Thirty-six studies (43 data sets) were included in the review. The total number of participants was unclear. Twenty-one studies assessed GDS30, 12 assessed GDS15 and three assessed GDS4/5. The overall prevalence of late-life depression was 29.2%. Most of the included studies used an acceptable reference standard, avoided verification biases and delays between index test and reference standard and interpreted tests with access to appropriate clinical information. None of the studies reported uninterpretable results. Avoidance of incorporation bias was unclear for all studies. Blinding of test interpretation was frequently unreported. Almost half of the included studies did not recruit a representative patient spectrum.
GDS30: Overall sensitivity was 81.9% (95% CI 76.47% to 86.9%) and specificity was 77.7% (95% CI 73.0% to 82.1%). There were no significant differences in test performance between study settings or when analyses were restricted to patients with cognitive impairment or with major depression.
GDS15: Overall sensitivity was 84.3% (95% CI 79.7% to 88.4%) and specificity was 73.8% (95% CI 68.0% to 79.2%). There was no significant difference in test performance when the analysis was restricted to patients with cognitive impairment, but specificity was significantly lower in outpatients with major depression. In in-patient settings, sensitivity was 32.2% (95% CI 13.3% to 54.7%) and specificity was 69.0% (95% CI 55.4% to 81.2%). In nursing home settings, sensitivity was 86.6% (95% CI 76.1% to 94.4%) and specificity was 72.3% (95% CI 50.6% to 89.6%).
GDS4/5: Overall sensitivity was 92.5% (95% CI 85.5% to 97.4%) and specificity was 77.2% (95% CI 66.6% to 86.3%). There were insufficient data for subgroup analyses.
Bayesian analysis found no significant difference in screening or case-finding ability between GDS30 and GDS15. There was a trend towards superior performance of GDS15 in nursing home settings.