Mixed-Effects Modeling for Censored and Complex Data

Mixed-effects modelling for censored and complex data unites fixed and random components to account for population-level effects and subject-specific variability when observations are only partially observed or possess intricate dependence structures. Censoring arises in contexts such as detection limits in environmental studies, time-to-event outcomes in clinical trials, and economic datasets with truncated responses. Recent advances have extended classical linear mixed models to accommodate left, right or interval censoring through Tobit-type formulations, survival-analysis frameworks and joint modelling of longitudinal and time-to-event data. Moreover, the integration of non-Gaussian distributions — including heavy-tailed scale mixtures of normals and multivariate t-distributions — has enhanced robustness to outliers and model misspecification. Methodological developments encompass adaptive expectation–maximisation algorithms, stochastic gradient techniques and Bayesian Markov chain Monte Carlo schemes; these allow efficient estimation in high-dimensional settings and facilitate flexible prior specification. Applications span biomedical longitudinal biomarker studies, ecological threshold detection and pharmacokinetics, where both parameter inference and individual predictive trajectories are of interest. Emerging software implementations harness parallel computation and subsampling to scale to large cohorts and delve into complex hierarchical structures. By balancing statistical rigour and computational tractability, mixed-effects models for censored and complex data constitute a vital toolkit across disciplines confronting incomplete, noisy or structured observations.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent work has proposed finite-mixture Tobit regression frameworks that flexibly address high degrees of censoring by blending component-specific intercepts and slopes, demonstrating reduced bias in heavily censored simulations and improved real-world environmental applications. Extensions to non-Gaussian hierarchies have been achieved through scale mixtures of normals, enabling any combination of random effects and residual terms to deviate from Gaussianity; subsampling-based stochastic gradient algorithms further accelerate maximum likelihood estimation for large longitudinal datasets. In biomedical contexts, fractional Brownian motion coupled with multivariate-t random-effect distributions has delivered biologically plausible trajectories and robust handling of dropout censoring, yielding more accurate predictions of clinical event timings than standard random-slopes models. Complementing these, asymmetric power-Student-t models have been introduced for censored regression to capture skewness and heavy tails, with simulation studies confirming enhanced parameter recovery under complex censoring schemes. Collectively, these contributions underscore the global applicability of advanced mixed-effects techniques to censored and multifaceted data, bridging methodology and practice across environmental, clinical and industrial domains.

Mixed-Effects Modeling for Censored and Complex Data publication trend

The graph below shows the total number of articles in mixed-effects modeling for censored and complex data across all publications each year (not limited to Nature Index journals).

Technical terms

Mixed-effects model: A statistical model combining fixed effects (common to all units) and random effects (unit-specific deviations) to capture hierarchical or clustered data structure.

Censoring: The condition where an observation is only partially known, as when values fall above, below or within detection limits.

Tobit model: A regression model designed for censored response variables, incorporating latent variables to handle limited-range outcomes.

Scale mixture of normals: A class of probability distributions formed by mixing normal distributions over a scaling parameter, often used to model heavy tails or robust residuals.

Expectation–maximisation algorithm: An iterative procedure for maximum likelihood estimation in models with latent variables or incomplete data, alternating between expectation and maximisation steps.

References

1. Finite mixture modeling of censored regression models. Statistical Papers (2013).
2. Linear Mixed Effects Models for Non-Gaussian Continuous Repeated Measurement Data. Journal of the Royal Statistical Society Series C (Applied Statistics) (2020).
3. Fractional Brownian motion and multivariate‐t models for longitudinal biomedical data, with application to CD4 counts in HIV‐positive patients. Statistics in Medicine (2015).
4. Flexible linear mixed models with improper priors for longitudinal and survival data. Electronic Journal of Statistics (2018).
5. The Asymmetric Power-Student-t Model for Censored and Truncated Data. Anais da Academia Brasileira de Ciências (2021).