Prevalence Estimation Using a Depression Screening Tool in the National Health and Nutrition Examination Survey: Comparison of Different Cutoffs

Abstract

Objectives

The National Health and Nutrition Examination Survey (NHANES) in the US relies on the depression screening tool PHQ-9 to assess depressive symptoms in the general population. For prevalence estimation, PHQ-9s imperfect diagnostic accuracy can be modeled with a Bayesian Latent Class Model. We investigate the impact of different cutoffs on prevalence estimation.

Methods

We used data from the 16-th wave of the National Health and Nutrition Examination Survey (NHANES). We assessed the joint posterior distribution to asssess the prevalence of major depression as well as sensitivity and specificity of the PHQ-9 at cutoffs 5 to 15. We also assessed the impact of weakly and strongly informative prevalence priors.

Results

Data from 9693 participants of the NHANES Wave 2019–2020 were analyzed. Under weakly informative prevalence priors, prevalence estimates ranged from 16.0% (95% CrI: 0.3%–87.8%) when using a cut-off of 5% to 3.9% (0.2%–12.7%) at 13. More informative prevalence priors led to narrower credible intervals, but the observed data was still in accordance with a wide range of possible MDD prevalence estimates.

Conclusions

Regardless of the cutoff and the prevalence prior chosen, prevalence estimation of major depressive disorders in the NHANES based on the PHQ-9 is imprecise.