Proof of Concept Example for Use of Simulation to Allow Data Pooling Despite Privacy Restrictions.

BACKGROUND: Integrating results from multiple samples is often desirable, but privacy restrictions may preclude full data pooling, and most datasets do not include fully harmonized variable sets. We propose a simulation-based method leveraging partial information across datasets to guide creation of synthetic data based on explicit assumptions about the underlying causal structure that permits pooled analyses that adjust for all desired confounders in the context of privacy restrictions. METHODS: This proof-of-concept project uses data from the Health and Retirement Study (HRS) and Atherosclerosis Risk in Communities (ARIC) study. We specified an estimand of interest and a directed acyclic graph (DAG) summarizing the presumed causal structure for the effect of glycated hemoglobin (HbA1c) on cognitive change. We derived publicly reportable statistics to describe the joint distribution of each variable in our DAG. These summary estimates were used as data-generating rules to create synthetic datasets. After pooling, we imputed missing covariates in the synthetic datasets and used the synthetic data to estimate the pooled effect of HbA1c on cognitive change, adjusting for all desired covariates. RESULTS: Distributions of covariates and model coefficients and associated standard errors for our model estimating the effect of HbA1c on cognitive change were similar across cohort-specific original and preimputation synthetic data. The estimate from the pooled synthetic incorporates control for confounders measured in either original dataset. DISCUSSION: Our approach has advantages over meta-analysis or individual-level pooling/data harmonization when privacy concerns preclude data sharing and key confounders are not uniformly measured across datasets.

Extension of Mendelian Randomization to Identify Earliest Manifestations of Alzheimer Disease: Association of Genetic Risk Score for Alzheimer Disease With Lower Body Mass Index by Age 50 Years.

Weight loss or lower body mass index (BMI) could be an early symptom of Alzheimer disease (AD), but when this begins to emerge is difficult to estimate with traditional observational data. In an extension of Mendelian randomization, we leveraged variation in genetic risk for late-onset AD risk to estimate the causal effect of AD on BMI and the earliest ages at which AD-related weight loss (or lower BMI as a proxy) occurs. We studied UK Biobank participants enrolled in 2006-2010, who were without dementia, aged 39-73, with European genetic ancestry. BMI was calculated with measured height/weight (weight (kg)/height (m)2). An AD genetic risk score (AD-GRS) was calculated based on 23 genetic variants. Using linear regressions, we tested the association of AD-GRS with BMI, stratified by decade, and calculated the age of divergence in BMI trends between low and high AD-GRS. AD-GRS was not associated with BMI in 39- to 49-year-olds (ß = 0.00, 95% confidence interval (CI): -0.03, 0.03). AD-GRS was associated with lower BMI in 50- to 59-year-olds (ß = -0.03, 95% CI: -0.06, -0.01) and 60- to 73-year-olds (ß = -0.09, 95% CI:-0.12, -0.07). Model-based BMI age curves for high versus low AD-GRS began to diverge after age 47 years. Sensitivity analyses found no evidence for pleiotropy or survival bias. Longitudinal replication is needed; however, our findings suggest that AD genes might begin to reduce BMI decades prior to dementia diagnosis.

Effect of reductions in amyloid levels on cognitive change in randomized trials: instrumental variable meta-analysis.

OBJECTIVE: To evaluate trials of drugs that target amyloid to determine whether reductions in amyloid levels are likely to improve cognition. DESIGN: Instrumental variable meta-analysis. SETTING: 14 randomized controlled trials of drugs for the prevention or treatment of Alzheimer's disease that targeted an amyloid mechanism, identified from ClinicalTrials.gov. POPULATION: Adults enrolled in randomized controlled trials of amyloid targeting drugs. Inclusion criteria for trials vary, but typically include adults aged 50 years or older with a diagnosis of mild cognitive impairment or Alzheimer's disease, and amyloid positivity at baseline. MAIN OUTCOME MEASURES: Analyses included trials for which information could be obtained on both change in brain amyloid levels measured with amyloid positron emission tomography and change in at least one cognitive test score reported for each randomization arm. RESULTS: Pooled results from the 14 randomized controlled trials were more precise than estimates from any single trial. The pooled estimate for the effect of reducing amyloid levels by 0.1 standardized uptake value ratio units was an improvement in the mini-mental state examination score of 0.03 (95% confidence interval -0.06 to 0.1) points. This study provides a web application that allows for the re-estimation of the results when new data become available and illustrates the magnitude of the new evidence that would be necessary to achieve a pooled estimate supporting the benefit of reducing amyloid levels. CONCLUSIONS: Pooled evidence from available trials reporting both reduction in amyloid levels and change in cognition suggests that amyloid reduction strategies do not substantially improve cognition.

