Search papers, labs, and topics across Lattice.
The paper introduces StructGP and LP-StructGP, continuous-time multi-task Gaussian processes that learn sparse, ordered directed acyclic graphs (DAGs) of inter-variable dependencies and latent pathways from irregular clinical time series data. These models use process convolutions with differentiable structure learning, trained with sparsity and acyclicity constraints via augmented Lagrangian and Adam optimization. Experiments on MIMIC-IV and PhysioNet datasets demonstrate improved short-horizon forecasting accuracy, calibration, and interpretability compared to unstructured kernels and graph neural networks.
Unlocking interpretable clinical forecasting: StructGP recovers causal relationships and patient progression patterns directly from irregular EHR data, outperforming black-box methods in accuracy and uncertainty calibration.
Background: Timely, uncertainty-aware forecasting from irregular electronic health records (EHR) can support critical-care decisions, yet most approaches either impute to a grid or sacrifice interpretability. We introduce StructGP, a continuous-time multi-task Gaussian process that couples process convolutions with differentiable structure learning to uncover a sparse, ordered directed acyclic graph (DAG) of inter-variable dependencies while preserving principled uncertainty. We further propose LP-StructGP, which augments StructGP with latent pathways-shared, temporally shifted trajectories inferred via subject-specific coupling filters and a softmax gating mechanism-to capture cross-patient progression patterns. Both models are trained under sparsity and acyclicity constraints (augmented Lagrangian, Adam) using scalable low-rank updates. Results: In simulations, the approach reliably recovers ground-truth graphs (Structural Hamming Distance approaching 0 as cohorts grow) and pathway assignments (high Adjusted Rand Index). On a MIMIC-IV septic shock cohort (n=1,008; norepinephrine, creatinine, mean arterial pressure), StructGP improves short-horizon (6 h) forecasting over independent-task baselines (average RMSE 0.68 [95%CI: 0.63--0.74] vs. 0.88 [0.83-0.94]) and, with 15 additional inputs, markedly outperforms unstructured kernels (0.63 [0.58-0.69] vs. 3.02 [2.85-3.18]) with superior calibration (coverage 0.96 vs. 0.84). On the PhysioNet Challenge (12k patients, 41 variables), StructGP attains competitive accuracy (MAE 3.72e-2) relative to a state-of-the-art graph neural model while maintaining calibrated uncertainty. Conclusion: These results show that structured process convolutions with latent pathways deliver interpretable, scalable, and well-calibrated forecasting for irregular clinical time series.
