Alexithymia and facial emotion recognition in patients with craniofacial pain and association of alexithymia with anxiety and depression: a systematic review with meta-analysis

Study selection

We included studies in the systematic review and meta-analysis if they met the following criteria: (1) an observational study with a cross-sectional, cohort or case-control design; (2) adult patients with CFP (headache or TMD) according to validated diagnostic criteria such as research diagnostic criteria for temporomandibular disorders or International Classification of Headache Disorders; (3) the degree of alexithymia and the facial discrimination of emotions was addressed and quantified; and (4) there was a control group for comparison.

Search strategy

We conducted a search of observational and comparative studies that included cohort, case-control and cross-sectional studies using MEDLINE, Scielo and Google Scholar. The last search was performed on 8 September 2021.

The search strategy of each database can be found in Supplemental information. The search was conducted by two independent reviewers (RL and FCM) using the same methodology, and disagreements were resolved by consensus with a third reviewer (LSM). Reference lists from original studies were screened manually. Bibliographic references of identified publications and published bibliographic reviews were searched by hand for potentially relevant articles.

Selection criteria and data extraction

Two independent reviewers (RL and FCM) performed the first phase: assessing the relevance of the studies. This first analysis was performed based on information from each study’s title, abstract and keywords. If the abstracts did not contain sufficient information, the full text was reviewed. During the second phase, we reviewed the full text and checked whether the studies met all of the inclusion criteria. A third reviewer (LSM) acted as mediator when there were differences between the 2 reviewers, with the 3 reviewers reaching a consensus (Furlan et al., 2009). The data described in the results were extracted by means of a structured protocol that ensured that the most relevant information from each study was obtained. To be able to perform the secondary analysis, data from the facial emotion recognition, anxiety and depression level variables were extracted from the studies that included them.

Methodological quality assessment

We assessed the methodological quality of the selected cross-sectional studies, using a modified version of the Newcastle-Ottawa Quality Assessment Scale (NOS) adapted for cross-sectional studies (Deeks et al., 2003; Wells et al., 2009), which has moderate interrater reliability (k = 0.75, 95% CI [0.65–0.81]) (Zhang et al., 2021). The NOS scale ranges from 0 to 10 stars out of a total of seven questions within three sections: selection, comparability, and outcome. The tallied stars provide five categories of study quality: (1) poor, 0 to 3 stars; (2) fair, 4 to 5 stars; (3) good, 6 to 7 stars; (4) very good, 8 to 9 stars; and (5) excellent, 10 stars (Wells et al., 2009).

Two independent reviewers examined the quality of the selected studies using the same methods; disagreements between the reviewers were resolved by consensus, which included mediation by a third reviewer. The inter-rater reliability was determined using the Kappa coefficient: (1) κ > 0.7 indicated a high level of agreement between the assessors; (2) κ = 0.5–0.7 indicated a moderate level of agreement; and (3) κ < 0.5 indicated a low level of agreement (Cohen, 1960).

Qualitative analysis

For the qualitative analysis of the selected cross-sectional studies, we employed an adaptation of the classification criteria provided by van Tulder et al. (2003) for randomized controlled trials. The results were categorized into 5 levels depending on the methodological quality: (1) strong evidence: consistent among multiple high-quality cross-sectional studies (at least 3); (2) moderate evidence: consistent findings from multiple good or very good-quality cross-sectional studies; (3) limited evidence: at least 2 fair quality cross-sectional studies; (4) conflicting evidence: inconsistent findings among multiple studies (cross-sectional studies); and (5) no evidence: no cross-sectional studies reported.

