Regional prevalence of generalized anxiety disorder in Germany: findings from the NAKO study

Abstract Background The German National Cohort (NAKO) offers a large population sample for estimating Generalized Anxiety Disorder (GAD) prevalence, but GAD-7-based estimates are limited by test accuracy. We applied a Bayesian framework to adjust for this and compared results with naive estimates using the conventional GAD-7 cutoff. Methods We analyzed data from the NAKO, a population-based study of adults aged 19-74 years, recruited between 2014 and 2019 across 18 centers in 16 German regions. Anxiety symptoms were assessed using the GAD-7. Bayesian modeling incorporated prior sensitivity (0.74, 95% CI: 0.61-0.84) and specificity (0.83, 95% CI: 0.68-0.92) estimates from a validation study. Bayesian-adjusted prevalence was compared with naive estimates based on the conventional GAD-7 cutoff (≥10). Design and calibration weights were applied to enhance regional representativeness. Results Of 204,783 NAKO participants, 15,799 (7.7%) were excluded due to missing GAD-7 responses. Bayesian modeling estimated a GAD prevalence of 1.2% (95% credible interval (CrI): 0.0-3.7%), significantly lower than the naive estimate of 5.2% (95% CI: 5.1-5.3%). Prevalence was higher among females (1.6% vs. 0.9% in males), non-native German speakers (2.2% vs. 1.1% in native speakers), and those with a history of anxiety (11.7% vs. 0.8% in those without). The prevalence of GAD decreased with age, with the highest estimates in young adults (19-29 years: 1.7%, 95% CrI: 0.1%-4.9%) and the lowest in older adults (70-79 years: 0.5%, 95% CrI: 0.0%-1.5%). Bayesian estimates were consistently lower than naive estimates across study centers (0.8-1.6% vs. 3.5-6.5%). Conclusions Estimating prevalence from screening scores without accounting for test accuracy can overestimate true prevalence and misallocate resources. Bayesian methods provide more robust estimates. Our findings highlight the importance of addressing diagnostic uncertainty to improve mental health surveillance and guide targeted interventions. Key messages • Bayesian methods yield more robust GAD prevalence estimates by accounting for GAD-7 screening tool limitations, improving mental health surveillance. • Ignoring test accuracy in screening-based prevalence estimates can significantly overestimate mental health disorder rates, risking resource misallocation.