1 INTRODUCTION

The Depression and Somatic Symptoms Scale (DSSS), a free and self-administered scale, simultaneously evaluates both depression and somatic symptoms (Hung et al., 2006a). The DSSS is composed of a 12-item depression subscale (DS) and a 10-item somatic subscale (SS), which includes five pain and five non-pain somatic symptoms. The DSSS was developed because previous depressive scales failed to bring the somatic aspect of depression to a level equal to that of psychological symptoms. The items of the DS were designed based on the criteria of a major depressive episode (MDE) in the DSM-IV, while the items of the SS were selected from common somatic symptoms of depression, which could reflect the severity of depression (Hung et al., 2006b). Moreover, many previous studies reported that somatic or pain symptoms among patients with depression were associated with a poor prognosis of depression (Hung et al., 2015; Jaracz et al., 2016). Therefore, a scale for depression with a somatic element might be more strongly associated with the prognosis of depression. In light of the design of the scale, we hypothesized that the DSSS could reflect the severity of depression, be significantly correlated with other depressive scales, be sensitive to changes to treatment for depression, be used as a tool for screening depression, and be associated with the prognosis of depression.

The reliability and validity of the Chinese, Korean, and English versions of the DSSS have been established (Hung et al., 2006a; Jeon et al., 2016; Tse et al., 2018). Several validities of the DSSS among patients with depression have been tested, as follows. (1) The DS score is significantly correlated with the scores of the Hamilton Depression Rating Scale (HAMD), Montgomery–Asberg Depression Rating Scale, Center for Epidemiologic Studies Depression Scale, and mental subscales of the Short-Form 36 (Hung et al., 2006a; Hung, Wang, & Liu, 2009; Jeon et al., 2016; Tse et al., 2018). (2) The DS is sensitive to pharmacotherapy—the improvement percentage of the DS score after four weeks of pharmacotherapy is correlated with that of the HAMD score (Hung et al., 2006a). (3) The cut-off points of the DS score of ≥9 and ≥19 for non-full remission and a MDE, respectively, among patients with major depressive disorder (MDD) are of good sensitivity and specificity (Hung et al., 2012). (4) The SS score is significantly correlated with the scores of the somatization subscale of the Symptom Checklist-90-Revised, the somatic component of the HAMD, and the physical subscales of the Short-Form 36 (Hung, Liu, Cheng, & Wang, 2009; Hung, Wang, & Liu, 2009; Jeon et al., 2016). (5) The SS score at baseline predicts the outcome of depression at the 2-year follow-up point (Hung et al., 2010). (6) Principal-axis factor analysis and Mokken scale analysis demonstrated that the DSSS is of an appropriate construct validity (Chou et al., 2017; Hung et al., 2006a); moreover, the validity and reliability of the DSSS in patients with lower back pain have also been established (Liu et al., 2019).

The Hospital Anxiety and Depression Scale (HADS), which includes seven items for depression (HADS-D) and seven items for anxiety (HADS-A), does not include any somatic component (Zigmond & Snaith, 1983). This design renders the HADS able to be used to evaluate anxiety and depression without being confounded by somatic symptoms. For this reason, the HADS is one of most commonly used scales in screening for anxiety and depression among patients with medical diseases (Annunziata et al., 2020; Nikayin et al., 2016). Previous studies have investigated the cut-off scores of the HADS subscales for anxiety and depression among patients with medical diseases (Annunziata et al., 2020; de Almeida Macedo et al., 2017; Nikayin et al., 2016).

Although several studies have reported on the validity of the DSSS, no study has investigated the associations of the DSSS, HAMD, and HADS with the outcome of depression over a 10-year period among patients with MDD. It is well-known that somatic symptoms, painful physical symptoms, and anxiety symptoms have negative impacts on the outcome of depression (Hung et al., 2019; Jaracz et al., 2016; Rosellini et al., 2018). This raises an interesting question: which scale or subscale is most strongly associated with the long-term outcome of depression when depression, anxiety, and somatic scales or subscales are compared? Investigation of the above is important, because (1) the results will inform physicians and psychologists as to which component of depression is most strongly associated with the long-term prognosis of depression, and (2) the results will further prove the validity of the DSSS.

