Abstract
PURPOSE: Exploratory eye movement (EEM) parameters, such as the number of eye fixations (NEF), the responsive search score (RSS), and the discriminant index (D score), serve as reliable biomarkers unaffected by medication or disease progression. However, the diagnostic accuracy of these parameters remains uncertain due to inconsistent thresholds and limited validation in small-sample studies. This study aimed to explore reliable diagnostic thresholds for NEF, RSS and D score values for clinical use. METHODS: Eight databases were used to identify relevant studies with tailored search strategies. Two independent reviewers screened all retrieved literatures according to pre-defined inclusion and exclusion criteria. Key information of included studies was extracted in standardized tables. Diagnostic Accuracy-2 instrument (QUADAS-2) was used to assess quality of included studies. Two-by-two tables were created and were used to recompute the sensitivity and specificity. The cut-off for the D score was pre-specified based on prior studies, sensitivity and specificity were calculated for all indicators, with heterogeneity assessed by Cochrane Q test. For the RSS and NEF, logistic regression was used to model the diagnostic prediction probability of schizophrenia. ROC curves were constructed by evaluating all possible thresholds, and Youden's Index was employed to determine optimal cut-offs. RESULTS: The D score was analysed on the basis of 1895 samples from 9 articles and achieved an overall accuracy of 90%, with a sensitivity of 79% and a specificity of 87%. On the basis of the 3158 predictive values from 21 articles, the NEF was shown to achieve the highest diagnostic rate of 70.1% when the threshold was set at 28.7, with a sensitivity of 63.38% and a specificity of 66.88%. On the basis of 3408 predictive values from 23 articles, the RSS was shown to achieve the highest diagnostic rate of 75.36% when the threshold was set at 8.05, with a sensitivity of 63.7% and a specificity of 73.2%. CONCLUSION: The study revealed that the D score had relatively high diagnostic power, while, the NEF and RSS had moderate overall diagnostic capacities at optimal cut-off points. However, reliability and generalizability of these result were comprised by using aggregated data for analysis in this study, and adoption of these result in clinical decision-making process should be cautious.