Rumeng Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Rumeng Li has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Hong Yu
University of Massachusetts Lowell
4 shared publications - Baotian Hu
Harbin Institute of Technology (Shenzhen) Shenzhen China
2 shared publications - David D. McManus
Program in Digital Medicine, Department of Medicine, University of Massachusetts Chan Medical School, Worcester, MA, United States
2 shared publications - Feifan Liu
University of Massachusetts Chan Medical School, Worcester, USA
2 shared publications - Francesca Cunningham
US Department of Veterans Affairs Hines Illinois USA
2 shared publications - Honghuang Lin
Department of Medicine, University of Massachusetts Chan Medical School, Worcester, MA, USA
2 shared publications
Latest Publications
- Open Access
- 28 September 2026
- Published by Elsevier in SSRN Electronic Journal
Longitudinal Clinical Note–Derived Signals Are Consistent Across Healthcare Systems And Support Alzheimer’s Disease Risk Prediction
Background: Alzheimer’s disease (AD) is often preceded by subtle cognitive, behavioral, and functional changes that emerge years before diagnosis. Clinical notes may capture these changes during care. We assessed whether AD-related signals in clinical notes were consistent across two US healthcare systems and supported prediction of future AD across settings. Methods: We conducted a cross-system validation study using longitudinal electronic health record (EHR) data from the Veterans Health Administration (VHA) and Mass General Brigham (MGB). The VHA derivation cohort included 61,537 AD cases and 234,105 matched controls; the MGB validation cohort included 2,074 AD cases and 7,054 matched controls. Using the same expert-curated AD-related keyword lexicon previously applied in VHA, we compared longitudinal keyword patterns before diagnosis across systems. We evaluated machine learning models using longitudinal keyword features within each system and directly applied VHA-trained models to MGB without retraining. Performance was compared with models using structured EHR features, including diagnosis codes, medications and abnormal laboratory measurements. Discrimination was assessed using area under the receiver operating characteristic curve (AUROC). Findings: During the 15 years preceding diagnosis, AD-related keyword frequencies increased more rapidly among individuals later diagnosed with AD than matched controls in both systems. Longitudinal case-to-control patterns were strongly correlated across systems (Spearman ρ=0·953). Across prediction horizons from 10 years to 1 day before diagnosis, random forest models using narrative keyword features achieved AUROCs of 0·542-0·756 in MGB, higher than models using structured EHR features. VHA-trained models retained measurable discrimination when applied directly to MGB (AUROC 0·533-0·706). Interpretation: Longitudinal narrative signals showed cross-system consistency and predictive value across two healthcare systems serving different patient populations, supporting their potential use for AD risk assessment while underscoring the need for local validation before clinical implementation. Funding: This research was supported by the National Institute on Aging (R01AG080670).
- 28 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessLongitudinal Clinical Note–Derived Signals Are Consistent Across Healthcare Systems And Support Alzheimer’s Disease Risk Prediction
Background: Alzheimer’s disease (AD) is often preceded by subtle cognitive, behavioral, and functional changes that emerge years before diagnosis. Clinical notes may capture these changes during care. We assessed whether AD-related signals in clinical notes were consistent across two US healthcare systems and supported prediction of future AD across settings. Methods: We conducted a cross-system validation study using longitudinal electronic health record (EHR) data from the Veterans Health Administration (VHA) and Mass General Brigham (MGB). The VHA derivation cohort included 61,537 AD cases and 234,105 matched controls; the MGB validation cohort included 2,074 AD cases and 7,054 matched controls. Using the same expert-curated AD-related keyword lexicon previously applied in VHA, we compared longitudinal keyword patterns before diagnosis across systems. We evaluated machine learning models using longitudinal keyword features within each system and directly applied VHA-trained models to MGB without retraining. Performance was compared with models using structured EHR features, including diagnosis codes, medications and abnormal laboratory measurements. Discrimination was assessed using area under the receiver operating characteristic curve (AUROC). Findings: During the 15 years preceding diagnosis, AD-related keyword frequencies increased more rapidly among individuals later diagnosed with AD than matched controls in both systems. Longitudinal case-to-control patterns were strongly correlated across systems (Spearman ρ=0·953). Across prediction horizons from 10 years to 1 day before diagnosis, random forest models using narrative keyword features achieved AUROCs of 0·542-0·756 in MGB, higher than models using structured EHR features. VHA-trained models retained measurable discrimination when applied directly to MGB (AUROC 0·533-0·706). Interpretation: Longitudinal narrative signals showed cross-system consistency and predictive value across two healthcare systems serving different patient populations, supporting their potential use for AD risk assessment while underscoring the need for local validation before clinical implementation. Funding: This research was supported by the National Institute on Aging (R01AG080670).
