The Beck Anxiety Inventory (BAI) is a prominent screening and outcome research tool for measuring the anxiety (1), and is validated in a number of languages, including German, French, Chinese, Spanish, Persian, Nepal, Icelandic, and others (2–7). Strong psychometric evidences have been established in diverse samples, including diverse clinical samples mixed psychiatric patients (8–14), panic disorder with and without agoraphobia (15), adolescent psychiatric patients (13, 16, 17), and older adult psychiatric patients (18) to non-clinical samples (19–21). The BAI requires about 5–10 min to administer (about 10 min for oral administration), and <5 min to interpret the scores. Among counselors in primary health care settings, BAI is reported to be the ninth most commonly used tools in the United States (22), owing to the advantages in cost-effectiveness and brevity in the application procedure. According to the meta-analysis of BAI (k = 117), BAI was reported to manifest an excellent internal consistency in clinical (0.91) and non-clinical sample (0.91) and a good test-retest reliability in clinical (0.66) and non-clinical (0.65) (1).

This 21-item self-report questionnaire was originally developed to assess clinical anxiety (i.e., an excess of normal anxiety resulting in significant distress and impairment of functioning (23, 24), differentiated from normal anxiety (i.e., an adaptive emotional responses to danger or threat, (25, 26), as well as the unique aspects of anxiety disorders that are assumed to differ from those of depressive disorders (27). While depression is defined as the experience of being sad, gloomy, and empty which is typically associated with events experienced in the past and decreased autonomic activity, anxiety is defined as feelings of fear and tension, and apprehension which is usually related to anticipation of future events and activation of autonomic nervous system (28).

However, some issues remain regarding the BAI's discriminant validity against depressive disorders. Distinguishing anxiety and depressive disorders through self-report measures has been the subject of debate due to a high rate of comorbidity or the possibility of a single, shared underlying mechanism, such as negative affect (28). Although Beck's original studies report significantly higher BAI mean scores for patients with anxiety disorders compared to those with depressive disorders (9, 27), other studies fail to replicate the results (29). In one study conducted in Korean sample, correlation coefficients of BAI with other depression assessments such as BDI and PHQ-9 are found to be even higher than other anxiety assessment tools (30). The ambiguity found in the mean scores difference of BAI and correlation sizes with other anxiety and depression measures questions the utility of BAI as measuring the general anxiety that are distinguishable from depression as Beck et al. (27). Because various anxiety and depression tools aim to assess the same construct with heterogeneous factors, more diverse assessments must be incorporated into an analysis to provide a comprehensive outlook on the divergent and convergent validity of the BAI.

Another issue concerns the BAI's clinical utility as an anxiety screening tool and a measurement of severity in primary care settings. Although the BAI was not originally developed as a diagnostic tool, it is essential to examine the degree of its diagnostic reliability and its score distribution in a sample before it can be utilized as a tool in anxiety screening, to track symptom changes, or as an outcome measure based on severity measurements. To date, 11 studies explored the diagnostic validity of BAI, optimal cutoff scores ranging from 7 to 26 depending on the diversity of studied samples (1). As the diagnostic cutoff score can be varying across the ethnicity and cultural background behind the research setting, undertaking the diagnostic validity study in a new ethnic sample can benefit the existing literature. Negative and positive predictive powers are critical sources for determining the clinical utility of a screening tool, especially for disorders in low prevalence rate (12). Providing such information would thus help researchers to decide optimal BAI cutoff scores for their individual purposes.

The BAI is a prominent anxiety assessment tool in Korea, with uses ranging from intervention outcome measures to mechanism studies. According to the epidemiological survey of mental disorders in Korea conducted by ministry of health and welfare, the estimated lifetime prevalence rate of anxiety disorders for Korean adults was 9.3% (male 6.7%, female 11.7%) and 1-year prevalence rate of anxiety disorders for Korean adults was 5.7% (male 3.8%, female 7.5%) (31). It was suggested that a social stigma against mental illness and inability to recognize the need of treatment may contribute to lower prevalence rates in Korea than that of Western countries, and hence the need to develop screening tools to identify the targets who need treatment was emphasized. Thus, far, the factor structure of BAI has been heterogeneously reported across studies from 2 factors to 4 factors (32, 33), which are comparable to the BAI literature conducted in other languages (1). Internal consistency and test-retest reliability of the Korean BAI have been reported as 0.91–0.93 and r = 0.84, respectively (30, 32, 34). Before publication of the formal translated version of the BAI by Korea Psychology Co., Ltd, which underwent several back-translation processes, all previous studies except but the most recent study (30) were based on a version of the BAI independently translated by Kwon (34). The study with newly translated version reported moderate correlation of the BAI with STAI-S, STAI-T, and PHQ-9, with excellent reliability (30). However, this was based on a stratified sample of community-dwelling people; no literature is yet available to determine the clinical utility of the BAI in Korean psychiatric populations.

