With this study, we aimed to investigate and quantify the entrenched relationship occurring between affective temperaments, anxiety, and depression on core aspects of AN body-related psychopathology, namely drive for thinness and body dissatisfaction. To do so, we assumed the causal models with multiple mediators depicted in Fig. 1 and applied a recently published multiple mediation statistical method . Three main findings emerged. First, all affective temperaments but the hyperthymic one were overall involved in the relationship with core aspects of AN. Second, only the anxious and, to a lesser significance level, the depressive temperaments had a significant direct–that is, unmediated by the measured mediators and thus directly influencing outcomes—on DT while only a trend towards significance was found for the relationship between anxious traits and BD. Third, concerning the mediated effect, state anxiety was the strongest mediator of the link between temperament and core AN body-related psychopathology. Depression showed intermediate results, mediating significantly (and negatively) only the relationship between hyperthymic traits and DT, and trait anxiety was not a significant mediator at all. Notably, these are novel data since represent an innovative empirical attempt not only to disentangle but also to quantify the role of anxious and depressive symptoms in the relationship between temperament and clinical characteristics of real-world patients with AN.
Altogether, these findings are in line with earlier data showing the affective temperaments as relevant in the framework of AN [36, 37]; in fact, all temperaments but the hyperthymic one reported a significant total effect on core body-related aspects of AN psychopathology. It should be also borne in mind that hyperthymic traits are more typical of healthy controls and somehow under-represented in AN , potentially contributing to explain this finding. In line with earlier data from our group , the irritable and the anxious temperaments showed a similar profile but irritable traits should be further studied given their potential connection with the painful dysphoria commonly experienced by AN sufferers .
Nevertheless, when considering specifically the direct effect of the affective temperaments on AN core symptomatology, only the anxious and, to a lesser extent, the depressive traits showed a significant direct effect on DT, a crucial aspect of patients’ clinical presentation. Although it should be remembered that these findings refer specifically to the presented models, thus the direct effect could be influenced by temperament via other pathways than the ones (through the mediators) included in our causal model, this represents a novel finding in the field of AN since no mediation analyses on these matters were conducted so far. Interestingly, concerning BD, the anxious temperament showed a trend towards significance while the depressive one did not reach the significance threshold. Taken together, these findings on one hand confirm the role of anxiety and depression in AN but on the other hand, provide the innovative measurement of their direct effect on AN psychopathology as well. In fact, notwithstanding the consolidated role of anxiety and depression in AN [14, 17, 18], findings from our models allow us to disentangle and quantify the role of the anxious and depressive temperament traits independently of current anxious and depressive symptoms. This was particularly true for the effect of anxious and depressive temperaments on DT while BD was found to be not directly influenced by these traits, with the anxious temperament showing only a trend towards significance not surviving statistical correction. These different mediation pathways (different direct effects on DT and BD) is of interest, since–according to this model—DT seems relatively more influenced by a direct effect of anxious and, to a lesser extent, depressive temperament traits than BD, independently of current anxious and depressive symptoms. This is an intriguing finding since BD would seem to be more symptom-dependent than DT. It is noteworthy that, although a genetic vulnerability has been proposed for both DT and BD [58,59,60], the latter showed to be markedly influenced by socio-cultural processes as well  thus potentially explaining the lack of a direct role of temperament in this regard. In fact, BD is so prevalent in Western Countries to be considered as a “normative discontent” ; relatedly, cultural influence resulted to strongly predict BD even after controlling for BMI  and influential models have been proposed to link socio-culturally mediated factors to body dissatisfaction [63, 64]. All in all, BD is confirmed as a multi-faceted construct potentially influenced by a myriad of factors that future research may want to analyze. Therefore, from a clinical standpoint, DT and BD may require a different approach in treatment given their partially different “response” to patients’ innate temperament.
Moreover, the other side of the previous finding on the direct effect is the measurement of the causal mediation effect of the mediators considered, namely trait and state anxiety, depressive symptoms, and their joint effect. When considering every single mediator, state anxiety reported the strongest effect (with estimates twice as high as those of depression), depressive symptoms showed intermediate results while trait anxiety did not show any mediation effect. However, their joint mediation effect was greater than that of any mediator as a stand-alone entity. Altogether, our findings show that these three major factors (i.e., temperament, anxiety/depression, and body-related symptoms of AN) are interwoven and should be considered as key elements in the development and maintenance of AN. On one hand, this finding provides useful parallelism to studies of network analysis showing that depressive and anxious symptoms are central in AN psychopathology [17, 18]. Nevertheless, this is the first time that the effect of these highly inter-correlated constructs [24, 65] is estimated. Also, the proposed models provided a scientific measure of an every-day clinical notion, since anxiety, both singularly and jointly measured, was found to significantly mediate psychopathological variables. Furthermore, anxiety has been linked to clinical severity as measured by BMI . Still, in the framework of the intriguing scientific debate on anxiety in AN [9, 14, 33], our findings support earlier literature suggesting severe malnutrition as exacerbating the eating core psychopathology aspects of patients with AN  in a fine-grained fashion. If the role of state anxiety was supported, trait anxiety, namely patients’ usual and stable levels of anxiety, did not result to be a significant mediator. This is an unexpected finding, given the relevance of anxiety in AN [15, 21, 67]. On one hand, it could be proposed that trait anxiety shows relevant “shared” and somehow overlapping characteristics with the anxious temperament (at least with respect to how the questionnaires address this construct) thus decreasing the chance to be “detected” by the applied statistical method. However, on the other hand, it could be argued instead that, the marked inpatients’ levels of state anxiety take the lion's share in the model, in keeping with the aforementioned hypothesis on the state-dependent aggravation of the anxiety diathesis . The mediation effect of BDI showed intermediate results instead. In fact, after intermediate Bonferroni's correction, depressive symptoms resulted to negatively mediate only the relationship between the hyperthymic temperament and DT. Although the literature on Cognitive Behavioral Therapy had already shown that depression significantly influences weight gain [25, 68], this is a novel finding that tends to dampen earlier findings [17, 18] since, according to our model, it can be surmised that current anxious symptoms impact more on DT and BD than depression in inpatients with severe AN.
