Quefeng Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Quefeng Li has a total of 134 co-authors. Here are the top co-authors by number of shared publications:
- Anil K. Gehi
University of North Carolina at Chapel Hill
3 shared publications - Hillary M. Heiling
Department of Data Sciences, Dana-Farber Cancer Institute, Boston, MA, USA
3 shared publications - Jiawei Zhou
Division of Pharmacotherapy and Experimental Therapeutics, Eshelman School of Pharmacy University of North Carolina at Chapel Hill Chapel Hill North Carolina USA
3 shared publications - Joseph G. Ibrahim
University of North Carolina at Chapel Hill
3 shared publications - Lindsey Rosman
Department of Medicine, Division of Cardiology, University of North Carolina at Chapel Hill , Chapel Hill, NC
3 shared publications - Xiaojing Zheng
Department of Pediatrics, University of North Carolina at Chapel Hill , Chapel Hill, North Carolina
3 shared publications
Latest Publications
- Open Access
- 20 November 2026
- Published by Taylor & Francis in Statistical Theory and Related Fields
- p. 1-22
Inference on the significance of modalities in multimodal generalized linear models
- 20 November 2026
- Published by Taylor & Francis in Statistical Theory and Related Fields
- p. 1-22
Open AccessInference on the significance of modalities in multimodal generalized linear models
- 12 March 2026
- Published by Wiley in Clinical Pharmacology & Therapeutics
- Vol. 119 (6) , 1638-1649
Quantitative Lesion‐Level Modeling Reveals Organ‐Dependent Therapeutic Response
Conventional RECIST criteria compress heterogeneous metastatic lesions into a single sum‐of‐diameters endpoint, which can obscure organ‐specific pharmacologic effects on response durability and thereby limit the interpretability of efficacy signals used for model‐informed drug development and clinical decision making. We developed a lesion‐level modeling framework to quantify compartment‐dependent therapeutic effects using routinely collected RECIST diameters. Lesion trajectories were characterized using nonlinear mixed‐effects modeling to estimate regression/kill (kkill), regrowth/progression (kge), and a resistance‐like fraction (Fx). We then fit pan‐cancer multivariable Cox models (adjusted for cancer lineage, treatment class, and clinical covariates) to quantify the independent contribution of metastatic site to response and progression hazards. Organ site emerged as a consistent determinant of response durability across lineages. Liver metastases frequently showed early shrinkage yet higher progression risk, whereas bone metastases, particularly in prostate cancer, tended to regress more slowly but exhibited more durable control once response occurred. Inter‐lesion heterogeneity in progression dynamics was greatest in breast and non‐small cell lung cancer and lowest in colorectal and pancreatic cancer. To contextualize site effects, we explored literature‐derived estimates of Vascular Perfusion and Leakiness Index (VaPLI) for each organ, along with a broad immune surveillance classification. VaPLI correlated with response probability (ρ = 0.44; P < 0.01), and immune‐tolerant sites showed higher progression hazards ( P < 0.05). With further verification, this analysis can support lesion site‐stratified efficacy evaluation by distinguishing true treatment effects from site‐driven bias, enabling more balanced interpretation of treatment effects and dose–response relationships.
