Publication
with Pengyuan Xu, Guang Yang, et al. (corresponding author)
Med Research, 2026
Abstract: Musculoskeletal disorders such as osteoarthritis (OA), rheumatoid arthritis (RA), and osteoporosis (OP) affect millions worldwide, often progressing silently until advanced stages. Accessible early risk stratification remains challenging because conventional diagnosis often depends on imaging, specialist evaluation, and fragmented clinical information. Herein, we developed Luma AI, a multimodal AI platform that integrates routine blood tests and demographic variables with curated patient‐reported symptom narratives. The text branch uses a domain‐specific clinical text encoder based on LoRA‐adapter‐fine‐tuned Bio_ClinicalBERT, whereas the structured branch uses optimized classical machine learning models. The two branches are fused through an explicit prediction‐score and text‐representation pipeline for early screening of OA, RA, and OP. Performance was validated using the U.S. NHANES and multi‐center Chinese cohorts. We evaluated more than 50 predictive models and applied Bayesian optimization to refine the structured models, achieving over 80% accuracy with structured data alone and exceeding 90% in the multimodal validation cohort when symptom narratives were included. The platform incorporates explainable AI tools to identify key risk factors and provides personalized post‐screening recommendations for lifestyle, diet, and medical follow‐up. This study demonstrates that Luma AI can support scalable low‐cost early risk screening and personalized management of OA, RA, and OP while remaining positioned as a risk‐stratification aid rather than a stand‐ alone diagnostic replacement.
From Regulatory Approvals to Patents: Cross-Domain Linking for Cardiovascular Device Traceability
with Qingqing Yang and Haijiang Liu. (corresponding author)
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), 2026
Abstract: Linking FDA-approved medical devices to their underlying United States Patent and Trademark Office (USPTO) patents enables critical applications such as recall root-cause analysis, M&A-driven IP discovery, and technology trajectory mapping. However, this cross-domain entity linking task remains unexplored due to severe semantic gaps: FDA documents focus on clinical outcomes, while patents describe technical mechanisms, yielding minimal lexical overlap. We formalize medical device-patent linking as a challenging cross-domain entity linking problem characterized by label scarcity and domain shifts. Using cardiovascular devices as a high-impact, representative showcase featuring diverse technologies, high recall rates, and abundant disclosures, we construct a benchmark with 434 devices, 698K patents, and 585 high-fidelity expert-verified pairs. To address these challenges, we propose Bridge-MedDevKG, a coarse-to-fine framework that integrates (1) MedDevOnto, a domain-adaptive ontology via expert-guided weight injection into UMLS; (2) Multi-signal candidate generation fusing company affiliation, semantic similarity, and ontology-weighted entity overlap; and (3) Heterogeneous reranking with multi-signal scoring and XGBoost classification on hard negatives. Our approach achieves 91.6% recall on the gold standard with 50.9% noise reduction, substantially outperforming LLM baselines under comparable evaluation. The resulting MedDevKG provides 6.8M high-confidence links, laying a scalable foundation for regulatory-IP integration across medical specialties.
Cigarette Taxes, Smoking, and Health in the Long-Run
with Andrew Friedson, Katherine Meckel, Daniel Rees, and Daniel W. Sacks
Journal of Public Economics, 2023
Abstract: Medical experts have argued forcefully that using cigarettes harms health, prompting the adoption of myriad anti-smoking policies. The association between smoking and mortality may, however, be influenced by unobserved factors, making it difficult to discern the underlying long-term causal relationship. In this study, we explore the effects of cigarette taxes experienced as a teenager, which are arguably exogenous, on adult smoking participation and mortality. A one-dollar increase in cigarette taxes experienced at ages 14-17 is associated with an 8 percent reduction in adult smoking participation and a 4 percent reduction in mortality. Mortality effects are most pronounced for heart disease and lung cancer.
Exposure to Cigarette Taxes as a Teenager and the Persistence of Smoking into Adulthood
with Andrew Friedson, Katherine Meckel, Daniel Rees, and Daniel W. Sacks
Health Economics, 2024
Abstract: Are teenage and adult smoking causally related? Recent anti-tobacco policy is predicated on the assumption that preventing teenagers from smoking will ensure that fewer adults smoke, but direct evidence in support of this assumption is scant. Using data from three nationally representative sources and instrumenting for teenage smoking with cigarette taxes experienced at ages 14-17, we document a strong positive relationship between teenage and adult smoking: deterring 10 teenagers from smoking through raising cigarette taxes roughly translates into 5 fewer adult smokers. We conclude that efforts to reduce teenage smoking can have important, long-lasting consequences on smoking participation and, presumably, health.
How Does Regional Integration Drive Energy Efficiency in Vertically Connected Industries?
with Hongyu Long and Chenyang Yu
The 7th IEEE International Conference on Universal Village, IEEE UV2024 2024
Can Developing Big Data Improve Urban Energy Efficiency? A Double Machine Learning Based Approach
with Hongyu Long and Chenyang Yu
The 7th IEEE International Conference on Universal Village, IEEE UV2024 2024
Working Papers
with Haizhen Lin and George Ball
Abstract: Medical device startups are a pivotal dimension of new product innovation in the U.S. healthcare industry. Numerous innovative medical devices originate from startups that rely on two primary sources of external funding: independent venture capital (IVC) and corporate venture capital (CVC). CVC firms bring unique industry knowledge and incentives to startups that may impact startup’s operational performance outcomes. Using difference-in-difference models and an instrumental variable strategy, we study the effect of CVCs on the operational outcomes of medical device startups. We find that CVCs foster more startup product approvals, but once launched, these products experience worse product quality, indicating a double-edged sword to CVC funding. Post-hoc analyses reveal that this negative quality effect for startups is accentuated when CVC firms lack prior experience in the startups’ product market, and when CVC firms have their own product quality problems.
Do Shareholders Really Not React to Medical Device Recalls?
with Qingqing Yang and Vivek Astvansh
Revisions Requested, Journal of Regulatory Economics
Work in Progress
Genealogical and Network Analysis on Medical Device Product Failures
with Jason Fletcher
Acquisition and Med-Tech Firm Performance
With Haizhen Lin