Introduction
Treatment-resistant depression (TRD) is common among individuals with major depressive disorder (MDD), with over one-third of patients being unable to reach an adequate level of symptomatic improvement after trying an antidepressant (AD) treatment ( ). However, it is unclear which factors may lead to poor response to treatment, categorically defined as “treatment resistance.” It is also unclear whether the lack of response to a medication or to a class of medications has clinical implications for long-term prognosis, or whether it has implications for the choice of the next AD treatment. Possible predictors of TRD have been investigated with different methodologies, from prospective clinical trials to large epidemiological studies; however, most of the available data addressing predictors of TRD is derived from post hoc analysis of data from randomized trials. The predictors derived from post hoc studies have significant limitations, including difficulties with replicating positive results across studies, and very few predictors have been tested prospectively. One of the major causes of conflicting results is the use of different definitions of TRD across studies, as the number of failed AD trials required to meet the definition of TRD varies substantially from “nonresponse to one prior treatment” to “nonresponse to three or more prior ADs.” Additional factors such as type of treatment considered (i.e., medications, psychotherapy, ECT), duration, and adequacy of doses, may also differ across studies, further complicating the definition of TRD ( ).
The purpose of this chapter is to review and summarize the literature reporting clinical and epidemiological factors that have been associated with treatment resistance in patients with MDD. Each of those factors has been investigated for possible correlation with the presence of the categorical phenotypes of “non-response” or “non-remission.”
Clinical predictors of TRD that have been commonly explored include those related to the timeline of the illness, such as age of onset, duration and number of episodes, and chronicity (i.e., symptoms of depression persisting unabated for more than 2 years). Some of the other factors considered are related to markers of depression severity including, for example, psychiatric hospitalizations, suicide attempts, the presence of psychotic symptoms, or patterns of general medical or psychiatric comorbidity. Other factors investigated are related to specific features of MDD (i.e., melancholic, atypical, or anxious depression), family history, and personality features. Finally, epidemiological predictors of TRD include gender, ethnicity, psychosocial and environmental factors (i.e., employment, marital status, educational status, adverse life events), and global level of functioning.
Clinical predictors
Clinical subtypes
Although some studies have suggested that individuals with different depressive subtypes, including anxious, melancholic, atypical, mixed, and psychotic depression, respond differentially to ADs, there are inconsistencies in the literature ( ). For example, regarding the “atypical” subtype of depression, in the 1970s researchers reported superior efficacy of MAOI compared to tricyclics for patients with this subtype, and additional trials were conducted in the 1980s ( ); however, this association has not been investigated with more recently developed ADs and it is of limited utility for clinicians due to the infrequent use of MAOIs, as discussed in greater detail elsewhere in this volume. Regarding “melancholic” depression, some studies showed that patients with this subtype and higher depression severity may not respond as well to SSRIs compared to venlafaxine or tricyclic ADs ( ) and that melancholic features are associated with TRD (defined as failure to respond to at least two consecutive AD trials) ( ), but these observations were not supported by other studies ( ). In an uncontrolled parallel design study using sertraline, citalopram, and venlafaxine, there was no clear association between depressive subtype and clinical outcome ( ). It should be noted that, often, the percentage of patients classified as having a pure-form subtype (i.e., melancholic) or more than one subtype (mixed) is fairly similar (approximately one-third of the sample), further complicating the classification ( ). The presence of anxiety symptoms in the context of a major depressive episode, or so-called “anxious depression,” seems to confer increased risk for nonresponse to ADs in a manner that is similar to having a comorbid anxiety disorder, and several groups have found that anxiety symptoms at baseline predicted poorer or delayed response to treatment ( ; ; ; ).
