Prevalence and prediction of dropout during depression treatment in routine outpatient care: an observational study

Despite the existence of efficacious treatments for major depressive disorder (MDD), premature treatment discontinuation (dropout) by patients restricts their effectiveness [1]. Treatment dropout should be distinguished from treatment refusal, where patients fail to start the intervention that was offered which is another type of problem with different rates and moderators [2]. Dropout is not only associated with worse short- and long-term clinical outcomes for patients [1] but also has a negative impact on their social functioning and care providers [3, 4]; and impedes effective use of scarce resources in mental health [5]. Regardless of slightly different definitions, most studies report that dropout occurs primarily early in treatment [6‐8]. Early identification of expected treatment dropout might reduce occurrence thereof and increase treatment effectiveness.

Dropout rates for psychotherapy in depression treatment are estimated to be around 20% (range 0–74%) [6, 9]. Rates for pharmacotherapy may vary from around 30% in clinical trials to as high as 60% in naturalistic settings [10‐12]. Many efforts have been made to identify factors that are associated with dropout of treatment, with inconsistent findings. The demographic characteristics race (non-Caucasian), less education and lower socio-economic status (SES) are regularly associated with more dropout [5, 6, 9, 10, 13‐15]. The results of studies examining the relationship between other demographic characteristics such as age, marital status and gender are less convincing. Male gender [16‐19], younger age [9, 13, 16‐20], and not being in a relationship [15, 17] have been associated with dropout. However, results are inconclusive according to others [6, 14, 21]. These inconsistencies also apply to the relationship between clinical features and the occurrence of dropout. The impact of higher baseline severity on the occurrence of dropout for example has not been consistently established, ranging from protective [5, 8, 22‐24] or no clear impact [13] to risk enhancing [19, 21, 25‐27]. Recurrent depression is found to be associated with lower dropout rates [28], whereas chronic depression may not affect the occurrence of dropout [21]. Co-morbidity such as personality disorders and eating disorders [6, 9, 29] are considered to be risk enhancing by some, but other studies describe no or even a protective effect of personality pathology [13, 21]. Inconsistencies are also apparent in studies on the influence of comorbid anxiety in MDD treatment outcomes including dropout [13, 30, 31]. Treatment-related variables like higher patient (pre-) treatment expectancies, patients receiving their preferred treatment and a good therapeutic alliance as well as a more experienced therapist are associated with less dropout. [9, 32‐35].

Studies directed at establishing treatment effectiveness reported higher dropout rates (26%) in comparison to randomized studies designed for treatment efficacy (17%) [9]. In fact, the majority of empirical knowledge on dropout may not be useful for daily clinical practice, being derived from randomized controlled trials (RCT’s). Hans & Hiller (2013) conducted a meta-analysis of non-randomized effectiveness studies of CBT for depression and reported a dropout rate of 24.6% with a wide range between studies (0–68%). Therefore, dropout from treatment in naturalistic treatment settings should be examined separately from investigations in traditional RCT oriented research contexts such as university affiliated settings [36, 37].

Identification of risk factors for dropout in routine care using a feasible measure will be helpful to prevent its occurrence and increase effectiveness of depression treatment. In the current retrospective study, we aimed to quantify risk factors for treatment dropout in a large naturalistic sample of outpatients with depression receiving treatment in a routine treatment facility in secondary care. We hypothesized that dropout rate would exceed 20% as typically found for effectiveness trials. Next, we hypothesized that known risk factors for treatment resistance (i.e., severity and duration of the depressed episode) would be associated with more dropout. Moreover, in line with mainly meta-analytic evidence, we expected male gender, younger age, lower social economic state, and the presence of a comorbid anxiety or personality disorder to be positively associated with dropout.

In this study, we examined prevalence and risk factors for dropout in routine care in a large sample of outpatients with depression receiving routine outpatient depression treatment. We found a dropout rate of 14.5%. Higher DM-TRD total scores were predictive for lower odds for dropout, although it was not possible to determine a cut-off score to predict accurately whether or not an individual would dropout. When analyzing separate risk factors for dropout, higher SES, higher severity of the depression and the presence of comorbid personality pathology and a comorbid anxiety disorder at baseline were associated with a decreased risk of dropout in our sample.