Data synthesis and analysis

The statistical analysis was performed using meta-analyses with interactive explanations (MIX, version 1.7), with the data comparing patients with CFP and alexithymia with asymptomatic participants (Bax et al., 2006). We employed the same inclusion criteria for the systematic review and the meta-analysis, but added two criteria: (1) the results section contained detailed information on the comparative statistical data (mean, standard deviation and/or 95% confidence interval) of the Toronto Alexithymia Scale (TAS), either TAS-26 (Taylor et al., 1988) or TAS-20 (Bagby, Parker & Taylor, 1994); and (2) data for the analyzed variables were represented in at least two studies. Meta-analyses of correlations of alexithymia with anxiety or depression were also performed, including those studies that presented statistical data for alexithymia with anxiety or depression. We present the summary statistics in the form of forest plots (Lewis & Clarke, 2001), which consist of a weighted compilation of all standardized mean differences (SMDs) and corresponding 95% confidence intervals (CIs) reported by each study, and provide an indication of heterogeneity among the studies. The statistical significance of the pooled SMDs was examined using Hedges’ g, to account for possible overestimation of the true population effect size in small studies (Hedges, 1982). The magnitude of g was interpreted according to a 4-point scale: (1) <0.20, negligible effect; (2) 0.20‒0.49, small effect; (3) 0.50‒0.79, moderate effect; and (4) ≥0.80, large effect (Cohen, 1992). We estimated the degree of heterogeneity among the studies by employing Cochran’s Q statistic test (p <0.1 was considered significant) and the inconsistency index (I²) (Higgins et al., 2003): I² > 25% is considered to represent low heterogeneity, I² > 50% is considered medium, and I² > 75% is considered to represent large heterogeneity (Huedo-Medina et al., 2006). The I² index is complementary to the Q test, although it has a similar power problem as the Q test with a small number of studies (Huedo-Medina et al., 2006). Therefore, a study was considered heterogeneous when it fulfilled one or both of the following conditions: (1) the Q-test was significant (p < 0.1); and (2) the result of I² was >75%. We performed a random-effects model in the meta-analysis of the heterogeneous studies to obtain a pooled estimate of effect (DerSimonian & Laird, 1986). To detect publication biases and to test the influence of each study, we performed a visual evaluation of the funnel plot and the exclusion sensitivity plot, searching for any asymmetry. We also employed Egger’s regression test to determine the presence of bias (Begg & Mazumdar, 1994; Sterne & Egger, 2001).

In addition, a meta-regression was performed using the OpenMEE program (Wallace et al., 2017) to assess whether Hedges’ g of alexithymia is linearly related to average normalized anxiety and depression. Previously, it was necessary to perform another meta-analysis including only the studies that assessed both the alexithymia variable and anxiety or depression.

Characteristics of the study population

A total of 1,228 participants with CFP were included in this study, of which the experimental group had TMD pain in two studies (Glaros & Lumley, 2005; Haas et al., 2013), six studies reported migraine (Muftuoglu et al., 2004; Vieira et al., 2013; Cerutti et al., 2016; Galli et al., 2017; Yalınay Dikmen et al., 2017; Shim, Park & Park, 2018) and two reported tension-type headaches (Yücel et al., 2002; Shim, Park & Park, 2018). Within the studies on migraines, all subdivided the sample according to type (chronic-episodic) except four studies, which generalized the sample with the name migraine (Muftuoglu et al., 2004; Vieira et al., 2013; Cerutti et al., 2016; Shim, Park & Park, 2018).

Within the total sample, 856 participants were women, representing 100% of the sample in two studies (Vieira et al., 2013; Cerutti et al., 2016); in the rest, the sample varied between 82.4% and 62% (Yücel et al., 2002; Muftuoglu et al., 2004; Glaros & Lumley, 2005; Galli et al., 2017; Yalınay Dikmen et al., 2017; Shim, Park & Park, 2018), and in 1 study, the female sample was 90% (Haas et al., 2013). The average age of the sample was 38.71 years, and the average age ranged from 29.06 to 46.89 years.

Results of the methodological quality

The agreement between the two evaluators according to the Kappa coefficient was high (κ = 0.72). The intervention of a third evaluator was necessary to achieve consensus on the quality of three studies.

With respect to the studies included, the average score was 7.40 ± 0.52. All studies showed very good methodological quality (7 or 8 stars) (Table 2).

Evidence of alexithymia in craniofacial pain

Nine studies analyzed the variable of alexithymia (Yücel et al., 2002; Muftuoglu et al., 2004; Glaros & Lumley, 2005; Vieira et al., 2013; Haas et al., 2013; Cerutti et al., 2016; Galli et al., 2017; Yalınay Dikmen et al., 2017; Shim, Park & Park, 2018) in patients with CFP (migraine, headache and temporomandibular disorder) vs healthy individuals using the TAS self-registration tool. In all the studies, significant differences between the symptomatic groups and the control group were reported, declaring higher scores in the total TAS.

