Social Network Structure, Wealth, and Wealth Inequality Across Cultures
研究通过分析全球46个社区的约3500户家庭数据，探讨了社会网络结构与财富不平等之间的关系，发现较贫穷的家庭在社会网络中与较富裕家庭的联系较少。
0 citationsRead paper
Selected work
Representative Papers
Information leakage from data revisions in retrospective forecasts
This study addresses a critical yet often overlooked source of information leakage in retrospective evaluations of AI-driven forecasting systems: the failure to account for data revisions. The authors systematically demonstrate how this oversight leads to inflated performance estimates and present, for the first time, a cross-domain cautionary framework to mitigate such biases. Through retrospective predictive analysis, explicit modeling of data revision processes, and rigorous evaluation protocols, they reveal that a previously reported AI system’s purported superiority over the CDC ensemble model stems not from genuine predictive gains but from information leakage introduced by unadjusted historical data revisions. These findings establish essential methodological corrections and evaluation standards for future research in AI-based forecasting, ensuring more reliable and reproducible assessments of predictive performance.
0 citationsRead paper
Complementary Roles of Image Classification and Vessel Segmentation in AI-Based Screening for Retinopathy of Prematurity Plus Disease in a Kenyan Preterm Cohort
This study addresses the scarcity of specialists and diagnostic subjectivity in screening for Plus disease associated with retinopathy of prematurity (ROP) in low-resource settings by systematically evaluating the complementary roles of image classification and retinal vessel segmentation in a Kenyan preterm infant cohort. We developed and compared multiple AI pipelines—including RGB-based classifiers, multiple instance learning, multitask joint models, and a segmentation-followed-by-classification pipeline—using patient-stratified nested cross-validation. The proposed probabilistic ensemble model achieved the best balanced performance at the eye level, with a sensitivity of 0.692, specificity of 0.914, and balanced accuracy of 0.803, significantly outperforming single-task classifiers. Additionally, vessel segmentation yielded a Dice coefficient of 0.533 and a high specificity of 0.979, effectively reducing unnecessary referrals.
0 citationsRead paper
Continuous-Time Bayesian Networks with Structured Shrinkage Priors for Modelling Multimorbidity Trajectories in Large-Scale Electronic Health Records
Existing approaches struggle to effectively model the complex interactions among disease progression, multimorbidity networks, and high-dimensional time-varying risk factors. This work proposes a structured Bayesian continuous-time Bayesian network to learn directed disease dependencies from longitudinal electronic health records, wherein transition intensities depend on existing conditions, pairwise interactions, and exogenous covariates. To mitigate parameter explosion in higher-order terms while preserving interpretability of main effects, the model incorporates order-dependent structured shrinkage priors—such as spike-and-slab and Bayesian LASSO—that selectively suppress redundant interactions. Simulation studies demonstrate that the spike-and-slab prior achieves superior performance in variable selection, network recovery, and false discovery control. Applied to UK Biobank data, the method successfully identifies a diabetes-centered metabolic module and a respiratory–atopic inflammatory module.
0 citationsRead paper
Interpreting Net Survival: What We Estimate Versus What We Think We Estimate
This study re-examines the conventional definition of net survival—commonly estimated via the Pohar-Perme estimator—under the widely adopted assumption that cancer patients experience the same non-cancer mortality risk as the general population. Challenging this assumption, the authors propose a theoretical framework decomposing total mortality into cancer-related, baseline health–related, treatment-induced, and background non-cancer components. Through mathematical derivation and empirical analysis, they demonstrate that the Pohar-Perme estimator inherently incorporates non-cancer mortality beyond background rates and thus should not be interpreted as a causal counterfactual quantity; rather, it represents a conditional survival probability adjusted only for background mortality. Empirical results reveal that for cancers such as head and neck, the relative risk of non-cancer mortality can exceed 4.0, leading to substantial underestimation of true cancer-specific survival whenever this risk surpasses unity, thereby exposing a systematic bias in current interpretations of net survival.
