Anxiety, Depression and Quality of Life—A Systematic Review of Evidence from Longitudinal Observational Studies

1The World Health Organization estimates that 264 million people worldwide were suffering from an anxiety disorder and 322 million from a depressive disorder in 2015, corresponding to prevalence rates of 3.6% and 4.4%. While their prevalence varies slightly by age and gender, they are among the most common mental disorders in the general population. During the COVID-19 pandemic, multiple challenges have arisen for many, such as loneliness or financial hardship. A meta-analysis showed a prevalence of anxiety of about 32% (95% CI: 28–37) and a prevalence of depression (n = 14 studies) of about 34% (95% CI: 28–41) in general populations during the COVID-19 pandemic.

2Anxiety and depression have been associated with adverse societal and individual correlates, including higher health care costs and an increased risk for physical comorbidities, such as cardiovascular illnesses. Moreover, they have been linked to a reduced quality of life (QoL) in numerous cross-sectional as well as longitudinal studies in which they significantly predicted QoL outcomes. Other studies have reported a reverse association, whereby QoL was predictive of mental health outcomes or a bi-directional association. Some very recent studies also examined these associations among quite different samples (e.g.,).

3Looking at longitudinal rather than cross-sectional data from observational studies has several advantages. It allows for the identification of trajectories over time within the same individuals rather than focusing on group differences at one point in time only. Moreover, when appropriate methods are applied to longitudinal data, intraindividual heterogeneity can be taken into account, resulting in more consistent estimates. This has previously been demonstrated in QoL research. A need to analyze longitudinal changes in QoL domains in QoL research in people with mental disorders has also been previously identified. Beyond individual longitudinal studies suggesting a link between anxiety or depression and QoL, several systematic reviews have synthesized longitudinal evidence on these associations and mostly reported negative associations between the variables. These reviews have tended to focus on specific age groups, such as older adults, samples with specific diseases, or have investigated the effect of specific treatments on QoL in patients with anxiety. Investigating these associations in samples without these limitations could reduce the effect of specific conditions and treatments on the association and strengthen the conclusions that can be drawn.

4In light of the previous findings, this study aims to add to the present literature by systematically synthesizing evidence from observational studies on the longitudinal association between anxiety, depression and QoL across all age groups in samples who do not have other specific illnesses and do not receive specific treatments.

5This review was registered with PROSPERO (CRD42018108008) and a study protocol was published.

6Five electronic databases from several fields of research (PubMed, PsycINFO, PSYNDEX, NHS EED and EconLit) were examined until December 2020. Where possible, search terms were entered as Medical Subject Headings (MeSH) or as keywords in the title/abstract. The PubMed search strategy was: (anxi*[Title/Abstract] or depress*[Title/Abstract] or anxiety disorder[MeSH] or depressive disorder[MeSH]) and quality of life[MeSH] and longitudinal study[MeSH]. Please note that “*” is a truncation symbol. Time or location were not restricted. In addition, we applied backward and forward reference searches of included studies to identify additional references. The forward reference search was conducted until January 2021 using Web of Science to identify cited papers.

7The study selection process is displayed in Figure 1. Most identified studies were screened in a two-step process (title/abstract; full-text screening) independently by two reviewers (J.K.H., E.Q.) against defined criteria (see Table 1). The last updated literature screening before submission was conducted by one reviewer (J.K.H.) and encompassed 9% of the studies included for title/abstract screening. Before the final criteria were applied, they were pretested and refined. Disagreements during the selection process were resolved through discussion or by the inclusion of a third party (A.H.) if a consensus could not be reached.

8We extracted information regarding the study design, operationalization of the variables, sample characteristics, statistical methods and results regarding the research question of interest. If several analyses were presented for the same research question, we extracted the final covariate-adjusted model for narrative synthesis. Data were extracted by one reviewer (J.K.H.) and cross-checked by a second reviewer (E.Q.). If needed, extracted data were standardized (e.g., by calculating the weighted average means when combining groups) to present comparable information. If clarification was needed, the corresponding authors were contacted.