The prevalence of dropout in our sample was lower compared to the overall prevalence rates of 20–25% as reported in recent meta-analyses [6, 9, 36]. However, it should be noted that these meta-analyses identified very wide ranges in observed dropout between 0 and 50–74%. As such, the dropout rate in our study appears low but in line with results from earlier work in RCTs and other naturalistic samples [22, 27, 58]. It is unknown whether our setting and therapeutic approaches play a role in our dropout rate. Our participating centers are part of a large secondary care mental health organization, a type of setting associated with lower dropout rates than academic or university affiliated centers [9]. More detailed comparison with other studies is impossible given the differences between definitions of dropout that are being used in the literature. We chose to define participants as dropouts when they had attended only 3 or less treatment sessions. Although some of the patients with MDD may benefit from psychotherapy within a few sessions [46], the majority of patients needs at least four sessions to achieve an adequate outcome [47, 48]. Additionally, previous work showed that, in the Netherlands, an acute phase pharmacological treatment for depression typically entails five or six visits to a psychiatrist [59]. It can be expected that raising the required ‘at least 4’ sessions or defining dropout as 'not attending any session after diagnostic assessment' would have increased dropout rates erroneously.

The DM-TRD is an easy-to-use clinician rated instrument combining clinical characteristics with information about previous treatments for the current depressive episode. We found that higher total scores on the DM-TRD were associated with less dropout. This was somewhat unexpected because the DM-TRD is designed to assess the risk for treatment resistance, and consists of items that score previous failed treatments for the index-episode, as well as clinical risk factors for treatment resistance including severity and duration of the episode and comorbid psychopathology. We hypothesized that higher scores on a composite of these items would be associated with less motivation for treatment and more pessimism about treatment success and as such expected an association with higher dropout rates. Our results, however, point in the opposite direction; one could speculate that higher levels of treatment resistance at baseline increase motivation and presumably the need for treatment in an otherwise unfavorable course of illness.

In line with previous findings [14, 15] higher SES was associated with less dropout in our sample. Inconsistencies about the association of dropout with demographical features age, gender and marital state that emerge from the literature (e.g., [6, 9, 13, 17]) may explain why these features were not predictive for dropout in our sample. In addition, this result can be attributed to the fact that the other variables contributed more to the dropout variance than these demographic characteristics.

Contrary to our hypothesis, the clinical variables severity at baseline and duration of the depressive episode were negatively or not associated with dropout. Comparison of this finding with meta-analyses that examined treatment dropout is difficult because baseline severity and episode duration were not assessed [6, 9, 36, 60]. As previously described, the association between severity and dropout is not unambiguous (e.g., [13, 23, 27]). The finding that more severe patients were less likely to drop out is in line with several studies [8, 22, 23]. A possible explanation for this finding is that this is a result of the way care is delivered in our setting. As proposed by Fernández et al. [8], a mismatch of treatment intensity with depression severity appears to be a causative factor for dropout. This is supported by the fact that patients with more severe depression are more likely to drop out from a self-guided online treatment intervention [16]. Our institution offers treatments from a multidisciplinary team consisting of psychiatrists, psychologists, nurses and other mental health professionals. It is therefore conceivable that in our setting a good match has been made with the needs of the patient, resulting in less dropout in the more severely depressed patients. In line with Machado et al., there was no association between more chronicity and the occurrence of dropout in our sample [21].

Previous meta-analyses showed higher dropout rates in patients with personality disorders [6, 9], while other studies showed that comorbid personality pathology had no negative effect on dropout rates in the treatment of depression [13, 21]. In our sample, the presence of comorbid personality pathology according to the SAPAS, but not the presence of a personality disorder as established by a structured clinical interview as the SCID-II, was associated with a lower risk for dropout. A probable explanation for this is that the patient's responses to the items of the questionnaire were negatively influenced by the severity of the depression due to its negative effects on self-perception (the state-effect) and memory processes [61, 62]. The reported personality pathology should thereby perhaps be seen as a proxy for the severity of the depression, which in turn was associated with lower dropout rates in our sample. The fact that there was no association present for an actual personality disorder according to a clinical interview supports this conjecture. Another relevant aspect is that the type of personality pathology appears to be relevant for the occurrence of dropout. For example, the presence of compulsive personality traits can have a protective effect on the occurrence of dropout [19, 21]. Since cluster C is the most common cluster of comorbid personality pathology in MDD [63], this could also explain the association of comorbid personality pathology with less dropout in our sample. Unfortunately, due to the way personality pathology was assessed in our sample, it was not possible to accurately distinguish between the various clusters.

