Fourteen studies that reported 17 analyses of the diagnostic performance of GDS were included in the review. Seven studies (n=3,012 participants) assessed GDS30 and 10 studies (n=1,762) assessed GDS15. Based on psychiatric interview, overall prevalence of depression was 17.1%. Six diagnostic interview-based studies that assessed general practitioner ability to detect late-life depression were included.
The pooled estimate of sensitivity for GDS30 was 77.4% (95% CI 66.3% to 86.8%) and the corresponding pooled estimate of specificity was 65.4% (95% CI 44.2% to 83.8%). The pooled estimate of sensitivity for GDS15 was 81.3% (95% CI 77.2% to 85.2%) and the corresponding pooled estimate of specificity was 78.4% (95% CI 71.2% to 84.8%). The positive clinical utility of both the GDS30 and GDS15 was rated as poor. However, the negative clinical utility of GDS15 was rated as good and that of GDS30 was rated adequate. Bayesian analysis suggested that GDS15 was the optimal test.
The pooled estimate of sensitivity for unassisted general practitioner diagnosis of depression was 56.3% (95% CI 40.0% to 72.0%) and specificity was 73.6% (95% CI 71.7% to 75.5%).
When detection of depression using GDS was compared with general practitioners' ability to diagnose late-life depression at a prevalence of 15%, GDS30 had no added benefit and GDS15 identified four additional cases per 100 and ruled-out an additional four non-cases per 100.