Bipolar disorder is a severe condition with estimated lifetime prevalence ranging from 1 to 5% across countries [1], and high risk of functional impairment and suicide [2]. Bipolar disorder is often unrecognized or misdiagnosed at its onset, since it may begin with a depressive episode that can be indistinguishable from unipolar depression and because hypomania or sub-threshold hypomania can go undetected or be attributed to substance use [3, 4]. Several screening measures have been developed to favour the early identification of bipolar spectrum disorders [5]. The Mood Disorder Questionnaire (MDQ) [6], is one of the most frequently used tools for the screening of bipolar spectrum disorders [7]. The core of the MDQ includes 13 yes/no questions on symptoms related to mania/hypomania (Table 1), plus two additional questions on the simultaneous occurrence of the symptoms and the impact of the symptoms on work, family life, legal troubles or getting into fights.

The psychometrics properties of the MDQ have been extensively investigated [8, 9]. At the conventional agreed cut-off of 7, the MDQ has a good capacity to detect people with probable bipolar disorder, with sensitivity around 0.60 and specificity around 0.85 [9]. Diagnostic accuracy of the MDQ did not differ significantly between studies from Eastern and Western countries, when taking into account clinical confounding variables [8, 9]. Overall, good accuracy of the MDQ was found in clinical psychiatric settings in the United Kingdom [10], in Italy [11], in France [12], in Spain [13, 14], and in Turkey [15], as well as in community samples in Italy [16] and France [17]. Accuracy was of a lower level in the United States in both the clinical setting [18], and in prison [19].

The MDQ items cover a broad range of content, from increased energy and risky behavior to distractibility and racing thoughts. The specificity of mania/hypomania of the symptomatic range covered by the MDQ has been questioned, since some of these symptoms may also occur in anxiety, in trauma-related, and in impulse control disorders [20]. However, some of this criticism may be misguided, since they overlook the high comorbidity of bipolar spectrum disorders with anxiety [21, 22], trauma-related [23], and impulsivity/dyscontrol disorders [24, 25]. Nonetheless, whether the MDQ taps into a single factor of mania/hypomania or whether it can be decomposed into different dimensions is still a matter of debate.

Several studies based on Exploratory Factor Analysis (EFA) provided evidence for two factors of the MDQ. However, the studies did not agree on the distribution of items by factor and on its meaning [26]. By taking into account several independent EFA Chinese studies, Massida et al. [26] identified three dimensions of the MDQ, labeled as “acceleration” (including items 1, 2, 4, 5, 6, and 7), “energy” (items 3, 8, 9, and 10), and “imprudence” (items 11, 12, and 13). This partition is the most coherent with a similar three-domain structure of mania, composed by a dimension of mental activation (including racing thoughts), a dimension of elated/high mood (including elevated/expansive mood), and a dimension of behavioral activation (including overactivity) [27]. However, the most commonly described structure of the MDQ purports a two-factor structure. Mangelli et al. [28], in an Italian sample of 1034 individuals recruited from a community via advertisement, have found two factors with EFA, one is the elated mood and increased energy (items 1, 3, 4, 8, 9, 10, and 11) and another is the racing thoughts, distractibility and risky behavior (items 2, 5, 6, 7, 12, and 13). Carta et al. [29], in a study including 291 participants, provided evidence with Confirmatory Factor Analysis (CFA) for two factors, a “self-confidence and energy” factor (items 3, 8, 9, 10, and 11) and a “acceleration, danger and irritability” factor (items 1, 2, 5, 6, 7, 10, 12, and 13). Item 10 “much more social” loaded on both factors, and item 4 “less sleep” was excluded from the CFA on the basis of a past EFA, which included 2278 Italian participants sampled from several Italian towns. Stanton and Watson [7] applied exploratory structural equation modeling (ESEM) to the MDQ responses of 700 participants reporting current psychiatric treatment and enrolled via the Amazon Mechanical Turk (presumably from the United States, but this is unclear). They found evidence for two factors, one is the “positive activation” (items 3, 4, 5, 8, and 9), and the other is “negative activation” (sic) (items 1, 2, 5, 6, 7, 10, 12, 13). Item 5 “more talkative than usual” was allowed to cross-load on both factors, while item 11 “more interested in sex than usual” did not load on any of the two factors. The factor structure of the MDQ is still confusing and the models that have been reported in the literature are still not reproduced so far as independent samples. Overall, there is some evidence that the MDQ might in fact have a one-dimensional structure (see CFA of the unidimensional model and Rasch analysis in Carta et al. [29]).

