Beck Anxiety Inventory (BAI) for detecting anxiety disorders in adults

Despite being highly prevalent mental health conditions, anxiety disorders frequently go undiagnosed. The Beck Anxiety Inventory (BAI) is a widely used self-report scale for measuring the severity of anxiety, which has also been used for anxiety screening.

To assess the test accuracy of the Beck Anxiety Inventory (BAI) in adults against structured or semi-structured diagnostic interviews for the following target conditions: any anxiety disorder (AAD), generalised anxiety disorder (GAD), and panic disorder (PD).

- To explore sources of heterogeneity according to setting, the prevalence of anxiety disorders, the reference standard and the risk of bias. - To investigate how test accuracy changes with the test threshold.

We searched Embase, MEDLINE, PubMed-not-MEDLINE subset and PsycINFO from 1990 to 12 July 2024. We checked the reference lists of included studies and review articles. We searched for errata or retractions of included studies via PubMed or Embase. We also checked relevant journal publishers' websites and conducted a search on the Retraction Watch database (retractiondatabase.org).

Studies that included adults presenting with reasons unrelated to mental distress were eligible for this review. These studies involved the administration of the BAI alongside structured or semi-structured diagnostic interviews, allowing for the generation of 2 x 2 tables. We excluded case-control studies and studies with a time gap exceeding four weeks between administering the index test and the reference standard.

At least two reviewers independently decided on the eligibility of the articles, extracted the data and assessed the methodological quality of the included studies using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. For each target condition, we present the sensitivity and specificity of each study along with 95% confidence intervals (CIs). We used the multiple thresholds model to obtain the summary test accuracy estimates of all available cut-offs.

We identified 14 studies encompassing data from 6232 participants that were available for the analyses. Ten studies contributed to the analysis of the BAI for detecting AAD, eight for GAD, and four for PD. The median prevalence of AAD, GAD and PD was 0.27, 0.12 and 0.10, respectively. Nine studies were conducted in diverse specialised clinical settings, two in non-clinical, and three in mixed settings. A considerable number of studies showed a high or unclear risk of bias in the domains of patient selection (n = 7), index test (n = 8) and flow and timing (n = 8). A major applicability concern was the presence of prediagnosed anxiety prior to undergoing BAI or the fact that this information may not have been collected. According to the multiple thresholds model, the BAI had a summary sensitivity of 0.54 (95% CI 0.43 to 0.64) and a specificity of 0.87 (95% CI 0.78 to 0.92) in detecting AAD at the cut-off ≥ 16. The summary sensitivity was 0.72 (95% CI 0.65 to 0.78) and the specificity 0.80 (95% CI 0.71 to 0.87) for detecting GAD, and the summary sensitivity was 0.72 (95% CI 0.50 to 0.87) and specificity was 0.77 (95% CI 0.55 to 0.90) for detecting PD. The area under the multiple thresholds summary receiver operating characteristic (mtsROC) curve (AUC) was 0.76 (95% CI 0.72 to 0.80) for detecting AAD, 0.83 (95% CI 0.80 to 0.86) for GAD and 0.80 (95% CI 0.74 to 0.87) for PD. A formal investigation of heterogeneity was not feasible due to the limited number of included studies.

The conclusions that can be drawn from this review are restricted due to the clinical heterogeneity, and limited quality and quantity of the included studies. Furthermore, the sources of heterogeneity remain incompletely understood. Given these limitations and the existence of shorter developed questionnaires (specifically for screening purposes), the utility of the BAI for detecting anxiety disorders is currently uncertain.

This Cochrane Review was funded by the German Federal Ministry of Education and Research (Grant Number: 01KG2105).

Protocol (2022): doi.org/10.1002/14651858.CD015292.