Applying CVR, composite value ratio, to brain MRI volume and thickness as an atrophy biomarker for Alzheimer’s disease clinical trials

Background

Biomarkers are key tools for clinical trials of putative therapeutics in Alzheimer’s Disease. Biomarkers with higher signal‐to‐noise ratio can improve the efficiency of drug development through reduced sample sizes and shorter trial durations. Image‐based biomarkers such as SUVR from PET and volumes from MRI have been used as secondary outcomes in many studies. Previously, we showed how our biomarker‐discovery algorithm CVR (combined value ratio) outperformed tau‐PET SUVR in clinical trial applications [Saguer et al., Alzheimer’s ∖& Dementia 2022]. Here, we apply the same methodology to brain volumes and cortical thickness, which are both cheaper than PET and routinely collected.

Methods

Data: We analyse longitudinal MRI data in ADNI from 1622 participants: 556 cognitively unimpaired (CU) and 1066 cognitive impaired (CI), keeping the latest three visits for CU, and earliest for CI, to better target the transition from CU to CI.

Model: A genetic algorithm explores combinations of 68 regions (averaged) defined by the Desikan‐Killiany atlas in FreeSurfer, separately for volume (including ventricles) and cortical thickness (all log‐transformed). We define CVR as the ratio of two data‐driven composite regions.

Evaluation: A linear mixed‐effects model estimates 1) group separation in atrophy (t‐statistic, CU vs CI fixed effect) and 2) longitudinal precision (std of model residuals). We calculate 3) sample size estimates (SSE) for a hypothetical 12 months‐long clinical trial involving CI individuals designed for 80% power, with 20% treatment effect. Metrics 1 and 3 drive the biomarker discovery algorithm.

Results

Table‐1 shows our CVR metrics wrt traditional biomarkers from the literature (whole‐brain and hippocampus volume, average cortical thickness of all regions and a subset). CVR improved group separation by 1.7x, longitudinal precision by 1.7x and reduced sample size by 5.1x.

Discussion

The data‐driven biomarker from volumetric measures is the ratio of two composites (Figure‐1): one is highly linked to neurodegeneration, while the other takes advantage of more stable or growing regions, like the ventricles. Having these opposite characteristics in a ratio boosts its performance.

Conclusion

Our algorithm discovered a ratio‐based imaging biomarker that could vastly improve clinical trials in Alzheimer’s disease using routinely collected MRI.