Two studies assessed patients with tension headache, the study by Yücel et al. (2002) observing that there were no significant differences between headache groups (episodic and chronic). Another study had found that the TAS scores were higher for the tension headache groups than for the migraine group, which also had higher pain intensity scores, correlating the impact of the headache with the total TAS (Shim, Park & Park, 2018). In 5 studies, the comparison group were patients with migraine (Muftuoglu et al., 2004; Vieira et al., 2013; Galli et al., 2017; Yalınay Dikmen et al., 2017; Shim, Park & Park, 2018), and the only study whose results differed from the rest was that by Yalınay Dikmen et al. (2017), who reported no significant differences between the 2 groups, even with the control group having higher mean scores than the sum of the total migraine group scores. For the migraine studies that assessed anxiety, this variable correlated with high levels of alexithymia, whereas depression was not correlated with this as a predictor variable (Muftuoglu et al., 2004; Vieira et al., 2013; Yalınay Dikmen et al., 2017; Shim, Park & Park, 2018); however, when added to the previous variables it did predict the individuals’ quality of life (Vieira et al., 2013). Cerutti et al. (2016) found significant differences between women with and without alexithymia for both depression and anxiety.

Finally, only two studies assessing individuals with TMDs had concluded that high TAS values were significant compared with the healthy group, directly correlating higher scores with greater pain intensity and greater depression levels (Glaros & Lumley, 2005). Haas et al. compared the TAS with facial emotion recognition in images, obtaining an inverse relationship between higher TAS levels with lower emotional recognition capacity (Haas et al., 2013).

There are three factors that subdivide the TAS tool (TAS-20 and TAS-26): the capacity to differentiate feelings and emotions (DDF), the capacity to describe feelings and emotions (DIF) and the capacity to express outwardly oriented feelings (EOT). It is important to note that the DDF provides much more information compared with the other factors, given that higher scores and statistically significant differences have been observed in relation to healthy individuals (Glaros & Lumley, 2005; Cerutti et al., 2016; Galli et al., 2017; Shim, Park & Park, 2018), this factor correlating only with pain severity. However, when controlling for the depression variable, the DDF changes to EOT (Glaros & Lumley, 2005). On the other hand, among symptomatic groups, individuals with tension headache obtain higher DIF and EOT scores, as shown in one study (Shim, Park & Park, 2018), just as there are significant differences in the DIF among migraine groups, with the less frequent temporal attack groups being those who report higher scores (Yalınay Dikmen et al., 2017). Finally, the study by Vieira et al. reveals a correlation with the total TAS as a predictive factor for quality of life. This study also includes EOT as a predictor for individuals with migraine and highlights the fact that for healthy groups, it is the DIF in the correlation that predicts the individuals’ quality of life (Vieira et al., 2013).

Quantitative analysis

TAS

The meta-analysis showed statistically significant differences, with higher values in patients compared with asymptomatic individuals, with a small clinical effect in 9 studies (n = 1519; SMD 0.46; 95% CI [0.22–0.71]; heterogeneity Q value 66.86; p < 0.001; inconsistency (I²) = 81%, 95% CI [69–89]). Egger’s test results suggested no significant evidence of publication bias for the TAS analysis (intercept = 3.28; t = 0.66; p = 0.53). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 2). The sensitivity exclusion analysis suggested that no study significantly affected the pooled SMD (Fig. S1).

TAS-DDF

The meta-analysis showed no statistically significant differences in 6 studies (n = 1,204; SMD 0.15; 95% CI [−0.03 to 0.33]; heterogeneity Q value 20.76; p = 0.01; I² = 57%, 95% CI [12–79]). Egger’s test results suggested no significant evidence of publication bias for the analysis of TAS-DDF (intercept = 3.19; t = 0.00; p = 1). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 3). The sensitivity exclusion analysis suggested that no study significantly affected pooled SMD (Fig. S1).

TAS-DIF

The meta-analysis showed statistically significant differences, with higher values in patients compared with asymptomatic individuals, with a large clinical effect in 6 studies (n = 1204; SMD 0.39; 95% CI [0.19–0.59]; heterogeneity Q value 25.17; p < 0.001; I² = 64%, 95% CI [30–82]). Egger’s test results suggested no significant evidence of publication bias for the analysis of the TAS-DIF (intercept = 3.80; t = 0.23; p = 0.82). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 4). The sensitivity exclusion analysis suggested that no study significantly affected the pooled SMD (Fig. S2).

TAS-EOT

The meta-analysis showed no statistically significant differences in 5 studies (n = 1204; SMD = 0.12; 95% CI [−0.09 to 0.32]; heterogeneity Q value = 26.81; p < 0.001; I² = 66%, 95% CI [34–83]). Egger’s test results suggested no significant evidence of publication bias for the analysis of TAS-EOT (intercept = 3.92; t = −0.22; p = 0.83). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 5). The sensitivity exclusion analysis suggested that no study significantly affected the pooled SMD (Fig. S2).

