Fomenko Alexey, Dümmler Daniel, Aktürk Zekeriya, Eck Stefanie, Teusen Clara, Karapetyan Siranush, Dawson Sarah, Löwe Bernd, Hapfelmeier Alexander, Linde Klaus, Schneider Antonius
Institute of General Practice and Health Services Research, TUM School of Medicine and Health, Department Clinical Medicine, Technical University of Munich, Munich, Germany.
Institute of Epidemiology and Social Medicine, University of Münster, Münster, Germany.
Cochrane Database Syst Rev. 2025 Jul 2;7(7):CD015456. doi: 10.1002/14651858.CD015456.
Despite being highly prevalent mental health conditions, anxiety disorders frequently go undiagnosed, prompting the use of questionnaires for anxiety screening as a potential solution. This review summarises the test accuracy of the Hospital Anxiety and Depression Scale Anxiety subscale (HADS-A) for screening purposes.
To assess the test accuracy of the HADS-A in screening for any anxiety disorder (AAD), generalised anxiety disorder (GAD) and panic disorder in adults, and to investigate how the test accuracy varies by sources of heterogeneity and across all cutoffs.
We searched Embase, MEDLINE, PubMed-not-MEDLINE subset and PsycINFO from 1990 to 10 July 2024. We checked the reference lists of included studies and review articles.
We included studies in adults in which the HADS-A was administered cross-sectionally alongside structured or semi-structured clinical interviews, allowing the creation of 2x2 tables. We excluded case-control studies, studies with a time gap exceeding four weeks between administering the HADS-A and the reference standard, and studies with diagnostic criteria based on the Diagnostic and Statistical Manual of Mental Disorders Third Edition or earlier versions. We also excluded studies involving people who were recruited based on mental health symptoms.
At least two review authors independently decided on the eligibility of the articles, extracted data, and assessed the methodological quality of the included studies using Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). For each target condition, we present the sensitivity and specificity of each study along with 95% confidence intervals (CIs). For the primary analyses, we used bivariate models to obtain summary estimates for the recommended HADS-A cutoff score of 8 or higher (≥ 8); if the bivariate models did not converge, we used multiple thresholds models. For the secondary analyses, we obtained summary estimates for all cutoffs using bivariate and multiple thresholds models. From the multiple thresholds model, we derived the summary estimates of all available cutoffs from the summary receiver operating characteristic (SROC) curve and the area under the curve (AUC) as a measure of overall accuracy. We explored sources of heterogeneity using meta-regression models.
We identified 67 studies, encompassing data from 18,467 participants that were available for the analyses. Fifty-four studies contributed to the analyses of HADS-A for detecting AAD, 35 for GAD, and 10 for panic disorder. The median prevalence of AAD, GAD and panic disorder was 17%, 7% and 6%, respectively. The included studies showed a wide spectrum of clinical and methodological differences. We considered the overall risk of bias to be low in 19 studies. The most frequent limitations were related to non-consecutive patient selection and to post-hoc cutoff determination. The applicability was of low concern across three domains in nine studies. The main limitations of applicability were the presence of prediagnosed anxiety (prior to undergoing HADS-A) or the fact that this information was not collected or reported. The estimates of both sensitivity and specificity varied strongly between studies. With regard to the recommended cutoff ≥ 8, the HADS-A subscale demonstrated a summary sensitivity of 0.74 (95% CI 0.70 to 0.78) and a summary specificity of 0.76 (95% CI 0.73 to 0.79) for detecting AAD; a summary sensitivity of 0.82 (95% CI 0.76 to 0.87) and a summary specificity of 0.74 (95% CI 0.70 to 0.77) for detecting GAD; and a summary sensitivity of 0.80 (95% CI 0.69 to 0.88) and a summary specificity of 0.66 (95% CI 0.55 to 0.76) for detecting panic disorder. Results from the multiple thresholds model showed an AUC of 0.81 (95% CI 0.79 to 0.82) for detecting AAD, 0.82 (95% CI 0.80 to 0.84) for GAD and 0.81 (95% CI 0.77 to 0.85) for panic disorder. The observed heterogeneity remained largely unexplained, except for the investigations of heterogeneity with regard to GAD, which showed that the setting had a significant impact on specificity; and prevalence and the reference standard had a significant impact on sensitivity. With respect to panic disorder, a formal heterogeneity assessment was not feasible.
AUTHORS' CONCLUSIONS: The use of the HADS-A for screening purposes with a cutoff ≥ 8 in an exemplary cohort of 1000 individuals with an AAD prevalence of 17% would result in 675 individuals testing negative, of whom 44 would be false negatives, while 325 would test positive. Of these, 199 would be false positives, potentially straining the available healthcare resources. However, caution is warranted in interpreting the review findings, as the strength of evidence was limited by the risk of bias, concerns regarding applicability and substantial, unexplained heterogeneity. The use of estimates derived from clinical populations in which HADS-A is applied would be a reasonable approach. However, subgrouping by clinical population is currently unfeasible due to the limited number of studies per population category. This represents an area of further exploration in future research. The unexplained heterogeneity makes it challenging to reliably predict the results of future studies. Given these limitations, the universal use of the HADS-A with a cutoff ≥ 8 for screening in different settings and populations is currently questionable.