Using unbiased clinical and non-clinical samples, this study aims to (1) determine the divergent and convergent validity of the BAI compared to an extended set of proxy self-report questionnaires designed to measure depression and anxiety symptoms in Korean clinical and non-clinical samples; (2) provide a comparison of BAI severity scores in anxiety disorder only, depressive disorder only, and comorbid anxiety and depression populations; and (3) reconsider the clinical utility of the BAI by determining cutoff scores and calculating specificity and sensitivity as well as positive and negative predictive power using Korean anxiety disorder base rates.

The present study illustrates the psychometric properties and diagnostic utility of the BAI in blind samples of clinical psychiatric and non-clinical Korean adult populations. The BAI demonstrated excellent internal reliability and significant difference in mean score between the anxiety disorder and non-clinical groups. However, the scale did not demonstrate adequate discriminant validity between anxiety and depressive disorders, nor diagnostic utility for anxiety disorders under DSM-IV.

Our correlational analyses and mean score comparisons showed that the BAI appears to possess good convergent validity with various anxiety measures such as the ASI-3, APPQ, and GAD-7, and discriminant validity for those with and without an anxiety disorder diagnosis. However, our data did not support divergent validity within samples for depressive disorders; the correlation of the BAI with depression-related scales were mostly found to be higher than with other anxiety-related scales in the Korean sample. Convergent validity with other anxiety scales was moderate to high, which is suitable considering that the various anxiety scales included in this study measure heterogeneous concepts of anxiety, such as worry, sensitivity to physiological anxiety, and fear responses to certain objects. ROC curve analysis revealed that none of the cutoff scores demonstrated adequate diagnostic sensitivity and specificity for screening participants with anxiety disorders from those without anxiety disorders, including both clinical and non-clinical participants.

The correlation of the BAI with depression scales was found to be slightly higher in our study than in previous literature (15, 61, 62). Whereas, the diagnostic criteria for depressive disorders are based on more homogenous and robust depressive symptomatology that span various depressive disorders differing in the duration, repetition, and severity of depressive episodes, anxiety disorders manifest considerable heterogeneity under DSM-5 categorization (29). For instance, severity and frequency in social anxiety disorder may be determined by the frequency of encountering threatening social contexts. Scales such as the ASI-3 isolate the social aspect of anxiety as a factor that may be more sensitive to social anxiety disorder than other disorders in the anxiety category in DSM-5. In contrast, a GAD patient with predominant and excessive worry may score high on worry-focused scales such as the PSWQ. The items in the BAI do not encapsulate these specificities, although it may successfully capture somatic or panic-related symptoms (12, 63).

Interestingly, in our Korean sample data, the depressive disorders only group showed higher BAI mean scores than the anxiety disorders only group. Previous studies have reported higher BAI mean scores for the anxiety disorders only group (27), or at least showed no significant difference between the two (29). We speculate that cross-cultural factors could have contributed to the high BAI mean score in the depressive disorder only group in our study. That is, because a majority of BAI items concern somatization (14 out of 21 items), it is possible that the Korean sample expressed more severe somatization symptoms. In fact, somaticizing depression is not a new concept in cross-cultural psychology (64). A number of Chinese studies indicate a tendency to deny depression and express it somatically, along with cultural terms such as neurasthenia and shenjing shuairuo (64, 65). A recent study in Korea that explores the factor structure of collapsed items of the BAI and BDI-II shows somatic anxiety and somatic depression as major factors of the two collapsed mood disorder assessments, and notes cognitive renderings of anxiety and depression as a main discriminant factor (66). Thus, the cultural prevalence of somaticized psychiatric symptoms may have obscured the discriminant validity of the BAI against the depressive disorder group.

None of the cutoff scores in our studies revealed acceptable diagnostic sensitivity and specificity. Whereas, the cutoff score of about 15 to 16 is suggested as an optimal score by the original study (27) as well as the meta-analysis of studies reporting cutoff scores (1), a cutoff score of 8 was found to be the most acceptable combination of sensitivity and specificity in our study. Considering that the maximum total BAI score is 63, this score may be too low to adequately distinguish between those with and without anxiety disorders. Positive Predictive Value (PPV) and Negative Predictive Value (NPV) illustrate the diagnostic utility of the BAI in a real setting: only two out of every 10 people identified with an anxiety disorder based on a cutoff score of 8 will actually have the disorder. However, such a low threshold may still have clinical utility in settings where detection of anxiety is of higher importance than actually having the disorder, such as in primary healthcare. Clinicians and researchers utilizing Korean samples may benefit from the cutoff score table, PPV, and NPV as references, selecting one according to their individualized settings and purpose for utilizing the tool.

