Comparison of the accuracy of latent factor and sum scoring of the hospital anxiety and depression scale to screen for major depression: An individual participant data meta-analysis

BACKGROUND: Latent factor scoring may provide more precise score estimates than sum scores, but this has not been evaluated for the Hospital Anxiety and Depression Scale (HADS). We investigated whether latent factor scores could improve HADS depression screening accuracy. METHODS: We used a HADS screening accuracy individual participant data meta-analysis (IPDMA) database. We included 42 studies (7982 participants; 12 to 1143 per study) with a semi-structured interview reference standard. We randomly split the database into calibration and validation datasets. In calibration, we estimated latent scores using one-factor models (14-item HADS total scale [HADS-T], 7-item depression subscale [HADS-D]) plus HADS-T two-factor and bi-factor (general factor and two specific factors) models. We estimated cut-offs that maximized combined sensitivity and specificity for each method. In validation, we compared screening accuracy between latent variable approaches and the HADS-D sum score. The process was repeated 1000 times to estimate 95% confidence intervals for parameters. RESULTS: After removing iterations with failed models in confirmatory factor analysis (N = 304) or IPDMA (N = 31), aggregated results showed that confidence intervals for sensitivity, specificity, and combined sensitivity and specificity included 0 for all comparisons between factor scores and sum scores. Statistically significant but minimal advantages appeared in the receiver operating characteristic curve for the two-factor and bi-factor models (0.01, 95% CI [0.01, 0.02]; 0.02, 95% CI [0.01, 0.02]). Sensitivity analysis confirmed findings. CONCLUSIONS: Latent factor scoring did not meaningfully improve HADS screening accuracy compared with sum scores. Sum scores may be preferred in applied settings for their simplicity and feasibility.