Acute Threat/“fear” construct (AT)

4 studies were identified that investigated the effectiveness of 5 antidepressants with FQ:Agoraphobia or FQ:Needle Injury, all of which suffer from significant risk of bias.

The following antidepressants were found to statistically significantly reduce AT:

• Paroxetine at 20-50mg for 8 weeks, with a 46.1% reduction in FQ:Agoraphobia score (from 15.6 +/- 11.9 to 8.4 +/- 10.4), and a 29.1% reduction in FQ:Blood Injury score (from 17.5 +/- 9.5 to 12.4 +/- 10.1, p values not stated), in a sample with DSM panic disorder with or without agoraphobia [19]. Unfortunately, the nature of the values after the mean score was not stated, so it was not possible to derive the standardised mean difference (SMD) between paroxetine and citalopram. Nonetheless, a significant reduction in FQ:Agoraphobia was replicated in a more DSM heterogeneous sample (comprising social phobia, unipolar depressive disorders, panic disorders, specific phobias, GAD, OCD and alcohol dependence) following administration of 10-60mg for 12 weeks (from 11.8+/-10.8 to 5.8 +/- 7.1, p = 0.017) [26]
• Citalopram at 20-50mg for 8 weeks, with a 36.9% reduction in FQ:Agoraphobia score (from 15.7 +/- 12.3 to 9.9 +/- 9.1, p value not stated), and a 46.2% reduction FQ:Blood Injury (from 15.6 +/- 9.9 to 8.4 +/- 10.4, p value not stated), in a sample with DSM panic disorder +/- agoraphobia [19]. A SMD could not be derived for the aforementioned reason.
• Fluvoxamine at 150mg for 12 weeks vs placebo (p<0.05), again in a sample with DSM panic disorder +/- agoraphobia [23]. Unfortunately, there was inadequate data from the study to derive a SMD.

The following antidepressants did not statistically significantly reduce AT:

• Escitalopram at 10-20mg for 12 weeks in a sample with DSM specific phobia [21], from 3.0 (SD = 2.4) to 1.2 (SD = 1.1) (p value not stated).
• Nefazodone at 200-600mg for 12 weeks in a sample with heterogeneous DSM diagnoses comprising generalised social phobia, unipolar depressive disorders, panic disorder, specific phobia, GAD, OCD and alcohol dependence, from 9.0 (SD = 7.8) to 6.4 (SD = 6.5) (p = 0.09) [25]

Potential Threat “anxiety” (PT)

3 studies were identified that investigated the effectiveness of 3 antidepressants with POMS:Tension, all of which suffered from significant risk of biases.

The following antidepressants statistically significantly reduced PT:

• Desvenlafaxine at 100mg or 150mg for 12 weeks in a sample with vasomotor symptoms and no DSM diagnoses vs placebo, with a 4.1 reduction in POMS:Tension score from a baseline of 9.3 (SD = 6.7) for 100mg dose, and a 3.9 reduction from a baseline of 9.0 (SD = 7.2) for 150mg dose (p<0.05) [20]. From the values provided for these dosages and for placebo, a SMD vs placebo of 0.26 for 100mg and 0.22 for 150mg were derived.
• Fluoxetine at 20mg for 6 months in a sample with PMS and no DSM-IIIR diagnoses vs placebo during the luteal phase only (p<0.005), when PT was found to be higher than during the follicular phase (T scores of 56.6 +/- 3.3 vs 40.5 +/- 3.6) [22]. From these values and those for placebo, a SMD of 1.1 vs placebo could be derived.
• Venlafaxine at 75-150mg for 8 weeks appeared to significantly reduce PT as well in a sample of DSM MDD where comorbid anxiety, phobic or personality disorders not excluded, but unfortunately no significance testing was performed to guide interpretation of the reduction in POMS:Tension score from 23.68 +/- 4.88 to 9.6 +/- 3.93 [31].
• Sertraline at 50mg or 150mg for 12 weeks appeared to significant reduce PT in a sample with DSM MDD that is potentially heterogeneous in nature due to exclusion criteria being acute or chronic organic mental disorder and “clinically significant psychiatric illness (including) active suicidality” by 5.9 (confidence interval not given) [30]. However, significance testing was limited to comparison with nortriptyline (p = 0.01).