Thus far, most studies on the factor structure of the MDQ have been carried out in European or Anglo-Saxon cultures. This is the first study to investigate the diagnostic validity and the factorial structure of the Arabic version of the MDQ. We also tested the discriminant validity of the Arabic version of the MDQ concerning its ability to delimitate patients with BD from patients with MDD, which is nuclear for its use as a first-step screening tool in patients with mood disorders.

The study has been conducted according to the guidelines of the 1995 Declaration of Helsinki and its revisions [30]. The Institutional Review Board (IRB) of Razi Hospital has approved the study protocol (authorization signed on 8 Oct 2014).

The sample included 151 patients with a current major depressive episode, 60 men (40%) and 91 women (60%). Mean age in the sample was 42 years (standard deviation [SD] = 9; range: 19 to 61 years old), with no difference by sex. The sample included 89 patients diagnosed with unipolar major depressive disorder (MDD; 59%), 62 diagnosed with a bipolar spectrum disorder, of whom 22 diagnosed with bipolar disorder type I (14.5%), 37 diagnosed with bipolar disorder type II (24.5%), and 3 diagnosed with Substance-induced Mood Disorder, type: manic (2%).

Overall, 69 patients had a university degree or higher education level (45.4%); the remaining patients (n=83; 54.6%) had a high school diploma or a lower level of education. There were 112 patients (74%) that reported being married; the remaining patients were single (n=27; 18%), divorced (n=10; 6%), widowed (n=1; 1%), or cohabiting with a partner (n=1; 1%).

Age of onset of the disorder was 33 years (SD=11; range: 12 to 58 years old), with no relevant differences by diagnosis.

Patients with bipolar spectrum disorders were more likely than patients diagnosed with MDD to have been admitted to a psychiatric hospital (n=22 [35.5% versus n=16 [18%]; χ²=5.05, p=0.025), to have attempted suicide (n=23 [37.7% versus n=15 [16.9%]; χ²=7.25, p=0.007), and to have a family history of bipolar disorder (n=16 [25.8% versus n=4 [4.6%]; χ²=12.24, p<0.0001). A family history of suicide attempt was also more likely in patients with bipolar spectrum disorders than in patients diagnosed with MDD, but the difference did not reach the pre-specified threshold for statistical significance (n=12 [19.4% versus n=8 [9.2%]; χ²=2.40, p=0.121).

3.1. Descriptive Statistics

Reliability, measured as internal consistency, was good: Cronbach’s aλπηα in the sample was 0.80 (95%CI: 0.75 to 0.85). Ordinal aλπηα was 0.88.

Positive replies on the core 13 items of the MDQ in the sample ranged from 33% (item 11, “much more interested in sex than usual”) to 79% (item 7, “easily distracted”). As expected for a mixed sample, there was a large dispersion by item, with a standard deviation ranging from 0.40 to 0.50 (Table 1).

Total score on the MDQ ranged from 0 (n=1) to 13 (n=11), with mean = 7.2 (SD=3.3).

As expected, patients with bipolar spectrum disorders scored higher than patients diagnosed with MDD on the MDQ: 10.0±2.5 vs 5.2±2.3; t = 11.54; p <0.0001.

Overall, the response profile on the MDQ items of the patients diagnosed with MDD was clearly different from that of the patients diagnosed with BD (Fig. 1).

Table 1. Descriptive statistics for the Mood Disorder Questionnaire items for the sample (n = 151).