尽管焦虑症是高度常见的心理健康状况，但往往未被诊断出来，这促使人们使用焦虑筛查问卷作为一种潜在的解决办法。本综述总结了医院焦虑抑郁量表焦虑分量表（HADS - A）用于筛查目的的测试准确性。
评估HADS - A在筛查成人的任何焦虑症（AAD）、广泛性焦虑症（GAD）和惊恐障碍方面的测试准确性，并研究测试准确性如何因异质性来源和所有临界值而异。
我们检索了1990年至2024年7月10日期间的Embase、MEDLINE、PubMed非MEDLINE子集和PsycINFO。我们检查了纳入研究和综述文章的参考文献列表。
我们纳入了针对成年人的研究，其中HADS - A与结构化或半结构化临床访谈同时进行横断面施测，从而能够创建2×2表格。我们排除了病例对照研究、在施测HADS - A和参考标准之间时间间隔超过四周的研究，以及基于《精神疾病诊断与统计手册》第三版或更早版本的诊断标准的研究。我们还排除了涉及基于心理健康症状招募的人群的研究。
至少两名综述作者独立决定文章的合格性，提取数据，并使用诊断准确性研究质量评估（QUADAS - 2）评估纳入研究的方法学质量。对于每个目标状况，我们呈现每项研究的敏感性和特异性以及95%置信区间（CI）。对于主要分析，我们使用双变量模型获得推荐的HADS - A临界值分数为8或更高（≥8）的汇总估计值；如果双变量模型未收敛，我们使用多阈值模型。对于次要分析，我们使用双变量和多阈值模型获得所有临界值的汇总估计值。从多阈值模型中，我们从汇总接受者操作特征（SROC）曲线和曲线下面积（AUC）得出所有可用临界值的汇总估计值，作为总体准确性的一种度量。我们使用元回归模型探索异质性来源。
我们识别出67项研究，涵盖来自18467名参与者的数据，这些数据可用于分析。54项研究为检测AAD的HADS - A分析做出了贡献，35项用于GAD分析，10项用于惊恐障碍分析。AAD、GAD和惊恐障碍的中位患病率分别为17%、7%和6%。纳入的研究显示出广泛的临床和方法学差异。我们认为19项研究的总体偏倚风险较低。最常见的局限性与非连续患者选择和事后临界值确定有关。在9项研究中，适用性在三个领域的关注度较低。适用性的主要局限性是存在预先诊断的焦虑（在进行HADS - A之前），或者未收集或报告此信息这一事实。研究之间敏感性和特异性的估计值差异很大。关于推荐的临界值≥8，HADS - A分量表在检测AAD时的汇总敏感性为0.74（95%CI 0.70至0.78），汇总特异性为0.76（95%CI 0.73至0.79）；在检测GAD时的汇总敏感性为0.82（95%CI 0.76至0.87），汇总特异性为0.74（95%CI 0.70至0.77）；在检测惊恐障碍时的汇总敏感性为0.80（95%CI 0.69至0.88），汇总特异性为0.66（95%CI 0.55至0.76）。多阈值模型的结果显示，检测AAD时的AUC为0.81（95%CI 0.79至0.82），检测GAD时为0.82（95%CI 0.80至0.84），检测惊恐障碍时为0.81（95%CI 0.77至0.85）。观察到的异质性在很大程度上仍无法解释，除了关于GAD的异质性研究表明，研究背景对特异性有显著影响；患病率和参考标准对敏感性有显著影响。关于惊恐障碍，进行正式的异质性评估不可行。
在一个AAD患病率为17%的1000人典型队列中，使用临界值≥8的HADS - A进行筛查，将导致675人检测为阴性，其中44人将为假阴性，而325人将检测为阳性。在这些阳性者中，199人将为假阳性，这可能会使可用的医疗资源紧张。然而，在解释本综述结果时应谨慎，因为证据的强度受到偏倚风险、对适用性的担忧以及大量无法解释的异质性的限制。使用从应用HADS - A的临床人群得出的估计值将是一种合理的方法。然而，由于每个群体类别的研究数量有限，目前按临床人群进行亚组分析不可行。这是未来研究中有待进一步探索的领域。无法解释的异质性使得可靠预测未来研究结果具有挑战性。鉴于这些局限性，目前在不同背景和人群中普遍使用临界值≥8的HADS - A进行筛查存在疑问。
