Purpose To establish a prediction model for premature ovarian insu ciency (POI) identi cation.Methods A total of 78 women with spontaneous POI and 48 healthy women were recruited from the A liated Shenzhen Maternity & Child Healthcare Hospital in the study. Retinal characteristics were analyzed using an automated retinal image analysis system. Binary logistic regression was used to identify POI cases and develop predictive models.Results Compared to the normal group, the POI group had larger central retinal artery equivalent (CRAE) (P=0.006), central retinal vein equivalent (CRVE) (P=0.001), index of venules asymmetry (Vasym) (P=0.000); larger bifurcation angles of arterioles (Aangle) (P=0.001), bifurcation coe cient of venule (BCV) (P=0.001) and more obvious arteriovenous nipping (Nipping) (P=0.005), but lower arteriole-tovenule ratio (AVR) (P=0.012). In the POI group, the odds ratio (OR) of Vasym was 6.72e-32 (95% C.I. 4.62e-49-9.79e-15, P=0.000), the OR of BCV was 5.66e-20 (95% C.I. 1.93e-34-.0000, P=5.66e-20) and the OR of Nipping was 6.65e-06 (95% C.I. 6.33e-10-.0698, P=0.012). Moreover, the area under the ROC curve for the binary logistic regression with retinal characteristics was 0.8582, and the tting degree of regression models was 60.48% (Prob> chi-square = 0.6048).
ConclusionThis study demonstrated that retinal image analysis can provide useful information for POI identi cation and certain characteristics may help with early clinical diagnosis of POI.