Using a nationwide database in the current study, one of three MDD patients have TRD. Furthermore, our results showed that patients with multiple psychiatric comorbidities have a lower ‘survival rate’ (not developing TRD) than those without these conditions (31% vs. 85%). The large registry-based claims data are generally representative as the registration rate is high (97%) and all data regarding prescribed medications are documented. This type of study is particularly beneficial when long-term clinical management is recorded, including the duration of depressive episodes within a pre-specified patient population and clinical characteristics (e.g., comorbid conditions) [6, 32]. However, claims data may not include the assessment of subjective treatment response and treatment adherence, and a poor response to medication may be underestimated. The definition of TRD in the current study was adapted from a previous study of claims data and was determined based on whether the patient’s antidepressant treatment regimen was altered two or more times [5-7]. However, we cannot dismiss the possibility that some patients who had continued taking medication still experienced severe depression. To complement the criteria of altering medication, this definition was further added: if the patient received two or more adequate antidepressant treatment regimens but was subsequently admitted in a psychiatric acute ward. This amended definition may represent a useful proxy to capture treatment failure and allow for better identification of TRD in patients who remain in clinical settings [7, 33, 34].
The current study identified the female sex as a risk factor of TRD (aHR=1.24). A cross-sectional study in the UK that used questionnaires to collect treatment history found that 70% of patients with TRD were females [34], which is consistent with our findings. Furthermore, increased vulnerability to depression in women begins at puberty and declines after menopause [35]. One possible explanation is that the hypothalamic–pituitary–adrenal function (cortisol levels) is more likely to fluctuate in response to stressors and during depressive episodes in women [36] . Redox imbalance elicited by estrogen has been revealed to weaken enzymatic antioxidant defenses, which may be linked to TRD [37, 38]. Although several physical comorbidities increase the risk of depression [39], it is unknown whether these comorbidities increase the risk of TRD. The current study illustrated that patients with TRD had more physical comorbidities than those without TRD during long-term follow-up, including diabetes, FGIDs, and thyroid dysfunction. Growing evidence suggests a bidirectional communication between the gastrointestinal tract and brain [40]. An unhealthy gastrointestinal tract may result in treatment complications and reduce responsiveness to antidepressant treatment. Other studies have found that antidepressants may be used to treat FGIDs [41]. Previous studies have also indicated that thyroid dysfunction, which is more prevalent in women, also influences treatment outcomes [37, 42]. The aHR of these physical comorbidities ranged from 1.09 (diabetes mellitus) to 1.19 (FGIDs). Taken together, these risk factors have a mild yet significant impact on the development of TRD and required further study.
We reported that long-term psychiatric comorbidities are important and independent risk factors of TRD development. In particular, anxiety disorders were observed in >80% of MDD patients in this nationwide database. We reported a 2.02-fold higher risk of TRD development in patients with comorbid anxiety disorders. In addition, our results revealed that anxiety is the strongest predictor of TRD with a high PAF. It is well-known that anxiety and depression/TRD often coexist [22, 23, 25], and that anxiety disorders have been identified as a risk factor of poor treatment response[43]. A previous study also found that anxiety disorders, particularly panic disorder, are associated with TRD [44]. There is evidence that depression and anxiety disorders have additive effects on increased disability and poorer recovery from medical illness [45]. Patients with comorbid depression and anxiety have more severe somatic symptoms than those MDD patients without anxiety [46], which may predispose them to sensitive to side effects from medication and dropping out of treatment prematurely [47]. In addition, comorbid anxiety in depressed patients is associated with poorer social function [45], which may correlate with poor social support. Besides, depression with comorbid symptoms of anxiety is associated with greater suicidal ideation [45]. Possibly as a result of increased illness severity, depressed patients are more likely to be prescribed benzodiazepines if they have anxiety symptoms [48]. It was also consistent with our finding of high usage of benzodiazepines use, because these drugs were prescribed two-fold more often in the TRD group than in the non-TRD group. This suggests that even in the early phase of MDD, comorbid anxiety disorders represent a red-flag in clinical settings that increases the difficulty of patient care and management and increases the risk of TRD development.
Except anxiety disorders, there are several psychiatric comorbidities correlated with TRD. A previous study found that certain personality traits, such as low cooperativeness and high neuroticism, positively correlate with TRD [13]. Personality disorders correlate with poor drug adherence, poor social function, and more SUDs, which may further result in TRD [49]. Our findings are in line with the observation that personality disorders are associated with a 1.23-fold higher risk of TRD. In addition, SUDs represent another important comorbidity condition of MDD [50]. One review article has found that mood disorders are common in patients with SUDs, whereas in many individuals, the mood disturbance cannot be attributed to the acute effects of substance use or withdrawal [51]. Mood disorders may trigger individuals to use drugs and alcohol to cope with negative affective states. Substance use is often viewed as a self-medication method to alleviate depression or anxiety. Meanwhile, substance use may mask underlying mood disturbances and aggravate symptoms, ultimately leading to TRD [52]. SUDs are also risk factors of several psychiatric disorders, including anxiety disorders; therefore, the effect on the increased risk of TRD may be confounded by SUDs. After adjustment in multiple Cox regression models, we reported that several psychiatric disorders remained independent risk factors.
