Mathematical model of the Alzheimer's disease biomarker cascade demonstrates statistical pitfall in identifying surrogates of cognitive reserve.

Statistical interaction analyses with biomarkers of pathology on cognitive outcome have been put forward to investigate neurobiological surrogates of cognitive reserve in Alzheimer's disease (AD). However, as these potential surrogates are likely affected by neurodegeneration during the course of AD, their joint alteration with biomarkers of pathology and cognitive outcome during disease progression may introduce misinterpretable interaction effects in cross-sectional studies. To demonstrate this, we conducted interaction analyses on synthetic data from a mathematical model of the AD biomarker cascade. When randomly sampling cross-sectionally, these gave interaction effects, which greatly reduced when controlling for the corresponding time point of each sampled data point. Cross-sectional studies investigating cognitive reserve using interaction analyses are advised to take into account the different time stages of the disease that individual data points represent.