Table 3. Association of CERAD (binary logistic regression analysis) with the classification of 113 controls and 49 MCI patients (CERAD composite score).

CERAD Subtest	B(SE)	Odds ratio (95% CI)	Sign.
Z verbal Fluency Animals	−0.644 (0.311)	0.5 (0.3 - 0.97)	0.038
Z Wordlist total	−2.222 (0.536)	0.1 (0.04 - 0.3)	3.4E-5
Z WL savings	−1.132 (0.359)	0.3 (0.2 - 0.7)	0.002
Z WL recognition	−0.790 (0.320)	0.5(0.2 - 0.9)	0.014
Z Figure recall	−1.125 (0.271)	0.3 (0.2 - 0.6)	3.2E-5

Stepwise logistic regression analysis (backward elimination, exclusion criterion: p ≥ 0.05; inclusion criterion: p < 0.05; cut-off value = 0.5) comparing 113 controls (MMSE ≥ 26) and 49 MCI patients (MMSE ≥ 26) was done on all 13 z -transformed CERAD subtests (excluding MMSE). Five subtests remained in the equation and performed much better (accuracy 89.5%) than the MMSE test (accuracy 74.1%); R² = 0.632 (Hosmer&Lemeshow), R² = 0.539 (Cox&Snell), R² = 0.763 (Nagelkerke), Model chi-squared = 125.53 and p = 2.1 E-25). The resulting CERAD composite score (0 to 1) is represented by the probability P for diagnosis of MCI: P(MCI)=1/(1+e^(− (− 1.718 - 0.644 * Z verbal Fluency Animals - 2.222 * Z Wordlist total - 1.132 * Z WL savings - 0.790 * Z WL recognition - 1.125 * Z Figure recall))), P(MCI) > 0.5 implies MCI classification.

Abbreviations: B…Beta value, SE…standard error, CI… confidence interval, Sign. … significance