Non-pharmacological treatment for depressed older patients in primary care: A systematic review and meta-analysis

Background Late-life depression is most often treated in primary care, and it usually coincides with chronic somatic diseases. Given that antidepressants contribute to polypharmacy in these patients, and potentially to interactions with other drugs, non-pharmacological treatments are essential. In this systematic review and meta-analysis, we aimed to present an overview of the non-pharmacological treatments available in primary care for late-life depression. Method The databases of PubMed, PsychINFO, and the Cochrane Central Register of Controlled Trials were systematically searched in January 2017 with combinations of MeSH-terms and free text words for “general practice,” “older adults,” “depression,” and “non-pharmacological treatment”. All studies with empirical data concerning adults aged 60 years or older were included, and the results were stratified by primary care, and community setting. We narratively reviewed the results and performed a meta-analysis on cognitive behavioral therapy in the primary care setting. Results We included 11 studies conducted in primary care, which covered the following five treatment modalities: cognitive behavioral therapy, exercise, problem-solving therapy, behavioral activation, and bright-light therapy. Overall, the meta-analysis showed a small effect for cognitive behavioral therapy, with one study also showing that bright-light therapy was effective. Another 18 studies, which evaluated potential non-pharmacological interventions in the community suitable for implementation, indicated that bibliotherapy, life-review, problem-solving therapy, and cognitive behavioral therapy were effective at short-term follow-up. Discussion We conclude that the effects of several treatments are promising, but need to be replicated before they can be implemented more widely in primary care. Although more treatment modalities were effective in a community setting, more research is needed to investigate whether these treatments are also applicable in primary care. Trial registration PROSPERO CRD42016038442.


Results
We included 11 studies conducted in primary care, which covered the following five treatment modalities: cognitive behavioral therapy, exercise, problem-solving therapy, behavioral activation, and bright-light therapy. Overall, the meta-analysis showed a small effect for cognitive behavioral therapy, with one study also showing that bright-light therapy was effective. Another 18 studies, which evaluated potential non-pharmacological interventions in the community suitable for implementation, indicated that bibliotherapy, life-review, problemsolving therapy, and cognitive behavioral therapy were effective at short-term follow-up. PLOS  Identification and selection of studies To be as comprehensive as possible, we decided not to restrict the searches to randomized controlled trials (RCTs). Results of other study types (e.g. cohort studies) were used to identify promising therapeutic strategies subject to future research. Therefore, we included all empirical studies that met the following criteria: (a) sample sizes !5 patients; (b) depression as the primary outcome; (c) a study population of adults aged !60 years at the moment of inclusion (or there were adequately reported sub-analyses of adults !60 years); (d) was conducted in a primary care or community setting; and (e) reported non-pharmacological treatments applicable in these settings. We set no language or date restrictions. Depression was defined as either an identified depressive disorder according to DSM or ICD criteria determined by a validated diagnostic interview or instrument, or as an elevated score on a screening tool. Since there is no known golden standard for the identification of depression in later life, we decided to include all studies focusing on depression, regardless of their depression inclusion criterion. The age cut-off of 60 was used, because this is the mostly used cut-off for late-onset depression [21]. In addition, the earlier review regarding treatment of depression in primary care [16] also included studies focusing on adults aged 60+.
Studies were excluded if they met the following criteria: (a) included bipolar disorder, psychotic depression, or depression with suicide ideation, which are considered indicative for referral to secondline treatment [22]; (b) focused on caregivers instead of patients; (c) studied the effect of a non-pharmacological intervention as an adjunct to pharmacotherapy; (d) studied the effect of service-level intervention, such as collaborative or stepped care; or (e) studied the effect of an intervention to prevent depression.
Studies were independently screened and selected for inclusion by two authors (FH and BM). First, titles were screened to exclude irrelevant papers, and the remaining abstracts were then scrutinized in detail. Of the potentially relevant papers, full texts were retrieved to determine whether the inclusion criteria were met. In cases of disagreement, consensus was reached based on discussion and, if necessary, consultation with a third author (PV). We searched for additional articles by studying published study protocols lacking published follow-up data, by checking the reference lists of the included publications and of relevant systematic reviews and meta-analyses [17][18][19][20], and by screening conference abstracts. If necessary, corresponding authors of possible relevant papers were contacted. the different studies, when interpreting the results, and when recommendations for future studies were formulated.
