Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Heliyon 9 (2023) e20951
Available online 17 October 2023
2405-8440/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Program of algorithm for pharmacological treatment of major
depressive disorder in China: Benets or not?
Yuncheng Zhu
a
,
b
,
c
,
1
,
*
, Fang Wang
d
,
1
, Fan Wang
b
,
1
, Hongmei Liu
b
, Xiaoyun Guo
b
,
Zuowei Wang
a
,
c
, Ruoqiao He
e
, Xiaohui Wu
b
, Lan Cao
b
, Zhiguo Wu
d
,
Daihui Peng
b
,
**
, Yiru Fang
b
,
c
,
f
,
g
,
h
,
***
a
Division of Mood Disorders, Shanghai Hongkou Mental Health Center, Shanghai, 200083, China
b
Clinical Research Center, Division of Mood Disorders, Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine,
Shanghai, 200030, China
c
Clinical Research Center for Mental Health, School of Medicine, Shanghai University, Shanghai, 200083, China
d
Shanghai Yangpu Mental Health Center, Shanghai, 200093, China
e
School of Social Work, New York University, New York, 10003, USA
f
Department of Psychiatry & Affective Disorders Center, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, 200025,
China
g
CAS Center for Excellence in Brain Science and Intelligence Technology, Shanghai, 200030, China
h
Shanghai Key Laboratory of Psychotic Disorders, Shanghai, 201108, China
ARTICLE INFO
Keywords:
Major depressive disorder
Chinese
Antidepressant agents
Treatment outcome
ABSTRACT
Background: This research was designed to investigate Algorithm Guided Treatment (AGT) and
clinical traits for the prediction of antidepressant treatment outcomes in Chinese patients with
major depressive disorder (MDD).
Methods: This study included 581 patients who had reached treatment response and 406 patients
remained non-responded observed after three months of treatment. Sociodemographic factors,
clinical traits, and psychiatric rating scales for evaluating therapeutic responses between the two
groups were compared. Logistic regression analysis was adopted to determine the risk factors of
unresponsive to antidepressant (URA) in MDD. Kaplan-Meier survival analysis was utilized to
compare the therapeutic response between AGT and treatment as usual (TAU).
Results: Compared to the MDD responsive to antidepressant (RA) group, the URA group had
signicantly lower rates of the following clinical traits: married status, anxious distress, moderate
to severe depressive symptoms, and higher rates of comorbidity (p-value <0.05). Logistic
Regression Analysis showed that eight clinical traits from psychiatric rating scales, such as
anxious characteristics, were correlated positively with URA, while the other eight symptoms,
such as autonomic symptoms, were negatively correlated. Time to symptomatic remission was
longer in TAU without statistically signicant (p-value =0.11) by log-rank testing.
* Corresponding author.Division of Mood Disorders, Shanghai Hongkou Mental Health Center, Shanghai, 200083, China
** Corresponding author.Clinical Research Center, Division of Mood Disorders, Shanghai Mental Health Center, Shanghai Jiao Tong University
School of Medicine, Shanghai, 200030, China
*** Corresponding author.Department of Psychiatry & Affective Disorders Center, Ruijin Hospital, Shanghai Jiao Tong University School of
Medicine, Shanghai, 200025, China
E-mail addresses: hellregenius@163.com (Y. Zhu), pdhsh@126.com (D. Peng), yirufang@aliyun.com (Y. Fang).
1
These authors contributed equally to this work.
Contents lists available at ScienceDirect
Heliyon
journal homepage: www.cell.com/heliyon
https://doi.org/10.1016/j.heliyon.2023.e20951
Received 11 January 2023; Received in revised form 11 October 2023; Accepted 11 October 2023
Heliyon 9 (2023) e20951
2
Conclusions: The factors may affect the therapeutic responses and compliance of patients,
increasing the non-response risk for antidepressants. Therapeutic responses might be improved by
increasing the clarication and elucidation of different symptom clusters of patients. Benets on
treatment response to AGT were not found in our study, indicating a one-size-ts-all approach
may not work.
Trial Registration: We registered as a clinical trial at the International Clinical Trials Registry
Platform (No. NCT01764867) and obtained ethical approval 2012-42 from SMHC.
1. Introduction
Currently, antidepressants are listed in the main treatment strategy for major depressive disorder (MDD). Second-generation an-
tidepressants are being recommended to be the rst-line medication with high safety and efcacy [1]. However, the treatment
outcome of depression was not optimistic in general. In the Sequenced Treatment Alternatives to Relieve Depression (STAR * D) study,
50 % of the patients did not respond to the initial trial [2]. Meanwhile, the efcacy of treatment decreases with its frequency. The nal
remission rate was only 70 %, even with the completion of potential sequenced treatment steps [3]. Approximately the remaining 30 %
patients would experience treatment-resistant depression (TRD) [4].
Although monoaminergic antidepressants have been under development for half of a century, there is no strong evidence of
sequenced treatment alternatives for implementing pharmacological strategies (monotherapy, combination, polypharmaceutical or
augmentation strategies) [5], as well as the effectiveness between any treatments at any treatment level [6]. Moreover, there is no
signicant difference between the different psychopharmacological classes used as augmenters in terms of symptom severity and
treatment response [7]. Therefore, the identication of risk factors for TRD may be useful to guide an initial trial, avoid inefcient
trial-and-error, and improve proper care for MDD [2,8]. Patients with MDD often experience side effects, inadequate responses, and
residual symptoms that interfere with compliance [9], all of which can lead to treatment discontinuation [10].
Antidepressant agents should be considered over evidence-based strategies [11], however, many physicians mostly rely on their
clinical experience for treatment decisions so that the treatment strategies are doomed to be different. Some physicians believe that
they cannot follow the guidelines for practical reasons. Compared with sequenced treatment, it is more important to adjust medication
to patient compliance for a better prognosis [12]. It may lead to inadequate or prolonged treatment (such as frequent changes among
different strategies), reducing the patient’s possibility of clinical cure.
On the other hand, choosing the most effective strategy to alleviate symptoms is an urgent direction of clinical researches on MDD.
Algorithm Guided Treatment (AGT) involves sequential progression and appropriate decision-making based on the results of clinical
evaluation, which may help improve treatment outcome, efcacy, and cost-effectiveness [13]. Therefore, the AGT was worth being
investigated by comparing with the treatment as usual (TAU), as a potential prognostic factor.