Depression severity
Among individuals treated with antidepressants, the severity of illness itself is commonly considered a factor contributing to resistance ( ; ; ). In a prospective study that included multiple next-step stages of treatment for depressed antidepressant-treated patients, found that higher depression severity at baseline was associated with poorer response to SSRIs (fluoxetine and paroxetine). Similar results were reported by who found that higher levels of depression severity at the beginning of follow-up was associated with TRD in a European multicenter study, in addition to other factors including comorbid anxiety disorders, comorbid personality disorders, presence of recurrent episodes, number of psychiatric hospitalizations, early age at onset of index depressive episode, and a lifetime history of poor response to the first antidepressant prescribed. However, the association between depression severity and TRD has not been consistently replicated. For example, in a post hoc analysis of pooled data from three randomized trials of pharmacotherapy, psychotherapy, or combination therapy for depression, reported that lower baseline depression severity, and not higher severity, was associated with non-response to treatment. Other researchers explored the relationship between symptom severity at baseline and response to escitalopram, compared with response to several other SSRIs and SNRIs combined (citalopram, duloxetine, fluoxetine, paroxetine, sertraline, and venlafaxine) ( ). In that study, higher depression severity was associated with lower response rates to all the other antidepressants combined, while there was no association between baseline symptom severity and response to escitalopram.
Chronicity
One of the more commonly reported factors associated with non-response to ADs is chronicity, or duration of the MDD episode of more than 2 years ( ; ; ; ). These findings of duration of episode overlap with those of , reporting nonresponse to psychotherapy associated with a longer duration of depression. Interestingly, this study explored psychotherapy, pharmacotherapy, and the combination of the two, and showed that patients with chronic depression exhibited a lower response to psychotherapy alone. However, the predictive effect of duration of episode was no longer significant when considering pharmacotherapy only. In a more recent, large, single-blind, 7-month prospective randomized trial, chronicity of depression did not appear to differentially impact acute or longer-term outcome with SSRI monotherapy or with combination of antidepressants ( ). In other studies the number of previous hospitalizations, a possible marker of both severity and chronicity, was significantly associated with non-response, however, this may lead to circular definition, as non-response itself may lead to an increase in the rate of hospitalizations ( ; ).
Treatment adherence
It is intuitive that patients who prematurely self-discontinue treatment, or have poor compliance, may be at higher risk for depressive relapse or for developing a more chronic symptomatic course. A repeated pattern of interruptions of pharmacologic treatment may be due to several factors including poor access to care or financial barriers, and these may overall lead to worse prognosis or be correlated with other adverse prognostic factors in people with depression. In an interesting study that explored the relationship between adherence to depression treatment guidelines (4–9 months of AD therapy following acute symptom remission) and recurrence of depression in a sample of Medicaid beneficiaries with diagnosed depression ( ), patients who stopped their treatment prematurely were at higher risk of experiencing a relapse, while those who continued with their AD treatment were the least likely to relapse. Similar findings were reported in a more recent study that suggested a fivefold increase in nonresponse associated with nonadherence to psychotherapeutic treatment ( ). Another variable to consider that is related to care not consistent with prescription guidelines and may be considered parallel to the general principle of adherence is the prescribing of inadequate doses of antidepressant treatment, which has been consistently associated with non-response ( ; ; ). However, in most cases, treatment non-adherence or under-prescribing (i.e., inadequate doses or inadequate duration of trials) are generally considered to be causes of “pseudo-resistance,” apparent non-response due to inadequately conducted therapeutic trials, as opposed to depression that has responded poorly to vigorous, evidence-based, guideline-consistent treatment.
Psychiatric comorbidities
Anxiety disorders
The association between co-occurring anxiety disorders and higher likelihood of TRD has been replicated in several studies ( ; ; ; ; ) and, specifically, the presence of one or more comorbid anxiety disorders has been repeatedly associated with lower likelihood of response to ADs ( ; ; ; ). As a specific example, the large multicenter study conducted by referenced earlier showed that poor response to antidepressant treatment was significantly associated with comorbid anxiety in general (OR 2.6) and panic disorder (OR 3.2) and social phobia (OR 2.1), specifically. These general observations were replicated in a large, multicenter, multistage clinical trial that was designed to study clinical and genetic factors associated with TRD, wherein comorbid anxiety disorders were only one of two significant clinical predictors (the other being lifetime history of non-response to an initial antidepressant trial) ( ).