Loss

7 studies were identified that investigated the effectiveness of 5 antidepressants wi8th POMS:Depression.

The following antidepressants were found to statistically significantly reduce Loss:

• Sertraline at 50/100/200mg for 6 weeks vs placebo (p<0.01) reduced POMS:Depression score by 0.1 (SD = 0.7) from a baseline of 2.7 (SD = 1.1) in a potentially heterogeneous sample of DSM MDD where comorbid diagnoses including anxiety, phobic, obsessive and personality disorders weren’t excluded [28]. From these values and those for the placebo group, a standardised mean difference vs placebo of 0.39 was derived. A reduction was replicated at 50-200mg for 8 weeks in a similar sample of those with DSM MDD where comorbid mild-moderate GAD or personality disorders weren’t excluded, but the reduction was much greater, by 1.0 (SE = 0.08) from a baseline of 1.9 (SE = 0.08) [29]. This was equivalent to a SMD of 0.56 vs placebo.

However, when the sample comprise those with MDD where comorbid mild neurocognitive disorder with MMSE score > = 23 or personality disorders were not excluded, the reduction from 1.5 (SD = 0.67) to 1.2 (SD = 1.13) did not reach statistical significance (p value not stated) [27].

• Desvenlafaxine at 100mg or 150mg for 12 weeks vs placebo for those with VMS but no DSM comorbidities, by 5.5 from a baseline of 8.3 (SD = 8.6) for 100mg and by 4.5 from a baseline of 8.2 (SD = 11.1) for 150mg [20]. From these values and those for the placebo group, a SMD of 0.22 for 100mg and 0.35 for 150mg vs placebo could be derived.
• Fluvoxamine at 50-200mg for 4 weeks in a heterogeneous Research Diagnostic Criteria MDD sample where the only exclusion was substance dependence, but statistical significance only reached at week 3 (p = 0.028, vs p = 0.6 at week 4) [24]. Unfortunately, there was inadequate data available from the study to derive a SMD.
• Fluoxetine at 20mg for 6 months vs placebo (p<0.005) in a sample with PMS but no DSM-IIIR comorbidities during the luteal phase only, when POMS:Depression was greater than during the follicular phase (T scores of 59.6+/-4.6 vs 44.8 +/- 5.2) [22]. From these values and those for placebo, a SMD of 0.93 vs placebo could be derived.
• Venlafaxine at 75-150mg for 8 weeks also reduced POMS:Depression in a sample with DSM MDD where comorbid diagnoses including anxiety, phobic and personality disorders were not excluded, seemingly significantly by 70.2% from 37.16 +/- 11.36 to 11.12 +/-9.11; however, significance testing was not performed [31].

Key conclusions

This is the first systematic review to investigate the effectiveness of current generation antidepressants using the RDoC paradigm. The review revealed that a remarkably limited proportion of studies included self-report psychometric measures proposed by Watson, Stanton and Clark [10] to correlate with the NVS constructs of AT, PT and Loss. The 13 studies identified used a total of three out of the seven instruments, despite the fact that most of them were developed a number of decades ago, likely reflecting the inherent primacy of DSM-validated measures in psychiatric research. The articles investigated 8 current generation antidepressants, and found that the majority were effective in reducing the severity in the construct(s) concerned (i.e. paroxetine, citalopram and fluvoxamine for AT, fluoxetine, desvenlafaxine and sertraline for PT, and sertraline, fluvoxamine, fluoxetine & desvenlafaxine for Loss), although the strength of this conclusion was limited by the high risk of bias within included studies.

Antidepressants lacking evidence of effectiveness along particular RDoC constructs

This review found a lack of evidence of effectiveness for 2 antidepressants: escitalopram and nefazodone for AT. The finding for nefazodone is particularly interesting because the study concluded that nefazodone was effective for generalised social phobia, based on a range of other psychometric measures such as the State-Trait Anxiety Inventory (STAI) that reported statistically significant improvement 25). However, discriminant validity is a pervasive issue with psychometric instruments for anxiety and depression [10] with psychometric instruments–such as the STAI–often measuring general distress/negative affectivity as opposed to more specific constructs of anxiety, fear or depression. Van Ameringen, Mancini [32] studied nefazodone again using a more rigorous randomised placebo-controlled design with a much larger sample size and a completely different set of outcome measures, and concluded that nefazodone was in fact ineffective for generalised social phobia. In light of trazodone’s comparatively poor receptor affinity for serotonin transporter relative to those antidepressants found to be effective for AT [33], and the established role of the serotonin system in threat regulation [34, 35], this finding is not surprising.

