JCJinying Chen

Jinying Chen

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Jinying Chen has a total of 153 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Linguistic features from constrained spontaneous speech as potential markers of preclinical Alzheimer’s disease

    by Haochun Huang, Zexu Li, Christina B. Young, Viktorija Smith, Ting Fang Alvin Ang, Vijaya B. Kolachalama, Rhoda Au, Jinying Chen

  • Open Access

    Detecting Misspelled Drug Names Using Transformer-Based Language Models: Model Development and External Validation

    by Jiayu Lu, Kevin W McConeghy, Andrew R Zullo, Jinying Chen

    Background: Misspellings in medication names can compromise patient safety, reduce data utility, and impede large-scale data initiatives that integrate medication information from electronic health records (EHRs). Existing methods for detecting misspelled medical terms are mostly dictionary-based and can lead to high false-positive rates when correctly spelled but previously unseen (out-of-vocabulary) terms are encountered. Objective: We aimed to develop and validate domain-specific, transformer-based language models for detecting misspelled drug names, with an emphasis on performance for unseen terms. Methods: Using RxNorm as a standardized drug vocabulary, we created an RxNorm-augmented training corpus and developed two BERT (Bidirectional Encoder Representations from Transformers)–based models—BERTDrug and CharBERTDrug—for misspelling detection. Specifically, we randomly split 69,824 RxNorm drug names into training, development, and test sets (3:1:1) and generated k misspellings per name using text-perturbation techniques (k optimized for training; fixed at 1 for development and test sets). The models were fine-tuned on the training and development sets and evaluated using the RxNorm test set and 3586 drug names from the Long-Term Care Data Cooperative (LTCDC) database (external validation). The RxNorm test set and out-of-vocabulary LTCDC dataset (1922 terms), neither overlapping with the RxNorm training data, were used to evaluate performance on unseen terms. SpellChecker served as a dictionary-based baseline, while fastTextML and BioWordVecML, which used different subword embeddings as inputs for machine learning, served as additional baselines. Additionally, we compared model performance with GPT-4o, a generative large language model (LLM), using 2200 randomly sampled test terms. Results: On the RxNorm test set, BERTDrug and CharBERTDrug outperformed the baseline models across most metrics. BERTDrug achieved the best overall performance (F1-score=0.859; area under the receiver operating characteristic curve [ROC-AUC]=0.947), followed by CharBERTDrug (F1-score=0.833; ROC-AUC=0.906). Both models also outperformed the baseline models on the out-of-vocabulary LTCDC dataset across most metrics, with CharBERTDrug performing best (F1-score=0.696; ROC-AUC=0.788), followed by BERTDrug (F1-score=0.669; ROC-AUC=0.786). In the secondary analysis, both models exceeded GPT-4o on most metrics (except Recall) for 2000 RxNorm terms. BERTDrug performed best (ROC-AUC=0.951; F1-score=0.855), followed by CharBERTDrug (ROC-AUC=0.911; F1-score=0.831) and GPT-4o (ROC-AUC=0.856; F1-score=0.721). In contrast, among 200 randomly selected LTCDC terms, GPT-4o performed best on most metrics except precision and specificity. Conclusions: Domain-specific language models improved detection of misspellings in out-of-vocabulary drug names and outperformed baseline models in both internal and external evaluations. The comparison with a generative LLM suggests that domain shift may substantially reduce the advantages conferred by domain-specific training. With further fine-tuning on diverse data that capture the terminology, formatting conventions, and spelling patterns encountered across real-world clinical settings, these models could be adapted for use in other clinical databases and EHR systems to improve medication data quality for research and to support future safety-focused applications.

  • Open Access

    Alzheimer’s disease risk prediction from clinical and social determinants of health: a machine learning cohort study in UK Biobank

    by Junming Hu, Simon Lu, Qi Zhang, Kehan Qian, Habbiburr Rehman, Congcong Zhu, John Farrell, Julio E Castrillon-Candas, Rhoda Au, Lindsay A Farrer, et al.