A multiple-treatments meta-analysis (117 studies, n =25,928) showed escitalopram, mirtazapine, venlafaxine, and sertraline are
more efcacious than other second-generation antidepressants and escitalopram was suggested the best prole of acceptability,
leading to signicantly fewer discontinuations [14]. A systematic review and network meta-analysis was updated (522 studies, n =
116,477), indicating escitalopram, mirtazapine, paroxetine, venlafaxine, agomelatine, amitriptyline, and vortioxetine are more
effective than other antidepressants [15]. Meanwhile, mirtazapine showed optimal acceptability in the lower range of their licensed
dose (about 30 mg), as well as selective serotonin reuptake inhibitors (SSRIs) with the lower licensed range between 20 mg and 40 mg
uoxetine equivalents [16]. Considering the most favorable balance between benets, acceptability, acquisition cost and
mechanism-based targeting of receptor for antidepressant, we chose escitalopram and mirtazapine as the antidepressant agents in
AGT. Although no advantage of treatment outcome was shown in head-to-head studies between the two antidepressants [15], mir-
tazapine was more effective in improving weight, poor appetite and biological rhythm symptoms of MDD [17].
Some studies suggested that baseline features and early symptoms could predict whether patients will respond to treatment [18,
19]. Before evaluating symptoms by clinical rating scales, the potential multicollinearity among them should be considered. In the
study design, variables from the overall symptomatologic dimension of MDD are supposed to be collected. At the same time, the
differentiated symptoms of MDD are considered able to efciently achieve a comprehensive assessment for predicting the prognosis.
Therefore, commonly used clinical rating scales are good candidates for this research. The selected scales should not only have good
discriminative validity and test-retest reliability, but also be convenient for clinical use. Thus, these corresponding results will be able
to benet internal and external uses, such as the ne distinctions made through these scales.
2. Materials and methods
2.1. Settings and participants
Data of participants were consecutively collected from four psychiatric hospitals and four psychiatric departments of general
hospitals, across seven cities from four provinces and a municipality directly under the central government of mainland China between
2012 and 2014. Patients with depressive episodes were diagnosed according to the Diagnostic and Statistical Manual of Mental
Disorders Fourth Edition Text Revision (DSM-IV-TR) criteria. The psychiatric hospitals included Shanghai Mental Health Center
(SMHC), Shanghai Changning Mental Health Center, Shanghai Hongkou Mental Health Center, Nanjing Brain Hospital, Wuhan Mental
Y. Zhu et al.
Heliyon 9 (2023) e20951
3
Health Center, the Second Afliated Hospital of Zhejiang University School of Medicine, Huzhou Third People’s Hospital and the First
Afliated Hospital of Kunming Medical University. The study was approved by the Institutional Ethical Committee for clinical research
of SMHC (2012BAI01B04) and all subcenters. The study was approved by the institutional review boards at each site, and written
informed consents were obtained from all participants prior to research entry according to the Declaration of Helsinki. This study has
been registered as a clinical trial at the International Clinical Trials Registry Platform (No. NCT01764867), and ethical approval
(2012–42) has been obtained from SMHC.
2.2. Inclusion and exclusion criteria
The inclusion criteria were：(1) aged 18–75 years, (2) meeting DSM-IV criteria for a current episode of depression, (3) Hamilton
Depression Rating Scale-17 items (HAMD-17) score ≥14 points, (4) sufcient to understand informed consent and research content;
(5) sufcient audio-visual level to complete the study, (6) necessary and suitable for receiving antidepressant treatment.
Overall exclusion criteria included：(1) physical illness (history of cerebral trauma, central nervous system disease, neuroendo-
crine disease, hepatic injury, renal injury, or heart disease), (2) severe suicide attempt (Item 3, “Suicide” score of HAMD-17 ≥3
points), (3) pregnant or lactating women, or those who have a plan for pregnancy, (4) substance abuse except nicotine, (5) drug
interaction among current medications for physical disease and the antidepressant, (6) those who have received modied electro-
convulsive therapy in the past six months, (7) history of treatment failure to any of the two protocol drugs (mirtazapine or escita-
lopram), (8) excessive use of tranquilizing drugs (≥15 mg/day diazepam equivalents).
2.3. Study design
The study was based on the Program of Algorithm Guided Treatment for Depression (PAGT*D), which is a 12-week randomized,
open label, parallel-group trail. The treatment of MDD was divided into two groups: AGT and TAU. The AGT group employed esci-
talopram and mirtazapine as protocol drugs, while TAU group was dened as an antidepressant monotherapy. A total of 987 patients
were randomized into AGT (n =663, escitalopram/mirtazapine =340/323) or TAU (n =324) group. Psychiatrists were allowed to
adjust the dose of the antidepressant within the therapeutic range. If no response was reached from the initial intervention for at least
six weeks, the treatment would be switched from mirtazapine to escitalopram and vice versa, and then a combination therapy of the
two drugs in the next treatment step for at least six weeks if necessary. Meanwhile, no more arrangements would be intervened in the
therapeutic alliance of the TAU group, starting from monotherapy. Participants were randomly allocated into one of the three
treatment strategies after screening. The measurement was administered at baseline, weeks 2, 4, 6, 8, and 12. The HAMD-17 [20] (9
items rated using a 5-point scale, from 0 =not at all to 4 =extreme and 8 items rated using a 3-point scale, from 0 =not at all to 2 =
major, with a score of >7 indicating depressive symptoms), the Hamilton Anxiety Rating Scale-14 items (HAMA-14) [21] (14 items
rated using a 5-point scale, from 0 =not at all to 4 =extreme, with a score of >7 indicating anxiety symptoms), the 6-Item Life of
Fig. 1. Flowchart of the screening process and data classication.
Y. Zhu et al.
Heliyon 9 (2023) e20951
4
Quality Questionnaires (QOL-6) [22](6 domains including physical health, psychological health, economic circumstances, work,
family relationships, and relationships with non-family associates, rated using a 5-point scale, from 1 =very poor to 5 =excellent), the
Depression and Somatic Symptoms Scale (DSSS) [23] (22 items rated using a 4-point scale, from 0 =not at all to 3 =major, with a
higher score suggesting greater severity level of depression and somatic symptoms), the Visual Analogue Scales (VAS) [24] (rated using
a 10-point scale, from 1 =no pain to 10 =severe pain, for estimating the severity of somatic pain) and the Self-report Version of the
Quick Inventory of Depressive Symptoms (QIDS-SR16) [25] (16 items rated using a 4-point scale, from 0 =not at all to 3 =major, three
depression clinical clusters—core emotional, atypical and sleep symptoms estimated by the sum of the corresponding items, with a
higher score implying greater severity level of corresponding symptoms) were used to evaluate the therapeutic outcome
retrospectively.
