LAUSR

Background: Cerebrospinal fluid (CSF) biomarkers are used to diagnose Alzheimer disease (AD), especially in atypical clinical presentations. No consensus currently exists regarding cut-off values. This study aimed, firstly, to define optimal cut-off values for CSF biomarkers, and secondly, to investigate the most relevant diagnostic strategy for AD based on CSF biomarker combinations. Methods: A total of 380 patients were prospectively included: 140 with AD, 240 with various neurological diagnoses (non-AD). CSF biomarkers were measured using ELISA. Univariate and multivariate analyses were performed using random forest and logistic regression approaches. Results: Univariate receiver operating curve curves analysis of T-Tau, P-Tau181, Aβ42, Aβ40 concentrations, and Aβ42/Aβ40 ratio levels showed AD cut-off values of ≥355, ≥57, ≤706, ≥10,854, and ≤0.059 ng/L, respectively. Multivariate analysis using random forest and logistic regression found that the algorithm based on P-Tau181, Aβ42 concentrations and Aβ42/Aβ40 ratio yielded the best discrimination between AD and non-AD populations. The cross-validation technique of the final model showed a mean accuracy of 0.85 and a mean AUC of 0.89. Conclusion: This study confirms that the Aβ42/Aβ40 ratio was more useful than the Aβ40 concentration in discriminating AD from non-AD populations in daily practice. These results indicate that the Aβ42/Aβ40 ratio should be assessed in all cases, independently of Aβ42 concentrations.