- 12 March 2026
- Published by Wiley in Clinical Pharmacology & Therapeutics
- Vol. 119 (6) , 1638-1649
Quantitative Lesion‐Level Modeling Reveals Organ‐Dependent Therapeutic Response
Conventional RECIST criteria compress heterogeneous metastatic lesions into a single sum‐of‐diameters endpoint, which can obscure organ‐specific pharmacologic effects on response durability and thereby limit the interpretability of efficacy signals used for model‐informed drug development and clinical decision making. We developed a lesion‐level modeling framework to quantify compartment‐dependent therapeutic effects using routinely collected RECIST diameters. Lesion trajectories were characterized using nonlinear mixed‐effects modeling to estimate regression/kill (kkill), regrowth/progression (kge), and a resistance‐like fraction (Fx). We then fit pan‐cancer multivariable Cox models (adjusted for cancer lineage, treatment class, and clinical covariates) to quantify the independent contribution of metastatic site to response and progression hazards. Organ site emerged as a consistent determinant of response durability across lineages. Liver metastases frequently showed early shrinkage yet higher progression risk, whereas bone metastases, particularly in prostate cancer, tended to regress more slowly but exhibited more durable control once response occurred. Inter‐lesion heterogeneity in progression dynamics was greatest in breast and non‐small cell lung cancer and lowest in colorectal and pancreatic cancer. To contextualize site effects, we explored literature‐derived estimates of Vascular Perfusion and Leakiness Index (VaPLI) for each organ, along with a broad immune surveillance classification. VaPLI correlated with response probability (ρ = 0.44; P < 0.01), and immune‐tolerant sites showed higher progression hazards ( P < 0.05). With further verification, this analysis can support lesion site‐stratified efficacy evaluation by distinguishing true treatment effects from site‐driven bias, enabling more balanced interpretation of treatment effects and dose–response relationships.
- Open Access
- 17 February 2026
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 15 (4) , e044151
Assessing Potential Benefit of Stroke Prophylaxis Therapy in Atrial Fibrillation and Atrial Flutter Using a Novel Competing Risk Scoring Tool
Background: The competing risk of nonstroke mortality may limit the potential benefit of stroke prophylaxis therapy in patients with atrial fibrillation or atrial flutter AF. Methods: Using a Medicare 20% sample, we identified a cohort of beneficiaries diagnosed with atrial fibrillation or atrial flutter from 2006 to 2019 using International Classification of Diseases, Ninth Revision ( ICD‐9 ) and Tenth Revision ( ICD‐10 ) codes. Fine and Gray regression analysis determined the hazard of stroke with a competing risk of nonstroke mortality, and Cox proportional hazard analysis determined risk of nonstroke mortality. A scoring tool stratified patients into low or high potential benefit for thromboembolic prophylaxis. Results: Among a total of 1 883 759 Medicare beneficiaries, 330 136 patients were included with median age of 79.7 years. 54% of patients had prior bleeding episodes. The median CHA 2 DS 2 ‐VASc score was 5. Of these patients, 211 791 (64%) died and 77 717 (24%) experienced an embolic stroke over median follow‐up of 7.3 years. In the high potential benefit group (26.1%), the risk of stroke was much higher than the risk of nonstroke mortality at 1 year (12.2% versus 7.1%); 3 years (22.7% versus 16.9%); 5 years (31.9% versus 26.1%). In the low potential benefit group (73.9%), the risk of stroke was much lower than the risk of nonstroke mortality at 1 year (11.5% versus 39.2%); 3 years (23.2 versus 57.2%); and 5 years (34.3% versus 69.4%). Conclusions: We propose a scoring tool to identify the potential benefit of thromboembolic prophylaxis therapy in older patients diagnosed with atrial fibrillation or atrial flutter. This tool can be used in shared decision‐making settings. Further studies to improve and validate this scoring tool are warranted.
- 17 February 2026
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 15 (4) , e044151
Open AccessAssessing Potential Benefit of Stroke Prophylaxis Therapy in Atrial Fibrillation and Atrial Flutter Using a Novel Competing Risk Scoring Tool
Background: The competing risk of nonstroke mortality may limit the potential benefit of stroke prophylaxis therapy in patients with atrial fibrillation or atrial flutter AF. Methods: Using a Medicare 20% sample, we identified a cohort of beneficiaries diagnosed with atrial fibrillation or atrial flutter from 2006 to 2019 using International Classification of Diseases, Ninth Revision ( ICD‐9 ) and Tenth Revision ( ICD‐10 ) codes. Fine and Gray regression analysis determined the hazard of stroke with a competing risk of nonstroke mortality, and Cox proportional hazard analysis determined risk of nonstroke mortality. A scoring tool stratified patients into low or high potential benefit for thromboembolic prophylaxis. Results: Among a total of 1 883 759 Medicare beneficiaries, 330 136 patients were included with median age of 79.7 years. 54% of patients had prior bleeding episodes. The median CHA 2 DS 2 ‐VASc score was 5. Of these patients, 211 791 (64%) died and 77 717 (24%) experienced an embolic stroke over median follow‐up of 7.3 years. In the high potential benefit group (26.1%), the risk of stroke was much higher than the risk of nonstroke mortality at 1 year (12.2% versus 7.1%); 3 years (22.7% versus 16.9%); 5 years (31.9% versus 26.1%). In the low potential benefit group (73.9%), the risk of stroke was much lower than the risk of nonstroke mortality at 1 year (11.5% versus 39.2%); 3 years (23.2 versus 57.2%); and 5 years (34.3% versus 69.4%). Conclusions: We propose a scoring tool to identify the potential benefit of thromboembolic prophylaxis therapy in older patients diagnosed with atrial fibrillation or atrial flutter. This tool can be used in shared decision‐making settings. Further studies to improve and validate this scoring tool are warranted.