Unlike trazodone, established evidence contrary to this review’s finding of a lack of effectiveness for escitalopram is based on psychometrics that measure agoraphobia like FQ:Agoraphobia (e.g. Panic and Agoraphobia Scale) [36–38]. Thus, the findings of Alamy, Wei [21] likely reflects risk-of-bias issues within the study, rather than the principle of assessing psychopathology in terms of behavioural-psychological constructs such as AT versus more heterogeneous DSM constructs such as “panic disorder”.

The lack of studies investigating other antidepressants on other constructs unfortunately preclude further consideration of the constructs’ biological underpinnings.

Extent of effectiveness

A few studies provided adequate data to allow SMDs vs placebo to be derived. These demonstrated that for PT, fluoxetine was very effective whilst desvenlafaxine was weakly effective (SMD = 1.1 vs 0.22–0.26 in POMS:Tension, respectively). For Loss, fluoxetine was similarly very effective whilst sertraline was moderately effective and desvenlafaxine again weakly effective (SMD = 0.93 vs 0.39–0.55 vs 0.22–0.26 in POMS:Depression, respectively).

Whilst the finding of effectiveness of antidepressants for both depression and anxiety is hardly surprising, the lack of overlap in confidence intervals amongst the antidepressants concerned is. Whilst caution is needed to interpret this given the low reliability of this finding (discussed below), it is worth noting that recent network meta-analyses of GAD [39] and MDD [40] found overlapping confidence intervals amongst all antidepressants such that the present review’s findings are not inconsistent with those reported.

It is tempting to consider desvenlafaxine’s comparatively poorer effectiveness to fluoxetine and sertraline in these constructs as a reflection of the effectiveness of targeting noradrenaline receptors in the treatment of PT and Loss. However, pharmacological studies serve as a reminder that the difference between SSRIs and SNRIs lie in their affinity for the noradrenaline transporter relative to the serotonin transporter, not in their absolute affinities for either receptors [33, 41–43]; for instance, sertraline is a more potent inhibitor of noradrenaline transporter than both venlafaxine and desvenlafaxine, and only has slightly lower affinity than duloxetine, not to mention its higher absolute affinity for the dopamine transporter.

Further, it is important to note that with regards to GAD, the DSM diagnosis most similar to the RDoC construct of PT, no studies to date have investigated the effectiveness of desvenlafaxine for GAD. Thus, the fact that a testable hypothesis for the effectiveness of an antidepressant for a particular disorder was able to be generated from this review, based on the results of a study investigating a sample with purely vasomotor symptoms and no DSM diagnoses, suggest the potential utility from both research and clinical perspectives of transdiagnostic constructs.

Risk of bias in included studies

Unfortunately, the quality of these studies were uniformly poor, and all were prone to significant risk of biases affecting the reliability of their findings and conclusions. However, this must be considered in the context of the wider clinical trial literature. In the aforementioned network meta-analyses, one identified only 18% of the 522 included trials as having a low risk of bias with the vast majority rated as moderate (73%) [40], whilst the other rated 84% of the 89 trials they included as a high risk of bias in at least one of the domains [39], meeting the criteria established by the authors of the tool for an overall high risk of bias. Similarly, in a systematic review of the efficacy and safety of adjunctive antidepressants in schizophrenia [44], of the 82 trials included, 60% met criteria for an overall high risk of bias. Again, in a systematic review of prophylactic antidepressant treatment following acute coronary syndrome [45], of the 6 studies included, all were rated as having a high risk of bias.