Personality disorders
A comorbid diagnosis of personality disorder has been previously found to be associated with AD non-response ( ; ); however, a straightforward clinical interpretation of these findings has been made more challenging by the results of a study by our group wherein nearly half of subjects classified as having maladaptive personality traits or having a diagnosis of personality disorder during an acute depressive episode will no longer meet criteria for personality disorder after successful treatment with antidepressants ( ). These results raise the question of whether maladaptive behaviors and cognitive patterns that are observed in at least a subset of acutely depressed patients are actually state-dependent depressive symptoms that respond well to AD treatment, rather than being trait-like characteristics that are more resistant to pharmacotherapy. Managing TRD in patients with comorbid personality disorders is discussed in greater detail elsewhere in this volume.
Childhood trauma
Childhood trauma has also been repeatedly associated with poor outcome with AD treatment and higher levels of depression severity in people with TRD. For example, in a sample of depressed adults, persons with TRD reported significantly greater levels of emotional abuse during childhood than individuals classified as having treatment-responsive depression ( ). More recently, reported very high rates of childhood abuse (64% of the sample) in a cohort of patients with TRD who were previously psychiatrically hospitalized in an inpatient affective disorders unit. In that study, childhood adversity was associated with having an earlier age of depression onset, more recurrent depressive episodes, and a lifetime history of suicide attempts, findings that were also consistent with the results of a cohort study that documented a significant association between childhood adversity and depression severity in adults with TRD ( ). Among participants in the International Study to Predict Optimized Treatment in Depression (i-SPOT) ( ), a larger comparative effectiveness study, a history of trauma was not associated with the clinical outcome; however, in a subsample of this study, a history of abuse in early childhood was associated with lower likelihood of response. Interestingly, the neurobiological mechanisms implicated in treatment non-response may be influenced by trauma. In fact, neuroimaging research studies have suggested that early life stress may impact brain circuit development that correlates with depression later in life and can potentially correlate with an individual’s capacity to respond to antidepressants ( ).
Substance use
Major depressive disorder often co-occurs with alcohol (AUD) or substance use disorders (SUD), and the longitudinal course of these diseases is intertwined, as discussed in greater detail in Chapter 31 . As appears to be the case with most other risk factors for TRD, associations with SUDs appear to vary between studies. For example, in a large, population-based registry study of people with diagnosed depression (13% classified as treatment-resistant) from Sweden, having any diagnosed SUD within 180 days preceding the initiation of AD treatment was associated with nearly double the risk for TRD (adjusted OR 1.86), with the greatest risk of treatment-resistance observed in those with diagnosed sedative, opioid, and mixed substance use disorders ( ). Similar results were reported in another claims-based retrospective cohort study from the United States ( ). However, prospective studies have yielded different findings. In STAR*D, no significant differences were observed between patients with or without SUD in regard to time to achieve response, or rate of response to citalopram; however, those patients who endorsed both alcohol and drug use had significantly reduced rates of remission and significantly increased times to reach remission compared to the MDD group without SUD ( ). In the Combining Medications to Enhance Depression Outcomes study (COMED) there were no significant differences between patients with MDD with or without SUD in terms of dose, time in treatment, response or remission ( ). Overall, no study has specifically and prospectively investigated the comparative effectiveness of different AD treatments in patients with comorbid MDD and SUD.
Suicide risk
A group using the data pool from the European “Group for the Study of Resistant Depression” (GSRD), a multi-center study, identified suicidality as a key variable in treatment outcome ( ). Cross-sectional data from the GSDR reported “suicide risk” as a risk factor linked to TRD ( ), with suicide risk defined according to the MINI diagnostic interview as the presence of one of the following items in the past month: believing it would be better being dead or wishing one’s death, wishing to inflict harm on oneself, having thoughts of suicide, having a plan for suicide, making a suicide attempt, and ever having at least one-lifetime suicide attempt. also found suicidal risk to be significantly associated with TRD. Also for this possible predictor, it is unclear whether it represents a marker of severity of depression or an independent risk factor.
