How Selection Over Time Contributes to the Inconsistency of the Association Between Sex/Gender and Cognitive Decline Across Cognitive Aging Cohorts.

The association between sex/gender and aging-related cognitive decline remains poorly understood because of inconsistencies in findings. Such heterogeneity could be attributable to the cognitive functions studied and study population characteristics, but also to differential selection by dropout and death between men and women. We aimed to evaluate the impact of selection by dropout and death on the association between sex/gender and cognitive decline. We first compared the statistical methods most frequently used for longitudinal data, targeting either population estimands (marginal models fitted by generalized estimating equations) or subject-specific estimands (mixed/joint models fitted by likelihood maximization) in 8 studies of aging: 6 population-based studies (the Advanced Cognitive Training for Independent and Vital Elderly (ACTIVE) Study (1996-2009), Personnes Âgées QUID (PAQUID; 1988-2014), the Reasons for Geographic and Racial Differences in Stroke (REGARDS) Study (2003-2016), the Three-City Study (Bordeaux only; 1999-2016), the Washington Heights-Inwood Community Aging Project (WHICAP; 1992-2017), and the Whitehall II Study (2007-2016)) and 2 clinic-based studies (the Alzheimer's Disease Neuroimaging Initiative (ADNI; 2004-2017) and a nationwide French cohort study, MEMENTO (2011-2016)). We illustrate differences in the estimands of the association between sex/gender and cognitive decline in selected examples and highlight the critical role of differential selection by dropout and death. Using the same estimand, we then contrast the sex/gender-cognitive decline associations across cohorts and cognitive measures suggesting a residual differential sex/gender association depending on the targeted cognitive measure (memory or animal fluency) and the initial cohort selection. We recommend focusing on subject-specific estimands in the living population for assessing sex/gender differences while handling differential selection over time.

Investigating the association between cancer and the risk of dementia: Results from the Memento cohort.

INTRODUCTION: Studies on the association of cancer and risk of dementia are inconclusive due to result heterogeneity and concerns of survivor bias and unmeasured confounding. METHODS: This study uses data from the Memento cohort, a French multicenter cohort following persons with either mild or isolated cognitive complaints for a median of 5 years. Illness-death models (IDMs) were used to estimate transition-specific hazard ratios (HRs) and 95% confidence intervals (CIs) for incident cancer in relation to dementia from time since study entry. RESULTS: The analytical sample (N = 2258) excluded 65 individuals without follow-up information. At the end of follow-up, 286 individuals were diagnosed with dementia, 166 with incident cancer, and 95 died. Incident cancer was associated with a reduced risk of dementia (HR = 0.58, 95% CI = 0.35-0.97), with a corresponding E-value of 2.84 (lower CI = 1.21). DISCUSSION: This study supports a protective relationship between incident cancer and dementia, encouraging further investigations to understand potential underlying mechanisms.

Joint latent class model for longitudinal data and interval-censored semi-competing events: Application to dementia.

Joint models are used in ageing studies to investigate the association between longitudinal markers and a time-to-event, and have been extended to multiple markers and/or competing risks. The competing risk of death must be considered in the elderly because death and dementia have common risk factors. Moreover, in cohort studies, time-to-dementia is interval-censored since dementia is assessed intermittently. So subjects can develop dementia and die between two visits without being diagnosed. To study predementia cognitive decline, we propose a joint latent class model combining a (possibly multivariate) mixed model and an illness-death model handling both interval censoring (by accounting for a possible unobserved transition to dementia) and semi-competing risks. Parameters are estimated by maximum-likelihood handling interval censoring. The correlation between the marker and the times-to-events is captured by latent classes, homogeneous sub-groups with specific risks of death, dementia, and profiles of cognitive decline. We propose Markovian and semi-Markovian versions. Both approaches are compared to a joint latent-class model for competing risks through a simulation study, and applied in a prospective cohort study of cerebral and functional ageing to distinguish different profiles of cognitive decline associated with risks of dementia and death. The comparison highlights that among subjects with dementia, mortality depends more on age than on duration of dementia. This model distinguishes the so-called terminal predeath decline (among healthy subjects) from the predementia decline.