2.4. Participant description
Candidates who have had at least one depressive episode from a total of 1746 patients were selected. Exclusion criteria of the
candidates included the following ve reasons: (1) bipolar disorder (BD), (2) minor depression, (3) other diagnoses, (4) screening
failure, or (5) data missing. Patients were divided into two groups by whether having reached a response by 50 % improvement of the
HAMD-17 score from baseline to endpoint after three months’ antidepressant treatment [26], as MDD responsive to antidepressant
(RA) or unresponsive to antidepressant (URA). With or without 50 % reduction rate of the HAMD-17 score was the second-level
observation of each individual. The HAMD-17 response (≥50 % score reduction) was widely evaluated in meta-analysis [27,28]
and clinical guidelines [1]. Enrolled patients were supposed to keep on drug adherence until they meet an adequate reason for
medication discontinuation. After excluding the unnecessary cases, a total of 581 MDD-respond patients and 406 MDD-non-respond
patients were selected. (see Fig. 1 for a owchart of sample selection).
2.5. Statistical methods
IBM SPSS version 17.0 for Windows (Chicago Inc., USA) was applied for statistical analysis. Treatment response was the outcome,
and the two treatments (AGT vs. TAU) were the main research factors under the consideration of the adjustment for untreated
duration, comorbidity, and severity of depressive symptoms. Multivariate Cox regression analysis (enter section) was employed to
compare the therapeutic effect. Kaplan-Meier survival analysis was utilized to compare different treatments. Continuous variables
were represented as mean ±SD for normality and homogeneity. The between-group variance with skewed distribution was compared
by the Mann-Whitney U test. Categorical variables were represented as n (%) and examined by the Chi-square test. Statistical sig-
nicance was dened as p value <0.05. An odds ratio (OR) was yielded if p value <0.05. We used the last observation carried forward
method, which is a statistical approach for longitudinal repeated measures data where data of observations might be missing in the 12-
weeks follow-up.
The formula method of Box-Tidwell was used to test the linear correlation between the log
e
(In)-transformed continuous variables
and the corresponding variables. A total of 180 (90* 2) items were involved in the linear test model, rst, including sociodemographic
factors (3), clinical traits (6), psychiatric rating scales (6), and their subscales (75) [HAMD-17 (17), QOL-6 (6), HAMA-14 (14), DSSS
(22), QIDS (16)]. We tried to assessed selected depressive symptoms with psychometrically acceptable properties, making them
suitable for use in busy practices [29–31]. All of the 180 items yielding linear relationships after Bonferroni correction were applied,
with the level of 0.0003 (0.05/180). Furthermore, outliers were deleted if their Studentized Residual was more than two standard
deviations away from the mean.
Before building the regression model, factors with multicollinearity under 0.1 of tolerance and above 10 of variance ination were
excluded, including the item of the total course of depression and all of the six psychiatric rating scales. We then compiled a total of 96
variables, with 83 continuous variables including sociodemographic factors (3), clinical traits (5), and subscales (75), and 13
dichotomous categorical variables extracted from clinical traits and symptomatic assessments except the item of hospitalization, which
were dichotomized into absent (score =0) vs. present (scores =1) for the estimation of frequency. For the identication of the
objective clinical traits of the MDD-non-respond patients, a backward Wald method in the Binary Logistic regression was performed,
with a p value criterion of 0.05 and 0.10 for entry and removal, respectively. Events per variable (EPV) of the effective sample size (n =
987) to the predictor variables (n =96) were beyond the criteria (≥10) [32,33].
Finally, it yielded 20 risk factors deployed in the binary logistic regression, including three dichotomous categorical variables and
17 In-transformed continuous variables. A receiver operating characteristic (ROC) curve was examined to discriminate the accuracy of
the model by the area under the curve (AUC) of the ROC curve [34]. The actual positive state was 2 =URA group. The ability of the
prediction models was calibrated via Hosmer-Lemeshow goodness of t test [35,36].
3. Results
3.1. Comparison of sociodemographic factors, clinical traits and psychiatric rating scales between responders and non-responders of MDD
patients
In the analysis of sociodemographic factors, age and age at onset of the RA group were 2.54 years and 2.39 years older than that of
the URA group, respectively(p-value<0.05), while higher number of depressive episodes and number of untreated episodes in the RA
group were observed compared to the URA group(p-value<0.05). No statistically signicant difference was found in the other items
Y. Zhu et al.
Heliyon 9 (2023) e20951
5
between the two groups(p-value>0.05). See Table 1.
In Table 2, the six psychiatric rating scales were selected, and no difference was found in any of the scales (HAMD-17, HAMA-14,
QOL-6, DSSS, QIDS-SR16, and VAS) (p-value>0.05).
Comparison of dichotomous categorical variables between responders and non-responders of MDD Patients.
The completion of the observation period (83.8 % vs. 18.7 %), married status (70.1 % vs. 62.5 %), with anxious distress (58.4 % vs.
51.5 %), moderate to severe depressive symptoms (81.4 % vs. 72.7 %) and hospitalization (23.3 % vs. 14.7 %) in the RA group was
signicantly higher than those in the URA group, while co-morbidity (19.8 % vs. 27.0 %) in the RA group was signicantly lower (p-
value<0.05). No statistical difference was found in any of the other categorical variables. See Table 3.
3.2. Outcome analysis of treatments utilizing Kaplan-Meier survival analysis
In this dataset, 42.8 % of AGT and 37.7 % of TAU in the duration of 12 weeks were censored, a level at which bias is negligible, even
when the survival distribution is highly skewed. The result of multivariate Cox regression analysis showed no difference (p =0.459),
and the two treatments had the same risk of affecting the efcacy, with a hazard ratio (HR) =0.931 (95%CI: 0.772–1.124). No sta-
tistical difference was found in the total number of weeks in treatment response comparing AGT (10.9 weeks) vs. TAU (11.3 weeks).
Time to symptomatic remission was longer in TAU using standard survival analyses, which showed no statistically signicant (p-value
=0.11) using log-rank testing. See Fig. 2.