We found that the presence of a comorbid anxiety disorder but not the presence of anxiety symptoms was predictive for lower dropout rates. This finding is not in line with previous studies reporting higher dropout rates in depression treatment in subjects with comorbid anxiety disorders [13, 30]. This inconsistency may be explained by the fact that combination treatments are mainly offered in our treatment setting. Combination therapy showed less dropout than pharmacotherapy alone in subjects with comorbid anxiety disorders [13]. In addition, some studies found that patients with MDD and comorbid anxiety disorders can experience a faster recovery from psychotherapy [29, 64]. Since patients who showed less symptom reduction during treatment for depression were more likely to drop out [1, 29], these factors may be related to the finding that a comorbid anxiety disorder is predictive of less dropout in our sample.

The divergence for both comorbidities may also be explained by the specialized secondary healthcare settings of PsyQ, where patients are referred for depression treatment specifically. When personality pathology or an anxiety disorder is strongly present, specialized treatment is offered aimed at these specific disorders. Moreover, therapists within these specialized treatment settings are used to offering MDD treatment to patients with comorbidities. By adapting the way the MDD treatment is offered by these therapists to the needs of the patient, the risk of dropout can be reduced, as suggested by Swift and Greenberg [65].

Several limitations need to be mentioned. First, although the naturalistic setting is an overall strength, this limited the possibility to obtain more information concerning the specific types of treatments, adherence and treatment integrity. We also do not have information on the content of the planned treatment and the reasons for dropout. It cannot be ruled out for certain that in some cases the treatment termination was a joint decision between patient and therapist. However, we do not consider this likely, as the population had severe complaints (QIDS-SR ≥ 11) that are not expected to remit within 1–3 sessions. In addition, subjects who had 1–3 therapy sessions, but whose follow-up measurement on the QIDS-SR showed remission, were defined as “justified quitters” and were therefore not considered to be dropouts. Unfortunately, we also do not have information on the course of the dropouts after the 6-month time frame that our study focused on. It is conceivable that some of these dropouts eventually returned with a renewed request for help in the same or another treatment setting. However, we lack this information and we think this could be a relevant question for future studies on this topic. Second and related, this lack of information on the content of the treatment and the reasons for dropout prevents a more detailed analysis of an association between treatment type and dropout, and their potential interaction with sociodemographic variables; this would be relevant given the difference in dropout between psychotherapy and pharmacotherapy [10, 60]. Third, it may be that an unmeasured variable (e.g., more previous episodes [28]) in more severe and chronic cases drives the association with less dropout. More recurrent episodes may make subjects realize treatment is necessary from past experience. Likewise, information on other relevant patient, treatment and therapist-related variables was absent [9, 32, 33, 35]. Previous work has shown that, apart from demographics and clinical characteristics, a variety of factors may contribute to dropout, although results are again not in complete accordance. These factors include patient and treatment-related variables such as patient’s treatment preference, (pre-) treatment expectancy, treatment setting, treatment format, manualized versus non-manualized treatment, treatment duration and therapeutic alliance, but also therapist-related information such as gender, age, patients’ treatment preference, experience level, and the therapist’s ability to repair alliance ruptures [9, 32, 33, 35, 66, 67]. Unfortunately, this information was not available. Fourth, the presence of personality pathology (in the absence of a SCID-II) was determined in our sample using a concise questionnaire, the SAPAS. Diagnostics based on a questionnaire can lead to over-reporting of personality pathology compared to diagnostics based on a clinical interview [63]. It is thereby conceivable that the way personality pathology was assessed in our sample had led to an overestimation of this comorbidity. In addition, it was not possible to distinguish between the specific types of comorbid disorders, both within the personality pathology and within the anxiety disorders. Fifth, although our sample was large, a substantial number of subjects could not be analyzed due to lacking baseline QIDS-SR scores according to our definition of 30 days around diagnostic work-up. In addition, it is conceivable that the lack of a baseline measurement is already indicative of the occurrence of dropout. On the other hand, there were no clinically relevant differences at baseline parameters between the analyzed sample and the subjects that were missing QIDS-SR scores. Fifth, it is conceivable that, due to missing follow-up QIDS-SR measurements, a part of our subjects was falsely qualified as “dropout”. However, given the natural course of MDD [68] and given the baseline severity of our subjects, it is not very likely that a substantial part of these patients would have recovered without adequate treatment.

In this observational study dropout from depression treatment was relatively low in comparison to previous studies and meta-analyses. We found that the DM-TRD was useful to assess dropout risk: higher total scores on this instrument were associated with a lower dropout rate. Several clinical variables from the DM-TRD, i.e., higher socio-economic state, higher severity of the depression and the presence of comorbid personality pathology and comorbid anxiety disorders, were significantly associated with a lower dropout risk. The DM-TRD may, therefore, be helpful to clinicians to assess risk factors for dropout before treatment commences.