- Open Access
- 29 March 2025
- Published by Oxford University Press (OUP) in NAR Genomics and Bioinformatics
- Vol. 7 (2) , lqaf046
ARTdeConv: adaptive regularized tri-factor non-negative matrix factorization for cell type deconvolution
Accurate deconvolution of cell types from bulk gene expression is crucial for understanding cellular compositions and uncovering cell-type specific differential expression and physiological states of diseased tissues. Existing deconvolution methods have limitations, such as requiring complete cellular gene expression signatures or neglecting partial biological information. Moreover, these methods often overlook varying cell-type messenger RNA amounts, leading to biased proportion estimates. Additionally, they do not effectively utilize valuable reference information from external studies, such as means and ranges of population cell-type proportions. To address these challenges, we introduce an adaptive regularized tri-factor non-negative matrix factorization approach for deconvolution (ARTdeConv). We rigorously establish the numerical convergence of our algorithm. Through benchmark simulations, we demonstrate the superior performance of ARTdeConv compared to state-of-the-art semi-reference-based and reference-free methods as well as its robustness under challenges to its assumptions. In a real-world application to a dataset from a trivalent influenza vaccine study, our method accurately estimates cellular proportions, as evidenced by the nearly perfect Pearson’s correlation between ARTdeConv estimates and flow cytometry measurements. Moreover, our analysis of ARTdeConv estimates in COVID-19 patients reveals patterns consistent with important immunological phenomena observed in other studies. The proposed method, ARTdeConv, is implemented as an R package and can be accessed on GitHub for researchers and practitioners.
- 29 March 2025
- Published by Oxford University Press (OUP) in NAR Genomics and Bioinformatics
- Vol. 7 (2) , lqaf046
Open AccessARTdeConv: adaptive regularized tri-factor non-negative matrix factorization for cell type deconvolution
Accurate deconvolution of cell types from bulk gene expression is crucial for understanding cellular compositions and uncovering cell-type specific differential expression and physiological states of diseased tissues. Existing deconvolution methods have limitations, such as requiring complete cellular gene expression signatures or neglecting partial biological information. Moreover, these methods often overlook varying cell-type messenger RNA amounts, leading to biased proportion estimates. Additionally, they do not effectively utilize valuable reference information from external studies, such as means and ranges of population cell-type proportions. To address these challenges, we introduce an adaptive regularized tri-factor non-negative matrix factorization approach for deconvolution (ARTdeConv). We rigorously establish the numerical convergence of our algorithm. Through benchmark simulations, we demonstrate the superior performance of ARTdeConv compared to state-of-the-art semi-reference-based and reference-free methods as well as its robustness under challenges to its assumptions. In a real-world application to a dataset from a trivalent influenza vaccine study, our method accurately estimates cellular proportions, as evidenced by the nearly perfect Pearson’s correlation between ARTdeConv estimates and flow cytometry measurements. Moreover, our analysis of ARTdeConv estimates in COVID-19 patients reveals patterns consistent with important immunological phenomena observed in other studies. The proposed method, ARTdeConv, is implemented as an R package and can be accessed on GitHub for researchers and practitioners.
