A total of 631 subjects were enrolled. Three patients did not complete the neuropsychological test, 71 exhibited depression, 14 exhibited vascular cognitive impairment (VCI), 14 exhibited vascular dementia (VaD), 50 exhibited mixed dementia and other types of dementia, while the remaining 249 were classified as NC, 134 were diagnosed with aMCI, and 97 were diagnosed with AD. The ‘all types of dementia’ group (n = 161) included the AD only group (n = 97), the VaD only group (n = 14) and the mixed dementia group (n = 50). The diagnostic algorithm for AD and aMCI and NC is shown in Figure
The demographic and neuropsychological characteristics of the different diagnostic categories are shown in Table
. There were significant differences between the four groups. Subjects in the all types of dementia group and AD group were significantly older than those in the aMCI group (P = 0.000 and P = 0.000 for all types of dementia and AD groups, respectively), and had fewer years of education (P = 0.000 and P = 0.000 for the all types of dementia and AD groups, respectively). In addition, the aMCI group was significantly older than the NC group (P = 0.000) and had fewer years of education (P = 0.000).
Demographic data and scores on neuropsychological tests
The results revealed a significant difference between the four groups with respect to MMSE scores, HVLT total learning, HVLT recognition and HVLT total score (P = 0.000, P = 0.000). The all types of dementia group and AD group exhibited lower performance than the aMCI group in total learning (P = 0.000, P = 0.000, respectively), recognition (P = 0.000, P = 0.000, respectively), total score (P = 0.000, P = 0.000, respectively) and MMSE score (P = 0.000, P = 0.000, respectively) (see also Table
), and aMCI patients exhibited poorer performance than the NC group regarding to all of the above tests. The ratio of HVLT total score / MMSE score exhibited a significant difference between the NC, MCI, AD and all types of dementia groups (P = 0.000). However, we found no significant differences between the AD group and the all types of dementia group in the HVLT total/MMSE ratio (P = 0.56). (P = 0.56).
There were no significant differences between the all types of dementia group and the AD group regarding age, sex, education, race, MMSE, and the HVLT total learning score, total scores (P > 0.05).
The ROC curves were produced by plotting the sensitivity against the 1-specificity for each score on the HVLT total learning, recognition score and total score in discriminating between cases of dementia versus NC, between AD cases versus NC and between cases of aMCI versus the NC group.
Impact of demographic factors on HVLT
A multiple linear regression analysis was conducted with the HVLT total learning score as the dependent variable and age and years of education as independent variables. The results revealed significant effects (F = 54.607, P = 0.000), accounting for 38.7% of HVLT total learning scores, age (standardized coefficients = −0.253, P = 0.000) and years of education (standardized coefficients = 0.273, P = 0.000) impacted on HVLT total learning score.
NC versus aMCI
shows the sensitivity and specificity for distinguishing the aMCI group from the NC group, for total learning, recognition score and total HVLT, with different cut-off scores. The area under the curve (AUC) was 0.774 for the HVLT total learning score, and 0.79 for the HVLT total score. When the cutoff score was 21.5 in the HVLT total learning, an optimal balance was obtained between the sensitivity and specificity (69.1% and 70.7%, respectively) in distinguishing aMCI and NC. With a cut-off value of 32.5, the HVLT total score exhibited sensitivity and specificity of 68.7% and 70.7%, respectively.
ROC curve of the HVLT discriminate amnestic mild cognitive impairment from normal cognition. Notes: HVLT = Hopkins verbal learning test, ROC = Receiver operating characteristic.
The AUC was 0.666 for the HVLT recognition score, 0.589 for sensitivity, and 0.699 for specificity, when the cutoff score was 11.5. The sensitivity and specificity for HVLT recognition score were relatively low, indicating that this was not an ideal tool for discriminating aMCI from NC in our sample.
Because the HVLT total learning score was associated with age, the cut-off score was calculated in different age groups. In 50 to 64-year-old subjects, when the cut-off score was 23.5, optimal sensitivity (0.700) and specificity (0.718) values were achieved in discriminating the NC and aMCI groups. In addition, optimal sensitivity (0.776) and specificity (0.562) values were obtained when the cutoff score was 18.5 in the 65–80 year-old patients.
NC versus AD and all types of dementia
Optimum sensitivity and specificity of HVLT total learning score, HVLT recognition score and HVLT total score for discriminating AD and all types of dementia from NC were determined by the ROC curve analysis using the most appropriate cut-off scores. The total AUC for total learning score was 0.976 for discriminating AD, and 0.982 for detecting all types of dementia. The total AUC for the HVLT total score was 0.982 for discriminating AD, and 0.986 for detecting all types of dementia. The total learning score exhibited optimal sensitivity and specificity using a cut-off score of 15.5 for detecting NC and AD and all types of dementia, with similar sensitivity for detecting AD and all types of dementia (94.7% for both), and specificity of 92.5% for AD, and 93.4% for all types of dementia. The HVLT total score exhibited better overall discrimination of AD and all types of dementia patients, using a cut-off score of 25.5, with 93.5% specificity and 95.5% sensitivity for detecting AD, and with 93.9% specificity and 96.0% sensitivity when the optimal cutoff score was 26.5 for detecting all types of dementia.
In the 50–64 years age-group optimal values for sensitivity (0.955) and specificity (0.921) were obtained for discriminating the NC and AD groups when the cutoff score was 18.5. Optimal values for sensitivity (0.948) and specificity (0.925) were obtained with a cutoff score of 14.5 in the 65–80 years age-group for discriminating the NC and AD groups.
The HVLT recognition scores for discriminating all types of dementia and AD were also calculated. The AUC was 0.96 for detecting AD, and 0.935 for detecting all types of dementia.
The sensitivity, specificity, PPV and NPV of HVLT total learning cut-off scores for different prevalence rates of all types of dementia are shown in Table
. In our sample, the prevalence of all types of dementia was 39.19%, and when the cut-off score the HVLT total learning score was 15.5, the PPV was 0.90, and NPV was 0.95. The PPV was impacted by the baseline prevalence, with PPV increasing when prevalence increased.
Sensitivity and specificity of HVLT total learning scores to discriminate all types of dementia and probabilities of dementia (PPV) and probabilities of no dementia (NPV) at different baseline rates