9For the narrative synthesis, all studies were first grouped by research question, e.g., whether disorders or the degree of symptoms were analyzed, which comparison groups were used, which QoL domains were considered, and at which waves the variables of interest were considered in the analyses. Because research questions and analyses were heterogeneous, a concise narrative synthesis of the main results of all studies was not feasible. Therefore, we provide an overview of all identified studies in the tables and a detailed narrative synthesis of those studies, analyzing trajectories of disorders or changes in symptoms in association with changes in QoL over time.

10Additionally, we examined whether data were appropriate for meta-analysis. The specific research questions, the operationalization of main variables and statistical methods were heterogeneous across studies and not all the statistical estimates needed could be obtained from covariate-adjusted analyses. Therefore, to enhance the comparability of the underlying data and the interpretation of the pooled estimates, we used descriptive information. Because most papers applied variations of the Short Form Health Survey and analyzed mental and physical component scores (MCS, PCS), we considered these studies as eligible for meta-analysis. The necessary information could be obtained for 8 publications. Random-effects meta-analysis was used for pooling. Heterogeneity was assessed by means of I2, with higher values representing a larger degree of heterogeneity in terms of variability in effect size estimates between studies. Pooled estimates are reported as Hedge’s g standardized mean difference (SMD), representing the difference in mean outcomes between groups relative to outcome measure variability. According to Cohen (as cited in), SMDs can be grouped into small ≤0.20, medium = 0.50 and large effects ≥0.80. Stata 16 was used for meta-analyses.

11Two reviewers (J.K.H., E.Q.) independently assessed the quality and risk of bias of the included studies using the Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies, which was developed by the National Heart, Lung, and Blood Institute.

12The literature search yielded 4027 unique references. After title/abstract screening, 215 studies were included for full-text screening. Finally, 47 publications were included in the final synthesis. During full-text screening, most studies were excluded because they exclusively analyzed data on a cross-sectional level (56.5%). For further details, see the PRISMA flow chart (Figure 1).

13Descriptive characteristics and quality/risk of bias assessment of the included studies are provided in Table S1 (Supplementary Material). In short, sample size ranged from 28 to 43,093. Most studies focused on adults; only four analyzed children/adolescents. Regarding the settings, 17 of the analyzed samples were exclusively recruited in a health care setting, 12 of the studies analyzed general population samples, 14 recruited in another or in several settings, and all studies on children/adolescents recruited in schools (n = 4). Twenty studies (42.6%) applied data from the same seven underlying datasets. Most studies reported on depression (n = 36), less reported on anxiety (n = 20) and some reported on the comorbidity between depression and anxiety (n = 7). To assess mental disorders, half (48.9%) used structured interviews. Regarding QoL, most studies applied variations of the Short Form Health Survey (SF, n = 27) or the WHOQOL (n = 12). A total of 38.3% of the studies were rated as “good”, 55.3% as “fair” and 6.4% as “poor” in the quality assessment.

14Detailed results on all studies investigating the association between anxiety/depression as independent variables and QoL outcomes are reported in Table 2. As described in the methods section, the following paragraphs give an overview of those studies focusing on disorder trajectories/changes in symptoms over time and changes in QoL outcomes over time, because they allow for more differentiated interpretations.

15Depression as independent variable and QoL as outcome. One study investigated QoL at several time points during the entire course of an episode of MD.

16Buist-Bouwman, Ormel, de Graaf and Vollebergh analyzed an MD group from a general population setting (NEMESIS) with data on SF-36 domains in the onset, acute and recovery phase of the depressive episode. The onset of MD was associated with a significant drop in several QoL domains and recovery with a significant increase. Pre- and post-morbid QoL levels were not significantly different for most domains, and post-morbid QoL was even higher for the psychological role functioning and psychological health domains. In comparison to a group without MD, pre- and post-morbid QoL levels in the MD group were significantly lower, except for the psychological role functioning domain, where no significant differences were found. Additionally, it should be noted that 40% of the sample had lower post-morbid QoL compared to pre-morbid levels.

17Two studies investigated changes in QoL for people experiencing an onset of depression relative to different comparison groups over two points in time.