3.3. Binary logistic regression analysis of factors for predicting responders or non-responders of MDD patients
In this research, we tried to associate clinical traits with 96 symptomatic predictors, as shown above, through the backward method
Wald of the binary logistic regression. After eliminating the missing items, a total of 694 cases were available (416 cases in the RA
group and 278 in the URA group). The regression model was constructed by X
1
=untreated duration, X
2
=co-morbidity (0 =no, 1 =
yes), X
3
=treatment regimen (0 =TAU, 1 =AGT), X
4
=severity of depressive symptoms (mild =0, moderate to severe =1), X
5
=
HAMD-17-3, X
6
=HAMD-17-4, X
7
=HAMD-17-14, X
8
=HAMD-17-15, X
9
=HAMA-8, X
10
=HAMA-13, X
11
=HAMA-14, X
12
=QOL-6-
2, X
13
=DSSS-1, X
14
=DSSS-9, X
15
=DSSS-15, X
16
=DSSS-16, X
17
=QIDS-SR16-2, X
18
=QIDS-SR16-3, X
19
=QIDS-SR16-6, X
20
=
QIDS-SR16-10. See Table 4. Details of these subscales are shown in Supplementary Material 1. Other variables did not enter the model.
The regression equation was nally observed as follows:
Logit (P) =0.310-0.005X
1
+0.630X
2
+0.378X
3
-0.552X
4
+0.337X
5
+0.295X
6
+0.182X
7
- 0.169X
8
-0.263X
9
-0.192X
10
-0.171X
11
-
0.343X
12
+0.238X
13
-0.194X
14
+0.242X
15
+0.196X
16
-0.190X
17
+0.178X
18
-0.409X
19
-0.213X
20
.
The regression model statistics was observed (Nagelkerke R
2
=0.183,
χ
2
=100.689, p <0.001).
The ROC curve in Fig. 3 showed a fair accuracy of this model, yielding an AUC of 0.701 (95 % CI, 0.665–0.738). The decision-rule
cut-off that optimizes the sensitivity/specicity tradeoff was 0.325. There was no statistical difference between the expected and the
observed values for the model via the Hosmer-Lemeshow Goodness-of-Fit Test, in which the calibration was satisfactory (p >0.05). See
Table 5.
4. Discussion
This research was carried out based on the PAGT*D, which was primarily tried in China. Some risk factors were found by comparing
the groups of responders vs. non-responders in the rst step. The average age and age at onset of non-responders were 2.5 years and 2.4
years younger than those of the responders, respectively. This study supported the view that antidepressant treatment might be
particularly effective in older patients with MDD [37,38], while Maarsingh OR et al. [39] found early onset of MDD could identify
patients at risk of an unfavorable outcome. Earlier onset is associated with greater illness burden across a wide range of indicators [40].
Besides, the number of depressive episodes and untreated episodes before diagnosis provided the same message as age and onset
age. Physicians have suggested that patients who have experienced more recurrences are at a greater risk of recurrence, and they can
continue to benet from medication during the rst year after recovery [41]. It was investigated that about three-quarters of patients
Table 1
Sociodemographic factors of MDD patients in RA and URA groups.
Variable (x ±SD) Total (n =987) Responders (n =581) Non-responders (n =406) z p
Sociodemographic factors
Age (y) 38.86 ±14.07 40.01 ±14.08 37.47 ±13.94 2.975 0.003
Age at onset (y) 35.48 ±13.74 36.46 ±13.77 34.07 ±13.59 2.826 0.005
Education Level (y) 12.20 ±3.89 12.01 ±4.03 12.48 ±3.67 −1.780 0.075
Clinical traits
Number of hospitalizations 0.30 ±0.72 0.33 ±0.68 0.27 ±0.78 0.677 0.498
Number of depressive episodes 1.99 ±3.71 2.03 ±4.29 1.94 ±2.70 1.998 0.046
Number of untreated episodes before diagnosis 1.17 ±1.33 1.19 ±1.56 1.15 ±0.92 2.623 0.009
Untreated duration (y) 1.87 ±3.66 2.08 ±4.20 1.57 ±2.67 0.629 0.529
Current course of the depressive episodes (y) 0.93 ±1.83 0.95 ±1.91 0.89 ±1.72 1.945 0.052
Total course of depression (y) 3.63 ±5.85 3.69 ±5.79 3.55 ±5.94 0.186 0.852
Y. Zhu et al.
Heliyon 9 (2023) e20951
6
discontinued antidepressants during maintenance therapy after 24 weeks [42], while the majority of non-responders discontinued
antidepressants during maintenance therapy in clinical practice [12].
There are some factors that can be predicted. For example, patients in married status run less risk of non-response to treatment (OR:
0.710 [95 % CI: 0.535 to 0.941]). More reduced levels of social functioning (unmarried, divorced, or widowed [43,44]) play an adverse
role in treatment responding. Emotional loneliness is associated with depressive symptoms [45]. This result also indicates that family
support from marriage is vital for a favorable outcome of the disease.
Table 2
Comparison of psychiatric rating scales between responders and non-responders.
Variable (x ±SD) Responders (n =581) Non-responders (n =406) z p
HAMD-17 21.27 ±4.51 20.85 ±4.61 1.613 0.107
HAMA-14 18.42 ±6.48 17.79 ±5.70 0.825 0.409
QOL-6 15.36 ±2.73 15.28 ±2.64 0.344 0.731
DSSS 26.21 ±9.34 27.11 ±9.42 −1.811 0.070
QIDS-SR16 19.54 ±6.06 19.28 ±5.60 1.058 0.290
VAS 3.68 ±3.07 3.52 ±3.13 0.992 0.321
HAMD-17: Hamilton Depression Scale-17 items; HAMA-14: Hamilton Anxiety Rating Scale-14 items; QOL-6: 6-item Life of Quality Questionnaires;
DSSS: Depression and Somatic Symptoms Scale; VAS: Visual Analogue Scales; QIDS-SR16: Self-report Version of the Quick Inventory of Depressive
Symptoms.
Table 3
Comparison of categorical variables between responders and non-responders.