    Objectives: Social determinants of health (SDOH) may improve Alzheimer’s disease (AD) risk prediction by capturing upstream contextual risk beyond routinely measured clinical variables. We aimed to develop and validate an accurate, interpretable machine-learning pipeline for AD risk prediction in UK Biobank using routinely collected data. Methods: Using data from 13 076 participants in the UK Biobank, we developed an automated machine-learning pipeline for AD risk prediction with feature selection and a C5.0 boosted-tree classifier. Data were split into training, development and test sets (7:2:1); missing values were imputed in the training data only, and feature selection, tuning and threshold calibration were performed using the training/development data, with final evaluation on the independent test set. Internal validation used repeated subsampling without replacement. Results: During up to 16 years of follow-up, 927 participants developed AD. Feature selection reduced 3590 variables to 26 predictors spanning age, APOE4, SDOH, medical history and routine clinical measures. The final model showed good discrimination (area under the precision–recall curve 0.89) and adequate calibration (Hosmer-Lemeshow p=0.71), with stable performance under repeated subsampling. Sex-stratified models showed similar patterns. Discussion: SDOH contributed useful predictive information, but their associations should be interpreted as predictive rather than causal and may reflect socioeconomic confounding and healthcare access. Conclusions: This model could support scalable AD risk screening using routinely collected data, but external validation and recalibration in non-UK populations are needed before broader application.

  • Open Access

    Gender differences in informal caregiving costs and burden in Alzheimer disease: a multinational cross-sectional analysis

    by Shubhi S. Jain, Ting Fang Alvin Ang, Preeti Sunderaraman, Rhoda Au, Jinying Chen

    Informal caregiving is a major driver of the societal impact of Alzheimer’s disease (AD) and has important implications for caregiver health, workfor


  • Open Access

    Detecting Misspelled Drug Names Using Transformer-Based Language Models: Model Development and External Validation (Preprint)

    by Jiayu Lu, Kevin W McConeghy, Andrew R Zullo, Jinying Chen

    BACKGROUND Misspellings in medication names can compromise patient safety, reduce data utility, and impede large-scale data initiatives that integrate medication information from electronic health records (EHRs). Existing methods for detecting misspelled medical terms are mostly dictionary-based and can lead to high false-positive rates when correctly spelled but previously unseen (out-of-vocabulary) terms are encountered. OBJECTIVE We aimed to develop and validate domain-specific, transformer-based language models for detecting misspelled drug names, with an emphasis on performance for unseen terms. METHODS Using RxNorm as a standardized drug vocabulary, we created an RxNorm-augmented training corpus and developed two BERT (Bidirectional Encoder Representations from Transformers)–based models—BERTDrug and CharBERTDrug—for misspelling detection. Specifically, we randomly split 69,824 RxNorm drug names into training, development, and test sets (3:1:1) and generated k misspellings per name using text-perturbation techniques (k optimized for training; fixed at 1 for development and test sets). The models were fine-tuned on the training and development sets and evaluated using the RxNorm test set and 3586 drug names from the Long-Term Care Data Cooperative (LTCDC) database (external validation). The RxNorm test set and out-of-vocabulary LTCDC dataset (1922 terms), neither overlapping with the RxNorm training data, were used to evaluate performance on unseen terms. SpellChecker served as a dictionary-based baseline, while fastTextML and BioWordVecML, which used different subword embeddings as inputs for machine learning, served as additional baselines. Additionally, we compared model performance with GPT-4o, a generative large language model (LLM), using 2200 randomly sampled test terms. RESULTS On the RxNorm test set, BERTDrug and CharBERTDrug outperformed the baseline models across most metrics. BERTDrug achieved the best overall performance (F1-score=0.859; area under the receiver operating characteristic curve [ROC-AUC]=0.947), followed by CharBERTDrug (F1-score=0.833; ROC-AUC=0.906). Both models also outperformed the baseline models on the out-of-vocabulary LTCDC dataset across most metrics, with CharBERTDrug performing best (F1-score=0.696; ROC-AUC=0.788), followed by BERTDrug (F1-score=0.669; ROC-AUC=0.786). In the secondary analysis, both models exceeded GPT-4o on most metrics (except Recall) for 2000 RxNorm terms. BERTDrug performed best (ROC-AUC=0.951; F1-score=0.855), followed by CharBERTDrug (ROC-AUC=0.911; F1-score=0.831) and GPT-4o (ROC-AUC=0.856; F1-score=0.721). In contrast, among 200 randomly selected LTCDC terms, GPT-4o performed best on most metrics except precision and specificity. CONCLUSIONS Domain-specific language models improved detection of misspellings in out-of-vocabulary drug names and outperformed baseline models in both internal and external evaluations. The comparison with a generative LLM suggests that domain shift may substantially reduce the advantages conferred by domain-specific training. With further fine-tuning on diverse data that capture the terminology, formatting conventions, and spelling patterns encountered across real-world clinical settings, these models could be adapted for use in other clinical databases and EHR systems to improve medication data quality for research and to support future safety-focused applications.

