Consistency between Treatment Effects on Clinical and Brain Atrophy Outcomes in Alzheimer’s Disease Trials

Background Longitudinal changes in volumetric MRI outcome measures have been shown to correlate well with longitudinal changes in clinical instruments and have been widely used as biomarker outcomes in clinical trials for Alzheimer’s disease (AD). While instances of discordant findings have been noted in some trials, especially the recent amyloid-removing therapies, the overall relationship between treatment effects on brain atrophy and clinical outcomes, and how it might depend on treatment target or mechanism, clinical instrument or imaging variable is not yet clear. Objective To systematically assess the consistency and therapeutic class-dependence of treatment effects on clinical outcomes and on brain atrophy in published reports of clinical trials conducted in mild cognitive impairment (MCI) and/or AD. Design Quantitative review of the published literature. The consistency of treatment effects on clinical and brain atrophy outcomes was assessed in terms of statistical agreement with hypothesized equal magnitude effects (e.g., 30% slowing of both) and nominal directional concordance, as a function of therapeutic class. Setting Interventional randomized clinical trials. Participants MCI or AD trial participants. Intervention Treatments included were those that involved ingestion or injection of a putatively active substance into the body, encompassing both pharmacological and controlled dietary interventions. Measurements Each trial included in the analysis reported at least one of the required clinical outcomes (ADAS-Cog, CDR-SB or MMSE) and at least one of the required imaging outcomes (whole brain, ventricular or hippocampal volume). Results Data from 35 trials, comprising 185 pairwise comparisons, were included. Overall, the 95% confidence bounds overlapped with the line of identity for 150/185 (81%) of the imaging-clinical variable pairs. The greatest proportion of outliers was found in trials of anti-amyloid antibodies that have been shown to dramatically reduce the level of PET-detectable amyloid plaques, for which only 13/33 (39%) of observations overlapped the identity line. A Deming regression calculated using all data points yielded a slope of 0.54, whereas if data points from the amyloid remover class were excluded, the Deming regression line had a slope of 0.92. Directional discordance of treatment effects was also most pronounced for the amyloid-removing class, and for comparisons involving ventricular volume. Conclusion Our results provide a frame of reference for the interpretation of clinical and brain atrophy results from future clinical trials and highlight the importance of mechanism of action in the interpretation of imaging results. Electronic Supplementary Material Supplementary material is available in the online version of this article at 10.14283/jpad.2023.92.

Figure S1 shows the schematic axes on which the data points were plotted, and interpretation of each quadrant.Figure S2 graphically illustrates the concept and construction of the "statistical distance" ds, whereby the distance of each data point to the line of identity is expressed as a fraction of the 95% confidence intervals on each variable.b) Each data point plotted as "statistical distance" ds (i.e., normalized to the 95% CI) for each trial in turn.The grey rectangle shows the range of |ds|< 95% CI.The green and blue points lying within the range of 95% CI are filled, whereas the red point, lying outside this range, is empty.The size of the points is inversely proportional to the 95% CI ellipses.These plotting conventions are also employed for the data in the main body of the paper.

Study selection
Figure S3 illustrates the study selection flow for the literature review performed in this study.

Concordance (data points > 25% from origin)
We also investigated nominal directional concordance between treatment effects on clinical outcomes and brain atrophy when considering only those data points lying greater than 0.25 from the origin, corresponding to effects of more meaningful magnitude and more likely to be clinically significant.Overall, 42/70 (60%) were directionally concordant.For specific classes, concordance rates were 6/7 (86%) for large molecule amyloid, 17/25 (68%) for small molecule amyloid, 3/3 (100%) for amyloid vaccines, 6/6 (100%) for small molecule tau, and 9/14 (64%) for the "other" class.Again, the large molecule amyloid remover class was lower, with 3/21 (14%) of the data points further from the origin being concordant.For all classes combined except the large molecule amyloid remover class, 39/49 (80%) of the data points were directionally concordant.

Statistical outliers (ds > 1)
Details of the outlier data points (i.e., those with ds > 1, or whose 95%CIs do not overlap the line of identity) are provided in Table S1.Overall, there were 35/185 (19%) outliers.The majority (20/35) of these were found in the amyloid remover class, for which 20/33 (61%) were outliers.

Distribution of ds values as a function of therapeutic class
As a supplement to Figure 3

Plots of individual clinical and imaging variable pairs
Figures S5-S13 depict the data points for each pair of clinical and imaging variables individually.The data are represented by the 95% CI ellipses (cf. Figure S2) with the point estimate depicted as the trial number corresponding to Table 1 in the main paper.

Figure S1 :
Figure S1: Schematic diagram illustrating the directionality of the axes used to plot treatment effects on pairs of imaging and clinical variables (i.e., the result of the comparison between the active treatment arm and the control arm).For example, the top right quadrant corresponds to concordant effects on both variables, in a direction favouring the active treatment.

Figure S2 :
Figure S2: Schematic diagrams illustrating statistical distance ds from the line of identity.(a) Three hypothetical data points with error bars equal to the 95% CI in the point estimate of the treatment effect on the imaging outcome (x-axis) and clinical outcome (y-axis), with those error bars defining ellipses describing the 95% CIs of the bivariate distributions.The ellipses for the green and blue points overlap the identity line, whereas those for the red point do not.Points to the lower right of the identity line are signed negative.(b) Each data point plotted as "statistical distance" ds (i.e., normalized to the 95% CI) for each trial in turn.The grey rectangle shows the range of |ds|< 95% CI.The green and blue points lying within the range of 95% CI are filled, whereas the red point, lying outside this range, is empty.The size of the points is inversely proportional to the 95% CI ellipses.These plotting conventions are also employed for the data in the main body of the paper.
in the main paper, Figure S4 shows density plots of the values of ds for each therapeutic class.The curve for the amyloid remover class is clearly right-shifted relative to the other classes, which are predominantly within |ds|<1.

Figure S4 :
Figure S4: Density plots of the distribution of ds within each of the therapeutic classes.The vertical dashed lines indicate ds=±1, so that values within the dashed lines correspond to points whose 95% CIs overlap the line of identity.

Figure S5 :
Figure S5: Scatter plot of treatment effects on data points corresponding to WBV vs. ADAS-Cog.

Figure S6 :
Figure S6: Scatter plot of treatment effects on data points corresponding to VV vs. ADAS-Cog.

Figure S7 :
Figure S7: Scatter plot of treatment effects on data points corresponding to HCV vs. ADAS-Cog.

Figure S8 :
Figure S8: Scatter plot of treatment effects on data points corresponding to WBV vs. MMSE.

Figure S9 :
Figure S9: Scatter plot of treatment effects on data points corresponding to VV vs. MMSE.

Figure S10 :
Figure S10: Scatter plot of treatment effects on data points corresponding to HCV vs. MMSE.

Figure S11 :
Figure S11: Scatter plot of treatment effects on data points corresponding to WBV vs. CDR-SB.

Figure S12 :
Figure S12: Scatter plot of treatment effects on data points corresponding to VV vs. CDR-SB.

Figure S13 :
Figure S13: Scatter plot of treatment effects on data points corresponding to HCV vs. CDR-SB.

Table S1 :
Data points not statistically consistent with equal magnitude effects on clinical and imaging (brain atrophy) variables