18One study investigated incident MD in a general population sample (NESARC; Rubio, Olfson, Perez-Fuentes, Garcia-Toro, Wang and Blanco). Here, incident MD (compared to those without a history of MD as well as to a group without any mental disorder) was associated with a significant drop in QoL (SF-12 MCS). Additionally, analyzing two waves, Pyne, Patterson, Kaplan, Ho, Gillin, Golshan and Grant compared the QoL (Quality of Well-Being scale) between MD patients and community controls. The patient group was further divided into those continuously not receiving an MD diagnosis, those who continuously received the diagnosis and those who only received the diagnosis at FU (onset). The authors found that changes in QoL did not differ between the groups. At both points in time, QoL scores differed significantly between the groups, except for the incident and the continuous depression group.

19Six studies investigated different courses of depression over time in people with depression at BL with or without a healthy comparison group as reference.

20Two primary care studies analyzed groups with clinical depression at BL with different FU depression statuses (remission, no remission). One study analyzed changes in generic QoL measures (SF-12, WHOQOL-Bref) and the disease-specific Quality of Life in Depression Scale. In this study, remission was associated with an improvement in all QoL domains, whereas QoL did not change significantly over time for the non-remitted group. Another study investigated SF-12 MCS and PCS scores and reported a significant increase in MCS over time in the remitting group. MCS scores in the continuously depressed group and PCS scores in both groups improved, albeit not significantly.

21Another study investigated whether chronic MD in a general population sample (NESARC) was associated with domain-specific reduced QoL (SF-12). They found that chronic MD was a significant risk factor for persistently reduced QoL in all domains and for the onset of reduced QoL at FU in all domains except for physical role.

22Two population-based studies further differentiated between the depressive disorders. Analyzing MCS scores (NESARC), Rubio, Olfson, Villegas, Perez-Fuentes, Wang and Blanco reported a significant increase in QoL for those who remitted from MD and from dysthymia relative to those who had a persistent disorder. Rhebergen, Beekman, de Graaf, Nolen, Spijker, Hoogendijk and Penninx differentiated between people with MD, double depression or dysthymia at BL who remitted until FU relative to a group without a mental health diagnosis (NEMESIS). Physical health (SF-36) was lowest at BL for double depression, dysthymia and then the MD group. Over time, the MD and double depression groups improved significantly in their physical health, while the dysthymia group did not improve significantly. QoL was significantly lower relative to healthy comparisons for all depression groups at all waves. There were no significant differences regarding physical health trajectories over time among the depressive disorder groups.

23Stegenga, Kamphuis, King, Nazareth and Geerlings investigated more than two MD course groups over time (remitted, intermittent and chronic MD) in association with SF-12 MCS and PCS over time in a primary care-recruited sample with BL MD (Predict study). MCS increased over time in all groups, while changes in PCS were small. Compared to those who remitted, MCS at BL was significantly lower for the chronic course group. While the intermittent group also displayed a lower mean MCS at BL, the coefficient was not significant.

24Three studies investigated changes in depressive symptom levels as the independent variable and changes in QoL as outcomes in adults.

25One study found no significant association between an initial change in depressive symptoms and subsequent change in QoL (EQ-VAS) in older adults recruited in primary care. The two other studies analyzed changes in depressive symptoms in samples with MD at BL. Chung, Tso, Yeung and Li found that changes in depressive symptom levels was associated with changes in several QoL domains (SF-36: general health, vitality, social functioning, mental health and MCS). Diehr, Derleth, McKenna, Martin, Bushnell, Simon and Patrick investigated whether quartiles of change in depressive symptoms were associated with changes in QoL (SF-12, QLDS and WHOQOL-Bref). Those without any change in depressive symptoms generally showed no change in QoL. For all QoL domains and scores except for SF-12 PCS, improvement in depressive symptoms over time was associated with a significant increase in QoL, while a reduction in depressive symptoms was associated with a significant reduction in QoL. Those who had the largest reduction in depressive symptoms also had the largest improvement in QoL measures.

26Anxiety as an independent variable and QoL as an outcome. Two publications used a general population sample (NESARC) to investigate incident anxiety disorders and the remission of anxiety disorders in association with SF-12 MCS. Both studies separated generalized anxiety disorder (GAD), social anxiety disorder (SAD), panic disorder (PD) and social phobia (SP). All incident disorders were associated with a significant reduction in QoL compared to people without a history of the specific disorders. When the analysis was restricted to incident cases without comorbidities, QoL levels were not significantly different compared to people without a history of any psychiatric disorder. Those who remitted from SAD showed a significant increase in QoL compared to persistent cases. While QoL improved for all remitting anxiety disorders, change scores for PD and SP were not significant.