0 citationsRead paper
Recent publications
Latest Papers
Social Network Structure, Wealth, and Wealth Inequality Across Cultures
研究通过分析全球46个社区的约3500户家庭数据，探讨了社会网络结构与财富不平等之间的关系，发现较贫穷的家庭在社会网络中与较富裕家庭的联系较少。
0 citationsRead paper
Information leakage from data revisions in retrospective forecasts
This study addresses a critical yet often overlooked source of information leakage in retrospective evaluations of AI-driven forecasting systems: the failure to account for data revisions. The authors systematically demonstrate how this oversight leads to inflated performance estimates and present, for the first time, a cross-domain cautionary framework to mitigate such biases. Through retrospective predictive analysis, explicit modeling of data revision processes, and rigorous evaluation protocols, they reveal that a previously reported AI system’s purported superiority over the CDC ensemble model stems not from genuine predictive gains but from information leakage introduced by unadjusted historical data revisions. These findings establish essential methodological corrections and evaluation standards for future research in AI-based forecasting, ensuring more reliable and reproducible assessments of predictive performance.
0 citationsRead paper
Complementary Roles of Image Classification and Vessel Segmentation in AI-Based Screening for Retinopathy of Prematurity Plus Disease in a Kenyan Preterm Cohort
This study addresses the scarcity of specialists and diagnostic subjectivity in screening for Plus disease associated with retinopathy of prematurity (ROP) in low-resource settings by systematically evaluating the complementary roles of image classification and retinal vessel segmentation in a Kenyan preterm infant cohort. We developed and compared multiple AI pipelines—including RGB-based classifiers, multiple instance learning, multitask joint models, and a segmentation-followed-by-classification pipeline—using patient-stratified nested cross-validation. The proposed probabilistic ensemble model achieved the best balanced performance at the eye level, with a sensitivity of 0.692, specificity of 0.914, and balanced accuracy of 0.803, significantly outperforming single-task classifiers. Additionally, vessel segmentation yielded a Dice coefficient of 0.533 and a high specificity of 0.979, effectively reducing unnecessary referrals.
0 citationsRead paper
Continuous-Time Bayesian Networks with Structured Shrinkage Priors for Modelling Multimorbidity Trajectories in Large-Scale Electronic Health Records
Existing approaches struggle to effectively model the complex interactions among disease progression, multimorbidity networks, and high-dimensional time-varying risk factors. This work proposes a structured Bayesian continuous-time Bayesian network to learn directed disease dependencies from longitudinal electronic health records, wherein transition intensities depend on existing conditions, pairwise interactions, and exogenous covariates. To mitigate parameter explosion in higher-order terms while preserving interpretability of main effects, the model incorporates order-dependent structured shrinkage priors—such as spike-and-slab and Bayesian LASSO—that selectively suppress redundant interactions. Simulation studies demonstrate that the spike-and-slab prior achieves superior performance in variable selection, network recovery, and false discovery control. Applied to UK Biobank data, the method successfully identifies a diabetes-centered metabolic module and a respiratory–atopic inflammatory module.
0 citationsRead paper
Interpreting Net Survival: What We Estimate Versus What We Think We Estimate
This study re-examines the conventional definition of net survival—commonly estimated via the Pohar-Perme estimator—under the widely adopted assumption that cancer patients experience the same non-cancer mortality risk as the general population. Challenging this assumption, the authors propose a theoretical framework decomposing total mortality into cancer-related, baseline health–related, treatment-induced, and background non-cancer components. Through mathematical derivation and empirical analysis, they demonstrate that the Pohar-Perme estimator inherently incorporates non-cancer mortality beyond background rates and thus should not be interpreted as a causal counterfactual quantity; rather, it represents a conditional survival probability adjusted only for background mortality. Empirical results reveal that for cancers such as head and neck, the relative risk of non-cancer mortality can exceed 4.0, leading to substantial underestimation of true cancer-specific survival whenever this risk surpasses unity, thereby exposing a systematic bias in current interpretations of net survival.
0 citationsRead paper