Data extraction. Two authors (FH and BM) independently extracted data from all included studies. The following data were extracted: year of study, study design, sample size, setting (primary care, community), population characteristics (age, gender, comorbidity), treatment type and characteristics (e.g. individual/group, number of sessions), diagnosis at baseline and diagnostic tool, main result, percentage that declined participation, duration of follow-up, percentage lost to follow-up, and percentage that adhered to treatment. Included studies were classified to the setting in which they were conducted, namely primary care setting or community setting. Studies were considered primary care studies if the study recruited participants in primary care and the intervention was delivered in that setting. Studies were considered community studies when participants were recruited from the community, for example, by means of self-referral.
The results were then summarized into the following three categories: (1) mean change, defined as the difference in depressive symptoms between baseline and follow-up measurement; (2) responders, defined as a !50% symptom reduction in the outcome measure between baseline and follow-up (unless stated otherwise); and (3) remission from depression at followup measurement. The definition of remission differed between studies. The mean change in depressive symptom scores was the primary outcome of this review.
Analysis. We narratively reviewed the included studies by type of treatment, with the results stratified by setting (primary care or community). Given that most studies in primary care have focused on the effect of cognitive behavior therapy (CBT), we chose to perform a meta-analysis for the effect of this intervention in the primary care setting. Because of expected heterogeneity of the studies, a random-effects model was used to pool the effect of CBT on depression. We calculated the standardized mean differences (SMD) using the mean scores and standard deviations immediately after treatment and at long-term follow-up for the intervention and control groups in the studies included in the meta-analysis. If these data were not available in the original articles, they were calculated by the researchers, using the published data. If studies varied in measurement time points during long-term follow-up, we calculated the SMDs for the points closest to 6 months. Statistical heterogeneity was evaluated by the chisquare and I 2 tests. Inter-study heterogeneity was considered significant for p < 0.1 and I 2 > 50%. The meta-analysis was conducted using Review Manager (Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). In total, 4027 references were screened and 273 full text papers were retrieved, of which 17 were identified through cited reference search. Of these, 31 were included that consisted of 29 different studies (27 RCTs, two cohort studies). Of two RCTs two references of each were included, one reporting short-term follow-up and the other long-term follow-up. Eleven primary care studies and 18 community studies were included. Table 1 shows the characteristics of the 11 included studies that were conducted in primary care settings (10 RCTs, 1529 patients; 1 cohort study, 14 patients). The interventions studied included CBT (n = 5), exercise (n = 2), problem-solving therapy (PST; n = 1), a combination of CBT and bibliotherapy (n = 1), behavioral activation (BA; n = 1), and bright-light therapy (n = 1). Follow-up ranged from 1 week up to 12 months. Table 2 shows the characteristics of the 18 studies recruiting in the community, most of which depended on self-referral by participants (17 RCTs, 1041 patients; 1 cohort study, 22 patients). The studied treatment modalities were CBT (n = 3), bibliotherapy (n = 4), life-review (n = 3), exercise (n = 4), PST (n = 3), and receiving postcards (n = 1). In addition, 1 study compared cognitive therapy, behavioral therapy, and brief psychodynamic therapy with patients on a waiting list. Follow-up period ranged from 4 weeks to 2 years.

Outcome data of primary care studies
The outcome data for studies conducted in a primary care setting are presented in Table 3, and the quality assessments with corresponding scores on the subscales are presented in Fig 2. The results of the meta-analysis are summarized in Fig 3. Cognitive behavioral therapy. Five studies assessed the effect of CBT on depression in older adults: three assessed the effect of CBT alone [25,28,31], one assessed its use in combination with self-management [29], and one compared CBT with clinical case-management [24]. CBT was delivered individually in three out of the five studies [28,29,31], and as a group therapy in the other two [24,25].
CBT delivered as individual therapy was more effective in reducing depressive symptoms at 4 and 12 months' follow-up compared with both control groups (talking control and care as usual) [31]. In a study where CBT was delivered by individual therapy, CBT was not effective at reducing depressive symptoms immediately after treatment or at 3 and 6 months' follow-up [28]. Equally, in another study where CBT was delivered by group therapy, it was no more effective in achieving response (determined by a decrease of !5 points in the PHQ-9 [Patient Health Questionnaire]) compared with care as usual at 12 weeks' follow-up [25]; however, this  later study did not report the mean change in depressive symptoms. In other research, clinical case-management was more effective than CBT at 12 months' follow-up [24]. By contrast, CBT in combination with self-management was shown to reduce depressive symptoms at 3 and 9 months' follow-up [29]. Fig 3 shows the results of the meta-analysis. Four out of the five studies focusing on CBT in primary care could be included in the meta-analysis; the fifth study could not be included because it did not report continuous baseline and follow-up data [25], and because the authors   To summarize, CBT was effective in two of the five studies, of which one was assessed to have the lowest risk of bias. This effect was confirmed in the meta-analysis at six months' follow-up. The two studies demonstrating a beneficial effect of CBT used individually delivered treatment rather than group therapy.
