We study heterogeneous treatment-effect (HTE) estimation in observational survival studies commonly associated with both censored outcomes and unmeasured confounding. We integrate causal survival forests (CSF) with negative controls (NC) from proximal causal inference and introduce Negative Control Causal Survival Forests (NC-CSF), a flexible nonparametric HTE learner for survival analysis. Our approach uses a loss that incorporates proxy variables and Neyman orthogonalization to train the random forest, thereby mitigating bias from unobserved confounding and gaining robustness to nuisance estimation. Through extensive simulations spanning varying levels of confounding, proxy relevance, and censoring mechanisms, we demonstrate that NC-CSF substantially reduces bias and estimation error relative to existing baselines. We further demonstrate the practical utility of our method on various clinical datasets, where it confirms several existing findings and also reveals new interpretable patterns of treatment-effect heterogeneity. To facilitate practical use, we provide an end-to-end Python implementation of NC-CSF that carefully handles implementation details such as nuisance estimation and clipping.
Zijun Gao, Kyoungeui Hong, Leyi Ma et al.· 0 citations
Hypertension is the leading preventable risk factor for cardiovascular disease, yet continuous blood pressure (BP) monitoring is still largely inaccessible outside of clinical settings. Photoplethysmography (PPG), collected via consumer wearables, offers a promising non-invasive method for continuous cuffless BP monitoring, but PPG signals lack the physiological information necessary for reliable BP estimation. Conversely, specific electrocardiogram (ECG)-level features, namely QRS timing and pulse transit time, show a direct correlation to BP. Unfortunately, most wearables lack the dedicated electrodes needed for accurate ECG measurement. Building on Ji and Zhou's (SenSys '24) demonstration that ECG can be reconstructed from PPG via a diffusion model, we address a gap left open by existing work: real-world wearable deployments introduce signal degradation, such as motion artifacts, sensor noise, and intermittent data loss that state-of-the-art frameworks do not account for, leaving the clinical reliability of diffusion-reconstructed ECG-based BP estimators unestablished. We propose replacing the BiLSTM, the state-of-the-art estimator used by Ji and Zhou, with a Transformer, whose self-attention captures long-range crossmodal dependencies between the generated ECG and PPG, and we introduce supervised contrastive learning to make the learned representations invariant to noise, missing data, and subject variation. On MIMIC-II and MIMIC-BP benchmarks, our Diffusion+Transformer framework achieves mean absolute errors of 4.50 mmHg (SBP) and 2.39 mmHg (DBP) on clean signals, improving over the BiLSTM baseline by 0.27 mmHg (SBP) and 0.28 mmHg (DBP), with substantially better robustness under temporal data removal across both datasets. Contrastive learning (SupCon) reduces error under additive noise conditions: at 50% noise, adding SupCon reduces SBP MAE from 25.65 to 11.94 mmHg and DBP MAE from 20.94 to 5.86 mmHg.
N. Valencia, J. T. Rodriguez, M. A. Rahman et al.· International Conference on...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