Variable (n, %) Responders Non-responders
χ
2 p OR (95% CL)
Observation period 413.4 <0.001 22.5 (16.1, 31.4)
Discontinuation 94 (16.2%) 330 (81.3%)
Adherence 487 (83.8%) 76 (18.7%)
Sex 192 (33.6%) 141 (35.2%) 0.267 0.606
Male 380 (66.4%) 260 (64.8%)
Female
Marital status 5.703 0.017 0.710 (0.535,0.941)
Unmarried 168 (29.9%) 131 (37.5%)
Married 394 (70.1%) 218 (62.5%)
Family history of mood disorders 0.172 0.679
No 511 (89.5%) 318 (90.3%)
Yes 60 (10.5%) 34 (9.7%)
Comorbidity 6.131 0.013 1.494 (1.086, 2.054)
No 388 (80.2%) 276 (73.0%)
Yes 96 (19.8%) 102 (27.0%)
Depressive episode 0.002 0.963
First 369 (64.6%) 228 (64.8%)
Recurrent 202 (35.4%) 124 (35.2%)
Treatment regimen 2.413 0.120 /
TAU 202 (34.8%) 122 (30.0%)
AGT 379 (65.2%) 284 (70.0%)
With anxious distress 4.204 0.040 0.754 (0.575, 0.988)
No 202 (41.6%) 184 (48.5%)
Yes 284 (58.4%) 195 (51.5%)
Severity 10.601 0.001 0.607 (0.449, 0.821)
Mild 108 (18.6%) 111 (27.3%)
Moderate to severe 473 (81.4%) 295 (72.7%)
Atypical symptom 0.068 0.794
No 513 (95.4%) 314 (95.7%)
Yes 25 (4.6%) 14 (4.3%)
Somatopathy 0.001 0.973 /
No 475 (81.8%) 293 (81.8%)
Yes 106 (18.2%) 65 (18.2%)
Seasonal depression 0.760 0.383 /
No 526 (93.6%) 322 (92.3%)
Yes 35 (6.2%) 27 (7.7%)
Morning depression 1.273 0.259 /
No 414 (77.4%) 242 (74.0%)
Yes 121 (22.6%) 85 (26.0%)
Hospitalization 10.185 0.001 0.563 (0.395,0.804)
No 431 (76.7%) 298 (85.2%)
Yes 131 (23.3%) 51 (14.7%)
TAU: treatment as usual; AGT: algorithm guided treatment; OR: odd radio; CL: condence interval.
Y. Zhu et al.
Heliyon 9 (2023) e20951
7
Other interesting ndings include that patients with anxious distress (OR: 0.754 [95 % CI: 0.575 to 0.988]) and moderate to severe
depressive symptoms (OR: 0.607 [95 % CI: 0.449 to 0.821]) are associated with less URA risks in the MDD-respond group than those in
the MDD-non-respond group. There are several reasons we may infer. First, the outcomes of MDD patients were evaluated by the 50 %
reduction rate of HAMD total score from baseline, thus patients with a higher baseline score may have a more substantial reduction
percent after treatment response [46]. The primary efcacy variable is the change of HAMD from baseline. More score reduction would
be reached within patients having higher baseline scores who achieved remission corresponding to HAMD scores ≤7. Second, the
average scores of HAMD and HAMA in the RA group were higher than those of the URA group (Table 2). Although there is no statistical
difference between the two groups, the higher average scores might partially explain these results. Third, due to the severer symptoms
of treatment responders, the probability of receiving hospitalization for them is higher (23.3 % vs. 14.7 %).
Discontinuing anti-depressive therapy within three months can be a strong risk factor for URA (OR: 22.5 [95 % CI: 16.1 to 31.4])
and comorbidity is also associated with a higher URA risk (OR: 1.494 [95 % CI: 1.086 to 2.054]). Patients would not give a denite
answer at the beginning of the treatment about whether they would adhere to the treatment or not. Therefore, medication adherence
and comorbidity should be considered as a factor affecting treatment outcome rather than a potential predictor in the regression model
[12]. We recommend that treatment strategies should be based on our results for mitigating symptoms.
Finally, twenty parameters were nally deployed for the predictive algorithms in the regression model, including three clinical
Fig. 2. Kaplan–Meier survival curve focusing on the occurrence and timing of treatment response.
Table 4
Binary logistic regression analysis (backward wald) of factors for predicting responders or non-responders of MDD patients.
Model β S.E. Wald p EXP(β) 95 % C.L. for EXP(B)
Lower Upper
Constant 0.310 0.499 0.385 0.535 1.363
Untreated duration −0.005 0.002 4.958 0.026 0.995 0.991 0.999
Co-morbidity 0.630 0.205 9.472 0.002 1.878 1.257 2.805
Treatment regimen 0.378 0.181 4.334 0.037 1.459 1.022 2.082
Severity of depressive symptoms −0.552 0.226 5.954 0.015 0.576 0.370 0.897
HAMD-17-3 0.337 0.110 9.472 0.002 1.401 1.130 1.736
HAMD-17-4 0.295 0.112 6.902 0.009 1.343 1.078 1.674
HAMD-17-14 0.182 0.103 3.108 0.078 1.199 0.980 1.467
HAMD-17-15 −0.169 0.100 2.859 0.091 0.844 0.694 1.027
HAMA-14-8 −0.263 0.098 7.173 0.007 0.768 0.634 0.932
HAMA-14-13 −0.192 0.105 3.362 0.067 0.825 0.672 1.013
HAMA-14-14 −0.171 0.096 3.192 0.074 0.843 0.699 1.017
QOL-6-2 −0.343 0.152 5.083 0.024 0.710 0.527 0.956
DSSS-1 0.238 0.103 5.289 0.021 1.269 1.036 1.554
DSSS-9 −0.194 0.110 3.126 0.077 0.824 0.664 1.021
DSSS-15 0.242 0.109 4.892 0.027 1.273 1.028 1.578
DSSS-16 0.196 0.110 3.156 0.076 1.216 0.980 1.509
QIDS-SR16-2 −0.190 0.091 4.423 0.035 0.827 0.692 0.987
QIDS-SR16-3 0.178 0.083 4.639 0.031 1.195 1.106 1.405
QIDS-SR16-6 −0.409 0.113 13.243 <0.001 0.664 0.533 0.828
QIDS-SR16-10 −0.213 0.111 3.669 0.055 0.808 0.649 1.005
HAMD-17: the Hamilton Depression Scale-17 items; HAMA-14: the Hamilton Anxiety Rating Scale-14 items; QOL-6: the 6-Item Life of Quality
Questionnaires; DSSS: the Depression and Somatic Symptoms Scale; QIDS-SR16: the Self-report Version of the Quick Inventory of Depressive
Symptoms.
Y. Zhu et al.
Heliyon 9 (2023) e20951
8
traits, one treatment regimen, and 16 subscales. The AUC was identied as a fair accuracy for predicting non-responding to the
treatment of MDD. The three parameters (untreated duration, comorbidity, and severity of depressive symptoms), which we have
discussed above, and the “Treatment regimen” entered the model. This model showed AGT could be a prognostic risk factor, while TAU
might be a prognostic protective factor. Namely, TAU based on the experience of doctors is likely to be more efcient than AGT,
suggesting that the immediate adjustment of treatment by the doctor’s instruction within three months can be more effective than a
xed AGT [47]. Another reason for the limited effect of AGT could be that the later treatment steps in the AGT were rarely utilized
because participants who did not receive any benet have dropped out early [48]. Even if the pharmacological strategies have been
proven to be clinically effective, they cannot be used directly without considering individual differences, which may be systematically
associated with the responses to antidepressants in MDD beyond placebo effects or statistical factors [49]. Time to symptomatic
remission by standard survival analyses was longer in TAU, although without statistically signicance.