Correlations between alexithymia and anxiety

The meta-analysis for the association between alexithymia and anxiety in patients with CFP showed statistically significant correlations with a small clinical effect in 3 studies (n = 678; SMD 0.31; 95% CI [0.10–0.50]; heterogeneity Q value 46.59; p < 0.001; I² = 87%, 95% CI [76–93]). Egger’s test results suggested no significant evidence of publication bias for the analysis of alexithymia and anxiety (intercept = 2.52; t = −2.06; p = 0.09). The shape of the funnel plot appeared to be asymmetrical in the dominant model (Fig. 6). The sensitivity exclusion analysis suggested that no study significantly affected the pooled SMD (Fig. S3).

Correlations between alexithymia and depression

The meta-analysis for the association between alexithymia and depression in patients with CFP showed statistically significant correlations with a small clinical effect in 4 studies (n = 825; SMD 0.37; 95% CI [0.23–0.51]; heterogeneity Q value 45.21; p < 0.001; I² = 80%, 95% CI [64–89]), Egger’s test results suggested no significant evidence of publication bias for the analysis of alexithymia and depression (intercept = 1.82; t = −1.94; p = 0.09). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 7). The sensitivity exclusion analysis suggested that no study significantly affected the pooled SMD (Fig. S3).

Meta-regression between alexithymia and anxiety or depression

The meta-regression showed significant relation between the TAS and anxiety (β = 0.020, 95% CI [0.003 to 0.04], p< 0.05) nut not between the TAS and depression (β = 0.021, 95% CI [−0.02 to 0.06], p > 0.05). The meta-regression showed no significant relation between the TAS-DIF and anxiety or depression (β = 0.033, 95% CI [−0.02 to 0.09], p > 0.05 and β = 0.009, 95% CI [−0.03 to 0.05], p > 0.05, respectively) (Figs. 8 and 9).

Facial emotion recognition

The facial emotion recognition test consists of presenting a series of images to the participants that show a face expressing the 6 basic emotions (happiness, sadness, anger, fear, disgust and surprise), as well as a neutral addition that they should identify as they see fit. Only 2 studies (Haas et al., 2013; von Piekartz et al., 2015) compared CFP (in this case TMDs) with asymptomatic individuals and with a transversal character. In both, individuals with temporomandibular pain were less accurate in identifying the presentation of images. In the Haas et al. study, participants identified significantly fewer emotions of sadness and fear, and the emotion of disgust was incorrectly identified significantly more often. For anger, happiness, disgust and surprise they found no statistically significant differences between groups (Haas et al., 2013). von Piekartz et al. (2015) had also documented the response time of the participants, finding that individuals in pain took longer to respond. Happiness, disgust and anger were more accurately identified than surprise, and disgust more accurately than sadness.

Quantitative analysis

The meta-analysis showed statistically significant differences, with greater difficulty of recognition in patients with CFP compared with asymptomatic individuals, with a large clinical effect in 2 studies (n = 78; SMD −1.17; 95% CI [−2.01 to −0.33]; heterogeneity Q value 2.97; p = 0.08; I² = 66%). The shape of the funnel plot appeared to be symmetrical in the dominant model (Fig. 10).

Limitations

This systematic review with meta-analysis presents a number of limitations. First, the design of the studies prevents a cause-effect relationship from being established. Future studies using prospective longitudinal designs are needed to better understand how alexithymia influences CFP. Second, the quantitative results of the meta-analysis showed the presence of high heterogeneity, which should be taken into consideration as an important limitation. Third, the outcome measures are all based on self-reported registers. Fourth, it could be possible that the databases used for the research could generate a limitation, given that they are not the common ones; however, the authors wanted to guarantee that the research could be freely replicable, thus also included the Portuguese and Spanish language databases. Finally, one meta-analysis consisted of only 2 studies; thus, the findings should be interpreted with caution.

Clinical implications and future research

The presence of alexithymia is related to biological dysfunctions and also to alterations in the psychosocial sphere. It is therefore that in the management of patients with CFP, the assessment of alexithymia should be considered clinically. Clinically, the deficit of facial emotion recognition might not be directly related to the presence of pain, but to the presence of alexithymia, which could be a better predictor of emotional recognition disruption. Future studies should aim to identify the associations between alexithymia with other variables and the timeline as well as calculating the odds ratio.