An interesting speculation found in our study is that the somatization measured in BAI may contribute to the high BAI mean scores in Korean depression disorder patients. Since BAI is a criterion-based measure rather than a diagnosis-based to anxiety disorders, its utility could be highlighted in domain-based descriptions of symptomatology rather than in categorical diagnosis. We failed to find a cutoff score to distinguish anxiety disorder patients from healthy population with adequate sensitivity and specificity values. In addition, BAI is not an effective tool to discriminate patients with anxiety disorders and depressive disorders. It is therefore suggested to use a more sensitive tool in primary health care settings, due to the high possibility to incorrectly diagnose anxiety disorders.

Some limitations of the study should be noted. We aggregated the anxiety disorder group as a whole, despite the diagnostic variation across sub-categorical anxiety disorders. Future studies with larger sample sizes will allow for comparison of BAI mean scores among different types of anxiety disorders. Furthermore, considering that depression and anxiety disorders are highly comorbid, separating anxiety disorder only and depressive disorder only groups may lead to laboratory-bound results. This may have contributed to the lower BAI mean score in the anxiety disorder only group than was found in previous literature, which mainly studied anxiety disorders without separating other comorbidities. However, dividing patients into anxiety disorder only, depressive disorder only, comorbid, and general psychiatric groups may provide incremental validity by enhancing the accurate portrayal of mean score among pure anxiety and depression groups, as well as allowing important comparisons between diagnostic-based patient differences.

The main limitations of self-reported questionnaires include the risks of self-recalled bias and its inability to measure objective biological parameters. The utilization of neuropsychological tools such as the functional near-infrared spectroscopy (fNIRS), electroencephalography (EEG) or event-related potential (ERP) measures have arisen as new potentials in academic and clinical psychiatry (67–69). For instance, EEG studies focus on asymmetric hemispheric activity in depression and anxiety (68, 70), and frontal lobe abnormalities, particularly the decrease in bilateral frontotemporal oxygenation, have been found in unipolar depression using the verbal fluency test using fNIRS (71). The implementation of neurophysiological and neuropsychological instruments may shed a light on finding anxiety- or depression-specific biomarkers.

Another means to overcome the vulnerability to distortion or biases of retrospective retrievals of episodic memory is applying current mobile technology. It is possible to measure anxiety symptoms in a real-time using ecological momentary assessment (EMA) method, applying the measure several times a day. Moreover, to address the burden of answering 21 questions of BAI at each time point of application, computerized adaptive testing (CAT) technique could be one solution. For instance, Gibbons, Weiss (72) developed an anxiety tool “CAT-ANX” using this technique, which automatically selects and presents the most adequate questions from the item bank with 431 items. An average of 12 questions could measure the anxiety symptom that is highly correlated with the result with the entire 431 questions (r = 0.94). Coupled with a mobile device, CAT technique could serve as a useful means to realize real-time, longitudinal, easily distributable assessments of anxiety symptoms with relatively fewer demands on service receivers.

Since anxiety and depressive disorders are highly comorbid (73, 74), psychological or biological treatments share common targets and mechanisms such as high emotional avoidance, rumination (75), and suppression of negative emotions (76). Despite the common features, specific treatment recommendations have been made: treatments for depressive disorders include diverse psychotherapy [e.g., interpersonal psychotherapy (IPT), cognitive-behavioral therapy (CBT)], psychopharmacological treatment (e.g., SSRI, MAOIs), transcranial magnetic stimulation (77), electroconvulsive therapy (78) For treatments of anxiety disorders, exposure-based psychotherapies (79), benzodiazepine, SSRIs, and serotonin-norepinephrine reuptake inhibitors (SNRIs) are commonly used (80). In addition, in a future study, it should be investigated whether distinctive treatment recommendations after diagnosing anxiety or depressive disorders may produce differential outcomes.

Despite these limitations, the strength of this study is in its unbiased sample, which strictly followed a robust quality assessment method using the QUADAS-2 tool (81). Our study satisfied all four domains in the risk of bias criterion. In addition, this study is the first to examine the clinical utility and psychometric properties of anxiety disorders involving both clinical and non-clinical samples in Korea. The data sample we used is assumed to reflect a natural primary healthcare setting, which presents a mixture of clinical, sub-clinical and non-clinical individuals. Additionally, the results from ROC curve analysis are expected to serve as a valuable reference in selecting cutoff scores for both clinicians and researchers according to their particular research design or clinical needs. However, attentive consideration must be made before utilizing the BAI in decision-making or outcome measures, in light of its equivocal convergent and discriminant validity.

HO, YYC, S-HL, and K-HC contributed to the conception and design of the study. K-HC supervised the overall study process. KP and HO performed data analysis. HO wrote the first draft of the manuscript. KP, SY, YK, and HO contributed to acquisition of data. All authors contributed to manuscript revision, and have read and approved the submitted version.