Item	Mean *	SD	Skewness	Kurtosis
1. So hyper to get into trouble	0.47	0.50	0.11	-2.01
2. Irritable	0.75	0.43	-1.16	-0.64
3. More self-confident	0.54	0.50	-0.16	-1.99
4. Less sleep	0.59	0.49	-0.37	-1.88
5. More talkative	0.53	0.50	-0.11	-2.01
6. Thoughts raced	0.74	0.44	-1.08	-0.84
7. Easily distracted	0.79	0.40	-1.45	0.10
8. Much more energy	0.60	0.49	-0.43	-1.84
9. Much more active	0.58	0.49	-0.31	-1.92
10. Much more social	0.35	0.47	0.64	-1.60
11. Much more interested in sex	0.33	0.47	0.71	-1.51
12. Excessive, foolish or risky things	0.35	0.48	0.61	-1.64
13. Spending money got into trouble	0.56	0.49	-0.25	-1.96

^*SD = Standard Deviation

^* For dichotomous items, item mean multiplied by 100 gives the percentage of “yes” responding.

Fig. (1). Profile plot of the distribution of the MDQ items’ scores by diagnosis. On the vertical axis (y-axis) it is reported the percentage of ‘yes’ endorsement of the item by diagnosis; on the horizontal axis (x-axis) there are the 13 items of the MDQ. Items for which a statistically significant difference at p<0.0001 was found on the Mann-Whitney U test were marked with an asterisk.

The largest effect size for the differences between MDD and BD patients were found in items 1, 3, 8, 9, and 10. Overall, patients with MDD had the lowest chance of endorsement for the items 10 (“Much more social”), 11 (“Much more interested in sex”), and 12 (“Greater involvement in excessive, foolish or risky things”). Conversely, patients with BD had the highest chance of endorsement for items 8 (“Much more energy”), and 9 (“Much more active”).

In particular, patients with MDD scored lower than patients with BD on most MDQ items, with the exception of items 2 (“Being irritable”), 6 (“Thoughts raced “), and 7 (“Easily distracted”) (Table 2).

Table 2. Distribution of scores for the Mood Disorder Questionnaire items by diagnosis.

	MDD n° = 89		BD n° = 62		Mann-Whitney U test	Cliff’s delta
Item	Mean *	SD	Mean *	SD
1. So hyper to get into trouble	0.25	0.43	0.81	0.39	Z=-6.74, p<0.0001	0.56
2. Irritable	0.70	0.46	0.84	0.37	Z=-1.94, p=0.052	0.12
3. More self-confident	0.33	0.47	0.85	0.35	Z=-6.28, p<0.0001	0.50
4. Less sleep	0.47	0.50	0.81	0.39	Z=-4.19, p<0.0001	0.34
5. More talkative	0.34	0.47	0.80	0.40	Z=-5.59, p<0.0001	0.44
6. Thoughts raced	0.69	0.46	0.81	0.39	Z=-1.65, p=0.098	0.12
7. Easily distracted	0.75	0.43	0.84	0.37	Z=-1.22, p=0.223	0.11
8. Much more energy	0.40	0.49	0.90	0.29	Z=-6.16, p<0.0001	0.49
9. Much more active	0.37	0.48	0.87	0.33	Z=-6.10, p<0.0001	0.51
10. Much more social	0.13	0.34	0.66	0.47	Z=-6.64, p<0.0001	0.51
11. Much more interested in sex	0.17	0.37	0.55	0.50	Z=-4.89, p<0.0001	0.39
12. Excessive, foolish or risky things	0.23	0.42	0.55	0.50	Z=-4.02, p<0.0001	0.30
13. Spending money got into trouble	0.42	0.49	0.75	0.43	Z=-4.08, p<0.0001	0.33

MDD = Major Depressive Disorder; BD = Bipolar Disorder.

SD = Standard Seviation.

^*For dichotomous items, item mean multiplied by 100 gives the percentage of “yes” responding.

^**Large effect sizes were marked in bold.