27Another study investigated different courses (intermittent, chronic or remitting) of obsessive compulsive disorder (OCD) and course in QoL (EQ-5D) as well as a comparison group from the general population. They found that the OCD groups mostly reported a lower QoL compared to the general population. Moreover, the course groups differed regarding their QoL over time, with remitters reporting small to moderate improvements compared to the chronic group.

28One study investigated changes in anxiety symptoms in association with changes in all SF-36 domains and both summary scores over time in a sample with MD at BL. Changes in anxiety symptoms were significantly associated with changes in bodily pain, general health and the mental health domain.

29Additionally, we identified publications operationalizing QoL as the independent variable and anxiety/depression as outcomes with details on all studies reported in Table 3. Only one study reported on change in QoL over time and change/trajectories in mental health outcomes over time. This study operationalized change in QoL as a predictor of future change in depressive symptoms over time and reported that an initial improvement in EQ-VAS was associated with a future reduction in depressive symptoms in older adults.

30In total, eight studies on adults were included in a supplementary meta-analyses of several research questions on SF PCS and MCS in association with anxiety and depressive disorders. Forest plots for the analyses are provided in the supplementary materials (Figures S1–S10).

31Differences in SF summary scores at FU among adults with and without depressive disorders at BL. Based on a pooling of four studies, those with depression at BL showed lower MCS scores at FU compared to a group without depression at BL with a large effect size (SMD = −0.96, 95% CI: −1.04 to −0.88, p < 0.001, I2 = 0.0%). PCS scores at FU were lower for the depression group compared to the non-depression group with a medium effect size (SMD = −0.68, 95% CI: −1.06 to −0.30, p < 0.001, I2 = 94.6%). Excluding the study rated “poor” in the quality/risk of bias assessment from the pooling did not substantially affect the results (MCS: SMD = −0.96, 95% CI: −1.03 to −0.88, p < 0.001, I2 = 0.01%; PCS: SMD = −0.63, 95% CI: −1.08 to −0.19, p < 0.01, I2 = 96.8%).

32BL differences in SF summary scores among adults with MD at BL with and without remitting courses over time. Based on a pooling of two studies of samples with MD at BL, those with persistent MD at FU had significantly lower MCS at BL (SMD = −0.25, 95% CI: −0.41 to −0.10, p = 0.001, I2 = 74.95) and PCS scores at BL (SMD = −0.24, 95% CI: −0.39 to −0.09, p = 0.002, I2 = 73.14) compared to those who achieved remission until FU. Effect sizes were small for both summary scores.

33FU differences in SF summary scores among adults with depressive and anxiety disorders at BL with and without remitting courses. Based on the pooling of two studies of samples with MD and/or dysthymia, the group where the disorder had persisted/a co-morbid condition was present/had a suicide attempt until FU had significantly lower MCS scores at FU compared to the group where the disorder had remitted without treatment until FU, with a medium effect size for depressive disorders (SMD = −0.59, 95% CI: −0.75 to −0.42, p < 0.001, I2 = 37.72) and a small effect size for anxiety disorders (SMD = −0.44, 95% CI: −0.58 to −0.30, p < 0.001, I2 = 58.87). The SMD for PCS scores at FU was negligible in terms of effect size for both disorder groups (depressive disorders: SMD = 0.02, 95% CI: −0.24 to 0.27, p = 0.90, I2 = 73.65; anxiety disorders: SMD = −0.09, 95% CI: −0.17 to −0.01, p = 0.03, I2 = 0.01).

34This review adds to the present literature by providing an overview of longitudinal observational studies investigating the association between depression, anxiety and QoL in samples without other specific illnesses or specific treatments. Additional meta-analyses investigated group differences according to SF MCS and PCS.