Association of genetic risk for Alzheimer disease and hearing impairment.

OBJECTIVE: To test the hypothesis that incipient Alzheimer disease (AD) may adversely affect hearing and that hearing loss may adversely affect cognition, we evaluated whether genetic variants that increase AD risk also increase problem hearing and genetic variants that increase hearing impairment risk do not influence cognition. METHODS: UK Biobank participants without dementia ≥56 years of age with Caucasian genetic ancestry completed a Digit Triplets Test of speech-in-noise hearing (n = 80,074), self-reported problem hearing and hearing with background noise (n = 244,915), and completed brief cognitive assessments. A genetic risk score for AD (AD-GRS) was calculated as a weighted sum of 23 previously identified AD-related polymorphisms. A genetic risk score for hearing (hearing-GRS) was calculated using 3 previously identified polymorphisms related to hearing impairment. Using age-, sex-, and genetic ancestry-adjusted logistic and linear regression models, we evaluated whether the AD-GRS predicted poor hearing and whether the hearing-GRS predicted worse cognition. RESULTS: Poor speech-in-noise hearing (>-5.5-dB speech reception threshold; prevalence 14%) was associated with lower cognitive scores (ß = -1.28; 95% confidence interval [CI] -1.54 to -1.03). Higher AD-GRS was significantly associated with poor speech-in-noise hearing (odds ratio [OR] 1.06; 95% CI 1.01-1.11) and self-reported problems hearing with background noise (OR 1.03; 95% CI 1.00-1.05). Hearing-GRS was not significantly associated with cognitive scores (ß = -0.05; 95% CI -0.17 to 0.07). CONCLUSIONS: Genetic risk for AD also influences speech-in-noise hearing. We failed to find evidence that genetic risk for hearing impairment affects cognition. AD disease processes or a that shared etiology may cause speech-in-noise difficulty before dementia onset.

Neuropsychological Test Performance and MRI Markers of Dementia Risk: Reducing Education Bias.

BACKGROUND: To use neuropsychological assessments for studying the underlying disease processes contributing to dementia, it is crucial that they correspond to magnetic resonance imaging (MRI)-based measures of dementia, regardless of educational level. METHODS: French 3-City Dijon MRI study cohort members (n=1782) with assessments of white matter lesion volume (WMLV), hippocampal volume (HCV), and cerebrospinal fluid volume (CSFV), and 6 waves of neuropsychological assessments over 11 years, including Mini-Mental State Examination (MMSE), plus 5 other tests combined using a Z-score or item-response theory (IRT-cognition) comprised the study cohort. We evaluated, testing interactions, whether education modified associations of MRI markers with intercept or rate of change of MMSE, Z-score composite, or IRT-cognition. RESULTS: In linear models, education modified the associations of WMLV and CSFV with MMSE and CSFV and Z-score composite. In mixed models, education modified the associations of WMLV and CSFV with level of MMSE and the association of HCV with slope of MMSE. Education also modified the association with CSFV and slope of Z-score composite decline. There was no evidence that education modified associations between MRI measures and level or slope of IRT-cognition. CONCLUSIONS: Longitudinal analysis of correctly scaled neuropsychological assessments may provide unbiased proxies for MRI-based measures of dementia risk.

"Liquid Biopsy" of White Matter Hyperintensity in Functionally Normal Elders.