Therefore, this study aimed to compare the associations of the DSSS, HAMD, and HADS with the outcome of depression over a 10-year period. We hypothesized that scales or subscales that include assessment of appropriate somatic symptoms might be more strongly associated with the long-term outcome of depression over a 10-year period, as previous studies demonstrated that somatic and pain symptoms were associated with a poor prognosis of depression (Jaracz et al., 2016).

2 METHODS 2.1 Subjects

The study was conducted in the psychiatric outpatient clinic of Chang Gung Memorial Hospital, a medical center in northern Taiwan. At baseline, the study enrolled patients from January 2004 to August 2007. The inclusion criteria were consecutive outpatients aged 18–65 years who (1) fulfilled the MDD criteria and were experiencing a MDE based on the DSM-IV-text revision (TR; American Psychiatric Association, 2000); and (2) had not taken antidepressants or other psychotropic drugs within the past 1 month. In order to prevent mental symptoms from being confounded, three exclusion criteria were used: (1) catatonic features, psychotic symptoms, or severe psychomotor retardation; (2) a history of substance abuse or dependence without full remission in the past one month; and (3) chronic medical diseases such as diabetes mellitus, hypertension, and others. The subjects were interviewed by a board-certified psychiatrist using the Structured Clinical Interview for DSM-IV-TR Axis I Disorders (First et al., 2002) to confirm psychiatric diagnoses.

At baseline, 290 subjects with MDD were enrolled; they were followed-up at the 6-month, 2-year, and 10-year points. The study was approved by the Institutional Review Board of the Chang Gung Memorial Hospital. The 10-year follow-up study was conducted from August 2014 to December 2016. Written informed consent was obtained from all subjects, based on the guidelines regulated in the Declaration of Helsinki.

2.2 Psychometric scales

The 17-item HAMD, DSSS, and HADS were used (Hamilton, 1967; Hung et al., 2006a; Zigmond & Snaith, 1983), all of which have different psychometric characteristics. First, the HAMD is an observer-rated scale and one of the most commonly used scales for the clinical evaluation of depression (Dunlop, Granros, et al., 2019; Dunlop, Parikh, et al., 2019; Nixon et al., 2020; Sawamura et al., 2018; Vindbjerg et al., 2019; Zimmerman et al., 2018), while the DSSS and the HADS are self-administered scales. Second, the three scales have different percentages of somatic components. In the 17-item HAMD, there are eight items for somatic symptoms, including initial, middle, and terminal insomnia, loss of weight, loss of appetite, loss of libido, somatic anxiety symptoms, and general somatic symptoms; therefore, the somatic symptoms component represents 34.6% (18/52) of the total possible HAMD score. The DS of the DSSS consists of 12 items, including four somatic items (insomnia, poor appetite, fatigue, and loss of sexual desire). This design allows the DS to be compatible with the criteria of a MDE and other scales for depression. All 10 items of the SS address somatic symptoms. Therefore, the somatic components of the DS and SS represent 33.3% and 100%, respectively, of the total possible subscale scores. Conversely, the HADS does not include any somatic symptoms. Owing to this characteristic, the HADS is one of the most commonly used scales for screening depression and anxiety among patients with medical illnesses (de Almeida Macedo et al., 2017; Nikayin et al., 2016). The SS therefore has the largest somatic component, followed by the HAMD and DS; the HADS has no somatic component. The total scores range from 0 to 52 for the HAMD, 0 to 36 for the DS, 0 to 30 for the SS, and 0 to 21 for both the HADS-D and the HADS-A. A higher score indicates a greater severity.

2.3 Procedure

At baseline and each follow-up point, subjects were requested to complete the self-administered DSSS and HADS. At baseline, the HAMD score was evaluated by one of two psychiatrists, who were blind to the results of the DSSS and HADS. The two psychiatrists had been trained together in evaluating the HAMD before the study (intraclass correlation coefficient = 0.87). At follow-up, the HAMD was evaluated by the same psychiatrist.