Medical comorbidities
The presence of medical comorbidities has been associated with worse overall prognosis during acute treatment of depression with ADs, and this result has been replicated independently in multiple studies ( ). In the STAR*D study, two-thirds of patients with MDD had at least one general medical condition ( ) and these patients showed poorer response to treatment. Physical disabilities and medical comorbidities are likely to impact MDD outcomes, particularly in elderly individuals ( ). In another study, higher burden of comorbid medical illness, as measured by the Cumulative Illness Rating Scale (CIRS) has been associated with poor treatment outcomes for major depression ( ). Another group found that patients with the melancholic subtype of MDD and with comorbid cardiovascular and endocrinological disorders experienced significantly worse response to antidepressant treatment compared to melancholic depression patients without these medical comorbidities ( ).
The presence of thyroid disease, even subclinical hypothyroidism (an abnormally elevated serum thyroid-stimulating hormone [TSH] but normal serum free thyroxine [fT4] level), has been repeatedly associated with poorer AD response and treatment-resistance in MDD ( ; ).
A recently published article from discussed the potential role of autoimmune diseases in particular, suggesting that diminished therapeutic effects of ADs may be related to immune system dysfunction or to neuroinflammation.
Neurological illnesses are also often associated with depression. For example, depressive symptoms or MDD are common in patients with Parkinson’s Disease (PD); however, very few studies investigated Parkinson’s Disease (PD) as a specific predictor of TRD. A study conducted by found that among PD patients more than one-third of those who were being treated with an AD continued to meet criteria for MDD; however, it is not clear whether the response rate is significantly different from depressed patients without PD. Also, elevated levels of physical pain at baseline predicted a longer time to remit from depression, which may indicate more difficult-to-treat depression and a need for more aggressive treatment ( ). Other conditions, such as sleep apnea, have been reported to be associated with nonresponse to AD treatment ( ).
AD response rates have been shown to be lower in patients affected by cardiovascular illness and, in turn, the lack of adequate response to AD treatment for post-myocardial infarction seems to be related to increased risk of subsequent cardiac events ( ). As an additional complicating factor, the presence of comorbid anxiety can, in turn, exacerbate cardiac outcomes ( ) and ultimately lead to inadequate response to AD treatment ( ; ). A number of studies investigated the link between TRD and metabolic syndrome; however, the association between the two is still not clear. While the prevalence of metabolic syndrome appears consistently higher than normal in MDD patients ( ), in a cross-sectional study did not detect an association between TRD and metabolic syndrome.
Similar to what has been observed in other medical conditions, although depression in patients with cancer is a common occurrence, in particular in some forms of cancer like pancreatic adenocarcinoma, no studies have investigated prospectively whether this may have prognostic implications, or to affect the decision-making process when choosing an antidepressant treatment.
Other factors
While family history of mood disorders is often associated with earlier age of onset, chronicity, suicidality and having multiple psychiatric comorbidities, it is unclear whether a positive family history of depression influences the effectiveness of treatment for MDD. In STAR*D, over half (55.6%) of the subjects reported a positive family history of depression; however, they did not differ in response or remission rates from those without family history ( ). Similarly, in a Chinese study, family history of affective disorders was not associated with the effectiveness of duloxetine in MDD ( ).
Epidemiological predictors
Socioeconomic status (SES)
Socioeconomic status (SES) has been shown to be associated with response to depression treatment in a study in older adults, or “late-life” depression ( ). Specifically, participants who lived in middle- or high-income neighborhoods were significantly more likely to respond to treatment than participants who lived in low-income neighborhoods ( ). found that treatment response rates were higher for residents of middle- and high-income, when aggregated, compared to those of low-income. Among patients treated with ADs, other social predictors of poor response to ADs or non-remission include lower level of functioning in both work and family life and being unmarried ( ). Other psychosocial and environmental factors associated with non-response were unemployment, lower educational status and adverse life events, while cognitive and global functioning were not found to be associated ( ).