  • Open Access

    Validating Digital Linguistic Features as Potential Biomarkers of Alzheimer's Disease

    by Haochun Huang, Zexu Li, Christina B. Young, Viktorija Smith, Ting Fang Alvin Ang, Vijaya B. Kolachalama, Rhoda Au, Jinying Chen

    Background: Speech features derived from verbal responses to cognitive tests have been shown to indicate mild cognitive impairment and later risk of Alzheimer's Disease (AD). However, research in validating speech features using established AD biomarkers remains limited. This study aims to validate linguistic features from digital audio recording against AD biomarkers. Method: We analyzed data from participants of the Framingham Heart Study who were cognitively intact and had (1) Tau PET and amyloid PET scans conducted on the same date or within a 3‐month interval; (2) a logical memory delayed recall (LMd) test within one year of the Tau PET scan; and (3) manually transcribed response from the LMd test. Using natural language processing methods and spaCy Python library, we extracted 52 linguistic features from transcribed LMd responses, which included measures of lexical density, syntactic complexity, and speech fluency. We validated these features against AD biomarkers: (1) beta‐amyloid status (+/‐) derived through Gaussian mixture modeling and (2) Tau PET signals in five brain regions: amygdala, entorhinal, inferior parietal (IP), inferior temporal (IT), and precuneus. First, we used multiple linear regression to assess the association between each feature (dependent variable) and the biomarker status (independent variable), adjusting for age, sex, and education. Features associated with an AD biomarker ( p p = 0.03) and higher Tau burden in entorhinal (Beta:‐0.148, p = 0.03), IT (Beta:‐0.143, p = 0.04), and Amygdala (Beta:‐0.139, p = 0.04). Reduced syntactic complexity (i.e., lower Yngve_avg scores) were associated with higher Tau burden in IP (Beta:‐0.168, p = 0.02), and IT (Beta:‐0.145, p = 0.04). Longer between‐utterance pause duration was positively associated with higher Tau burden in IP (Beta:0.322, p p = 0.01). Conclusion: Linguistic‐based speech features are associated with amyloid positivity and tau accumulation and can potentially serve as digital biomarkers for preclinical AD but need further validation.

  • Open Access

    A natural language processing approach to support biomedical data harmonization: Leveraging large language models

    by Zexu Li, Suraj P. Prabhu, Zachary T. Popp, Shubhi S. Jain, Vijetha Balakundi, Ting Fang Alvin Ang, Rhoda Au, Jinying Chen

    Background: Biomedical research requires large, diverse samples to produce unbiased results. Retrospective data harmonization is often used to integrate existing datasets to create these samples, but the process is labor-intensive. Automated methods for matching variables across datasets can accelerate this process, particularly when harmonizing datasets with numerous variables and varied naming conventions. Research in this area has been limited, primarily focusing on lexical matching and ontology-based semantic matching. We aimed to develop new methods, leveraging large language models (LLMs) and ensemble learning, to automate variable matching. Methods: This study utilized data from two GERAS cohort studies (European [EU] and Japan [JP]) obtained through the Alzheimer’s Disease (AD) Data Initiative’s AD workbench. We first manually created a dataset by matching 347 EU variables with 1322 candidate JP variables and treated matched variable pairs as positive instances and unmatched pairs as negative instances. We then developed four natural language processing (NLP) methods using state-of-the-art LLMs (E5, MPNet, MiniLM, and BioLORD-2023) to estimate variable similarity based on variable labels and derivation rules. A lexical matching method using fuzzy matching was included as a baseline model. In addition, we developed an ensemble-learning method, using the Random Forest (RF) model, to integrate individual NLP methods. RF was trained and evaluated on 50 trials. Each trial had a random split (4:1) of training and test sets, with the model’s hyperparameters optimized through cross-validation on the training set. For each EU variable, 1322 candidate JP variables were ranked based on NLP-derived similarity scores or RF’s probability scores, denoting their likelihood to match the EU variable. Ranking performance was measured by top-n hit ratio (HR-n) and mean reciprocal rank (MRR). Results: E5 performed best among individual methods, achieving 0.898 HR-30 and 0.700 MRR. RF performed better than E5 on all metrics over 50 trials (P < 0.001) and achieved an average HR-30 of 0.986 and MRR of 0.744. LLM-derived features contributed most to RF’s performance. One major cause of errors in automatic variable matching was ambiguous variable definitions. Conclusion: NLP techniques (especially LLMs), combined with ensemble learning, hold great potential in automating variable matching and accelerating biomedical data harmonization.