3.2. Principal Component Analysis

Bartlett’s test of sphericity was 481.07 (p < 0.0001), Kaiser-Meyer-Olkin adequacy value was 0.80. The matrix, thus, can be factorized. In a parallel analysis, two components had eigen values superior to both simulated and resampled data (Fig. 2).

The subsequent PCA with varimax rotation to maximize the separation of the components revealed that the two components explained 59% of the variance in the data (Table 3).

The first component can be considered similar but not coinciding with the “elated mood and increased energy” dimension of Mangelli et al. [28], while the second component can be equated to the “racing thoughts, distractibility and risky behavior” dimension of Mangelli et al. [28].

Item 11 “Much more interested in sex” was less clearly separated between the two components (Table 3 and Fig. 3).

Fig. (2). Parallel analysis applied to the items of the Mood Disorder Questionnaire (n = 13). Plot of the eigenvalues calculated on the basis of the actual data and of the simulated and resampled data. The number of dimensions to retain corresponds to the number of eigenvalues that have a higher value than the corresponding eigenvalues calculated on the basis of the simulated and resampled data. In this case, the number of dimensions to retain is 2 and is marked by a vertical dashed line.

Fig. (3). Separation plot of the results of the principal component analysis applied to the items of the Mood Disorder Questionnaire (n = 13). Items are plotted on the basis of their loadings on the two extracted main dimensions. An item is as much “separated” along the two dimensions as much its loading differs across the two dimensions, i.e., it is very high on one dimension and very low on the other dimension.

Table 3. Principal component analysis of the Mood Disorder Questionnaire in the sample (n= 151).

Item	First component*	Second component*
1. So hyper to get into trouble	0.66	0.52
2. Irritable	-0.06	0.72
3. More self-confident	0.85	0.16
4. Less sleep	0.54	-0.10
5. More talkative	0.61	0.33
6. Thoughts raced	0.20	0.53
7. Easily distracted	-0.03	0.84
8. Much more energy	0.92	0.03
9. Much more active	0.83	0.21
10. Much more social	0.81	0.33
11. Much more interested in sex	0.56	0.48
12. Excessive, foolish or risky things	0.28	0.75
13. Spending money got into trouble	0.25	0.52
Proportion of variance	35%	24%

^*Item is assigned to the component on which it has the highest loading

3.4. Roc Analysis

The MDQ had a good capacity of distinguishing patients diagnosed with BD from those diagnosed with MDD (Fig. 4).

The AUC was good: 0.88 (95%CI: 0.81 to 0.96). Sensitivity was also good (0.87), while specificity was acceptable but not optimal (0.77). According to the Youden method, the suggested cut-off was 7, which represents a good compromise between sensitivity and specificity (Fig. 5).

The positive predictive value was 0.66 and the negative predictive value was 0.92. The positive diagnostic likelihood ratio was 3.8, meaning that patients with a mood disorder scoring 7 or higher on the MDQ were about 4 times more likely to have a bipolar disorder than a major depressive disorder.

In the sample, 77 patients reached or surpassed the 7-point threshold on the MDQ, qualifying as a probable case of BD: 55 (out of 62; 89%) among those diagnosed with BD; 22 (out of 89; 25%) among those with MDD (odds ratio = 23.9 [95%CI: 9.5 - 60.2]; z = 6.75, df=1, p<0.0001).

Fig. (4). Receiver operator characteristic (ROC) curve of the predictive capacity of the MDQ in differentiating patients with BD from patients with MDD. Sensitivity and specificity are reported as percentages, with a cross indicating on the curve the best compromise between them (corresponding to the cut-off). The area under the ROC curve (AUC) is reported alongside its 95% confidence interval.

Table 4. Goodness-of-fit indices of the tested models.