- Open Access
- 12 January 2026
- Published by Springer Nature in Communications Medicine
- Vol. 6 (1) , 23
Early prediction of Alzheimer’s disease using longitudinal electronic health records of US military veterans
Background: Early prediction of Alzheimer’s disease is important for timely intervention and treatment. We examine whether machine learning on longitudinal electronic health record notes can improve early prediction of Alzheimer’s disease. Methods: From Veterans Health Administration records (2000 to 2022), we studied 61,537 individuals diagnosed with Alzheimer’s disease and 234,105 without, aged 45–103 years, 98.4% were male. From clinical notes, we quantified the frequency of subjective cognitive decline and Alzheimer’s disease-related keywords, and applied statistical machine learning models to assess their ability to predict future diagnosis. Results: Here we show that Alzheimer’s-related keywords (e.g., “concentration,” “speaking”), occur more often in notes of individuals who later develop Alzheimer’s disease than in controls. In the 15 years preceding diagnosis, cases demonstrate an exponential increase in keyword mentions (from 9.4 to 57.7 per year), whereas controls show a slower, linear increase (8.2 to 20.3). These trends are consistent across demographic subgroups. Random forest models using these keywords for prediction achieve an area under receiver operating characteristic curve from 0.577 at ten years before diagnosis to 0.861 one day before diagnosis, consistently outperforming models using only structured data. Conclusions: Signs and symptoms of early Alzheimer’s disease are reported in clinical notes many years before a clinical diagnosis is made and the frequency of these signs and symptoms, approximated by keywords, increases the closer one is to the diagnosis. A simple keyword-based approach can capture these signals and can help identify individuals at high risk of future Alzheimer’s disease.
- 12 January 2026
- Published by Springer Nature in Communications Medicine
- Vol. 6 (1) , 23
Open AccessEarly prediction of Alzheimer’s disease using longitudinal electronic health records of US military veterans
Background: Early prediction of Alzheimer’s disease is important for timely intervention and treatment. We examine whether machine learning on longitudinal electronic health record notes can improve early prediction of Alzheimer’s disease. Methods: From Veterans Health Administration records (2000 to 2022), we studied 61,537 individuals diagnosed with Alzheimer’s disease and 234,105 without, aged 45–103 years, 98.4% were male. From clinical notes, we quantified the frequency of subjective cognitive decline and Alzheimer’s disease-related keywords, and applied statistical machine learning models to assess their ability to predict future diagnosis. Results: Here we show that Alzheimer’s-related keywords (e.g., “concentration,” “speaking”), occur more often in notes of individuals who later develop Alzheimer’s disease than in controls. In the 15 years preceding diagnosis, cases demonstrate an exponential increase in keyword mentions (from 9.4 to 57.7 per year), whereas controls show a slower, linear increase (8.2 to 20.3). These trends are consistent across demographic subgroups. Random forest models using these keywords for prediction achieve an area under receiver operating characteristic curve from 0.577 at ten years before diagnosis to 0.861 one day before diagnosis, consistently outperforming models using only structured data. Conclusions: Signs and symptoms of early Alzheimer’s disease are reported in clinical notes many years before a clinical diagnosis is made and the frequency of these signs and symptoms, approximated by keywords, increases the closer one is to the diagnosis. A simple keyword-based approach can capture these signals and can help identify individuals at high risk of future Alzheimer’s disease.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4945-4949
Improving Rare and Common ICD Coding via a Multi-Agent LLM-Based Approach
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4945-4949
Open AccessImproving Rare and Common ICD Coding via a Multi-Agent LLM-Based Approach
- Open Access
- 8 February 2019
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 7 (1) , e10788
Detection of Bleeding Events in Electronic Health Record Notes Using Convolutional Neural Network Models Enhanced With Recurrent Neural Network Autoencoders: Deep Learning Approach
Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Bleeding events are common and critical which may cause significant morbidity and mortality. Studies show that high incidences of bleeding events are associated with cardiovascular disease (CVD) patients on anticoagulant therapy. Prompt and accurate detection of bleeding events are essential for preventing serious consequences. As bleeding events are often described in clinical notes, automatic detection of bleeding events from Electronic Health Record (EHR) narratives has the potential to improve drug safety surveillance and pharmacovigilance. Objective: We developed a natural language processing (NLP) system to automatically classify whether an EHR note sentence contains a bleeding event. Methods: We expert-annotated 878 EHR notes (76,577 sentences and 562,630 word tokens) for identifying bleeding events at the sentence-level. This annotated corpus was then used to train and validate our NLP systems. We developed an innovative hybrid CNN and LSTM Autoencoder model (HCLA), which integrates a convolutional neural network architecture (CNN) with a bidirectional Long-short term memory (BiLSTM) autoencoder model to leverage large unlabeled EHR data. Results: HCLA achieved an F-score of 93.79% for identifying whether a sentence contains a bleeding event, surpassing the strong baseline SVM and other CNN models. Conclusions: By incorporating a supervised CNN model with a pre-trained unsupervised BiLSTM Autoencoder, HCLA achieved a high performance in detecting bleeding events.
