In order to achieve comparability between the Italian translation of BDRS and the original instrument, we used a multiple-phase translation process based on the 5 cross-cultural criteria of Flaherty (Flaherty et al., 1998) and general guidelines for translating study instruments. Firstly two independent teams (Rome and Bari) translated the original scale into Italian; then the translations were compared and integrated into a single version. The next step was a back translation into English by a native speaker, followed by a consistency check by the original authors.

The Italian version of the BDRS, as the original instrument, consists of 20 items each with a score ranging from 0 to 3, with a minimum of 0 and a maximum score of 60.

Subjects were recruited from July 2011 to July 2015 among outpatients referred to the Bipolar Disorder Unit of the A. Gemelli Hospital in Rome and among inpatients and outpatients admitted to the Psychiatric Ward of the Hospital Policlinico Consorziale in Bari.

Inclusion criteria were: (Judd et al., 2002) currently meeting DSM-IV-TR criteria for Bipolar Disorder I (BP-I), Bipolar Disorder II (BP-II) (2) age 18–75 years; (3) current experience of depressive symptoms (but not necessarily fulfilling criteria for a major depressive episode); (4) native Italian speakers, with mastery of spoken and written Italian language.

Subject were excluded if any of the following conditions were present: (1) a diagnosis of mental retardation or documented IQ < 70; (2) any other DSM diagnosis (3) unstable general medical conditions; (3) clinically significant pre-study physical exam, electrocardiogram, laboratory or urinalysis abnormalities indicating serious medical disease impairing evaluation; (4) pregnant or breast-feeding women. A sample of patients with a current diagnosis of Major Depressive Disorder (MDD), fulfilling the same inclusion/exclusion criteria was recruited.

At evaluation, patients followed a naturalistic maintenance treatment, with typical or atypical antipsychotics (Asenapine, Aripiprazole, Clotiapine, Clozapine, Haloperidole, Olanzapine, Quetiapine, Risperidone, and Paliperidone), mood stabilizers or antiepileptic drugs (Lithium, Valproate, Carbamazepine, Lamotrigine, and Oxcarbazepine), antidepressants (SSRI, SNRI, NaSSA, and other unspecific antidepressants) and Benzodiazepines or Hypnotics.

A BP diagnosis was established by trained psychiatrists using the Structured Clinical Interview for DSM-IV Axis I Disorder (SCID-I) (First et al., 1996). Personality disorders were excluded through the Structured Clinical Interview for DSM-IV Axis II Disorders (SCID-II) (First et al., 1997). An anamnestic interview was administered in order to obtain sociodemographic information and psychiatric history.

All participants were interviewed by specifically trained psychiatrists using the Bipolar Depression Rating Scale (BDRS), HDRS 21-item version (HAM-D), Montgomery Asberg Depression Rating Scale (MADRS) and Young Mania Rating Scale (YMRS). Good inter-rate reliability has been found on all instruments (Fleiss' coefficient 0.82). Patients were further interviewed using the Italian version of the Temperament and Character Inventory revised version (TCI-R), a self-administered interview for personality characteristics (Martinotti et al., 2008).

Anonymity was guaranteed to all the participants: data were de-identified before any further data manipulation from the coordinating center of the Catholic University of the Sacred Heart. Then the database was sent to the University of Bologna, University of Florence and the LUMSA University of Rome for the statistical analysis, ensuring an adequate level of protection using a double level of data encryption.

The distribution of the I-BDRS scores was performed, in order to test whether each item had a normal distribution, using the Kolmogorov-Smirnov one-sample test (K–S test). Descriptive statistics and Pearson's correlation tests between the different rating scales were also used. A distribution can be considered approximately normal whether skewness and kurtosis indices are between −1 and +1 (Joanes and Gill, 1998). Pearson's correlation coefficients between I-BDRS sub-scales, MADRS, HAM-D and YMRS were calculated to assess the convergent validity. In order to test discriminant validity, Student t-tests for independent samples were carried out between the bipolar sample and unipolar sample.

Then an exploratory factor analysis (EFA) was carried out of the I-BDRS through the principal component analysis method using IBM SPSS Statistics 21 software. To test sampling adequacy, we measured Kaiser–Meyer–Olkin (KMO) and Bartlett's test of sphericity. A sample can be considered numerically adequate when KMO is close to 1 or Bartlett's test is significant. The number of components extracted was based on the percentage of variance accounted for by the Kaiser-Guttman method and, overall, the scree plot (Mazza et al., 2012). After the selection of the number of components, we verified, by the means of the communality matrix, whether the factor model adequately represented each of the initial variables (each variable should have a score ≥0.10; a score <0.10 indicates that the variable is not properly reproduced in the factor solution). Subsequently, we verified the component loading of each variable, using the component loading matrix, in order to organize and accomunate each item with his latent factor consinstency. Then we performed a confirmatory factor analysis and carried out structural equation modeling by the use of M-PLUS. This analysis allows to test the goodness of the factor structure emerging from an exploratory model. To test the goodness of fit, we considered absolute fit indices as standardized root mean square residual (SRMR) and root mean square error adjustment (RMSEA) and incremental fit indices as comparative fit index (CFI), incremental fit index (IFI) and non-normed fit index (NNFI). Furthermore, we considered chi-square divided by degrees of freedom (χ²/df). As suggested by Byrne (Byrne, 1998), a model can be considered reasonably standard if SRMR and RMSEA are lower than 0.08, if CFI, IFI, and NNFI are higher than 0.90 and if χ²/df is <3.

