Parkinson's disease phenotype recovery from sparse longitudinal trajectories using multiclass likelihood contrast classification

Sparse and irregular longitudinal measurements are common in clinical studies but are difficult to incorporate into conventional classifiers without interpolation or substantial feature reduction. We developed an interpretable multiclass likelihood contrast framework for classifying unaligned longitudinal trajectories and evaluated it using simulation experiments and Parkinson's disease motor-UPDRS data. The proposed approach fits class-specific mixed-effects models and classifies each held-out subject according to the largest marginal likelihood, using a quadratic mean trajectory with subject-specific random intercepts and slopes. Performance was assessed using subject-level cross-validation and compared with linear mixed-likelihood classification, functional k -nearest neighbors, random forest, support vector machine, polynomial-feature classification, and grid-based principal component analysis classifiers. In the Parkinson's disease experiment, dense longitudinal histories were used to define three mathematical trajectory phenotypes, after which approximately 90% of observations were masked to evaluate phenotype recovery from sparse records. Despite this substantial sparsification, the proposed model achieved an accuracy of 0.8095, macro-F1 of 0.8160, Matthews correlation coefficient of 0.7107, and macro-area under the receiver operating characteristic curve of 0.9268. Validation and held-out test macro-F1 values were 0.7980 and 0.8160, respectively, suggesting consistent performance across data partitions. These findings demonstrate that multiclass likelihood contrast learning can preserve informative temporal structure in sparse, irregular longitudinal records without reducing individual trajectories to a small number of summary features. The Parkinson's disease findings represent recovery of data-derived trajectory phenotypes rather than prospective clinical prediction or diagnostic validation. Further evaluation in larger independent cohorts with clinically defined outcomes and multimodal longitudinal biomarkers is needed to establish clinical utility.

Authors

Institutions

Publication Details

Journal
Frontiers in Artificial Intelligence
Published
2026-11-18
DOI
https://doi.org/10.3389/frai.2026.1968393
Primary Topic
Statistical Methods and Bayesian Inference
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Parkinson's disease phenotype recovery from sparse longitudinal trajectories using multiclass likelihood contrast classification

Fahad B. Mostafa, Md Rejuan Haque, Sangam pangeni
Frontiers in Artificial Intelligence
Statistical Methods and Bayesian Inference
article

Parkinson's disease phenotype recovery from sparse longitudinal trajectories using multiclass likelihood contrast classification

Fahad B. Mostafa, Md Rejuan Haque, Sangam pangeni
article en

Abstract

Sparse and irregular longitudinal measurements are common in clinical studies but are difficult to incorporate into conventional classifiers without interpolation or substantial feature reduction. We developed an interpretable multiclass likelihood contrast framework for classifying unaligned longitudinal trajectories and evaluated it using simulation experiments and Parkinson's disease motor-UPDRS data. The proposed approach fits class-specific mixed-effects models and classifies each held-out subject according to the largest marginal likelihood, using a quadratic mean trajectory with subject-specific random intercepts and slopes. Performance was assessed using subject-level cross-validation and compared with linear mixed-likelihood classification, functional k -nearest neighbors, random forest, support vector machine, polynomial-feature classification, and grid-based principal component analysis classifiers. In the Parkinson's disease experiment, dense longitudinal histories were used to define three mathematical trajectory phenotypes, after which approximately 90% of observations were masked to evaluate phenotype recovery from sparse records. Despite this substantial sparsification, the proposed model achieved an accuracy of 0.8095, macro-F1 of 0.8160, Matthews correlation coefficient of 0.7107, and macro-area under the receiver operating characteristic curve of 0.9268. Validation and held-out test macro-F1 values were 0.7980 and 0.8160, respectively, suggesting consistent performance across data partitions. These findings demonstrate that multiclass likelihood contrast learning can preserve informative temporal structure in sparse, irregular longitudinal records without reducing individual trajectories to a small number of summary features. The Parkinson's disease findings represent recovery of data-derived trajectory phenotypes rather than prospective clinical prediction or diagnostic validation. Further evaluation in larger independent cohorts with clinically defined outcomes and multimodal longitudinal biomarkers is needed to establish clinical utility.

Frontiers in Artificial IntelligenceVol. 9
University of New Mexico (US), Gallup Indian Medical Center (US), The Ohio State University (US), Arizona State University (US)
Good health and well-being
Statistical Methods and Bayesian Inference
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