Validity of shape as a predictive biomarker of final infarct volume in acute ischemic stroke.

BACKGROUND AND PURPOSE: This study examines whether lesion shape documented on magnetic resonance diffusion-weighted imaging during acute stroke improves the prediction of the final infarct volume compared with lesion volume only. METHODS: Diffusion-weighted imaging data and clinical information were retrospectively reviewed in 110 consecutive patients who underwent (n=67) or not (n=43) thrombolytic therapy for acute ischemic stroke. Three-dimensional shape analysis was performed on admission diffusion-weighted imaging data and 5 shape descriptors were developed. Final infarct volume was measured on T2-fluid-attenuated inversion recovery imaging data performed 30 days after stroke. RESULTS: Shape analysis of acute ischemic lesion and more specifically the ratio of the bounding box volume to the lesion volume before thrombolytic treatment improved the prediction of the final infarct for patients undergoing thrombolysis (R(2)=0.86 in model with volume; R(2)=0.98 in model with volume and shape). CONCLUSIONS: Our findings suggest that lesion shape contains important predictive information and reflects important environmental factors that might determine the progression of ischemia from the core.

Bayesian profile regression for clustering analysis involving a longitudinal response and explanatory variables.

The identification of sets of co-regulated genes that share a common function is a key question of modern genomics. Bayesian profile regression is a semi-supervised mixture modelling approach that makes use of a response to guide inference toward relevant clusterings. Previous applications of profile regression have considered univariate continuous, categorical, and count outcomes. In this work, we extend Bayesian profile regression to cases where the outcome is longitudinal (or multivariate continuous) and provide PReMiuMlongi, an updated version of PReMiuM, the R package for profile regression. We consider multivariate normal and Gaussian process regression response models and provide proof of principle applications to four simulation studies. The model is applied on budding yeast data to identify groups of genes co-regulated during the Saccharomyces cerevisiae cell cycle. We identify 4 distinct groups of genes associated with specific patterns of gene expression trajectories, along with the bound transcriptional factors, likely involved in their co-regulation process.

Quantile regression for incomplete longitudinal data with selection by death.

Quantile regressions are increasingly used to provide population norms for quantitative variables. Indeed, they do not require any Gaussian assumption for the response and allow to characterize its entire distribution through different quantiles. Quantile regressions are especially useful to provide norms of cognitive scores in the elderly that may help general practitioners to identify subjects with unexpectedly low cognitive level in routine examinations. These norms may be estimated from cohorts of elderly using quantile regression for longitudinal data, but this requires to properly account for selection by death, dropout and intermittent missing data. In this work, we extend the weighted estimating equation approach to estimate conditional quantiles in the population currently alive from mortal cohorts with dropout and intermittent missing data. Suitable weight estimation procedures are provided for both monotone and intermittent missing data and under two missing-at-random assumptions, when the observation probability given that the subject is alive depends on the survival time (p-MAR assumption) or not (u-MAR assumption). Inference is performed through subject-level bootstrap. The method is validated in a simulation study and applied to the French cohort Paquid to estimate quantiles of a cognitive test in the elderly population currently alive. On one hand, the simulations show that the u-MAR analysis is quite robust when the true missingness mechanism is p-MAR. This is a useful result because computation of suitable weights for intermittent missing data under the p-MAR assumption is untractable. On the other hand, the simulations highlight, along with the real data analysis, the usefulness of suitable weights for intermittent missing data. This method is implemented in the R package weightQuant.

Interpretation of mixed models and marginal models with cohort attrition due to death and drop-out.

Mixed models estimated by maximum likelihood and marginal models estimated by generalized estimating equations are the standard methods for the analysis of longitudinal data. However, their use is highly debated when attrition may be due to death. While some authors consider that mixed model estimates are interpretable only in an immortal cohort, we show that their subject-specific interpretation still holds in the population currently alive, but their population-averaged interpretation is valid only in the immortal cohort. We propose an approximation of the population-averaged mean among the population alive that highlights the difference with the population-averaged mean in the immortal cohort. The interpretation of ML estimates of mixed models and joint models for the marker and the time-to-death as well as unweighted and weighted GEE of marginal models is then illustrated in a simulation study and in an application regarding cognitive decline in the elderly.