- Open Access
- 4 March 2025
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 14 (5) , e038756
Atrial Fibrillation Treatment Pathway in the Emergency Department Reduces Median 30‐Day Health Service Charges
Background: The economic burden of atrial fibrillation (AF) continues to increase. AF treatment pathways have been shown to reduce avoidable admissions, but the effects on health care costs are not understood. This study sought to assess the impact of an AF treatment pathway on health service charges and emergency department (ED) discharge rates. Methods: An AF treatment pathway was implemented at 7 hospital EDs in North Carolina between 2017 and 2020. Thirty‐day health service charges were calculated for ED visits, hospitalizations, and outpatient clinic appointments. A quasi‐experimental design was used to assess changes in health service charges and ED discharge rates following implementation of the AF treatment pathway. Adjusted quantile and negative binomial regressions were used to analyze changes in median 30‐day health service charges and discharge rates from the ED, respectively. Results: Among 12 504 patients that met eligibility for study inclusion (preperiod: n=3893; postperiod: n=8611), implementation of the AF treatment pathway significantly reduced 30‐day health service charges (preperiod: $11 922; postperiod: $9219; P P =0.04). Additionally, the adjusted predicted probability of ED discharge increased from 65.3% to 70.0% after the AF treatment pathway implementation, a 4.7 percentage point increase (95% CI, 1.4–8.0; P <0.001). Conclusions: The implementation of an AF treatment pathway in the ED was associated with a decrease in median 30‐day health service charges and an increase in ED discharge rate for patients presenting with AF.
- 4 March 2025
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 14 (5) , e038756
Open AccessAtrial Fibrillation Treatment Pathway in the Emergency Department Reduces Median 30‐Day Health Service Charges
Background: The economic burden of atrial fibrillation (AF) continues to increase. AF treatment pathways have been shown to reduce avoidable admissions, but the effects on health care costs are not understood. This study sought to assess the impact of an AF treatment pathway on health service charges and emergency department (ED) discharge rates. Methods: An AF treatment pathway was implemented at 7 hospital EDs in North Carolina between 2017 and 2020. Thirty‐day health service charges were calculated for ED visits, hospitalizations, and outpatient clinic appointments. A quasi‐experimental design was used to assess changes in health service charges and ED discharge rates following implementation of the AF treatment pathway. Adjusted quantile and negative binomial regressions were used to analyze changes in median 30‐day health service charges and discharge rates from the ED, respectively. Results: Among 12 504 patients that met eligibility for study inclusion (preperiod: n=3893; postperiod: n=8611), implementation of the AF treatment pathway significantly reduced 30‐day health service charges (preperiod: $11 922; postperiod: $9219; P P =0.04). Additionally, the adjusted predicted probability of ED discharge increased from 65.3% to 70.0% after the AF treatment pathway implementation, a 4.7 percentage point increase (95% CI, 1.4–8.0; P <0.001). Conclusions: The implementation of an AF treatment pathway in the ED was associated with a decrease in median 30‐day health service charges and an increase in ED discharge rate for patients presenting with AF.
- Open Access
- 27 February 2025
- Published by American Association for Cancer Research (AACR) in Clinical Cancer Research
- Vol. 31 (9) , 1580-1586
Revolutionizing Patient-Reported Outcomes Analysis for Oncology Drug Development Using Population Models
Patient-reported outcomes (PRO) play a crucial role as clinical endpoint in oncology trials. Traditional statistical methods, such as hypothesis testing, have been commonly used by pharmaceutical industry and regulators to evaluate treatment efficacy on PRO endpoints. However, the analysis of PRO data remains challenging because of high variability and missing data issues. In this study, we will present examples in which inappropriate statistical analyses of PRO data can confound treatment efficacy analyses. To overcome these challenges, we propose the application of individual participant data and population models. Population models have been extensively used in pharmacokinetics and pharmacodynamics analyses and are well accepted by regulators. However, their potential in PRO data analyses, particularly in the field of oncology, remains largely untapped. This perspective article aims to highlight the value of population modeling approaches in PRO data analyses for oncology clinicians and researchers. Population models integrate individual participant data and can effectively handle the substantial variability in PRO measurements by incorporating covariates, between-subject variability, and accounting for measurement noise. By leveraging information from the population, this approach also provides accurate estimations for participants with missing data or sparse sampling. Moreover, these models could be applied to predict long-term PRO dynamics. If used appropriately, population modeling approaches could revolutionize the analysis of PRO data in oncology drug development, enabling a more comprehensive understanding of the impact of treatment on patients’ lives. Our aim is to encourage stakeholders to consider population modeling as a standard and effective tool to enhance decision-making and ultimately improve patient care.