Exercise. Two studies assessed the effect of exercise [32,33]. Compared to a control group receiving information about exercise and local exercise options, progressive resistance training was not more effective in reducing depressive symptoms at 10 weeks' and 6 months' follow-up [32]. But, high and low intensity training were both more effective in reducing depressive symptoms at 8 weeks' follow-up based on self-reported, but not observer-rated, measures [33]. The risk of bias was assessed as moderate for both studies.
Other. Treatment modalities in the "other" category included PST, CBT-based bibliotherapy, behavioral activation, and bright-light therapy; all four were delivered individually. Of the two studies with a low risk of bias, bright-light therapy was effective [30], whereas PST was not [34]. In a study of moderate quality, CBT-based bibliotherapy was shown to be no more effective than care as usual [27]. Behavioral activation, which was only studied in a pilot cohort, was found to reduce symptoms of depression at 4 weeks' follow-up [26].

Outcome data of studies in community settings
Outcome data for studies conducted in the community are presented in Table 4, and the quality assessment with corresponding scores on the subscales is presented in Fig 2. Cognitive behavioral therapy. Three RCTs [39,53,54] studied the effect of CBT on depressive symptoms. One RCT demonstrated that CBT group therapy was more effective than remaining on a waiting list, but found no difference between the effects of CBT and exercise [39]. Another RCT showed that, after treatment, CBT group therapy was effective at reducing depressive symptoms among participants suffering from depression with comorbid Control condition entered in meta-analyses specified by study: Arean (2005) [24] used clinical casemanagement; but, Laidlaw (2008) [28], Lamers (2010) [29], Serfaty (2009) [31] used care as usual. One of the two reported outcome measurements by Laidlaw was used in the meta-analyses, namely HRSD.
https://doi.org/10.1371/journal.pone.0184666.g003  anxiety [54]. Individual CBT delivered through the internet was also more effective at reducing depressive symptoms after treatment than care as usual [53]. In the latter RCT, this effect was maintained at 3 months' follow-up, although this was not compared to a control condition. In summary, individual CBT tended to be an effective treatment for reducing depressive symptoms compared with inactive control conditions among older adults, but the risk of bias ranged from low to moderate in the included studies. Bibliotherapy. Four RCTs investigated the effect of individual bibliotherapy [37,44,47,48], and had low to moderate risk of bias. All RCTs showed that bibliotherapy was effective at reducing depressive symptoms at 4 weeks' follow-up compared with remaining on a waiting list and being given a control form of bibliotherapy.
Life-review. All three RCTs investigating the effect of individual life-review on depression in older adults found a positive effect on depressive symptoms from 2 to 8 weeks' follow-up [35,45,50]. One RCT [45] also reported a further improvement of depressive symptoms at 3 months' follow-up, but did not compare this with a control condition. The risk of bias did differ a little between the included studies, ranging from high to moderate.
Exercise. Three RCTs [39,43,51] investigated the effect of exercise on depressive symptoms, with risk of bias assessments ranging from high to moderate. Compared with an active control group, one RCT did demonstrate an effect of exercise on depressive symptoms [51].  Non-pharmacological treatments for late-life depression However, although two other RCTs did find that exercise had an effect compared with inactivity, no difference was observed between the exercise group and the active controls in each study (e.g., CBT [39] and social contact control [43]). In addition, one cohort study [46] studied the effect of exercise on depressive symptoms. It was not reported whether the intervention was delivered as group or as individual therapy, but showed reduced depression scores at 12 and 24 weeks' follow-up compared with baseline.
Problem-solving therapy. Three RCTs studied the effect of individually delivered PST [36,41,42]. They all demonstrated that PST reduced depressive symptoms. Two of these RCTs [36,42] delivered PST in community agencies, and one combined PST with engagement in social activities [36]. The risk of bias varied from low to moderate.
Other. One RCT investigated the effect of receiving postcards on depressive symptomatology [55], but showed no effect at follow-up. Another RCT showed that three interventions (cognitive therapy, behavioral therapy, and brief psychodynamic therapy) had beneficial effects compared with controls (waiting list) [52] at 6 weeks (mid-treatment), but showed no differences between these three interventions after treatment (16 weeks). However, the effects of the three interventions were not compared with the control group after treatment, because those on the waiting list had started treatment.