The course of depression fluctuates, and depressive severities at the follow-up points might be unable to represent the longitudinal course of depression. At the 10-year follow-up point, subjects were asked to grossly estimate the percentages of time spent experiencing the following depressive symptoms over the past 10 years: depressed mood, anxiety, diminished motivation, insomnia, poor appetite, fatigue, decreased concentration, poor memory, and guilty feelings.

At follow-up, the HAMD score was used as the major indicator of the outcome of depression, because the HAMD is one of the gold-standard scales used for the assessment of depression (Worboys, 2013). The DS score, HADS-D score, and percentages of time spent experiencing depressive symptoms over the past 10 years as assessed at the follow-ups were considered as secondary outcomes of depression.

The subjects accepted pharmacotherapy after enrollment. Some patients might have quit pharmacotherapy at the follow-up points, but for those who still accepted pharmacotherapy, psychiatric medications were not controlled, because this study was an observational study. Pharmacotherapy without controlling for dosages and kinds of medication might confound the severity of depression at follow-up. Subjects who were undergoing pharmacotherapy at the index month of the follow-up point were classified as the treatment group, while those who were not were classified as the non-treatment group. Therefore, subjects were divided into treatment and non-treatment groups for analysis of the correlations of scale or subscale scores at baseline with the outcome of depression at follow-up. In analyzing the correlations of scale or subscale scores at baseline with the percentages of time spent experiencing the various depressive symptoms over the past 10 years, subjects were not divided into the two groups, because the percentages of time over the past 10 years were a gross estimation of the long-term course of depression, and subjects might intermittently accept treatment over the 10-year period.

At baseline and each follow-up point, the investigators checked all collected data to avoid the problem of missing data after each subject had been assessed.

2.4 Statistical methods

At baseline, we hypothesized that 50% of the subjects would not attend the 10-year follow-up. Using G*Power v3.1.9.2 (Faul et al., 2009), the sample size was required to be greater than 123 under the conditions of alpha level = 0.05, power = 0.8, number of predictors = 11, and effect size = 0.15. Therefore, the sample size should be greater than 246 at baseline.

SPSS for Windows 20.0 was used for statistical analyses. The Chi-square test and the independent t test were used when appropriate. Pearson correlation or Spearman correlation were used to test the correlations of the scores of the three scales at baseline with the HAMD score at follow-up and the percentages of time with depressive symptoms over the past 10 years.

Multiple linear regressions with forward selection were employed to identify the scale or subscale score at baseline most strongly associated with the HAMD score at the three follow-up points. This method was used for two reasons: (1) it can prevent multicollinearity; and (2) the forward method will select the most powerful factor associated with the independent variable into the regression model, followed by the second most powerful factor, then the others; therefore, it can identify the scale or subscale score that is most strongly associated with the dependent variable. The Durbin–Watson test was used to detect autocorrelation in the residuals. Collinearity is an important issue for regression models (Lang & Altman, 2015); this was examined using the Variance Inflation Factor (VIF).

Two regression models were used to compare the associations of (1) three depressive scale or subscale (HAMD, DS, and HADS-D) scores at baseline, and (2) five scale or subscale (HAMD, DS, SS, HADS-D, and HADS-A) scores at baseline with the outcomes of depression at the follow-up points. In the two regression models, the dependent variable was the HAMD score at the three follow-up points. In the first model, the independent variables were the scores of the HAMD, DS, and HADS-D at baseline, with pharmacotherapy or not at the follow-up point, in addition to five demographic variables, including gender, age, marital status, duration of education, and occupation. In the second model, the independent variables were all of the independent variables in the first model with the addition of the SS and HADS-A scores at baseline.