The Internal consistency reliability of the scale was tested through Cronbach's alpha coefficient and, in order to explore further variables, it was calculated separately for gender and for each subtype of BD. For Cronbach's alpha interpretation, George and Mallery (George and Mallery, 2003) provided the following rules: α > 0.9 = Excellent, α > 0.8 = Good, α > 0.7 = Acceptable, α > 0.6 = Questionable, α > 0.5 = Poor, and α < 0.5 = Unacceptable.

The analysis of variance (ANOVA) was used to correlate the scores of specific sub-groups, distinguishing between BD I, BD II and unipolar patients. Finally, considering that we had more than two raters, the inter-rater reliability was measured using Fleiss' coefficient (which measures the agreement between a fixed number of raters) instead of the classic Cohen's kappa (valid only on two raters), by the independent rating of five audio-recorded interviews.

On a total of 250 subjects consecutively screened, 188 patients were selected using the aforementioned criteria. Among 188 depressed patients, 63 meet criteria for MDD (33.3%), 62 for BD I (32.8%), and 63 for BD II (33.3%); 14 patients (7.4%) display some significant manic symptoms (YMRS≥12) indicating a mixed state. Bipolar patients gender ratio is 55M/69F whilst in unipolar depression there are more woman (17M/46M) and the age of onset differ from 28.3 years old of bipolar disorder vs. 38.7 years old of major depression. Other clinical characteristics of the sample are shown in Table 1, detailed for different types of depression (Bipolar and Unipolar) and sites of screening. Table 2 shows pharmacological treatment.

Mean, standard deviation, skewness and kurtosis for each item were calculated, in order to test whether the distribution is approximately normal (see Figure 1 and Table 3). Results showed that no item had extreme means or standard deviation close to zero; furthermore, skewness and kurtosis were between −1 and +1, except that for six items for which they were slightly lower or higher. These results suggested that the item distribution could be considered approximately normal.

A summary statistic of the scales shown a mean of 24.32, a Standard Deviation of 8.46, with a variance of 71.62. The Cronbach's alpha is 0.81 and the Standardized alpha results in 0.82.

We tested the psychometric properties and the dimensional structure of the BDRS through the principal component analysis. In the preliminary analysis, KMO is equal to 0.85 and Bartlett's test is significant (χ(190)2 = 1277.59; p < 0.001), suggesting our sample results are adequate. The scree plot suggests that three factors should be extracted; they have eigenvalues > 1 and explain 47.38% of the total variance. Communalities are observed between 0.19 and 0.66, suggesting that each item is well-represented by the factorial model. Table 4 shows the component loading matrix (just loads >0.30), eigenvalues and the percentages of variance accounted for each dimension.

The first factor was denominated “primary depression symptoms” because it concerns aspects such as suicidal ideation, depressed mood, appetite disturbance and impaired concentration. The second factor was denominated “mixed symptoms” because it concerns aspects such as lability, increased motor drive, increased speech or agitation. Finally, third factor was denominated “secondary depression symptoms” because it concerns aspects such as reduced motivation, anhedonia, affective flattening, or worthlessness.

On the basis of the results of the component analysis, the hypothetical model (Figure 2) consists of 3 latent variables (primary depression symptoms, mixed symptoms and secondary depression symptoms, represented in the ellipses) and 20 observed variables (the items, represented in a box).

The model reached the following fit indices: SRMR = 0.080, RMSEA = 0.071, CFI = 0.861, and NNFI = 0.853; furthermore, the χ²/df was 1.68 (279.77/167). Therefore, the results suggested that the model fits the data reasonably well, although CFI an d NNFI were slightly lower than 0.90.

The Italian BDRS total has strong positive correlation coefficients with HAM-D (r = 0.81, p < 0.001) and with MADRS (r = 0.67, p < 0.001) and, in contrast to the original validation sample of Berk et al. (2007), a good correlation with YMRS (r = 0.46, p < 0.001). The analysis of variance (ANOVA) between total scores in BD I, BD II, and MDD patients is shown in Table 5.