- 8 February 2019
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 7 (1) , e10788
Open AccessDetection of Bleeding Events in Electronic Health Record Notes Using Convolutional Neural Network Models Enhanced With Recurrent Neural Network Autoencoders: Deep Learning Approach
Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Bleeding events are common and critical which may cause significant morbidity and mortality. Studies show that high incidences of bleeding events are associated with cardiovascular disease (CVD) patients on anticoagulant therapy. Prompt and accurate detection of bleeding events are essential for preventing serious consequences. As bleeding events are often described in clinical notes, automatic detection of bleeding events from Electronic Health Record (EHR) narratives has the potential to improve drug safety surveillance and pharmacovigilance. Objective: We developed a natural language processing (NLP) system to automatically classify whether an EHR note sentence contains a bleeding event. Methods: We expert-annotated 878 EHR notes (76,577 sentences and 562,630 word tokens) for identifying bleeding events at the sentence-level. This annotated corpus was then used to train and validate our NLP systems. We developed an innovative hybrid CNN and LSTM Autoencoder model (HCLA), which integrates a convolutional neural network architecture (CNN) with a bidirectional Long-short term memory (BiLSTM) autoencoder model to leverage large unlabeled EHR data. Results: HCLA achieved an F-score of 93.79% for identifying whether a sentence contains a bleeding event, surpassing the strong baseline SVM and other CNN models. Conclusions: By incorporating a supervised CNN model with a pre-trained unsupervised BiLSTM Autoencoder, HCLA achieved a high performance in detecting bleeding events.
- Open Access
- 15 April 2018
- Published by JMIR Publications Inc. in JMIR Preprints
Detection of Bleeding Events in Electronic Health Record Notes Using Convolutional Neural Network Models Enhanced With Recurrent Neural Network Autoencoders: Deep Learning Approach (Preprint)
BACKGROUND Bleeding events are common and critical and may cause significant morbidity and mortality. High incidences of bleeding events are associated with cardiovascular disease in patients on anticoagulant therapy. Prompt and accurate detection of bleeding events is essential to prevent serious consequences. As bleeding events are often described in clinical notes, automatic detection of bleeding events from electronic health record (EHR) notes may improve drug-safety surveillance and pharmacovigilance. OBJECTIVE We aimed to develop a natural language processing (NLP) system to automatically classify whether an EHR note sentence contains a bleeding event. METHODS We expert annotated 878 EHR notes (76,577 sentences and 562,630 word-tokens) to identify bleeding events at the sentence level. This annotated corpus was used to train and validate our NLP systems. We developed an innovative hybrid convolutional neural network (CNN) and long short-term memory (LSTM) autoencoder (HCLA) model that integrates a CNN architecture with a bidirectional LSTM (BiLSTM) autoencoder model to leverage large unlabeled EHR data. RESULTS HCLA achieved the best area under the receiver operating characteristic curve (0.957) and F1 score (0.938) to identify whether a sentence contains a bleeding event, thereby surpassing the strong baseline support vector machines and other CNN and autoencoder models. CONCLUSIONS By incorporating a supervised CNN model and a pretrained unsupervised BiLSTM autoencoder, the HCLA achieved high performance in detecting bleeding events.
- 15 April 2018
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessDetection of Bleeding Events in Electronic Health Record Notes Using Convolutional Neural Network Models Enhanced With Recurrent Neural Network Autoencoders: Deep Learning Approach (Preprint)
BACKGROUND Bleeding events are common and critical and may cause significant morbidity and mortality. High incidences of bleeding events are associated with cardiovascular disease in patients on anticoagulant therapy. Prompt and accurate detection of bleeding events is essential to prevent serious consequences. As bleeding events are often described in clinical notes, automatic detection of bleeding events from electronic health record (EHR) notes may improve drug-safety surveillance and pharmacovigilance. OBJECTIVE We aimed to develop a natural language processing (NLP) system to automatically classify whether an EHR note sentence contains a bleeding event. METHODS We expert annotated 878 EHR notes (76,577 sentences and 562,630 word-tokens) to identify bleeding events at the sentence level. This annotated corpus was used to train and validate our NLP systems. We developed an innovative hybrid convolutional neural network (CNN) and long short-term memory (LSTM) autoencoder (HCLA) model that integrates a CNN architecture with a bidirectional LSTM (BiLSTM) autoencoder model to leverage large unlabeled EHR data. RESULTS HCLA achieved the best area under the receiver operating characteristic curve (0.957) and F1 score (0.938) to identify whether a sentence contains a bleeding event, thereby surpassing the strong baseline support vector machines and other CNN and autoencoder models. CONCLUSIONS By incorporating a supervised CNN model and a pretrained unsupervised BiLSTM autoencoder, the HCLA achieved high performance in detecting bleeding events.