35While a concise synthesis of all the identified studies is challenging due to heterogeneity, the following picture emerges from studies investigating change–change associations: before the onset of disorders, QoL is already lower in disorder groups in comparison to healthy comparisons. The onset of the disorders further reduces the QoL. Remission is associated with an increase in QoL, mostly to pre-morbid levels. Additionally, some studies show that remission patterns are relevant for QoL outcomes as well. Moreover, a bi-directional effect was reported, whereby QoL is also predictive of mental health outcomes.

36Evidence for a bi-directional association as well as studies showing lower QoL across the entire course of the disorders (before onset, during disorder, after disorder) relative to a healthy comparison group seem to suggest that impairments in QoL may result from a certain pre-disorder vulnerability in these groups. Longitudinal studies using general population data have investigated different hypotheses on (QoL) impairments after remission of anxiety disorders and MD. One hypothesis suggests that impairments after the illness episode reflect a pre-disorder vulnerability (vulnerability or trait hypothesis), while the another states that impairments develop during the mental health episode and remain as a residual after recovery (scar hypothesis). Generally, both studies favored the vulnerability hypothesis. For subgroups with recurrent anxiety disorders, scarring effects were also found for mental functioning. Yet, it has to be noted that it was not the aim of our review to gather evidence for these hypotheses using QoL as an indicator, which represents an opportunity for future research.

37To be able to investigate possible domain-specific differences across studies, we aimed to conduct a meta-analysis on all studies on the same research question which reported on QoL subdomains (e.g., using WHOQOL and SF). However, as described in the Methods section above, only eight studies reported comparable information on different research questions and could be included in meta-analyses. Due to the limited number of studies included in each meta-analysis, the focus on SF MCS and PCS scores, and most studies reporting on depression, the results of the meta-analyses should be viewed with caution. Keeping this in mind, our results indicate that both mental and physical QoL are significantly impacted by anxiety and depressive disorders and that the course of the disorder is also relevant for QoL outcomes. Not surprisingly, effect sizes for MCS were larger compared to PCS for most research questions. A pooling of two studies on different courses of anxiety and depressive disorders found that effect sizes for MCS at FU were of moderate size for depressive (SMD = −0.59) and of small size for anxiety disorders (SMD = −0.44), while SMDs for PCS at FU were negligible in size.

38Overall, effect sizes from meta-analyses ranged from negligible to large, and heterogeneity varied considerably (I2 between 0% and 95%). Because of the small number of studies, possible influential study-level factors (e.g., setting, operationalization of the variables, length of FU) could not be investigated in further detail by means of a meta-regression, which remains a question for future research.

39Based on the results described and study heterogeneity discussed above, we provide recommendations for future research.

40First recommendation: future research should differentiate between individual disorders and focus on anxiety disorders. The majority of the studies investigated depressive disorders or symptoms. On the level of individual disorders, most focused on MD, while two studies additionally reported on dysthymia. One of these investigated double depression. On the level of anxiety disorders, three publications differentiated between individual anxiety disorders within the same study. While it was not possible to conduct a meta-analysis comparing different anxiety disorders in our case, individual studies suggest possible disorder-specific differences when analyzing changes in QoL over time: Rubio, Olfson, Villegas, Perez-Fuentes, Wang and Blanco suggest that QoL significantly improved for those remitting from GAD and SAD (compared to non-remission). QoL improved for PD and SP as well, but differences in change scores were smaller and did not reach statistical significance. The incidences of all of these disorders were associated with a significant drop in QoL. In summary, future longitudinal studies should focus on anxiety disorders and generally differentiate between individual disorders to investigate possible disorder-specific differences.

41Second recommendation: future research should consider trajectories of disorders/change in symptoms and changes in QoL over time. We would have liked to include a meta-analysis of disorder trajectories and change scores in QoL over time. Because of the small, diverse number of studies on this association in general and the number of assumptions that would have had to have been made for a meta-analysis, we refrained from pooling effects for this research question. In total, 17 studies investigated changes in independent variables associated with changes in outcomes. This approach has several advantages. On the one hand, different disorder or symptom trajectories can be identified. Several studies reported that QoL outcomes differ according to disorder course and the degree of change in symptoms. The focus on the change in characteristics over time in future research could additionally reduce the problem of unobserved time-constant heterogeneity in observational studies when appropriate methods are applied.