Generalized Estimating Equation (GEE) models, which were fitted with robust error estimation and an unstructured covariance matrix, were used to estimate the associations of the three psychiatric scales or subscales at baseline with the outcomes of depression at the follow-up points. The dependent variable in the GEE models was the HAMD score, while the independent variables included eight variables at baseline (gender, duration of education, marital status, HAMD score, DS score, SS score, HADS-D score, and HADS-A score) and four at each follow-up point (age, pharmacotherapy or not, employed or not, and visit). Baseline, 6-month, 2-year, and 10-year follow-up points were considered as the first, second, third, and fourth visits. Insignificant factors were removed from the model step by step until all independent variables were significant. One of the advantages of the GEE model is that it can handle imputation for missing data; that is, the GEE model remains a robust statistical method when data are missing at random. No important bias was observed with levels of loss that varied from 5% to 60% (Kristman et al., 2004; Seaman & Copas, 2009).

A p-value < 0.05 was considered statistically significant in the statistical analyses. Moreover, Bonferroni correction was used in multiple linear regressions, and a p-value < 0.017 was considered statistically significant.

3 RESULTS 3.1 Subjects

At baseline, 290 MDD patients with a current MDE were enrolled. Table 1 shows the percentages of participants remaining, demographic variables, and psychometric scale scores at baseline and the three follow-up points. At the 6-month and 2-year follow-up points, 36 (12.4%) and 53 (18.3%) participants did not attend, respectively, due to “being unable to be contacted by mail or phone” (n = 15 and 27, respectively) and “refusing to participate in the follow-up study” (n = 21 and 26, respectively). At the 10-year follow-up point, 153 (52.8%) subjects did not participate, for the following reasons: 99 (34.1%) could not be contacted; 49 (16.9%) refused to participate in the follow-up program; and 5 (1.7%) for other reasons. There were no significant differences in the five demographic variables between the subjects who did and did not attend follow-up at the three points, with the exception of age at the 10-year follow-up point (with vs. without follow-up: 41.3 [8.1] vs. 39.3 [8.2] years, p = 0.04).

TABLE 1. Demographic variables and psychometric scores at baseline and three follow-up points Time point Baseline Six months Two years Ten years Case number 290 254 237 137 Follow-up participation (%) − 87.6 81.7 47.2 Male (%) 28.6 29.1 30.8 30.7 Age (years) 30.2 (8.2) 30.6 (8.2) 32.5 (8.4) 41.0 (8.1) Education (years) 13.2 (2.4) 13.3 (2.4) 13.3 (2.4) 13.3 (2.5) Employed (%) 57.2 56.7 55.7 73.0 Married (%) 42.1 42.1 43.0 52.6 With pharmacotherapy (%) 0 47.6 27.4 27.7 HAMD score 23.4 (4.2) 10.6 (7.8) 10.4 (7.4) 9.4 (6.4) DS score 25.5 (5.3) 11.4 (8.5) 10.8 (8.0) 11.2 (8.0) SS score 16.1 (6.6) 8.2 (6.6) 8.2 (6.2) 8.9 (5.6) HADS-D score 14.2 (3.4) 7.7 (5.0) 7.0 (5.0) 7.2 (5.1) HADS-A score 15.0 (3.3) 8.6 (4.8) 9.0 (4.5) 8.4 (4.7) Abbreviations: DS, depression subscale of the Depression and Somatic Symptoms Scale (DSSS); HADS-A, anxiety subscale of the HADS; HADS-D, depression subscale of the Hospital Anxiety and Depression Scale (HADS); HAMD, Hamilton Depression Rating Scale; SS, somatic subscale of the DSSS. a Continuous variables are presented as the mean (SD).

For clarity, the four footnotes “(B),” “(6M),” “(2Y),” and “(10Y)” are used to represent data collected at baseline, 6 months, 2 years, and 10 years, respectively.

At the 6-month, 2-year, and 10-year follow-up points, the mean (SD) scores of the HAMD in the 2 groups (subjects with pharmacotherapy vs. without pharmacotherapy) were 8.6 (6.5) vs. 12.4 (8.5; p < 0.001), 11.5 (7.2) vs. 10.0 (7.4; p = 0.15), and 13.1 (6.7) vs. 8.0 (5.7; p < 0.001), respectively.