In addition, we employed six scales, which are widely used in clinical practice, and tried to nd the relationship between clinical
characteristics and the efcacy of medications. In the remaining 16 subscales after modeling, the interpretation of the “Suicide” factor
is limited because we have excluded patients with high risk of suicide attempts (Item 3, “Suicide” score of HAMD-17 ≥3 points). Then,
“Early insomnia” and “Waking up too early” are URA predictors. We observed that patients often regard insomnia symptoms as an
assessment of whether they are remitted during the clinic service process. Treating insomnia in patients with depression has a positive
effect on mood [50]. Other symptoms of these risk factors tend to be summarized as anxiety characteristics, from which patients are
more difcult to recover. Anxiety was associated with an ineffective treatment response in MDD regardless of the treatment type [51,
52] in which insomnia plays an important role. Therefore, it is essential for the prognosis of the disease to formulate an early
insomnia-specic intervention to improve depressive and anxious mood [50]. Furthermore, “Headache” was involved. Increasing
intensity of headache is associated with comorbidities related to depression, anxiety, and insomnia [53]. Co-morbid and co-occurring
conditions in sleep disturbance might increase the risk of intensity and frequency of headache. Many antidepressant agents have
anti-anxiety effects, however, we found that “Anxious or nervous” is still a risk factor for URA. Therefore, sleep disturbance and anxiety
are prior to treatment targets for MDD with comorbidity. These predictive risk factors should be recognized frequently during clinical
interviews.
Among the protective factors, the symptoms tend to have typical depressive characteristics with the absence of comorbid illness
[38]. Patients with these protective factors (eight clinical symptoms) are more likely to be cured. It’s an important nding that the
variable of the “Hypochondriasis” is able to increase the positive treatment response. Although hypochondriasis has low recovery rates
(30%–50 %) [54], the related symptoms at baseline can be actually benecial to treatment responding where treatment compliance
Fig. 3. ROC curve of the logistic regression model for prediction of URA
Table 5
Discrimination and calibration of the recognition algorithms for predicting MDD with or without response.
Youden index AUC (95%CI) Sensitivity Specicity Balanced Accuracy Calibration
a
χ
2 p
Backward Wald 0.325 0.701 (0.665, 0.738) 0.708 0.617 0.663 10.102 0.258
a
Hosmer-Lemeshow Goodness-of-Fit Test; AUC, the area under the curve.
Y. Zhu et al.
Heliyon 9 (2023) e20951
9
may take effect. It may be the reason for becoming a protective factor although itself is a symptom. Regarding sleep disturbance, on the
contrary to “Early insomnia” and “Waking up too early”, “Sleep during the night (sleep quality)” is acceptable for patients and does not
affect the outcome of treatment for MDD. Everitt H et al. [55]. found a moderate improvement in subjective sleep quality over placebo.
A better prediction of prognosis may result from perceived sleep quality. Reduction of anticipatory stress is associated with improved
subjective sleep quality on a day-to-day basis, regardless of the severity of insomnia [56]. Other protective factors are closely related to
the effectiveness of the antidepressant agents on the above symptoms.
5. Conclusions
At present, many psychiatric rating scales are used clinically. However, comprehensive evaluation for each patient based on scales
is far from cost-effectiveness, which cannot be used comprehensively in clinical practice. Baseline characteristics allow prediction of
non-response, which could be sufciently certain for physicians to identify patients with prolonged exposure to ineffective treatment,
thereby personalizing depression management as well as saving time, cost, and medical resources [18]. Therefore, the 16 subscales we
extracted from the six commonly used scales would greatly save valuable time for clinicians. We found that the baseline scores of these
encoded items: Suicide, Early insomnia, Genital symptoms, Mental state in the last month (reverse scoring of severity), Headache,
Shortness of breath or difculty breathing, Anxious or nervous, and Waking up too early contributed to non-responding to treatment,
while Hypochondriasis, Somatic (sensory), Autonomic symptoms, Behavior at interview, Dizziness, Sleep during the night, Decreased
appetite and Concentration/decision-making are protective factors of treatment responding. A fairly accurate model was constructed
for the prognosis by assessing the 3-months treatment. Under the consideration of the unfavorable prognostic factors, combination
drug therapy, physical therapy, or psychological therapy could be considered able to improve the outcome of MDD patients by
increasing the clarication and elucidation of different symptom clusters to them. Measurement-based care is dened as the clinical
practice where physicians collect patient data through validated outcome scales and use the results to guide their decision-making
processes [57] may increase chances of achieving remission of MDD. Benets on treatment response to AGT were not found in our
study, indicating a one-size-ts-all approach may not work.
6. Limitations
There were several main limitations of the cohort study. First, the relatively high drop-out rates were observed, especially in the
AGT group (42.8 % of AGT and 37.7 % of TAU at week 12). The strict-guided strategy may be responsible for the high drop-outs in the
AGT that required patients to stick to at least six weeks of the initial intervention [58]. Although a statistical method was conducted for
missing value inputation, loss to follow-up bias cannot be ruled out. Second, misdiagnosis rate of depressive episode (unipolar or
bipolar) still existed even after we excluded the patients with BD (n =198). The misdiagnosis and BD-conversion will increase with
observation time [59]. Third, the AUC of 0.701 which is dened as a balanced statistics, presents the authenticity of the detection
method. The discrimination accuracy of the model was examined by using the AUC of the ROC curve and was categorized as fair
(0.70–0.80), almost poor (0.60–0.70), indicating the limited use in clinical practice. Further researches are required to validate our
results and explore the predictors or moderators with a long follow-up period, adjusted parameters, and quality control of drop-outs.
Funding information
The work was supported by the National Key Research and Development Program of China (2016YFC1307100, 2016YFC1307105),
the National Natural Science Foundation of China (81801338, 81771465, 81930033,81971269), Shanghai Key Medicine Specialties
Program (ZK2019A06), Shanghai Clinical Research Center for Mental Health (SCRC-MH, 19MC1911100),Shanghai Science and
Technology Committee (YDZX20213100001003), the Special Project for Clinical Research in Health Industry of Shanghai Municipal
Health Commission (20204Y0025), the National Key Technologies R&D Program of China (2012BAI01B04), the Major Project of
Scientic Research of Shanghai Hongkou District Health Commission (2101-03), the Excellent Talent Training Program of Shanghai
Hongkou Mental Health Center (2023XKDTR01) and the Innovative Research Team of High-level Local Universities in Shanghai. The
authors would also like to acknowledge and thank all participants in this study and all members of the PAGT*D team.