  • Open Access

    Electronic health record activity changes around new decision support implementation: monitoring using audit logs and topic modeling

    by Jinying Chen, Sarah L Cutrona, Ajay Dharod, Adam Moses, Aaron Bridges, Brian Ostasiewski, Kristie L Foley, Thomas K Houston, iDAPT (Implementation & Informatics Developing Adaptable Processes and Technologies for Cancer Control) Implementation Science Center for Cancer Control

    Objectives: To develop and test a novel machine learning approach for monitoring impact of computerized clinical decision support (CDS) tools on clinicians’ electronic health record (EHR) activities. Materials and Methods: Our CDS monitoring approach leverages topic modeling, a latent-variable statistical machine learning method, to infer health providers’ EHR activities from EHR audit logs. We applied this approach to monitor the impact of a tobacco cessation support CDS tool newly implemented in 5 cancer clinics (2018-2021). We trained the topic model on EHR audit log data from 3445 encounters (pre-CDS-implementation: 1734, post-CDS-implementation: 1711) for patients with active smoking status. The number of topics was automatically determined based on within-topic coherence and across-topic divergence, and the identified topics were assigned clinically relevant EHR activity labels by 4 domain experts. Results: The topic model identified 2 distinct activities focusing on CDS (act on CDS, bypass/postpone CDS), 2 activities related to CDS (review patient records and address alerts, use note templates and acknowledge the completion of CDS), 6 related to accessing (access patient station) and reviewing patient data (external records, synopsis data, snapshot of patient data, problem list/diagnosis/notes, treatment plan), and 4 related to modifying EHR (modify diagnosis/problem lists, document visit with record review, perform administrative activities for visit and billing, and document follow-up care plan). Comparing matched 1-hour after-check-in windows post-implementation (n = 841) versus pre-implementation (n = 841) of CDS, the mean prevalence (expressed as proportions out of 1.0) of providers’ EHR-use activity increased on CDS-focused activities (0.073, 95% CI, 0.066-0.079) and CDS-related activities (0.098, 95% CI, 0.089-0.106) and decreased on modifying EHR (−0.113, 95% CI, −0.124 to −0.102) and reviewing patient data (−0.058, 95% CI, −0.072 to −0.044). Discussion: Our topic model-based CDS monitoring approach can identify shifts in prevalence of EHR-use activities pre-implementation versus post-implementation. This approach can be applied to detect unintended changes in EHR activities on a large population scale following CDS implementation, providing valuable insights to guide focused qualitative investigations for CDS improvement or de-implementation. Conclusion: Our approach offers a scalable, data-driven framework for evaluating the real-world impact of EHR-embedded CDS tools. Built on a generic machine learning framework, this approach could be adapted to explore impact of other healthcare quality improvement strategies using EHR-integrated CDS interventions.