3.2 Correlations of HAMD, DS, SS, HADS-D, and HADS-A scores at baseline with outcomes of depression at the three follow-up points

At baseline, the HAMD(B) score was significantly (all p < 0.001) correlated with the DS(B) (correlation co-efficient r = 0.60), SS(B) (r = 0.46), HADS-D(B) (r = 0.40), and HADS-A(B) (r = 0.41) scores. The DSSS(B) score was also significantly (all p ≤ 0.001) correlated with the HADS(B) score (DS(B) and HADS-D(B), r = 0.54; DS(B) and HADS-A(B), r = 0.46; SS(B) and HADS-D(B), r = 0.20; SS(B) and HADS-A(B), r = 0.37).

Table 2 shows the correlations of scale or subscale scores at baseline with the outcomes of depression at the three follow-up points. In the non-treatment group, the HAMD(B) score was correlated with the three depressive scale (the HAMD, DS, and HADS-D) scores at the 6-month and 2-year follow-up points. The DS(B) score was correlated with the three depressive scale scores at the three follow-up points, with the exception of the HAMD(2Y) score. The SS(B) score was correlated with the HAMD and DS scores at the three follow-up points and the HADS-D(6M) score. The HADS-D(B) score was not significantly correlated with the scores of the three depressive scales at the three follow-up points, with the exception of the HADS-D(10Y) score. The HADS-A(B) score was significantly correlated with the HAMD(6M), DS(6M), and DS(2Y) scores.

TABLE 2. Correlations of depression, anxiety, and somatic severities at baseline with depressive severities at three follow-up points Without pharmacotherapy With pharmacotherapyb HAMD(B) DS(B) SS(B) HADS-D(B) HADS-A(B) HAMD(B) DS(B) SS(B) HADS-D(B) HADS-A(B) HAMD(6M) 0.38** 0.26** 0.36** 0.12 0.27** 0.11 0.09 0.27** −0.10 0.10 HAMD(2Y) 0.31** 0.12 0.23** −0.04 0.14 0.14 0.14 0.25* 0.14 0.06 HAMD(10Y) 0.17 0.25* 0.31** −0.06 0.11 0.46** 0.47** 0.35* 0.42** 0.40* DS(6M) 0.34** 0.25** 0.32** 0.12 0.23** 0.20* 0.21* 0.31** −0.05 0.14 DS(2Y) 0.31** 0.21** 0.25** 0.02 0.19* 0.10 0.17 0.20 0.04 0.05 DS(10Y) 0.15 0.26* 0.37** −0.04 0.12 0.39* 0.51** 0.37* 0.31 0.36* HADS-D(6M) 0.20* 0.20* 0.21* 0.16 0.17 0.11 0.19* 0.30** 0.04 0.13 HADS-D(2Y) 0.22** 0.15* 0.15 0.09 0.10 0.02 0.16 0.11 0.14 0.02 HADS-D(10Y) 0.12 0.26** 0.16 0.24* 0.01 0.39* 0.49** 0.37* 0.44** 0.25 Note: **p < 0.01; *p < 0.05. Abbreviations: DS, depression subscale of the Depression and Somatic Symptoms Scale (DSSS); HADS-A, anxiety subscale of the HADS; HADS-D, depression subscale of the Hospital Anxiety and Depression Scale (HADS); HAMD, Hamilton Depression Rating Scale; SS, somatic subscale of the DSSS. a “(B),” “(6M),” “(2Y),” and “(10Y)” represent data collected at baseline and at the 6-month, 2-year, and 10-year follow-up points, respectively. b Subjects undergoing pharmacotherapy in the index follow-up month.

In the treatment group, the SS(B) score was correlated with the HAMD score at the three follow-up points; however, the HAMD(B) and DS(B) scores were not significantly correlated with the HAMD(6M) and HAMD(2Y) scores. All HAMD(B), DS(B), and SS(B) scores were correlated with the HAMD(10Y), DS(10Y), HADS-D(10Y), and DS(6M) scores. The HADS-D(B) and HADS-A(B) scores were significantly correlated with the HAMD(10Y) score, but not with the three depressive scale scores at the 6-month and 2-year follow-up points.