The internal consistency reliability of the scale, assessed by calculation of Cronbach's alpha, as mentioned shows that I-BDRS has a score of 0.82. This good result is coupled with the excellent values obtained on the total index of skewness and kurtosis, both below zero with a decline of tails better than average. The Kolmogorov-Smirnov test gave a similar result (d = 0.057; p = 0.92), confirming a normal distribution (see Figure 1).

In order to explore the possible effect of gender or the impact of sub-diagnosis on internal consistency of the instrument, the Cronbach's alpha of the different subgroups is shown in Table 6.

Pearson's correlation coefficient between each item and the total BDRS score (item total correlations) is shown in Table 7: all items significantly correlated with the total score at the 0.05 level of significance. For a better understanding of the instrument's reliability and sensibility, the same analysis was performed for the 3 sub-samples (BD I, BD II, and MDD patients), results are shown in Table 7.

Table 8 reports the Cronbach's alpha for each scale of the BDRS and the Pearson correlations between the BDRS's subscales and other measures.

The I-BDRS's subscales had good internal consistency coefficients and therefore they can be considered reliable, as suggested by Cronbach's alpha coefficients between 0.75 and 0.84. Furthermore, the measures concerning depression (MADRS and HAM-D) were positively related to the I-BDRS's subscales, but mostly to the two subscales measuring depression (primary and secondary). Similarly, the measure concerning mania was positively related to I-BDRS's subscales, but mostly to subscale measuring mixed symptoms. Moreover I-BDRS scores, controlling for age and sex, were significantly higher in BD1 (F = 7.34 p = 0.001). Higher scores in BD I were observed also on MADRS scores (p < 0.001) and HAMD (p = 0.001). The BD I group also had higher scores on the YMRS (p < 0.001).

We then tested differences on I-BDRS single items (BD I vs. BD II vs. MDD), controlling for initial severity, other than age and sex. In BD I we found significantly higher scores on the psychotic item (p = 0.001) and agitation item (p = 0.001) whilst BD II patients scored higher on flattened affect (p = 0.001) and anhedonia (p = 0.003). MDD patients showed higher scores on hopelessness (p = 0.002).

Finally, in order to test discriminant validity, we compared the scores of the bipolar sample with the score of unipolar sample considering the 3 subscales. In particular, we hypothesized that the bipolar group would have higher scores than the unipolar group in terms of mixed symptoms, while there would not be significantly differences in primary and secondary depression. The results of t-tests for independent samples supported these hypotheses; in primary and secondary depression the means of the bipolar group (M = 1.28 and sd = 0.54; M = 1.45 and sd = 0.64, respectively) were similar to the unipolar group (M = 1.41 and sd = 0.57; M = 1.48 and sd = 0.61, respectively).

In mixed symptoms, the mean of the bipolars group (M = 0.72 and sd = 0.52) was significantly higher (p < 0.05) than the unipolar group (M = 0.56 and sd = 0.45), suggesting that the BDRS was able to distinguish between unipolar and bipolar depressed patients. These positive correlations supported the convergent and discriminant validity of the I-BDRS.

The correlation between the total score of BDRS and TCI-R subscales is shown in Table 9. No evident correlation was found in any of the TCI-R subscales.

The results show that the I-BDRS is a valid scale for the measurement of depression in patients with Bipolar Disorder, with a considerable internal consistency (Cronbach's α 0.82), a significant item-total correlations (although in a range from 0.18 to 0.56) and strong positive correlation with the depressive symptom severity measured by the other administered scales (MADRS r = 0.67, p < 0.001; HAM-D r = 0.81, p < 0.001; YMRS r = 0.46 p < 0.0001), including the YMRS (different to other samples in the literature and to the original validation sample) (Berk et al., 2007). There is a good inter-rater reliability, measured by Fleiss' coefficient with a mean result of 0.81. The Cronbach's alpha remains high even in the different subsamples, confirming that there is no impact of gender as well as Bipolar I or II diagnosis. These robust results are noteworthy, expecially when considering the notable representativeness of demographic distributions of the sample that includes bipolar inpatients and outpatients, tested in two separated Italian centres, with quite different catchment areas. The clinical characteristics of the sample had favorable elements including an even BP I/BP II ratio (62/63) and expected differences between unipolar and bipolar patients (higher number of females in MDD, more episodes and a longer illness in BD), confirming patterns seen in other samples reported in literature (Weissman et al., 1996; Kessler et al., 2003; Berk et al., 2007; Mazza et al., 2011, 2012; Nivoli et al., 2011).

We showed that a three-cluster structure of the instrument is the most parsimonious model, in line with the original construction of the scale that included two sections for depression and one for mixed symptoms.