Should these data be confirmed, some interesting clinical implications could be proposed. In fact, our data suggest that—when treating AN sufferers in a severe phase of their disorder–close attention should be paid to patients’ constellation of comorbid psychiatric symptoms. Also, it could be raised that hypothesis that, from a clinical standpoint, working on DT and BD could require different treatment approaches, given the different influence of patients’ innate characteristics on DT and BD. In this vein, temperament-based treatments have been authoritatively proposed and are currently being validated . Additionally, therapeutic interventions may want to tackle the more punctiform anxiety and depressive components in treatment as well. For example, although no medications can reverse the core symptoms of AN, pharmacotherapy could be helpful in this regard [70, 71]. Additionally, from a psychotherapy standpoint, the management of anxiety and depression can improve patients’ motivation and quality of life .
In closing, prompted by the state-of-the-art supporting the entrenched nature of AN psychopathology and anxious and depressive symptoms, we designed models trying to disentangle this relationship and, importantly, measure the effect of each component involved. We found that temperament per se has a relevant impact on body-related core components of AN, namely DT and BD. However, a clear direct effect could be identified only for the anxious and, to a lesser extent, the depressive temperaments on DT. Also, state anxiety was the strongest mediator as a standalone component when compared to depression and trait-anxiety. In keeping with earlier literature , the joint effect of all mediators resulted to be even stronger than state anxiety alone. The proposed models of mediation of anxiety and depression between affective temperaments and body-related psychopathology of AN allowed to quantify direct and indirect effects that were found to be consistent with every-day clinical practice. In fact, clinicians and researchers who work with AN sufferers know well how innate characteristics (i.e., temperament) and current clinical symptoms (i.e., anxiety and depression) can impact on core symptoms of AN. Therefore, if confirmed by further studies, our data contribute to the ongoing scientific debate aiming to shed light on this complex picture also providing some intriguing clinical implications.
Strength and limits
Overall, this study has some relevant strengths including the innovative statistical method allowing to measure multiple mediators and their joint effect, the recruitment of “real world” patients with severe AN (no insurance/financial barriers exist, according to the Italian National Health System), the pure and full-blown diagnosis of AN, and the aforementioned clinical implications. Nevertheless, some limitations should be acknowledged as well: we recruited only inpatients so this could hamper data generalizability, the sample could be larger, and a cross-sectional design self-report assessments have been used. Even though we applied an advanced method for mediation inference, future studies on larger samples are needed to expand knowledge on its application. It is important to point out that for this study we did not carry out power or sample size calculations as no methods currently exist adapted to our causal multiple mediation setting. Moreover, we reiterate that our quantitative results are valid as long as our assumption of the causal models relating all the variables at stake that is depicted by the DAG in Fig. 1 holds true, with recent advances in the causal inference literature potentially expanding our knowledge in this regard. The statistical method that we applied relies on strong assumptions among which the lack of measured or unmeasured confounders of the relationships between candidate mediators and response variables . Accordingly, we assumed that temperaments do not have an impact on BMI. Even though there is no literature supporting the finding that temperament itself is able to modify, from a causal standpoint, individuals’ BMI, it is important to put forward this possibly controversial assumption. Another hypothesis we made is that candidate mediators do not causally affect each other. Even though this strong assumption can be seen as a limitation, we highlight that we referred to symptoms (of anxiety and depression) rather than to a formal diagnosis of anxiety or depression and, at such a symptom level, it is reasonable to believe that state and trait anxiety and depression can be much causally unrelated. Finally, we assumed the five temperaments as independent. With that being said, if these findings will be confirmed by further studies, these data could be much informative from a clinical perspective.
What is already known on this subject?
Research consistently supported the role of temperament as a vulnerability and maintaining factor for anorexia nervosa (AN). Also anxiety and depression are key-elements in AN even from an outcome perspective. However, currently little is known about the role of temperament as compared to that of anxiety and depressive symptoms on core aspects of AN psychopathology (i.e., drive for thinness [DT] and body dissatisfaction [BD]).
What this study adds?
An innovative statistical method allowing to measure multiple mediators and their joint effect was applied to investigate and quantify the relationship between affective temperaments, anxiety, and depression on DT and BD in AN. This study clarified that affective temperaments impacted on body-related core components of AN; a clear direct effect could be identified only for the anxious and depressive temperaments. Also, state anxiety resulted to have the strongest mediator effect.