Age
Age was consistently found not to be associated with outcome when evaluating non-response to different ADs, or rate of remission ( ; ; ; ; ; ; ). Only a few studies found a correlation between advanced age and poorer outcome ( ; ; ; ; ) and for this reason, some authors still consider age as a potential negative predictor ( ; ).
Gender
Epidemiological studies have repeatedly reported notable gender differences in AD response in MDD. It is well known that the prevalence of MDD in women is nearly twice that in men ( ; ), and depressive symptoms and illness trajectory appear to be different between men and women ( ). Studies dating back to the 1970s have reported gender differences in AD response, and this was supported by some more recent research ( ). Studies have shown women receive more benefit from some AD treatments including sertraline ( ; ), fluoxetine ( ), and citalopram ( ) while men respond preferentially to other drugs like imipramine ( ) and maprotiline ( ). Similar findings have been consistent with STAR*D results, which reported that females were more likely to respond to citalopram compared to males ( ).
These findings are complicated by the lack of reporting of subgroups, such as menopausal status, which is at times included as a covariate or moderator of the effects of gender on AD response ( ). Within a study assessing post-menopausal women, high levels of luteinizing hormone (LH) and follicular stimulating hormone (FSH) were reported to be associated with poorer SSRI response ( ). In other studies, AD response rates have been shown to improve with hormone replacement therapy (HRT), estrogen replacement therapy (ERT), or estrogen replacement ( ).
Nevertheless, the data on association between gender and AD response is still controversial. When nine AD studies were reevaluated by , there were no reported sex differences in treatment efficacy. No gender differences were found in later AD efficacy studies ( ), including a prospective and retrospective study investigating SSRI and TCA response ( ). A review article published in 2016 found that several additional studies occurring between 2002 and 2012 found no predictive role of gender for response or remission ( ).
Race and ethnicity
Race and ethnicity have been shown to play a role in AD response rates; however, it is very difficult to draw any conclusion as the majority of the clinical trials enroll participants who are not representative of different races and ethnicities. The STAR*D study demonstrated black participants experienced poorer AD responses than white participants ( ). In a secondary analysis of the STAR*D study ( ), differences in response or remission rates were not found to be significantly different. Overall data on potential associations between ethnicity and non-remission are inconclusive due to a low number of studies and lack of diversity within samples.
Discussion
The research for predictors in TRD has been complicated by the use of inconsistent definitions of TRD and retrospective analyses, while the prospective studies, or studies in which patients are randomized to a certain treatment based on the presence or absence of a sociodemographic variable are difficult to conduct. It is also likely that multiple predictors interact with each other, increasing or decreasing the likelihood of response to treatment. There are no predictors clearly associated with choosing a specific type of treatment, such as augmentation with atypical antipsychotic, or switching to TMS or ECT. Inconsistent findings have made the translation of findings from research studies difficult to translate to clinical care or programs, policies, and other public health interventions. Some of the predictors of poor prognosis have been independently replicated and may be clinically useful, such as the presence of comorbid anxiety disorders, medical comorbidities, poor adherence to treatment, and the presence of suicidality. Unfortunately, most of the other variables have been only inconsistently associated with TRD. Moreover, some of the “predictors” (i.e., chronicity) are detected in patients who already have failed several treatments, who may have more severe illness or are dealing with adverse circumstances, but it is not clear whether the illness was already severe since the onset of it was modified by several attempts at treating it. Finally, while many of the predictors of TRD reviewed in this chapter may characterize subtypes of depression that is more difficult to treat, they are usually not helpful, in-and-of themselves, for choosing the next-step of treatment. Further research is needed in the form of well-designed, large-scale, prospective comparative effectiveness studies in well-characterized and broadly-phenotyped samples of depressed patients.