- 27 February 2025
- Published by American Association for Cancer Research (AACR) in Clinical Cancer Research
- Vol. 31 (9) , 1580-1586
Open AccessRevolutionizing Patient-Reported Outcomes Analysis for Oncology Drug Development Using Population Models
Patient-reported outcomes (PRO) play a crucial role as clinical endpoint in oncology trials. Traditional statistical methods, such as hypothesis testing, have been commonly used by pharmaceutical industry and regulators to evaluate treatment efficacy on PRO endpoints. However, the analysis of PRO data remains challenging because of high variability and missing data issues. In this study, we will present examples in which inappropriate statistical analyses of PRO data can confound treatment efficacy analyses. To overcome these challenges, we propose the application of individual participant data and population models. Population models have been extensively used in pharmacokinetics and pharmacodynamics analyses and are well accepted by regulators. However, their potential in PRO data analyses, particularly in the field of oncology, remains largely untapped. This perspective article aims to highlight the value of population modeling approaches in PRO data analyses for oncology clinicians and researchers. Population models integrate individual participant data and can effectively handle the substantial variability in PRO measurements by incorporating covariates, between-subject variability, and accounting for measurement noise. By leveraging information from the population, this approach also provides accurate estimations for participants with missing data or sparse sampling. Moreover, these models could be applied to predict long-term PRO dynamics. If used appropriately, population modeling approaches could revolutionize the analysis of PRO data in oncology drug development, enabling a more comprehensive understanding of the impact of treatment on patients’ lives. Our aim is to encourage stakeholders to consider population modeling as a standard and effective tool to enhance decision-making and ultimately improve patient care.
- Top Cited PapersOpen Access
- 6 August 2024
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 13 (15) , e033750
Wearable Devices, Health Care Use, and Psychological Well‐Being in Patients With Atrial Fibrillation
Background: Wearables are increasingly used by patients with atrial fibrillation (AF) for symptom monitoring and health management, but their impact on patient health care use and psychological well‐being is not well understood. Methods and Results: In this retrospective, propensity‐matched study of patients with AF, survey and electronic health record data were merged to compare AF‐specific health care use (outpatient/inpatient visits, rhythm‐related testing, and procedures) and informal health care use (telephone calls and patient portal messages) over a 9‐month period between wearable users and nonusers. We also examined the effects of wearable cardiac monitoring features (eg, heart rate alerts, irregular rhythm notification, and ECG) on patient behavior and well‐being. Of 172 patients with AF in this analysis (age, 72.6±9.0 years; 42% women), 83 used a wearable. Compared with nonusers, wearable users reported higher rates of symptom monitoring and preoccupation ( P =0.03) and more AF treatment concerns ( P =0.02). Moreover, 20% of wearable users experienced anxiety and always contacted their doctors in response to irregular rhythm notifications. After matching, AF‐specific health care use was significantly greater among wearable users compared with nonusers ( P =0.04), including significantly higher rates of ECGs, echocardiograms/transesophageal echocardiogram, and ablation. Wearable users were also significantly more likely to use informal health care resources compared with nonusers ( P =0.05). Conclusions: Wearables were associated with higher rates of symptom monitoring and preoccupation, AF treatment concerns, AF‐specific health care use, and use of informal health care resources. Prospective, randomized studies are needed to understand the net effects of wearables and their alerts on patients, providers, and the health care system.
