Accounting for latent class uncertainty in trajectory-specific risk estimation: the Targeted Compositional Corner Estimator Trajectory-specific risk estimation under latent class uncertainty

Abstract Background In many chronic diseases, risk depends on how a biomarker evolves over time rather than on its value at a single visit. These longitudinal patterns are commonly summarized as latent trajectory groups, but the uncertainty in class membership is usually ignored by assigning each individual to the single most likely group, which can bias trajectory-specific risk estimates, especially for small groups. Methods We developed the Targeted Compositional Corner Estimator (TCCE), a method that retains the full posterior class-membership probabilities and estimates trajectory-specific counterfactual risks using compositional data methods, mapping the probabilities to an unconstrained space with the isometric log-ratio transformation and evaluating risk at the corners representing pure membership. We compared the TCCE with hard assignment, probability-weighted averaging, and targeted maximum likelihood estimation with hard labels in simulations under well-classified, misclassification-stress, and nonlinear-interaction scenarios (200 replicates; n = 2000), including additional conditions with correlated covariates and small, unbalanced classes. We then applied all methods to linked clinical and administrative records from 1,614 Singaporean adults with diabetes, estimating the association between three-year HbA1c trajectories and severe periodontitis, with reduced dentition as a secondary outcome. Results Across all simulation scenarios, the TCCE achieved the lowest overall root mean squared error, with the greatest gains under nonlinear interaction, and its advantage persisted under correlated covariates and smaller, unbalanced samples. The point estimates were invariant to the corner offset. In the application, the high-increasing HbA1c trajectory carried the highest risk of severe periodontitis, and the TCCE produced the narrowest compatibility intervals. Conclusions The TCCE provides a practical approach for estimating trajectory-specific risks while accounting for latent class uncertainty, and it applies more broadly to exposures represented as probabilities over categories.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11276349/v1
Primary Topic
Advanced Causal Inference Techniques
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Accounting for latent class uncertainty in trajectory-specific risk estimation: the Targeted Compositional Corner Estimator Trajectory-specific risk estimation under latent class uncertainty

Bibhas Chakraborty, Pinyan Liu, Marco Aurelio Peres, Murthy Mittinty
2 links
Research Square
Advanced Causal Inference Techniques
preprint

Accounting for latent class uncertainty in trajectory-specific risk estimation: the Targeted Compositional Corner Estimator Trajectory-specific risk estimation under latent class uncertainty

Bibhas Chakraborty, Pinyan Liu, Marco Aurelio Peres, Murthy Mittinty
preprint en

Abstract

Abstract Background In many chronic diseases, risk depends on how a biomarker evolves over time rather than on its value at a single visit. These longitudinal patterns are commonly summarized as latent trajectory groups, but the uncertainty in class membership is usually ignored by assigning each individual to the single most likely group, which can bias trajectory-specific risk estimates, especially for small groups. Methods We developed the Targeted Compositional Corner Estimator (TCCE), a method that retains the full posterior class-membership probabilities and estimates trajectory-specific counterfactual risks using compositional data methods, mapping the probabilities to an unconstrained space with the isometric log-ratio transformation and evaluating risk at the corners representing pure membership. We compared the TCCE with hard assignment, probability-weighted averaging, and targeted maximum likelihood estimation with hard labels in simulations under well-classified, misclassification-stress, and nonlinear-interaction scenarios (200 replicates; n = 2000), including additional conditions with correlated covariates and small, unbalanced classes. We then applied all methods to linked clinical and administrative records from 1,614 Singaporean adults with diabetes, estimating the association between three-year HbA1c trajectories and severe periodontitis, with reduced dentition as a secondary outcome. Results Across all simulation scenarios, the TCCE achieved the lowest overall root mean squared error, with the greatest gains under nonlinear interaction, and its advantage persisted under correlated covariates and smaller, unbalanced samples. The point estimates were invariant to the corner offset. In the application, the high-increasing HbA1c trajectory carried the highest risk of severe periodontitis, and the TCCE produced the narrowest compatibility intervals. Conclusions The TCCE provides a practical approach for estimating trajectory-specific risks while accounting for latent class uncertainty, and it applies more broadly to exposures represented as probabilities over categories.

Research Square
Flinders University (AU), Duke-NUS Medical School (SG)
Duke-NUS Medical School, National Medical Research Council
Good health and well-being
Advanced Causal Inference Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon