The association of sensitivity and specificity with disease prevalence: Analysis of 5,925 diagnostic test accuracy studies

Date & Time
Wednesday, September 6, 2023, 12:30 PM - 2:00 PM
Location Name
Session Type
Statistical methods
Murad MH1, Lin L2, Chu H3, Hasan B1, Alsibai R1, Abbas A1, Mustafa R4, Wang Z1
1Mayo Clinic, USA
2University of Arizona, Tucson, AZ, USA
3Statistical Research and Data Science Center, New York, USA, USA
4University of Kansas Health System, USA

Background: Sensitivity and specificity are characteristics of a diagnostic test and are not expected to change as the prevalence of the target condition changes.
Objectives: To empirically determine if sensitivity and specificity are associated with prevalence.
Methods: We retrieved data from diagnostic test accuracy meta-analyses published in the Cochrane Database of Systematic Reviews between January 2003 and January 2020. The association of sensitivity and specificity with prevalence was evaluated using a mixed-effects random intercept linear regression model to jointly evaluate the association between prevalence and log-transformed sensitivity and specificity. The model evaluated all meta-analyses as nested within each systematic review.
Results: We analyzed 5,925 diagnostic test accuracy studies included in 525 meta-analyses. Compared with the lowest quartile of prevalence, the second, third, and fourth quartiles had increasing average sensitivities by 1.99% (-0.25%, 4.28%), 3.56% (1.34%, 5.82%), and 3.96% (1.63%, 6.36%), respectively. The corresponding reduction in average specificities were -2.38% (-4.10%, -0.63%), -3.91% (-5.56%, -2.23%), and -8.36% (-10.02%, -6.67%), respectively. Results are summarized in Table 1.
Conclusions: In this large sample of diagnostic studies, higher prevalence was associated with higher sensitivity and lower specificity. Clinicians should consider the implications of disease prevalence and disease spectrum when interpreting the results of diagnostic test accuracy studies. Systematic reviewers should explore the effect of prevalence on test accuracy measures or use trivariate models that condition test accuracy measures on prevalence. Patient, public, and/or healthcare consumer involvement: Patients/healthcare consumers were not involved in this research.