- 6 August 2024
- Published by Wolters Kluwer Health in Journal of the American Heart Association
- Vol. 13 (15) , e033750
Top Cited PapersOpen AccessWearable Devices, Health Care Use, and Psychological Well‐Being in Patients With Atrial Fibrillation
Background: Wearables are increasingly used by patients with atrial fibrillation (AF) for symptom monitoring and health management, but their impact on patient health care use and psychological well‐being is not well understood. Methods and Results: In this retrospective, propensity‐matched study of patients with AF, survey and electronic health record data were merged to compare AF‐specific health care use (outpatient/inpatient visits, rhythm‐related testing, and procedures) and informal health care use (telephone calls and patient portal messages) over a 9‐month period between wearable users and nonusers. We also examined the effects of wearable cardiac monitoring features (eg, heart rate alerts, irregular rhythm notification, and ECG) on patient behavior and well‐being. Of 172 patients with AF in this analysis (age, 72.6±9.0 years; 42% women), 83 used a wearable. Compared with nonusers, wearable users reported higher rates of symptom monitoring and preoccupation ( P =0.03) and more AF treatment concerns ( P =0.02). Moreover, 20% of wearable users experienced anxiety and always contacted their doctors in response to irregular rhythm notifications. After matching, AF‐specific health care use was significantly greater among wearable users compared with nonusers ( P =0.04), including significantly higher rates of ECGs, echocardiograms/transesophageal echocardiogram, and ablation. Wearable users were also significantly more likely to use informal health care resources compared with nonusers ( P =0.05). Conclusions: Wearables were associated with higher rates of symptom monitoring and preoccupation, AF treatment concerns, AF‐specific health care use, and use of informal health care resources. Prospective, randomized studies are needed to understand the net effects of wearables and their alerts on patients, providers, and the health care system.
- Open Access
glmmPen: High Dimensional Penalized Generalized Linear Mixed Models
Generalized linear mixed models (GLMMs) are widely used in research for their ability to model correlated outcomes with non-Gaussian conditional distributions. The proper selection of fixed and random effects is a critical part of the modeling process since model misspecification may lead to significant bias. However, the joint selection of fixed and random effects has historically been limited to lower-dimensional GLMMs, largely due to the use of criterion-based model selection strategies. Here we present the R package glmmPen, one of the first to select fixed and random effects in higher dimension using a penalized GLMM modeling framework. Model parameters are estimated using a Monte Carlo Expectation Conditional Minimization (MCECM) algorithm, which leverages Stan and RcppArmadillo for increased computational efficiency. Our package supports the Binomial, Gaussian, and Poisson families and multiple penalty functions. In this manuscript we discuss the modeling procedure, estimation scheme, and software implementation through application to a pancreatic cancer subtyping study. Simulation results show our method has good performance in selecting both the fixed and random effects in high dimensional GLMMs.
Open AccessglmmPen: High Dimensional Penalized Generalized Linear Mixed Models
Generalized linear mixed models (GLMMs) are widely used in research for their ability to model correlated outcomes with non-Gaussian conditional distributions. The proper selection of fixed and random effects is a critical part of the modeling process since model misspecification may lead to significant bias. However, the joint selection of fixed and random effects has historically been limited to lower-dimensional GLMMs, largely due to the use of criterion-based model selection strategies. Here we present the R package glmmPen, one of the first to select fixed and random effects in higher dimension using a penalized GLMM modeling framework. Model parameters are estimated using a Monte Carlo Expectation Conditional Minimization (MCECM) algorithm, which leverages Stan and RcppArmadillo for increased computational efficiency. Our package supports the Binomial, Gaussian, and Poisson families and multiple penalty functions. In this manuscript we discuss the modeling procedure, estimation scheme, and software implementation through application to a pancreatic cancer subtyping study. Simulation results show our method has good performance in selecting both the fixed and random effects in high dimensional GLMMs.