Summary of main findings
Of the five treatments studied in primary care, a meta-analysis on CBT yielded a significant result, indicating its potential benefit in primary care settings. There was also a positive effect with bright-light therapy, and although this is promising, it needs replication in a second trial in primary care before recommendations for implementation can be made. Unfortunately, we did not find convincing evidence in favor of exercise, PST, or behavioral activation for the treatment of depressive symptomatology in primary care, but better quality research is needed before we can reach any definitive conclusions. In addition, community-based studies showed promising short-term results for bibliotherapy, life-review, PST, behavioral therapy, brief psychodynamic therapy, and cognitive therapy, which might, therefore, be suitable for use as treatment strategies in primary care.

Comparison with existing literature
Previous systematic reviews and meta-analyses focusing on the use of non-pharmacological treatment for depression in older patients have reported different findings to those in our review [17][18][19][20]. Two recent systematic reviews [18,20], for example, concluded that psychological treatments may be feasible for late-life depression (65+), but they did not perform formal meta-analyses. However, both of these reviews questioned the generalizability and efficacy because of the wide diversity of interventions, the low number of studies per intervention, and the poor quality of studies included. Moreover, neither review was limited to the primary care setting, and studies were excluded if they had a low quality assessment, leading to the exclusion of 73.9% [18] and 36.4% [20] of the identified studies, respectively. To be more comprehensive, we decided not to restrict ourselves to RCTs and not to exclude studies based on the quality assessment. This not only ensured that we could summarize all available evidence but also enabled us to formulate explicit targets for future research, such as instances where an included study was of poor quality but focused on a promising intervention.
Another two reviews included formal meta-analyses of the research [17,19], and they indicated that psychological treatments were moderately effective in the treatment of late-life depression. Specifically, one showed that CBT, life-review, and PST [17] were effective, while the other showed that CBT was more effective than a non-active control group [19]. However, these meta-analyses included studies conducted in clinical settings and with middle-aged participants (50/55+). These differences might explain why we could not replicate the finding that PST was an effective treatment for late-life depression in primary care; also, it should be noted that life-review therapy has been studied as treatment for late-life depression in primary care to date. Nonetheless, we confirmed the positive results for life-review and PST on depressive symptoms in community settings. We could also replicate the finding that CBT was an effective treatment modality for late-life depression at 6 months' follow-up, though with a small effect size (SMD -0.21 [-0.40 to -0.03]) comparable to that reported in one of the previous studies [17]. The other meta-analysis demonstrated a much larger effect size (-1.35) when CBT was compared with inactive controls, but did not find an effect when comparing CBT to active controls [19]. The fact that we analyzed the effect of CBT compared with both active and inactive controls might explain this difference. Although one might question the clinical relevance of this small effect of CBT, it might be partly caused by a floor-effect of treatment associated with milder forms of depression as seen and treated within primary care.
Another systematic review found that physical exercise may be effective for late-life depression [56]. We could not replicate this finding, irrespective of the quality assessments of these studies, but it should be noted that the previously conducted review included studies recruiting non-depressed adults, and that none of the studies included in the earlier review [56] was conducted in a primary care setting.
Several differences can be seen when comparing the studies conducted in primary care with those conducted in the community. First, although more treatment modalities have been studied in community settings, it is questionable whether these treatment modalities are applicable in general practice. For example, creating a life-story book with personalized pictures [35] is overly time-consuming for most GPs or practice nurses. Second, the follow-up periods of the community-based studies were shorter than those conducted in primary care. Because none of the studies included a control condition beyond the assessment when treatment ended, no data is available on the sustainability of the effects. Third, most of the community studies only included self-referred participants, thereby introducing selection bias. Self-referred participants show the initiative to seek out interventions targeting depression, whereas in general, depressed older adults are more likely to be reluctant to seek help [57]. This purported selection bias might also explain some of the low percentages lost to follow-up in the self-referral studies performed in the community. Although it is conceivable that community-based interventions would also be effective in primary care, further research is needed to confirm this assumption. Finally, among the therapist-guided interventions, almost half were delivered by a postgraduate therapist or clinical psychologist in the community studies, while only one-third included a psychologist in the primary care studies. Because it is questionable whether clinical psychology services could be successfully embedded in general practice, due for example to higher costs for patients and/or insurances, future research should determine whether these interventions can be successfully given by a practice nurse or other allied healthcare professionals.