Background and Objective: In the aging brain, increased blood-brain barrier (BBB) leakage and white matter hyperintensity (WMH) on MRI are frequently presumed secondary to cerebral small vessel disease (cSVD) or endotheliopathy. We investigate this association in vivo by quantifying protein cargo from endothelial-derived exosomes (EDE), and comparing levels between two groups of functionally normal elders with and without WMH. In addition, we study associations of EDE proteins with upstream and downstream factors, such as inflammation and neurodegenerative changes, respectively. Methods: Twenty six neurologically normal older adults completed general health questionnaires, neuropsychological and physical examinations, and brain MRI. WMH was visually graded with modified Fazekas score of 2 or greater used to classify 11 subjects as cases, and 15 without WMH as controls. Plasma total exosomes were precipitated and EDEs enriched by sequential immuno-precipitations. In addition, we quantified three inflammatory cytokines from plasma and imaging variables on MRI. Group means were compared, the discriminant functions of biomarkers calculated, and the association of EDE biomarkers with plasma inflammatory markers, cognition, and imaging outcomes assessed via regression modeling. Results: Plasma levels of EDE cargo proteins GLUT1, LAT1, P-GP, and NOSTRIN were significantly higher in subjects with WMH in comparison to those without. In contrast, EDE levels of the marker with low expression in brain (VCAM1) were equal between groups. The effect sizes for each of the brain-expressed cargo proteins (GLUT1, LAT1, and P-GP) were such that age-adjusted logistic regressions revealed areas under the curve (AUC) with range of 0.82-0.89, differentiating subjects with WMH from those without. VCAM1 poorly discriminated between groups (AUC:0.55). Higher levels of all brain-expressed EDE proteins were also associated with lower cognitive function, unrelated to burden of WMH. Levels of LAT1 and P-GP were significantly inversely associated with global gray matter volumes, and EDE GLUT1, LAT-1, and P-GP concentrations were significantly associated with systemic IL-6 levels. Conclusion: In a case control study of clinically normal adults with and without WMH, concentrations of EDE proteins were significantly higher in subjects with WMH in comparison to controls. This work is a first step toward in vivo dissection of molecular changes in endothelia of functionally normal subjects with radiographic evidence of age-associated white matter disease.

Does selective survival before study enrolment attenuate estimated effects of education on rate of cognitive decline in older adults? A simulation approach for quantifying survival bias in life course epidemiology.

Background: The relationship between education and late-life cognitive decline is controversial. Selective survival between early life, when education is typically completed, and late life, when cognitive ageing studies take place, could attenuate effect estimates. Methods: We quantified potential survival bias (collider-stratification bias) in estimation of the effect of education on late-life cognitive decline by simulating hypothetical cohorts of 20-year-olds and applying cumulative mortality from US life tables. For each of four causal scenarios (2000 replications each), we compared the estimated versus causal effect of education on cognitive decline over 9 years, starting at age 60, 75 or 90 in random samples of n = 2000 people who survived to each age. Results: Effects of education on cognitive decline were underestimated when both education and U, another determinant of cognitive decline, influenced mortality (collider-stratification bias). The magnitude of bias was sensitive to the magnitude of the effect of U on cognitive decline and whether there was a multiplicative interaction between education and U on mortality. For example, when there was a multiplicative interaction between education and U on mortality, 95% confidence interval coverage of the causal effect ranged from 83.4% to 50.4% at age 60 and 25.8% to 0.2% at age 90. Conclusions: Selective survival could lead to underestimation of effects of education on late-life cognitive decline. Our simulations map survival bias to testable assumptions about underlying causal structures.

OrbId: Origin-based identification of microRNA targets.

MicroRNAs coordinate networks of mRNAs, but predicting specific sites of interactions is complicated by the very few bases of complementarity needed for regulation. Although efforts to characterize the specific requirements for microRNA (miR) regulation have made some advances, no general model of target recognition has been widely accepted. In this work, we describe an entirely novel approach to miR target identification. The genomic events responsible for the creation of individual miR loci have now been described with many miRs now known to have been initially formed from transposable element (TE) sequences. In light of this, we propose that limiting miR target searches to transcripts containing a miR's progenitor TE can facilitate accurate target identification. In this report we outline the methodology behind OrbId (Origin-based identification of microRNA targets). In stark contrast to the principal miR target algorithms (which rely heavily on target site conservation across species and are therefore most effective at predicting targets for older miRs), we find OrbId is particularly efficacious at predicting the mRNA targets of miRs formed more recently in evolutionary time. After defining the TE origins of > 200 human miRs, OrbId successfully generated likely target sets for 191 predominately primate-specific human miR loci. While only a handful of the loci examined were well enough conserved to have been previously evaluated by existing algorithms, we find ~80% of the targets for the oldest miR (miR-28) in our analysis contained within the principal Diana and TargetScan prediction sets. More importantly, four of the 15 OrbId miR-28 putative targets have been previously verified experimentally. In light of OrbId proving best-suited for predicting targets for more recently formed miRs, we suggest OrbId makes a logical complement to existing, conservation based, miR target algorithms.