3.3 Correlations of the five scale or subscale scores at baseline with self-reported percentages of time spent experiencing depressive symptoms over the past 10 years

The mean (SD) self-reported percentages of time spent experiencing depressive symptoms over the past 10 years were 42.3 (26.6) for depressive mood, 38.4 (28.3) for anxiety, 37.4 (28.5) for diminished motivation, 40.0 (31.7) for insomnia, 16.6 (21.8) for poor appetite, 48.8 (28.3) for fatigue, 38.7 (28.1) for decreased concentration, 44.3 (29.9) for poor memory, and 30.2 (30.3) for guilty feelings.

Table 3 shows the correlations of the five scale or subscale scores at baseline and the 10-year follow-up point with the self-reported percentages of time spent experiencing depressive symptoms over the past 10 years. Both the DS(B) and SS(B) scores were significantly correlated with eight of the nine symptoms, while the HAMD(B) score was significantly correlated with five symptoms. The HADS-D(B) and HADS-A(B) scores were not significantly correlated with any of the nine symptoms, with the exception of the HADS-A(B) score being correlated with guilty feelings.

TABLE 3. Correlations of depression, anxiety, and somatic scores at baseline and the 10-year follow-up point with self-reported percentages of time spent experiencing depressive symptoms over the past 10 years HAMD(B) DS(B) SS(B) HADS-D(B) HADS-A(B) HAMD(10Y) DS(10Y) SS(10Y) HADS-D(10Y) HADS-A(10Y) Depressed mood 0.06 0.28** 0.24** 0.10 0.10 0.62** 0.66** 0.46** 0.55** 0.57** Anxiety 0.22* 0.28** 0.28** −0.03 0.16 0.59** 0.60** 0.53** 0.41** 0.59** Diminished motivation 0.15 0.31** 0.25** 0.15 0.10 0.68** 0.70** 0.47** 0.60** 0.57** Insomnia 0.14 0.16 0.23** 0.16 0.13 0.52** 0.51** 0.47** 0.39** 0.46** Poor appetite 0.19* 0.27** 0.13 0.09 0.03 0.43** 0.39** 0.25** 0.34** 0.40** Fatigue 0.16 0.26** 0.31** 0.08 0.10 0.62** 0.65** 0.55** 0.53** 0.50** Decreased concentration 0.21* 0.31** 0.32** 0.12 0.16 0.62** 0.67** 0.53** 0.56** 0.57** Poor memory 0.20* 0.34** 0.33** 0.13 0.10 0.59** 0.61** 0.46** 0.53** 0.52** Guilty feelings 0.25** 0.35** 0.32** 0.10 0.24** 0.69** 0.71** 0.49** 0.56** 0.65** Note: **p < 0.01; *p < 0.05. Abbreviations: DS, depression subscale of the Depression and Somatic Symptoms Scale (DSSS); HADS-A, anxiety subscale of the HADS; HADS-D, depression subscale of the Hospital Anxiety and Depression Scale (HADS); HAMD, Hamilton Depression Rating Scale; SS, somatic subscale of the DSSS. a “(B)” and “(10Y)” represent data collected at baseline and at the 10-year follow-up point, respectively.

All of the self-reported percentages of time spent experiencing depressive symptoms over the past 10 years were significantly correlated with the scores of the five scales or subscales at the 10-year follow-up point.

3.4 Associations of scale or subscale scores at baseline with the HAMD score at the three follow-up points

Table 4 shows the results of the first regression model. The HAMD(B) score appeared to be most strongly associated with the HAMD(6M) and HAMD(2Y) scores among the three depressive scale scores at baseline after controlling for demographic variables and pharmacotherapy. The DS(B) score was most strongly associated with the HAMD(10Y) score after controlling for other factors.

TABLE 4. Independent factors associated with depressive severities in the first regression model at three follow-up points, Independent variable B R2 change 95% B confidence interval p value HAMD(6M) HAMD(B) 0.48 0.08