Complicating matters is the fact that assessments of risk of bias is subjective, and despite the development of tools and associated guidance documents to facilitate in the analysis, it has been found that 45% of subfertility trials included in more than 1 Cochrane review received differing risk of bias judgements from different groups of authors, with greater agreement in random sequence generation (71%) and incomplete outcome data (79%) and less agreement in blinding (35%) [46]. A strength of this study is the use of the RoB2 and ROBINS-I risk of bias assessment tools for analysing randomised and non-randomised trials, respectively, due to their comprehensive scope and detailed guidance for grading levels of risk.

It can thus be said that risk of bias issues are not limited to the articles identified within this review, but pervade the clinical literature at large. Hence, a useful rule of thumb for interpreting findings of any study or review is to be ever-vigilant and thoughtfully consider them in the context of its methodology and one’s clinical question of interest. Authors of systematic reviews can facilitate readers by being more overt about their risk of bias judgements, for example by moving the risk of bias table from the appendix or supplementary material to the body of the paper, and by providing the underlying rationale in addition to the summary judgement within each domain.

Nevertheless, it was noteworthy that the few times the same antidepressant was investigated within the same construct, the findings were replicated by different investigators (i.e. paroxetine and AT, sertraline and Loss). This is even more remarkable considering the heterogeneous nature of the samples’ DSM diagnoses. Whilst a caveat must be given since the association between sertraline and loss in an older population with likely mild cognitive impairment did not reach statistical significance, the results nonetheless suggest further research is warranted to corroborate the findings of this review with the view to furthering the evidence base with respect to transdiagnostic constructs.

Construct validity of “potential threat” and “loss” in the negative valence domain

This review identified 3 studies that investigated outcomes for specific antidepressants along multiple constructs; all were along both PT and Loss constructs. 2 were in samples with PMS/VMS, and both reported similar scores with significant overlap in 95% confidence intervals in both measures at baseline and endpoint [22], during follicular and luteal phases [20], and they reported similar levels of improvements in both measures with antidepressant treatment. The third study [31] reported baseline POMS:Depression that appeared to be higher than POM Tension, but a small overlap in confidence interval still existed; endpoint POMS measures after antidepressant were similar with considerable overlap in confidence intervals. There is thus a need to further study the ability of POMS-Depression and POMS-Tension to differentiate amongst different phenomenological manifestations of the pathophysiological processes underlying PMS/VMS and MDD, or of antidepressant treatment.

Whilst the 65 adjectives that constitute POMS were found to cluster into 6 distinct factors/subscales, concerns had been raised that the original list of adjectives from which the factors were derived were created with these 6 factors in mind [47]. Moreover, factor analyses of categorical psychiatric disorders have failed to demonstrate that MDD and GAD (the DSM disorder that best represent the Response to PT construct) belong to different factors, either within psychiatric samples [48–50], community samples [51–53], child and adolescent samples [54–56] or cross-cultural samples [57]. This phenomenon persisted when factor analysis occurred on a symptom level, with “internalising disorders”–a factor which influenced RDoC’s NVS [58]–consistent with other factor analyses, found to be better described by a “fear” factor and a “general distress” factor [59].

Further, the neurobiological substrates subserving responses to PT and loss appear to share considerable overlap. Indeed, it is still not possible to differentiate between GAD from MDD based on a review of a multimodal neuroimaging evidence to date although the same study found a correlation between limbic/paralimbic activity and the combined sample including healthy controls based on levels of general distress [60]. Of course, similarity between the DSM diagnoses of MDD and GAD and the RDoC constructs of response to Loss and PT do not constitute equivalence, but evidence using POMS:Tension and Depression also suggest a convergence in correlation between PT and Loss on the one hand, and impaired activity of the anterior cingulate cortex [61, 62] and amygdala [63], as well as reduced GABA-A receptor binding in the posterior cingulate cortex and left superior frontal regions [64] on the other. Increasing evidence that physiological paradigms such as startle response are a reliable differentiator of fear or distress [65] suggests that a more biologically-grounded conceptualisation of NVS constructs is simply AT vs non-AT. This is further supported by evidence implicating only serotonin (and not dopamine or noradrenaline) in the startle response [66], as well as meta-analytic studies reporting reboxetine to be effective in MDD [40], but not panic disorder [67].

