Summary Alzheimer’s Disease (AD) research increasingly recognizes the importance of characterizing the joint and relative dynamics of biomarker trajectories during the preclinical phase. Jack’s model stands as the most well-known theoretical framework of these processes. While it offers conceptual guidance on modeling, it is not a directly implementable model for observed data. Key challenges include: (i) biomarker measurements are often noisy, depend on individual characteristics, and often exhibit complex temporal and cross-sectional correlations; (ii) data are sampled sparsely and may cover only a limited window of disease progression; and (iii) the temporal scale of the disease may not be synchronized across individuals. To address these challenges, we introduce a multivariate (for the multiple biomarkers), sparse (for the longitudinal sampling mechanism), nonlinear (for the hypothesized S-shape of biomarker evolution) latent variable model that accounts for complex correlation structures and individual factors. Crucially, we identify and address substantial practical and computational issues due to the noisy, sparse, and limited longitudinal data. Our contribution includes using a penalized likelihood to address estimates being close to the boundary due to degenerate cases, implementing a Gaussian Quadrature Newton-Raphson algorithm for efficient estimation under flexible correlation structures, and proposing a practical solution for potentially incomplete biomarker trajectories. We evaluate the finite-sample performance of the proposed method through simulations designed to mimic real-world data. We then apply the model to 2 longitudinal AD cohorts, BIOCARD and ADNI, to estimate biomarker trajectories across imaging, cerebrospinal fluid, and cognitive domains. These applications provide insights into the temporal dynamics of AD biomarkers, supporting improvements in preclinical diagnosis and guiding future research.
Summary Time-lag bias occurs when studies with larger or statistically significant effects appear in print sooner than those with smaller or null effects, which can inflate early meta-analytic estimates and mislead decision makers. Existing diagnostics typically rely on cumulative meta-analysis or meta-regression of effects on publication time. These approaches are vulnerable to repeated-testing inflation and loss of power when temporal trends are nonlinear. We propose a family of flexible nonparametric statistics that summarize standardized pairwise differences between early and later studies, each emphasizing different magnitudes of temporal contrast, and a hybrid test that adaptively combines evidence across these differences via permutation. The procedure controls type I error under exchangeability of publication order and is sensitive to diverse temporal patterns, including linear trends, early outliers, and oscillations in effect sizes over time. Simulations across 9 scenarios, with and without heterogeneity, show that the hybrid test maintains the nominal error rate and improves statistical power relative to linear meta-regression when temporal trends are nonlinear. In 2 clinical meta-analyses, our method detects pronounced time-lag patterns that standard regression misses. We provide guidance for practice and discuss practical considerations for detecting and interpreting time-lag bias in meta-analyses.
Summary Social determinants of health (SDOH) are key drivers of cardiovascular disease (CVD), but their relationships differ substantially across regions, necessitating the design of locally tailored intervention strategies. In this work, we propose a novel clustering approach to partition US counties into different clusters based on similarities in their SDOH–CVD associations, providing a data-driven regional partition to inform region-specific interventions. To enhance the interpretability of the clustering result, we adopt the SDOH framework from the Agency for Healthcare Research and Quality to group SDOH variables into 5 domains, and build on this to develop an SDOH-domain-aware regularization (SDAR), which identifies the most influential variables within each SDOH domain tailored for each cluster. By embedding SDAR into the clustering process via a regularized expectation-maximization algorithm, our approach enables simultaneous clustering, parameter estimation, and variable selection, thereby mitigating error propagation across stages. We validate the method through theoretical developments and simulations. Applied to a national longitudinal dataset of US county-level SDOH and CVD mortality, our approach discovered 7 distinct clusters, each characterized by a unique set of SDOH drivers spanning multiple SDOH domains. Notably, 2 of these clusters exhibited elevated CVD mortality and inverse SDOH–CVD associations compared to the other 5. These findings reveal complex regional heterogeneity in SDOH–CVD associations, yielding insights to guide multi-domain, region-specific strategies for CVD management.
Summary Understanding how cells interact to coordinate tissue function and respond to disease remains a central challenge in biology. Recently developed multiplex bioimaging technologies [multiplex immunofluorescence (MxIF)] can quantify multiple protein markers at a single-cell resolution in spatially preserved tissue, offering the opportunity to understand how cellular organization relates to disease mechanisms and treatment response. Spatial generalized linear mixed models for cell co-localization offer a flexible framework for these analyses, accommodating cell-, image-, or subject-level predictors while accounting for correlation arising from repeated cells per image, repeated images per subject, and spatial dependence among nearby cells. However, as MxIF data typically have hundreds of thousands to millions of cells per study, fitting the full spatial model is computationally infeasible due to the large, dense, spatial covariance matrix. We present a scalable inference strategy that uses parallelizable eigen-decomposition of an empirically estimated spatial correlation matrix to approximate the full spatial model. This approach enables fast estimation of fixed effects of interest without requiring formal estimation of spatial parameters or surface prediction. In simulations, our proposed method could be fitted to large datasets ($ \sim $ 285,000 observations) in under a minute, with runtime linearly increasing by only $ \sim $ 7 s per 50,000 additional observations. This scalability makes cell-level modeling in the presence of spatial correlation feasible for real-world MxIF datasets, as we demonstrate through applications to metastatic melanoma and high-grade serous ovarian carcinoma.
Summary Accurate estimates of microbial species abundances are needed to advance our understanding of the role that microbiomes play in human and environmental health. However, laboratory-constructed microbiomes demonstrate that intuitive estimators of microbial relative abundances are biased. To address this, we propose a method to estimate relative abundances, species detection effects, and/or cross-sample contamination in microbiome experiments. We show that certain experimental designs result in identifiable model parameters, and present consistent estimators and asymptotically valid inference procedures that are robust to misspecification of the working likelihood. Notably, our procedure can estimate relative abundances on the boundary of the simplex. We demonstrate the utility of the method for comparing experimental protocols, removing cross-sample contamination, and estimating species’ detectability.