Data availability statement
Data will be made available on request.
CRediT authorship contribution statement
Yuncheng Zhu: Writing – review & editing, Writing – original draft, Investigation. Yuncheng Zhu: Performed the experiments and
wrote the paper. Fang Wang: Analyzed and interpreted the data，and wrote the paper. Fan Wang: Performed the experiments.
Hongmei Liu: Performed the experiments. Xiaoyun Guo: Performed the experiments. Zuowei Wang: Performed the experiments.
Ruoqiao He: contributed reagents, materials, analysis tools or data. Xiaohui Wu: Performed the experiments. Lan Cao: Performed the
experiments. Zhiguo Wu: Performed the experiments. Daihui Peng: conceived and designed the experiments. Yiru Fang: Conceived
and designed the experiments.
Y. Zhu et al.
Heliyon 9 (2023) e20951
10
Declaration of competing interest
The authors declare that they have no known competing nancial interests or personal relationships that could have appeared to
inuence the work reported in this paper.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.heliyon.2023.e20951.
References
[1] S.H. Kennedy, et al., Canadian network for mood and anxiety treatments (CANMAT) 2016 clinical guidelines for the management of adults with major
depressive disorder: section 3. Pharmacological treatments, Can J Psychiatry 61 (9) (2016) 540–560.
[2] B.N. Gaynes, et al., Treating depression after initial treatment failure: directly comparing switch and augmenting strategies in STAR*D, J. Clin.
Psychopharmacol. 32 (1) (2012) 114–119.
[3] H.N. Chan, et al., Pharmacological treatment approaches to difcult-to-treat depression, Med. J. Aust. 199 (S6) (2013). S44-S47.
[4] D.F. Ionescu, J.F. Rosenbaum, J.E. Alpert, Pharmacological approaches to the challenge of treatment-resistant depression, Dialogues Clin. Neurosci. 17 (2)
(2015) 111–126.
[5] Y. Zhu, et al., Clinical guideline (CANMAT 2016) discordance of medications for patients with major depressive disorder in China, Neuropsychiatric Dis. Treat.
19 (2023) 829–839.
[6] M. Sinyor, A. Schaffer, A. Levitt, The sequenced treatment alternatives to relieve depression (STAR*D) trial: a review, Can J Psychiatry 55 (3) (2010) 126–135.
[7] M. Dold, et al., Pharmacological treatment strategies in unipolar depression in European tertiary psychiatric treatment centers - a pharmacoepidemiological
cross-sectional multicenter study, Eur. Neuropsychopharmacol 26 (12) (2016) 1960–1971.
[8] Z. Nie, et al., Predictive modeling of treatment resistant depression using data from STAR*D and an independent clinical study, PLoS One 13 (6) (2018),
e0197268.
[9] Q. Zhou, et al., Clinical characteristics associated with therapeutic nonadherence of the patients with major depressive disorder: a report on the National Survey
on Symptomatology of Depression in China, CNS Neurosci. Ther. 11 (10) (2018), 13030.
[10] M.H. Trivedi, How can measurement-based care help improve treatment outcomes for major depressive disorder in primary care? J. Clin. Psychiatry 81 (2)
(2020), UT17042BR2C.
[11] Y. Zhu, et al., Hypothalamic-pituitary-end-organ axes: hormone function in female patients with major depressive disorder, Neurosci. Bull. 37 (8) (2021)
1176–1187.
[12] Y. Zhu, et al., Causes of drug discontinuation in patients with major depressive disorder in China, Prog. Neuro-Psychopharmacol. Biol. Psychiatry 96 (2020),
109755.
[13] M. Bauer, et al., Algorithms for treatment of major depressive disorder: efcacy and cost-effectiveness, Pharmacopsychiatry 52 (3) (2019) 117–125.
[14] A. Cipriani, et al., Comparative efcacy and acceptability of 12 new-generation antidepressants: a multiple-treatments meta-analysis, Lancet 373 (9665) (2009)
746–758.
[15] A. Cipriani, et al., Comparative efcacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a
systematic review and network meta-analysis, Lancet 391 (10128) (2018) 1357–1366.
[16] T.A. Furukawa, et al., Optimal dose of selective serotonin reuptake inhibitors, venlafaxine, and mirtazapine in major depression: a systematic review and dose-
response meta-analysis, Lancet Psychiatr. 6 (7) (2019) 601–609.
[17] H. Huang, et al., Difference in the regulation of biological rhythm symptoms of Major depressive disorder between escitalopram and mirtazapine, J. Affect.
Disord. 296 (2022) 258–264.
[18] A.Y. Kuk, J. Li, A.J. Rush, Recursive subsetting to identify patients in the STAR*D: a method to enhance the accuracy of early prediction of treatment outcome
and to inform personalized care, J. Clin. Psychiatry 71 (11) (2010) 1502–1508.
[19] F. Wang, et al., Association between olfactory function and inhibition of emotional competing distractors in major depressive disorder, Sci. Rep. 10 (1) (2020)
6322.
[20] S. Rachel, The Hamilton rating scale for depression, Occup. Med. 65 (7) (2015) 340.
[21] E. Thompson, Hamilton rating scale for anxiety (HAM-A), Occup. Med. 65 (7) (2015) 601.
[22] S.C. Gill, et al., Validity of the mental health component scale of the 12-item Short-Form Health Survey (MCS-12) as measure of common mental disorders in the
general population, Psychiatr. Res. 152 (1) (2007) 63–71.
[23] C.I. Hung, et al., Depression and somatic symptoms scale: a new scale with both depression and somatic symptoms emphasized, Psychiatr. Clin. Neurosci. 60 (6)
(2006) 700–708.
[24] Y.T. Sung, J.S. Wu, The visual Analogue scale for rating, ranking and paired-comparison (VAS-RRP): a new technique for psychological measurement, Behav.
Res. Methods 50 (4) (2018) 1694–1715.
[25] Y. Feng, et al., The psychometric properties of the Quick inventory of depressive symptomatology-self-report (QIDS-SR) and the patient health questionnaire-9
(PHQ-9) in depressed inpatients in China, Psychiatr. Res. 243 (2016) 92–96.