- Open Access
- 29 January 2024
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 80 (1)
Efficient computation of high-dimensional penalized generalized linear mixed models by latent factor modeling of the random effects
Modern biomedical datasets are increasingly high-dimensional and exhibit complex correlation structures. Generalized linear mixed models (GLMMs) have long been employed to account for such dependencies. However, proper specification of the fixed and random effects in GLMMs is increasingly difficult in high dimensions, and computational complexity grows with increasing dimension of the random effects. We present a novel reformulation of the GLMM using a factor model decomposition of the random effects, enabling scalable computation of GLMMs in high dimensions by reducing the latent space from a large number of random effects to a smaller set of latent factors. We also extend our prior work to estimate model parameters using a modified Monte Carlo Expectation Conditional Minimization algorithm, allowing us to perform variable selection on both the fixed and random effects simultaneously. We show through simulation that through this factor model decomposition, our method can fit high-dimensional penalized GLMMs faster than comparable methods and more easily scale to larger dimensions not previously seen in existing approaches.
- 29 January 2024
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 80 (1)
Open AccessEfficient computation of high-dimensional penalized generalized linear mixed models by latent factor modeling of the random effects
Modern biomedical datasets are increasingly high-dimensional and exhibit complex correlation structures. Generalized linear mixed models (GLMMs) have long been employed to account for such dependencies. However, proper specification of the fixed and random effects in GLMMs is increasingly difficult in high dimensions, and computational complexity grows with increasing dimension of the random effects. We present a novel reformulation of the GLMM using a factor model decomposition of the random effects, enabling scalable computation of GLMMs in high dimensions by reducing the latent space from a large number of random effects to a smaller set of latent factors. We also extend our prior work to estimate model parameters using a modified Monte Carlo Expectation Conditional Minimization algorithm, allowing us to perform variable selection on both the fixed and random effects simultaneously. We show through simulation that through this factor model decomposition, our method can fit high-dimensional penalized GLMMs faster than comparable methods and more easily scale to larger dimensions not previously seen in existing approaches.
- Open Access
- 1 January 2024
- Published by Statistica Sinica (Institute of Statistical Science) in Statistica Sinica
- Vol. 34 (2) , 527-546
Multi-response Regression for Block-missing Multi-modal Data without Imputation
Multi-modal data are prevalent in many scientific fields. In this study, we consider the parameter estimation and variable selection for a multi-response regression using block-missing multi-modal data. Our method allows the dimensions of both the responses and the predictors to be large, and the responses to be incomplete and correlated, a common practical problem in high-dimensional settings. Our proposed method uses two steps to make a prediction from a multi-response linear regression model with block-missing multi-modal predictors. In the first step, without imputing missing data, we use all available data to estimate the covariance matrix of the predictors and the cross-covariance matrix between the predictors and the responses. In the second step, we use these matrices and a penalized method to simultaneously estimate the precision matrix of the response vector, given the predictors, and the sparse regression parameter matrix. Lastly, we demonstrate the effectiveness of the proposed method using theoretical studies, simulated examples, and an analysis of a multi-modal imaging data set from the Alzheimer’s Disease Neuroimaging Initiative.
- 1 January 2024
- Published by Statistica Sinica (Institute of Statistical Science) in Statistica Sinica
- Vol. 34 (2) , 527-546
Open AccessMulti-response Regression for Block-missing Multi-modal Data without Imputation
Multi-modal data are prevalent in many scientific fields. In this study, we consider the parameter estimation and variable selection for a multi-response regression using block-missing multi-modal data. Our method allows the dimensions of both the responses and the predictors to be large, and the responses to be incomplete and correlated, a common practical problem in high-dimensional settings. Our proposed method uses two steps to make a prediction from a multi-response linear regression model with block-missing multi-modal predictors. In the first step, without imputing missing data, we use all available data to estimate the covariance matrix of the predictors and the cross-covariance matrix between the predictors and the responses. In the second step, we use these matrices and a penalized method to simultaneously estimate the precision matrix of the response vector, given the predictors, and the sparse regression parameter matrix. Lastly, we demonstrate the effectiveness of the proposed method using theoretical studies, simulated examples, and an analysis of a multi-modal imaging data set from the Alzheimer’s Disease Neuroimaging Initiative.