42Third recommendation: future research should investigate individual QoL domains. Several systematic reviews on cross-sectional studies found that effect sizes differed by QoL domains. For example, Olatunji, Cisler and Tolin reported that health and social functioning were most impaired for anxiety disorders (compared to non-clinical controls). Comparing individuals with diabetes and depressive symptoms to those with diabetes only, Schram, Baan and Pouwer reported that while SF pain scores were mild to moderately impaired, role and social functioning displayed moderate to severe impairments in those with comorbid depressive symptoms. The other scores were moderately impaired. As described above in detail, a meta-analysis using all subdomains was not feasible in this review. Further research differentiating between QoL domains would thus allow future meta-analyses to investigate whether the observed domain-specific differences reported in previous reviews of cross-sectional data can be observed in longitudinal studies as well.

43Fourth recommendation: future research should consider bi-directional effects. While investigating QoL as the outcome measure and anxiety/depression as independent variables seems relatively straightforward, ten studies investigated QoL as the independent variable and anxiety/depression as outcomes. In light of possible bi-directional effects and pre-existing vulnerability suggested by individual studies, future research considering QoL as an independent variable could inform a deeper understanding of this complex association.

44A strength of this work is the transparent methodological process: the review was prospectively registered with PROSPERO and a study protocol was published. Two reviewers were included in screening, data extraction and quality assessment processes. There were no limitations regarding the time or location of the publications. Moreover, all versions of the ICD/DSM and validated questionnaires were considered eligible to identify anxiety or depression. Another strength is the thorough literature search that enabled us to identify all relevant studies. Additionally, we did not limit the age range and were therefore able to shed light on studies investigating children/adolescents. Moreover, some studies could be pooled using random-effects meta-analyses, which allows for stronger conclusions regarding effect sizes compared to individual studies. Besides the content analysis, this review emphasizes difficulties in meta-analysis from observational, longitudinal studies. We hope that our work can facilitate discussion on this topic.

45The study has some limitations. We did not limit our search to specific research questions, which led to the inclusion of heterogeneous studies. Heterogeneity particularly stemmed from the operationalization of the variables of interest. Due to this, a concise narrative synthesis of all results was not feasible. The positive aspect of this broad focus is that it allowed us to provide an overview of studies and research questions analyzed and to formulate more nuanced recommendations for future research. We have to acknowledge that there is an abundance of QoL assessments used in medicine and health sciences. The list applied in this work was derived with respect to previous relevant reviews on QoL research. It was not designed to be fully comprehensive or exhaustive. Rather, it provided us with a working definition for this review and helped to enhance the transparency of our selection processes. Additionally, because we included validated QoL measures frequently used in research, we assume that exclusion would particularly have been the case for novel or study-specific measures. Finally, the focus on peer-reviewed literature means that studies in other languages and gray literature were not considered. Nonetheless, this focus on literature published in peer-reviewed journals should ensure a certain scientific quality.

46Overall, the results indicate that QoL is lower before the onset of anxiety and depressive disorders, further reduces upon onset of the disorders and generally improves with remission to pre-morbid levels. Moreover, disorder course (e.g., remitted, intermittent, chronic) seems to play an important role; however, only a few studies analyzed this. Changes in anxiety and depressive symptoms were also associated with changes in QoL over time. Meta-analyses found that effect sizes were larger for MCS relative to PCS, highlighting the relevance of differentiation between QoL domains. While our review identified some gaps in the current literature and made recommendations for future research, the following should be noted for clinical practice. On the one hand, an improvement in mental health is associated with better QoL, which emphasizes the relevance of support during the disorders. This is also shown by meta-analyses, which show that cognitive behavioral therapy additionally improves QoL. Moreover, the results indicate reduced QoL even before disorder onset, highlighting the relevance of early preventive measures in vulnerable groups. In line with this, studies on school-based prevention programs show a significant reduction in anxiety and depressive symptoms, and psychosocial prevention programs may additionally improve QoL.

47During the COVID-19 pandemic, it is of high relevance to tackle the arising challenges associated with this pandemic. For example, it is important to face the high prevalence rates of both depression and anxiety with appropriate measures.