[26] T. Guo, et al., Measurement-based care versus standard care for major depression: a randomized controlled trial with blind raters, Am. J. Psychiatr. 172 (10)
(2015) 1004–1013.
[27] E.S. Weitz, et al., Baseline depression severity as moderator of depression outcomes between cognitive behavioral therapy vs pharmacotherapy: an individual
patient data meta-analysis, JAMA Psychiatr. 72 (11) (2015) 1102–1109.
[28] M.M. Maslej, et al., Individual differences in response to antidepressants: a meta-analysis of placebo-controlled randomized clinical trials, JAMA Psychiatr. 78
(5) (2021) 490–497.
[29] S. Ma, et al., The patient health questionnaire-9 vs. The Hamilton rating scale for depression in assessing major depressive disorder, Front. Psychiatr. 12
(747139) (2021).
[30] S. Li, et al., Development and preliminary validation of the 6-item short form of the Wisconsin stone quality of life questionnaire, Urology 7 (23) (2023).
[31] A.J. Rush, et al., Psychometric and clinical evaluation of the clinician (VQIDS-C(5)) and self-report (VQIDS-SR(5)) versions of the very Quick inventory of
depressive symptoms, Neuropsychiatric Dis. Treat. 18 (2022) 289–302.
[32] M. Pavlou, et al., Review and evaluation of penalised regression methods for risk prediction in low-dimensional data with few events, Stat. Med. 35 (7) (2016)
1159–1177.
[33] M. van Smeden, et al., Sample size for binary logistic prediction models: beyond events per variable criteria, Stat. Methods Med. Res. 28 (8) (2019) 2455–2474.
[34] D.W. Joyce, J. Geddes, When deploying predictive algorithms, are summary performance measures sufcient? JAMA Psychiatr. 77 (5) (2020) 447–448.
[35] P. Paul, M.L. Pennell, S. Lemeshow, Standardizing the power of the Hosmer-Lemeshow goodness of fit test in large data sets, Stat. Med. 32 (1) (2013) 67–80.
Y. Zhu et al.
Heliyon 9 (2023) e20951
11
[36] Y. Zhu, et al., Biochemical and endocrine parameters for the discrimination and calibration of bipolar disorder or major depressive disorder, Front. Psychiatr. 13
(2022), 875141.
[37] L. van Diermen, et al., Prediction of electroconvulsive therapy response and remission in major depression: meta-analysis, Br. J. Psychiatry 212 (2) (2018)
71–80.
[38] G.I. Keitner, et al., Recovery and major depression: factors associated with twelve-month outcome, Am. J. Psychiatr. 149 (1) (1992) 93–99.
[39] O.R. Maarsingh, et al., Development and external validation of a prediction rule for an unfavorable course of late-life depression: a multicenter cohort study,
J. Affect. Disord. 235 (2018) 105–113.
[40] S. Zisook, et al., Effect of age at onset on the course of major depressive disorder, Am. J. Psychiatr. 164 (10) (2007) 1539–1546.
[41] R. Dawson, et al., Maintenance strategies for unipolar depression: an observational study of levels of treatment and recurrence, J. Affect. Disord. 49 (1) (1998)
31–44.
[42] W.Y. Jung, et al., Times to discontinue antidepressants over 6 Months in patients with major depressive disorder, Psychiatry Investig 13 (4) (2016) 440–446.
[43] N. Markkula, et al., Prevalence and correlates of major depressive disorder and dysthymia in an eleven-year follow-up–results from the Finnish Health 2011
Survey, J. Affect. Disord. 173 (2015) 73–80.
[44] A. Lasalvia, et al., Global pattern of experienced and anticipated discrimination reported by people with major depressive disorder: a cross-sectional survey,
Lancet 381 (9860) (2013) 55–62.
[45] M. Stroebe, W. Stroebe, G. Abakoumkin, The broken heart: suicidal ideation in bereavement, Am. J. Psychiatr. 162 (11) (2005) 2178–2180.
[46] S. Leucht, et al., Translating the HAM-D into the MADRS and vice versa with equipercentile linking, J. Affect. Disord. 226 (2018) 326–331.
[47] A. Hanbury, et al., Immediate versus sustained effects: interrupted time series analysis of a tailored intervention, Implement. Sci. 8 (2013) 130.
[48] A. Yoshino, et al., Algorithm-guided treatment versus treatment as usual for major depression, Psychiatr. Clin. Neurosci. 63 (5) (2009) 652–657.
[49] M.M. Maslej, et al., Individual differences in response to antidepressants: a meta-analysis of placebo-controlled randomized clinical trials, JAMA Psychiatr. 77
(6) (2020) 1–12.
[50] M.A. Gebara, et al., Effect of insomnia treatments on depression: a systematic review and meta-analysis, Depress. Anxiety 35 (8) (2018) 717–731.
[51] J.W. Tiller, Depression and anxiety, Med. J. Aust. 199 (S6) (2013) S28–S31.
[52] J. Deckert, A. Erhardt, Predicting treatment outcome for anxiety disorders with or without comorbid depression using clinical, imaging and (epi)genetic data,
Curr. Opin. Psychiatr. 32 (1) (2019) 1–6.
[53] D.C. Buse, et al., Comorbid and co-occurring conditions in migraine and associated risk of increasing headache pain intensity and headache frequency: results of
the migraine in America symptoms and treatment (MAST) study, J. Headache Pain 21 (1) (2020) 23.
[54] T.C. olde Hartman, et al., Medically unexplained symptoms, somatisation disorder and hypochondriasis: course and prognosis. A systematic review,
J. Psychosom. Res. 66 (5) (2009) 363–377.
[55] H. Everitt, et al., Antidepressants for insomnia in adults, Cochrane Database Syst. Rev. 5 (5) (2018) CD010753.
[56] V.J. Block, et al., The effect of anticipatory stress and openness and engagement on subsequently perceived sleep quality-An Experience Sampling Method study,
J. Sleep Res. 18 (10) (2019), 12957.
[57] R.H. Hong, et al., Implementing measurement-based care for depression: practical solutions for psychiatrists and primary care physicians, Neuropsychiatric Dis.
Treat. 17 (2021) 79–90.
[58] L. Yang, et al., Predictors and moderators of quality of life in patients with major depressive disorder: an AGTs-MDD study report, J. Psychiatr. Res. 138 (2021)
96–102.
[59] Y. Zhu, et al., Employing biochemical biomarkers for building decision tree models to predict bipolar disorder from major depressive disorder, J. Affect. Disord.
308 (2022) 190–198.
Y. Zhu et al.
