The Use, Standardization, and Interpretation of Brain Imaging Data in Clinical Trials of Neurodegenerative Disorders

Abstract

Imaging biomarkers play a wide-ranging role in clinical trials for neurological disorders. This includes selecting the appropriate trial participants, establishing target engagement and mechanism-related pharmacodynamic effect, monitoring safety, and providing evidence of disease modification. In the early stages of clinical drug development, evidence of target engagement and/or downstream pharmacodynamic effect-especially with a clear relationship to dose-can provide confidence that the therapeutic candidate should be advanced to larger and more expensive trials, and can inform the selection of the dose(s) to be further tested, i.e., to "de-risk" the drug development program. In these later-phase trials, evidence that the therapeutic candidate is altering disease-related biomarkers can provide important evidence that the clinical benefit of the compound (if observed) is grounded in meaningful biological changes. The interpretation of disease-related imaging markers, and comparability across different trials and imaging tools, is greatly improved when standardized outcome measures are defined. This standardization should not impinge on scientific advances in the imaging tools per se but provides a common language in which the results generated by these tools are expressed. PET markers of pathological protein aggregates and structural imaging of brain atrophy are common disease-related elements across many neurological disorders. However, PET tracers for pathologies beyond amyloid β and tau are needed, and the interpretability of structural imaging can be enhanced by some simple considerations to guard against the possible confound of pseudo-atrophy. Learnings from much-studied conditions such as Alzheimer's disease and multiple sclerosis will be beneficial as the field embraces rarer diseases.

Keywords: Clinical trials; Imaging; MRI; Neurology; PET.

Fig. 1

Schematic overview of applications of imaging in clinical trials for drug development. (This is intended to indicate typical scenarios, but exceptions may occur.)

Fig. 2

Time-dependent SUVR plots provide a means to assess the temporal stability of SUVR measurements, can help identify an optimal static scanning window, and can highlight instances where deviations in acquisition time might contribute to additional variability. (a) Schematic showing the ideal case where a quasi-steady-state of SUVR(t) is obtained at a certain time post-injection. (b) Average SUVR(t) curves across small cohorts of subjects at different disease stage for [¹⁸F]flortaucipir in the lateral temporal lobe, indicating that a quasi-steady-state is not achieved (on average) in more advanced disease stages. (c) SUVR(t) curves from a single Alzheimer’s disease individual using [¹⁸F]PI-2620, where each color represents a different brain region, indicating that a quasi-steady-state is not achieved in regions with higher tau burden (a was originally published in the Journal of Cerebral Blood Flow and Metabolism [130]© SAGE Publishing; b was originally published in the Journal of Nuclear Medicine [131] © SNMMI; c was originally published in the Journal of Nuclear Medicine [96] © SNMMI)

Fig. 3

(a) Mapping of [¹⁸F]flutemetemol amyloid PET SUVR values (x-axis) into Centiloids (y-axis) [158]. (b) Side-by-Side comparison of [¹⁸F]florbetapir SUVR values and Aβ_L values across Alzheimer’s disease stages from the ADNI database [160]. Note that Aβ_L values have well-defined floor and ceiling levels. (c) Side-by-Side comparison of longitudinal change in global TauL and regional SUVR values from [¹⁸F]flortaucipir scans in the ADNI database (a was originally published in the European Journal of Nuclear Medicine and Molecular Imaging [158], reproduced under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/); b was originally published in the Journal of Nuclear Medicine [160] © SNMMI; c was originally published in the Journal of Nuclear Medicine [162] © SNMMI; axis labels have been redrawn for legibility)

Fig. 4

Meta-analysis of reported treatment effects on T2 active lesion load from MRI versus treatment effects on clinical relapses, showing similar relationships for (a) placebo-controlled and active-controlled trials, and for (b) phase 2 and phase 3 trials. Figure reproduced with permission from [127]

Fig. 5

(a), (b) Theoretical dependence of absolute and relative differences between volumetric changes in treatment and placebo arms on change in the placebo arm for (a) the case where the reduction in volume loss is directly proportional to the rate of change in the placebo arm, consistent with a slowing of neurodegeneration, and (b) the case where the reduction in volume loss is independent of the rate of change in the placebo arm, such as might result from a non-specific inflammatory effect. Each dot in these ensemble plots represents a different brain region (the data points in (a, b) are illustrative only). In the case of region-proportional slowing (or acceleration) (a), this framework yields three estimates of the relative slowing parameter, r. These are the slope of the regression line from the analysis of absolute change, and the average and y-intercept from the analysis of relative change. Here, r represents the fractional slowing of atrophy in the treatment arm relative to the placebo arm (e.g., r = 0.25 would correspond to a 25% slowing). In the case of region-independent slowing (or acceleration) (b), this framework yields three estimates of the absolute slowing parameter, a. These are the average and y-intercept of from the analysis of absolute change, and the coefficient of the inverse relationship from the analysis of relative change. Here, a represents the absolute slowing of atrophy in the treatment arm relative to the placebo arm (e.g., a = 0.01 would correspond to 1% absolute slowing of brain volume loss). If the treatment arm evidences faster volume loss than the placebo arm, δ_abs or δ_rel are negative. (c, d) Ensemble plots and regression analysis for vMRI data reported for vMRI outcome measures from the EXPEDITION3 trial of solanezumab. The regressions indicate that in this case the overall pattern of the treatment effect on brain atrophy is most consistent with the region-proportional scenario (a). With all 12 vMRI metrics included (c), the three estimates of the relative rate of slowing were 3.9%, 3.8%, and 3.8%, highly consistent, with a coefficient of variance of only 1%. When the ventricles were excluded (d), the three estimates were 4.5%, 3.8%, and 3.3%, still consistent, with a coefficient of variance of 13%

Fig. 6

Schematic illustrating the interpretive value of intermediate scans, similar to data observed in [187]. (a) A transient increased volume loss relative to placebo immediately after treatment initiation, with no long-term change in the rate of atrophy, can be detected and its time course well understood if intermediate scans are acquired. (b) If only baseline and endpoint scans are acquired, the transient nature of the treatment effect is lost, and the data may be interpreted differently, as an ongoing acceleration of brain atrophy