As with other scales such as the Beck Depression Inventory (BDI) and other international validations of the BDRS (Galvão et al., 2013; Sarró et al., 2015) the factors related to a psychological component of depression (Anhedonia, Reduced Motivation, Flattened Affect, Worthlessness, Social Impairment, Reduced Energy) are associated in a specific sub-scale, and the impact of the core component of depression (Mood, Suicidal Ideation, Change in Appetite and Sleep) is expressed by another sub-scale. Other items typical of mixed states (Motor Drive, Agitation, Lability, increased speech, Irritability and Psychotic Symptoms) were carried in the same cluster, constituting the subscale for mixed depression.

The patterns of item's mean scores give an indication of how the symptoms perform in terms of salience for this sample. The highest mean load fell between the mild and moderate anchor point. The most highly expressed items were anxiety, anedonia, social impairment, depressed mood and reduced activity and the least loaded items were irritability, increased motor drive, increased speech, and agitation. This is compatible with literature suggesting that mixed states are present in a significant minority of individuals with bipolar depression (Benazzi, 2004a).

MDD patients showed higher mean in MADRS and HAM-D total scores but similar scores on the BDRS compared to Bipolar Disorders. A possible explanation could be addressed to the structure of the scales, expecially if we take into account the results of the item-total correlation of MDD patients in BDRS: the main factor loading for the total score is the summary of classical depressive symptoms (like reduced concentration, helplessness, guilt, social impairment and reduced motivation) and physical symptoms (linked biological and circadian rhythms such as appetite and sleep disturbances). The patterns were completely different among the BD sample, where the factor loadings exclude appetite and sleep and are more weighted to mental symptoms and mixed symptoms of depression.

It is interesting that, in MDD patients, anxious, and pure depressive symptoms are the lowest contributors to the total: this could explain the higher scores on the MADRS vs. HAM-D (considering that MADRS items are more focused on mood symptoms while HAM-D has more items for evaluation of anxious symptoms). Regarding the ANOVA symptoms analysis between BD I and BD II, BD I patients are different in terms of psychotic symptoms and agitation, while Bipolar II patients more frequently express anhedonia and flattened affect. MDD patients are characterized by more hopelessness than BP patients. Another interesting finding is that unipolar depressed patients showed a lower load of anxious and pure depressive symptoms on the total scores of the BDRS, in agreement with other studies (Katz et al., 1982).

Our study confirmed that anhedonia could have a role in discriminating between MDD and BD depressed patients, as reported by several authors (Endicott et al., 1985; Coryell et al., 1989; Parker et al., 2000). This observation could be a possible explanation of the fact that, even if the I-BDRS scores were similar between unipolar depressed patients and Bipolar I depressed patients, MADRS and HAM-D scores differed between the two samples. Moreover, Bipolar I patients showed higher scores on the different psychometric scales, with similar means compared to those of unipolar depressed patients (except for MADRS scores that were higher in unipolar depressed patients with a higher standard deviation).

As for BDRS item correlation in the different subsamples it is noted that Bipolar I patients had few factors with a high loading (social impairment, reduced concentration, worthlessness, anxiety, helplessness, and guilt) whilst Bipolar II patients showed a more varied picture with a widespread factor loading that excludes, as in unipolar depressed patients, only anxious symptoms. From this point of view, it seems that Bipolar II depression, excluding mixed symptoms, is more similar to unipolar depression while Bipolar I depression is characterized by fewer but more expressed symptoms. Besides, patients with Bipolar I Depression patients seem to present more anxiety. It is not clear if depression with psychomotor activation and anxiety can be ascribed purely to bipolar disorder, but mixed depression can be represented as a symptomatic continuum between unipolar depression and mania, with variable expressions of bipolarity representing dimensions of underlying pathophysiologic processes (Benazzi, 2007).

As reported in literature (Galvão et al., 2013; Hirschfeld, 2014) psychometric instruments that can differentiate mixed symptoms could be very useful in the diffential diagnosis of Bipolar or Unipolar Depression. A Portuguese version of the BDRS (BDRS-P) already demonstrated validity in screening bipolar patients (Galvão et al., 2013). Our study contributes to outline the clinical usefulness of this instrument and confirms that BD patients score higher on the mixed symptoms subscale.

Although the sample is balanced and thus could be representative of the different characteristics of the depressed population, we have not found noticeable associations between the BDRS total scores and the subscales of the TCI-R.

The limited statistical power may have not allowed us to detect temperamental and characterial nuances of the patients. We did however exclude personality disorder, which may have weakened the capacity to find such factors. The main limitation is sample size, which reduces the strength of our findings. Another limitation could be that the assessors were not blind about the clinical history of each patient. Furthermore, the naturalistic characterization of the sample might have introduced a number of confounding factors (e.g., treatment options), though we systematically controlled for some of them.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.