Table 5.
Here, we test the cut-off Alzheimer’s disease (AD) score value of 0.5 in the NACC dataset by applying linear analysis of the relationship between AD score and cognitive scores using Bayesian piecewise linear regression analysis
| Variable breakpoint | |||||
|---|---|---|---|---|---|
| BP | Slope < BP (CI) | Slope > BP (CI) | Slope diff (CI) | ELPD diff (se) | |
| MMSE | 0.67 | −0.19 (−1.8, −0.94) | −1.4 (−1.8, −0.94) | −1.2 (−1.7, −0.65) | 0.0 (0.0) |
| MoCA | 0.6 | −0.18 (−0.35, 0.03) | −1.40 (−1.8, −1.0) | −1.2 (−1.8, −0.77) | 0.0 (0.0) |
| Backward digit span | 0.56 | −0.08 (−0.25, 0.08) | −0.51 (−0.96, −0.13) | −0.43 (−1.0, 0.10) | 0.0 (0.0) |
| Forward digit span | 0.59 | 0.09 (−0.08, 0.26) | −0.62 (−1.1, −0.2) | −0.71 (−1.3, −0.16) | 0.0 (0.0) |
| Semantic fluency | 0.59 | −0.22 (−0.34, −0.10) | −1.0 (−1.3, −0.74) | −0.8 (−0.18, −0.44) | 0.0 (0.0) |
| Trails B | 0.44 | 0.2 (0.079, 0.35) | 0.66 (0.45, 0.89) | 0.46 (0.13, 0.79) | 0.0 (0.0) |
| WAIS | 0.53 | −0.24 (−0.42, −0.07) | −0.59 (−0.97, −0.25) | −0.35 (−0.82, 0.13) | −0.1 (0.1) |
| Boston naming task | 0.66 | −0.08 (−0.23, 0.056) | −1.3 (−1.7, −0.81) | −1.2 (−1.70, −0.64) | 0.0 (0.0) |
| Fixed breakpoint (0.5) | |||||
|---|---|---|---|---|---|
| BP | Slope < BP (CI) | Slope > BP (CI) | Slope diff (CI) | ELPD diff (se) | |
| MMSE | [0.5] | −0.17 (−0.32, −0.02) | −1.12 (−1.4, −0.82) | −0.95 (−1.4, −0.53) | −1.3 (1.0) |
| MoCA | [0.5] | −0.16 (−0.34, 0.03) | −1.3 (−1.6, −1.0) | −1.1 (−1.6, −0.72) | −0.5 (0.6) |
| Backward digit span | [0.5] | −0.08 (−0.26, 0.09) | −0.47 (−0.81, −0.13) | −0.38 (−0.87, 0.10) | −0.3 (0.2) |
| Forward digit span | [0.5] | 0.10 (−0.07, 0.28) | −0.53 (−0.87, −0.2) | −0.63 (−1.1, −0.15) | −0.3 (0.3) |
| Semantic fluency | [0.5] | −0.21 (−0.34, −0.08) | −0.94 (−1.20, -0.72) | −0.74 (−1.1, −0.41) | −0.5 (0.4) |
| Trails B | [0.5] | 0.22 (0.08, 0.36) | 0.67 (0.46, 0.89) | 0.46 (0.13, 0.77) | 0.0 (0.2) |
| WAIS | [0.5] | −0.24 (−0.40, −0.07) | −0.58 (−0.87, −0.24) | −0.34 (−0.79, 0.13) | 0.0 (0.0) |
| Boston naming task | [0.5] | −0.07 (−0.23, 0.09) | −1.0 (−1.3, −0.7) | −0.94 (−1.4, −0.49) | −1.7 (0.8) |
| Linear model (no breakpoint) | ||
|---|---|---|
| Slope (CI) | ELPD diff (se) | |
| MMSE | −0.47 (−0.55, −0.39) | −10.0 (5.3) |
| MoCA | −0.60 (−0.69, −0.51) | −12.4 (6.8) |
| Backward digit span | −0.2 (−0.29, −0.12) | −0.5 (1.6) |
| Forward digit span | −0.1 (−0.18, −0.02) | −2.6 (2.6) |
| Semantic fluency | −0.46 (−0.52, −0.40) | −9.6 (4.5) |
| Trails B | 0.39 (0.36, 0.45) | −2.7 (2.8) |
| WAIS | −0.35 (−0.43, −0.26) | −0.3 (1.4) |
| Boston naming task | −0.37 (−0.45, −0.29) | −9.6 (4.7) |
For the slope estimates, we include the 95% Credible Interval (CI). Given that the data are z-scored, we use 0.1 as the Region of Practical Equivalence (ROPE) which represent 0.1 of the standard deviation of the data. If the CI lies outside −0.1 to 0.1, we consider there is good evidence of a relationship between the AD score and clinic score, indicated by values in bold. For models with variable breakpoints, the breakpoint values obtained were similar to 0.5, and comparing models with the Expected Log Pointwise Predicted Density (ELPD) the difference between variable and fixed breakpoint models was negligible, except for the Boston naming task. There was good evidence for a breakpoint in MMSE, MoCA, forward digit span, semantic fluency, and the Boston naming task, and further evidence supporting no difference between the two breakpoint models; however, both being superior to the non-breakpoint model. These findings support our use of a breakpoint of 0.5.
BP breakpoint, MMSE Mini Mental State Examination, MoCA Montreal Cognitive Assessment, WAIS Wechsler Adult Intelligence Scale.