	Goodness of fit indicators							Wald Rank Rule
Model	|2	df	p	CFI	RMSEA (90%CI)	SRMR	McDonald’s ω	n columns	Rank
Unidimensional model	95.96	65	0.008	0.96	0.058 (0.031 – 0.081)	0.082	0.81	26	26	Identified
Mangelli et al. (2005) two factor model	87.14	64	0.029	0.97	0.050 (0.017 – 0.075)	0.078	0.82	27	27	Identified
Carta et al. (2014) two factor model	77.52	52	0.012	0.97	0.059 (0.028 – 0.085)	0.080	0.82	26	26	Identified
Massidda et al. (2016) three factor model	81.44	62	0.050	0.98	0.047 (0.002 – 0.073)	0.076	0.82	29	29	Identified
Threshold for fit			p > 0.05	≥ 0.90	≤ 0.08	≤ 0.09	≥ 0.90	n =	Rank

Fig. (5). Operating characteristics of the MDQ for various threshold scores among 151 Tunisian patients diagnosed with a current episode of major depressive disorder either in the course of a unipolar or bipolar mood disorder as diagnosed with the SCID.

This is the first study exploring the diagnostic validity and the factorial structure of the MDQ in an Arabic speaking sample of patients with mood disorders. The study provided evidence of good reliability of the Arab version of the MDQ, albeit with suboptimal values for a screening tool (<0.90; see Kottner et al. [35]). Reliability of the MDQ was surprisingly overlooked in past investigations. However, when tested, the Cronbach’s alpha appeared to be acceptable for group comparisons, with values ranging from 0.78 in a general population Hong Kong community sample [46], to 0.82 in Chinese patients diagnosed with mood disorders [47], 0.87 in a mixed sample of Brazilian patients with mood disorders and controls with no psychiatric disorder [48], 0.88 in Korean patients diagnosed with bipolar disorder [49], 0.89 in a French sample of patients diagnosed with MDD or bipolar spectrum disorders [12]. In this study, too, when tested with a procedure that is more suited for dichotomous (yes/no) data, reliability was 0.88. Nonetheless, when estimated from CFA with McDonald’s omega, reliability was still below the optimal threshold suggested for the screening tools [35]. Further investigation of the reliability of the MDQ is warranted. To the best of our knowledge, the only other screening tool for bipolar disorder so far validated in the Arabic language is the Arabic adaptation of the Hypomania Check List-32, second revision (HCL-32-R2) by Fornaro et al. [50], which shows a PPV for BD (I and II) of 0.93 and a NPV of 0.73. However, in the Fornaro study, the proportion of people with BD was inflated by active enrollment, therefore the information on the PPV and NPV should be taken with caution. Our study design attempted to represent “real life conditions”, as newly presenting patients were enrolled, while it was initially unknown whether their major depressive episode was uni- or bipolar.

In this study, the sensitivity of the MDQ was better (0.87) than in the past investigations: it was 0.69 in the meta-analysis of Carvalho et al. [8], and 0.76 in the meta-analysis of Wang et al. [9]. Specificity was 0.77 in this study, it was 0.79 in the meta-analysis of Carvalho et al. [8], and it was 0.88 in the meta-analysis of Wang et al. [9]. However, in primary care or general population settings, the sensitivity of the MDQ was calculated to be low (0.43), albeit with excellent specificity (0.95) [8]. This may depend on the tendency of people with hypomania to overlook their own symptoms and underreport them. Nevertheless, when compared with another frequently used screening tool for hypomanic/manic symptoms, the HCL-32, the MDQ revealed a better profile. In nine studies with available data, the two screening tools showed comparable sensitivity (0.82; 95%CI: 0.59-0.71 for the HCL-32, and 0.80; 0.71-0.86 for the MDQ), but the MDQ showed a better specificity (0.70; 0.59-0.71) than the HCL-32 (0.57; 0.48-0.66) [51]).