The idea of a NVS “distress” construct is actually not new. The initial NIMH workgroup responsible for creating RDoC initially proposed that the NVS comprise of “distress”, “fear”, and “anger” constructs [1]. However, an expert workshop convened to discuss these proposals instead decided to split the “distress” construct into response to “PT”, “ST” and “loss” because the former was found to be too “vague and diffuse”, and deemed the revisions better to “accommodate a wide range of experiences and situations that logically fall under the Construct” [1]. Interestingly, the workshop conceded “uncertainty about whether ST should be considered as a separate Construct, or as variations… that impact the circuits involved in AT and potential harm”, but “decided to include [it]” with the caveat that “further clarification is needed” [1]. There did not seem to be much debate about the “loss” construct, as it was “the most frequently nominated Construct of the pre-workshop survey”; depression was considered the “sustained” response to loss [1].

However, it must be said that the workgroup acknowledged the list of Constructs created was “not intended to be definitive or all-inclusive” [1]. Similarly, when pondering the future in light of the body of research that led to the development and publication of the DSM-III, Spitzer wrote that it was “only one still frame in the ongoing process of attempting to better understand mental disorders” [68]. A similar sentiment was cited by Kozak and Cuthbert [58] to warn against the reification of RDoC constructs that had befallen DSM-III diagnostic categories.

Psychometric score interpretation

When validated against a representative community sample, normal POMS:Depression score was 7.5 (SD = 9.2) for men and 8.5 (SD = 9.4) for women, and normal POMS:Tension score was 7.1 (SD = 5.8) for men and 8.2 (SD = 6.0) for women [69]. For FQ, the normal score in a representative community sample was 7.9 (SD = 7.1) for men, 14.8 (SD = 8.5) for women, and 11.8 (SD = 8.6) overall [70]. Therefore, with the exception of the only study performed in an inpatient population [31], all the other studies identified by this systematic review reported mean baseline values of their subjects that were within the “normal” range.

Given that the inclusion criteria for these study were based on clinical judgements according to the DSM or related psychometric instruments, this finding raises questions about the relationship between observer and self reports as well as the relationship between DSM diagnoses and RDoC.

With regards to the former, whilst a number of questions about the validity of HAMD as a measure of depression severity have been raised [71], it should be pointed out that POMS asks subjects to rate how strongly a word/phrase describes them over a period of time [12]. However, none of the studies identified reported the timeframe, making it more difficult to interpret the meaning of these results although the low values suggest a timeframe much shorter than that used by the instruments used by clinicians. Nonetheless, underreporting of mood had been described in the literature due to response bias motivated by perceived social desirability [72, 73]. Thus, for RDoC research to be clinically useful, considering expanding the “self-report” unit of analysis to include observer-rated elements of such self-report is warranted.

With regards to the latter, it could be observed from the 3 studies included in this systematic review that reported baseline FQ:Agoraphobia scores that their samples consisted primarily of generalised social phobia [25, 26] and panic disorder [19]. Given the FQ:Agoraphobia subscale assesses the severity of the respondent’s behavioural avoidance and not affective distress, these samples’ normal scores can be seen as a reflection of DSM’s heterogeneous criteria for severity, namely “a) persistent concern… or b) worry about the implications (of the [panic] attack)” versus “significant change in behaviour” for panic disorder, and “the feared social or performance situations are avoided or else are endured…” for social phobia. In light of evidence that generalised social phobia correlate with “distress” disorders whereas specific social phobia correlate with “fear” disorders, it is thus more understandable why samples of the former [25, 26] scored within the normal range but much lower that of the latter [19], whose mean baseline scores were higher than the normative means. Further, it suggests the need to further study the potential for RDoC’s dimensionally-based constructs to differentiate amongst not just normal and pathological, but amongst different types of pathology as well.

Another observation that can be made from the 3 studies that measured FQ:Agoraphobia is the effectiveness of antidepressants along this dimension in spite of the “normality” of the scores. This finding is suggested by results of a recent meta-analysis concluding that antidepressants are effective across all severity levels of anxiety disorders [74], and reinforces earlier findings that antidepressants can reduce attentional vigilance to threat in healthy volunteers [75]. These findings are consistent with the dimensional conceptualisation of psychobehavioural constructs spanning normality to pathology. Further study could shed light on whether antidepressants have the potential to be used as emotional nootropic agents.