitalized, naturalistic treated and extensively reevaluated patients with major depression is still one of the largest European follow-ups.
Since there were broad inclusion and only few exclusion criteria, patients who would have been excluded in most randomized controlled trials, were included in our study. This is a limitation and strength at the same time. The results of this study could be more generalizable to routine clinical practice and exert high external validity for adult inpatients with major depressive disorder. On the other hand internal validity is reduced due to the lack of any control group. Therefore conclusions regarding treatment effects are very limited. In addition these results might not be easily generizable to outpatient populations or elderly patients [39].
We also strictly focused on remission on certain points in time without applying any duration thresholds. Several authors suggested that a patient should at least remain in remission for at least 8 consecutive weeks before he can be considered as recovered [2, 9]. Moreover, since there was no course interview implemented, we cannot rule out that a patient who is in remission at all three visits may have experienced episodes or more illness activity in between the visits. Thus our rates of remission are, if anything, optimistic. Modern outcome measures like ecological momentary assessment techniques (EMA) could provide promising tools which might complement such traditional outcomes measures.
Thus, in summary the results of the 3-Year completer sample, even if positively biased, are rather disappointing with 12 % showing a chronic course, 52 % a highly fluctuating course and only 36 % percent a stable remission and call for future strategies enhancing long-term outcome.
A comment on missing data in long-term trials - Last observation carried forward (LOCF) and observed cases (OC)
Valid analyses of longitudinal data are complex and difficult, especially if data are missing for reasons that are related to the outcome. We used two traditional methods to address this problem: 1) the LOCF method which imputes data by carrying the last observation forward and 2) the completer or observed case analysis by only including those patients who had an observation at each visit and at endpoint. It is often argued that the bias in LOCF leads to a “conservative” (under-) estimation of treatment effects. On the other hand, the completer analysis or observed case analysis is assumed to lead to an overestimation of treatment effects. Therefore a combined approach can help to get a realistic idea of the outcome rates.
In the present study LOCF indeed estimated absolute remission at the predefined time points rates more conservatively but overestimated the stability over the long term course due to the high rate of patients lost to follow-up. For example stable remitters were higher in the LOCF analysis than with the OC method (43 %, LOCF vs 36 % OC). In contrast, the observed case analysis showed less conservative absolute outcomes (e.g. yielding higher remission rates after 1, 2 and 3 years) but allows a more realistic look on the development over the long term course.
In recent years modern imputation methods like mixed models for repeated measures are becoming more and more common and are often recommended as preferable method. On the other hand it has been emphasized that they are not the cure for all problems associated with missing data [40, 41].