It should be noted that in this study, the large majority of the patients who were SCID diagnosed with a BD were detected as probable cases of BD by the MDQ at the suggested threshold of 7. However, about 25% of the patients SCID diagnosed with unipolar MDD screened positive for probable BD. As a matter of fact, when a patient presents with a major depressive episode, it cannot be excluded that it will manifest symptoms of hypomania or mania in the future. Indeed, a fraction of patients with unipolar MDD may switch to bipolar disorder over time, more often as an effect of antidepressant-associated mood-switching [52]. Moreover, subsyndromal and major depression are the known precursors of BD [53]. Thus, it cannot be excluded that a subsample of the patients SCID diagnosed with MDD in this study might have a propensity towards progressing into BD that was detected by the MDQ. Long-term longitudinal studies are necessary to test whether those patients with MDD who are detected as at risk of BD by the MDQ will evolve into a frank BD syndrome at follow-up.

As in past investigations (for a review, see Massidda et al. [26]), the data spontaneously distributed into two separated dimensions, one is elated mood and increased energy, and the other one is behavioral activation, secondary to racing thoughts and greater involvement into risky behavior. However, when CFA was applied to the data, a reasonable fit was found for several models of the MDQ. A marginally better fit was found for a three-factor model derived from studies that were done in China [26]. Indeed, a three-factor model is the most compatible with a similar three-dimension model of mania as composed by mental activation, elated/high mood, and behavioral activation [27]. Nevertheless, the unidimensional model also had a reasonably good fit to the data, as expected for a screening tool.

This is the first study to provide evidence for reproducibility of some of the models of MDQ factor structure that have been described in the literature. It should be noted that the sample was a mixed sample of patients diagnosed with affective disorders, including both patients diagnosed with MDD and patients with bipolar spectrum disorders. This may have led to some dispersion of the data, since patients with MDD, as expected, were much less likely to endorse the items of the MDQ. Moreover, the sample size was acceptable for principal factor analysis, with about 10 cases per item, but it was overall rather small for extended investigations of additional models (e.g., the second-order, hierarchical and bifactor implementation of the models).

Some findings deserve attention. As in some non-Western studies [46-48, 54, 55], item 11 on a greater interest in sex was connected both to elated mood and to risky behavior. In all likelihood, there are patients that feel the increased sexuality that is related to hypomania/mania as troublemaking rather than egosyntonic, i.e., acceptable to the needs and goals of the subject. Indeed, it cannot be excluded that the spontaneous distribution of the items into two dimensions at the PCA might be a reflection of the propensity of the patients of giving prominence to the more egosyntonic symptoms of hypomania/mania (those that grouped under the first dimensions of elated/high mood), or, conversely, of their tendency to principally report the most disturbing ones (those grouped under the second dimensions of racing thoughts and risky behavior). This may be related to the personal attitudes of the patients and their socio-cultural background rather than a real bipartition of the MDQ’s items. As a matter of fact, we agree with the consideration of Carta et al. [29], who considered that with such a small number of items, parsimony is preferable, thus favoring the unidimensional structure of the MDQ. Such a structure is coherent with the main purpose of the MDQ, which is not to measure the latent dimension of hypomania/mania along its continuum, but to identify which persons are located on its positive side, i.e., to identify clinically at-risk subjects and discriminate them from those who do not have bipolar disorder.

Small sample size for more complex analysis and the mixed nature of the sample represent the main limitations of the study. Convergent and divergent validity was not explored in this study, as it is customary for adaptations of questionnaires in another language (e.g., see Ioannou et al. [56],). However, a ROC analysis was applied to assess the capacity of the MDQ to distinguish patients with BD from patients with MDD, and this is a more robust method to assess the discriminative validity of a test than simply a correlational analysis. Additional studies, with larger samples and from more diverse socio-cultural and ethnic populations are necessary for a full understanding of the factorial structure of the MDQ, in particular to explore the measurement invariance of the MDQ across diagnosis (BD versus MDD).

The Arabic version of the MDQ showed good measurement properties in terms of reliability and factorial validity, suggesting that its items measure a single latent trait that clusters into two dimensions of elated mood and behavioral activation. In this study, the sensitivity and specificity of the MDQ were reasonably good, with a robustly informative diagnostic likelihood ratio of nearly 4.

The study is a contribution to a more comprehensive understanding of hypomania/mania across different ethnic and cultural enclaves, from the perspective of a real worldwide investigation of psychiatric disorders.