Several non-pharmacological treatments for late-life depression studied a community setting seem promising for implementation in primary care. First of all, PST demonstrated a beneficial effect in the community [36,41,42], but the only RCT conducted in primary care demonstrated no effect on depressive symptoms [34]. However, the risk of bias was lower in two of the community studies [36,42] compared with the study conducted in primary care [34], and among middle-aged adults the effectiveness of PST in primary care has been confirmed [58]. Due to the positive results of PST in the community setting and among middleaged adults, we recommend a second RCT in primary care focusing on PST with a longer follow-up duration than the study included in our review (11 weeks) [34]. Moreover, the control group in this primary care study [34] existed of paroxetine or a placebo, while an attention control form of therapy would have been more adequate. Furthermore, bibliotherapy [37,44,47,48] and life-review [35,45,50] have demonstrated beneficial effects in community settings, although follow-up duration was short (maximum of 8 weeks). Before implementation in primary care, life-review and bibliotherapy need to be studied among primary care patients with a longer follow-up duration. Furthermore, these studies had some issues regarding their risk of bias, with a high risk of bias for the bibliotherapy studies [37,44,47,48] and a moderate risk of bias for two of the life-review studies [35,45], and these concerns need to be addressed in a future RCT conducted in primary care. Finally, the effect of behavioral activation therapy seems promising in a pilot cohort study conducted in primary care [26], and is currently being investigated in a well-designed RCT in primary care [59]. In addition to this latter RCT, also PST, bibliotherapy, and life-review should be studied in a RCT in primary care among depressed patients confirmed by a diagnostic interview and with at least a one-year follow-up.

Limitations
First, although we decided to review the results narratively, we did diverge from the published protocol to perform a meta-analysis concerning the effect of CBT in primary care. This was because most of the included studies in primary care focused on CBT and the combined results of the individual studies were inconclusive. Although only two of the five individual studies indicated a beneficial effect of CBT, the meta-analysis confirmed a small but beneficial effect. Too few studies focusing on other non-pharmacological treatment options were conducted to perform a meta-analysis for these interventions; for example, two studies focused on exercise and both concluded that it was ineffective at follow-up, whereas only single studies were conducted for the other treatment modalities. Since we aimed to present an overview of the evidence for non-pharmacological treatments for late-life depression within primary care, we decided not to perform meta-analyses of studies conducted in the community, but to narratively review these studies in order to identify promising non-pharmacological treatments. Second, one of the search terms was "general practice OR synonym," so we only found a few studies that were conducted in the community in the primary search. Although these settings were not the focus of our review, we wanted to include all studies that focused on non-pharmacological treatment options in primary care. Due to careful selection of studies from previous systematic reviews and meta-analyses, we could find and included more studies conducted in a community (n = 15) setting, consistent with the aim of our review (Fig 1). However, we cannot ignore the possibility that we did not include all studies focusing on non-pharmacological interventions for late-life depression conducted in a community setting. Third, included studies differed in their depression inclusion criterion, which may have introduced heterogeneity and thus may have affected the results of this review. However, the observed heterogeneity in our meta-analysis was small (I 2 = 0%). Finally, limitations of included studies should also be acknowledged; such as the low number of included participants in primary care studies and the short follow-up period in community studies.

Conclusion
Through this systematic review and meta-analysis, we aimed to provide general practitioners with a comprehensive summary of the available evidence for non-pharmacological treatments in late-life depression in primary care. We found a limited amount of studies studying a wide variety of non-pharmacological interventions. Moreover, these studies differed in their definition of depression, definition of remission, and follow-up duration. Although this limits the evidence for specific interventions, it does give merit for several promising therapeutic options for treatment of late-life depression within primary care. CBT was the only treatment option meeting the highest level of evidence according to the GRADE (Grading of Recommendations Assessment, Development and Evaluation) criteria, with a small but beneficial effect after meta-analysis. However, a wealth of alternative options were identified that could be delivered by well-trained nurses based on evidence that exists from studies in a community setting. This review indicates that bibliotherapy, life-review, PST, and behavioral activation therapy are the options most likely to be of benefit in primary care settings, but the paucity of high-quality research means that we can only conclude that these options warrant further investigation in RCTs performed in primary care.
Supporting information S1 Appendix. Search strategy for different databases. Search strings for each part were combined using the "AND" Boolean statement. (DOCX) S1 PRISMA Checklist. PRISMA checklist. (PDF)