Search

The increasing prevalence of Alzheimer’s disease poses a considerable socioeconomic challenge in the years ahead. There are few clinical treatments available and none capable of halting or slowing the progressive nature of the condition. Despite decades of experimental research and testing over 300 interventions in transgenic mouse models of the condition, clinical success has remained elusive. Deepening our understanding of how such studies have been conducted is likely to provide insights which could inform future preclinical and clinical research. Therefore I performed a systematic review and meta-analysis on interventions tested in transgenic mouse models of Alzheimer’s disease. My systematic search was performed by electronically searching for publications reporting the efficacy of interventions tested in transgenic models of Alzheimer's disease. Across these publications I extracted data regarding study characteristics and reported study quality alongside outcome data for pathology (i.e. plaque burden, amyloid beta species, tau, cellular infiltrates and neurodegeneration) and neurobehaviour. From these data I calculated estimates of efficacy using random effects meta-analysis and subsequently investigated the potential impact of study quality and study characteristics on observed effect size. My search identified 427 publications, 357 interventions and 55 transgenic models representing 11, 688 animals and 1774 experiments. There were a number of principal concerns regarding the dataset: (i) the reported study quality of such studies was relatively low; less than 1 in 5 publications reported blinded assessment of outcome or random allocation to group and no studies reported a sample size calculation, (ii) the depth of data on any individual intervention was relatively poor-only 16 interventions had outcomes described in 5 or more publications and (iii) publication bias analyses suggested 1 in 5 pathological and 1 in 7 neurobehavioural experiments remain unpublished. Where I inspected relationships between outcomes, meta-regression identified a number of notable associations. Changes in amyloid beta 40 were reflective of changes in amyloid beta 42 (R2 = 0.84, p<0.01) and within the Morris water maze changes in the ‘training’ acquisition phase could explain 44% of the changes in the probe ‘test’ phase (p<0.05). Additionally, I identified measures of neurodegeneration as the best pathological predictors of changes in neurobehaviour (R2 = 0.72, p<0.01). Collectively this work identifies a number of potential weaknesses within in vivo modelling of Alzheimer’s disease and demonstrates how the use of empirical data can inform both preclinical and clinical studies.