  • Association between education and rate of cognitive decline among individuals with Alzheimer's disease: A multi-national European observational study

    by Zachary T Popp, Ting Fang Alvin Ang, Phillip H Hwang, Rhoda Au, Jinying Chen

    Background: The cognitive reserve (CR) hypothesis presumes higher tolerance of Alzheimer's disease (AD)-related pathology without functional decline for those with high education and more rapid decline after AD onset. Evidence supporting the second part of the hypothesis has been largely confined to U.S.-based studies. Objective: To assess the relationship between education and cognitive decline in a multi-national European cohort of older adults living with AD. Methods: We analyzed data from participants recruited into the GERAS-EU cohort study from AD clinics in the United Kingdom, Germany, and France. Linear mixed models were employed to assess the relationship between education (dichotomized using a 12-year cutoff) and cognitive decline measured by Mini-Mental State Examination (MMSE) scores during 1.5 to 3 years of follow-up, adjusting for age, sex, time from formal diagnosis, country, comorbidities, and AD treatment. Results: A total of 1313 participants were analyzed, with mean age of 77.3 years (SD = 7.6), 715 (54.5%) females, and 378 (28.8%) with high education (≥12 years). Participants with high education experienced a 0.19-point greater decline (versus low education group) in MMSE scores every 6 months during follow-up (95% Confidence Interval: 0.03–0.35, p = 0.02). The secondary analyses (stratified by disease severity, sex, or country) showed a consistent direction of the association, although only significant in the severe AD group (p = 0.01). Conclusions: Our findings provide partial support for the CR hypothesis. Delayed AD diagnosis in individuals with high education may contribute to faster decline after diagnosis, highlighting the importance of sensitive screening for early signs of cognitive impairment.


  • Open Access

    Implementation of Health IT for Cancer Screening in US Primary Care: Scoping Review

    by Constance Owens-Jasey, Jinying Chen, Ran Xu, Heather Angier, Amy G Huebschmann, Mayuko Ito Fukunaga, Krisda H Chaiyachati, Katharine A Rendle, Kim Robien, Lisa DiMartino, et al.

    Background: A substantial percentage of the US population is not up to date on guideline-recommended cancer screenings. Identifying interventions that effectively improve screening rates would enhance the delivery of such screening. Interventions involving health IT (HIT) show promise, but much remains unknown about how HIT is optimized to support cancer screening in primary care. Objective: This scoping review aims to identify (1) HIT-based interventions that effectively support guideline concordance in breast, cervical, and colorectal cancer screening provision and follow-up in the primary care setting and (2) barriers or facilitators to the implementation of effective HIT in this setting. Methods: Following scoping review guidelines, we searched MEDLINE, CINAHL Plus, Web of Science, and IEEE Xplore databases for US-based studies from 2015 to 2021 that featured HIT targeting breast, colorectal, and cervical cancer screening in primary care. Studies were dual screened using a review criteria checklist. Data extraction was guided by the following implementation science frameworks: the Reach, Effectiveness, Adoption, Implementation, and Maintenance framework; the Expert Recommendations for Implementing Change taxonomy; and implementation strategy reporting domains. It was also guided by the Integrated Technology Implementation Model that incorporates theories of both implementation science and technology adoption. Reporting was guided by PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews). Results: A total of 101 studies met the inclusion criteria. Most studies (85/101, 84.2%) involved electronic health record–based HIT interventions. The most common HIT function was clinical decision support, primarily used for panel management or at the point of care. Most studies related to HIT targeting colorectal cancer screening (83/101, 82.2%), followed by studies related to breast cancer screening (28/101, 27.7%), and cervical cancer screening (19/101, 18.8%). Improvements in cancer screening were associated with HIT-based interventions in most studies (36/54, 67% of colorectal cancer–relevant studies; 9/14, 64% of breast cancer–relevant studies; and 7/10, 70% of cervical cancer–relevant studies). Most studies (79/101, 78.2%) reported on the reach of certain interventions, while 17.8% (18/101) of the included studies reported on the adoption or maintenance. Reported barriers and facilitators to HIT adoption primarily related to inner context factors of primary care settings (eg, staffing and organizational policies that support or hinder HIT adoption). Implementation strategies for HIT adoption were reported in 23.8% (24/101) of the included studies. Conclusions: There are substantial evidence gaps regarding the effectiveness of HIT-based interventions, especially those targeting guideline-concordant breast and colorectal cancer screening in primary care. Even less is known about how to enhance the adoption of technologies that have been proven effective in supporting breast, colorectal, or cervical cancer screening. Research is needed to ensure that the potential benefits of effective HIT-based interventions equitably reach diverse primary care populations.