The relationships between comorbidities and TRD have been examined in a cross-sectional manner without clear temporal consequences in the literature [6, 12, 13]. In the present study, using Cox regression models and plots with survival function, our results demonstrated the temporal relationship that early and long-term psychiatric comorbidities significantly increase the risk of subsequent treatment resistance events. The aHRs of psychiatric comorbidities ranged from 1.19 to 2.02; the point estimates are generally higher than those of physical comorbidities. Furthermore, approximately 70% of patients with multiple lifetime psychiatric comorbidities developed TRD in this study. On the other hand, non-organic psychosis, and severe depression at the first year of diagnosis reduced the risk of TRD. It is possible that patients with initial severe and psychotic depression may have been treated more thoroughly or prescribed with a higher dose of antidepressant medication. Taken together, the comorbidity profiles of TRD are important and warrant further evaluation for the early detection of TRD.
In the current study, the patients in the TRD group visited non-psychiatric clinics more often than those in the non-TRD group. Patients in the TRD group more commonly visited non-psychiatric clinics than psychiatric clinics during their first year of treatment. Interestingly, 65% of patients with depression sought help at general medical clinics prior to a psychiatric clinic [53]. Another clinical study reported that somatic symptoms prevailed in a great majority of depressed patients ; these patients may initially seek non-psychiatric medical care before psychiatric care [54]. Patients with depression may delay proper psychiatric treatment, resulting in an increased risk of treatment resistance. Therefore, it is important to evaluate patients with depression who have frequently visit non-psychiatric clinics and provide them suitable treatment. Furthermore, both health-seeking behavior and drug compliance may be correlated with treatment resistance. Adherence was found to play an important role in TRD development and increased the risk of mortality [8, 14, 26]. In the claims database, adherence was not easy to assess. On the other hand, patients with very poor drug adherence were unlikely to fulfill the definition of TRD for having a number of antidepressant trials. Our analysis revealed no significant difference in the mortality rate between the TRD and non-TRD groups, possibly due to the relatively young age of the patients and relatively short follow-up period in the database.
We also found that patients in the TRD group had significantly higher doses of psychotropic prescriptions and a greater incidence of long-term use than those in the non-TRD group within 1 year of MDD diagnosis. The antidepressants dose remained significantly associated with TRD in multiple analysis. Long-term treatment with antidepressants might have contributed to the subsequent treatment resistance [55]. There are some possible explanations for this phenomenon. First, treating patients with mood disorders using antidepressants may have a paradoxical effect (exacerbate depression) [56, 57]. Second, previous studies have described antidepressant-induced switching and cycle acceleration in patients with BpD [58]. In the current study, patients whose diagnosis was converted to BpD experienced a 1.72-fold higher risk of developing TRD. Third, tolerance to antidepressants has been reported [59]. Patients with TRD also had more complex treatment regimens, because it is a common strategy to use higher doses and adjunctive psychotropics to counter poor response to medication [60]. A recent meta-analysis of placebo-controlled trials demonstrated that adjunctive antipsychotics were effective for the treatment of TRD [61]. Another meta-analysis demonstrated the efficacy of lithium augmentation with antidepressants compared with the placebo for the treatment of TRD [62]. However, because the current study was not a randomized, placebo-controlled trial, we did not compare a combined treatment regimen with monotherapy. Rather, we found that patients eventually developing TRD required early attention and more complex treatment at an early disease stage. The results of the current study suggest that early prescription patterns in patients represents a proxy for the combined treatment regimen and poor response/prognosis.
Current treatment regimens for depression sometimes have limited efficacy. Novel antidepressants with different mechanisms are currently used as treatment options for TRD, including agents targeting the hypothalamic–pituitary–adrenal axis, glutamate/GABA system, or opioid receptors [63]. One promising agent is ketamine, which is a noncompetitive NMDA receptor antagonist that can produce rapid and stable antidepressant effects [64]. Several ongoing clinical trials of ketamine for the treatment of TRD are ongoing in Taiwan; they have revealed satisfactory effects [65, 66]. Other widely discussed antidepressant agent are cannabinoid compounds because the endocannabinoid system is widely distributed throughout the body and serves to modulate mood symptoms [67].
This study has several limitations that need to be considered when interpreting the results. First, structured interview data were not available because we solely relied on claims data. Furthermore, more detailed clinical features and the assessments of disease severity were not available. Second, we could not obtain information on uninsured subjects. However, this study is highly representative of the target population because approximately 97% of the population was insured. Third, we targeted patients who had ever been admitted to a hospital because of depressive episodes. The results may not be extrapolated to all MDD patients, particularly to those who had never been admitted. Fourth, it was not possible to evaluate patients’ true treatment compliance. Nevertheless, we attempted to exclude those with poor compliance (e.g., patients filling prescriptions for less than 56 days). Fifth, we intended to investigate the early features of the TRD group; however, we did not have information on the subjects before they sought medical help. Sixth, we did not consider the patient’s lifestyle and personal history (i.e., coping strategies, exercise habits, smoking, psychological trauma, etc.), which might have influenced the incidence of TRD. Finally, the mortality rate may be underestimated because death records were missing from the dataset and we cannot clarify the primary causes of death. Therefore, future studies are warranted to link this database with national death records to answer such questions.