Zhenke Wu
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
Research Subjects
Zhenke Wu has a total of 240 co-authors. Here are the top co-authors by number of shared publications:
- Srijan Sen
Michigan Neuroscience Institute, University of Michigan, Ann Arbor
11 shared publications - Jennifer A. Miner
University of Michigan
9 shared publications - Noelle E. Carlozzi
University of Michigan
9 shared publications - Sung Won Choi
Department of Pediatrics, University of Michigan, Ann Arbor, MI, USA
9 shared publications - Angelle M. Sander
TIRR Memorial Hermann
6 shared publications Irena Chen
Max Planck Institute for Demographic Research
6 shared publications
Latest Publications
- Open Access
Multimodal Artificial Intelligence for Cause-of-Death Assignment From Verbal Autopsy Records: A National Benchmark Study in South Africa
Background: Timely cause-of-death (COD) information guides health policy, yet many deaths in low- and middle-income countries are not medically certified. In Africa, fewer than 10% of deaths are. Verbal autopsy (VA) yields a structured symptom questionnaire, and a free-text narrative. Mortality surveillance increasingly records the narrative alongside the questionnaire, but the algorithms used to produce VA-based statistics were designed for the structured responses only, so narratives inform physician COD assignment but do not contribute to algorithmic COD assignment. We evaluated whether supervised models could use both modalities to improve COD assignment. Methods: We analyzed 4,999 VA records for deaths at ages 12 years or older from a national sample of registered deaths in South Africa (RSA), using physician-assigned CODs as the reference standard. We trained unimodal and multimodal pipelines on the labelled records: pretrained language models (PLMs) fine-tuned on narratives alone, machine learning (ML) models on structured responses alone, and the two combined at the data, feature and decision levels. We also prompted a locally deployed generative large language model (LLM) on the same unimodal and multimodal inputs, without parameter updates. All were compared with InSilicoVA, the operational comparator representing current automated practice. We evaluated individual-level assignment, population-level agreement, calibration, and generalization across provinces in RSA, reporting these metrics rather than ranking pipelines on any one of them. Results: Every supervised pipeline improved on InSilicoVA by 8.5–15.9 percentage points in top-1 accuracy and 3.5–7.1 in cause-specific mortality fraction (CSMF) accuracy. The narrative and structured pipelines performed similarly in aggregate but differed by cause: 63.9% of deaths received identical predictions, 11.6% were correct only from the narrative pipeline and 13.6% only from the questionnaire pipeline, with the narrative pipeline more accurate for noncommunicable causes and the questionnaire pipeline for communicable causes. Against the unimodal approaches, the decision-fusion pipelines added 4.2 to 5.4 percentage points in top-1 accuracy, concentrated in the most common causes, whereas feature fusion with cross-modal attention improved the less common causes instead, adding 6.6 (1.7–11.5) points in balanced accuracy and 5.5 (0.6–10.3) in macro F1. Feature fusion reached the highest CSMF accuracy (95.7%). Paired differences from the better unimodal representative ranged from − 0.2 to +3.4 points, with all intervals including zero. The LLM exceeded InSilicoVA on individual-level metrics and reached the highest balanced accuracy of any pipeline (60.6%), while the supervised pipelines retained higher CSMF accuracy. Supervised pipelines had lower expected calibration error than InSilicoVA. Supervised gains held across all nine held-out provinces, including every top-1 comparison and nearly every macro F1 and CSMF comparison. Conclusions: PLMs using narratives alone substantially improved on the current best practice and matched ML pipelines using structured responses, while multiple multimodal strategies produced further gains in individual COD assignment. The zero-shot LLM showed particular strength on rare causes, whereas supervised pipelines better recovered the population cause distribution. These findings support incorporating narratives as a core input to automated VA systems and evaluating representative fusion strategies for the intended output. External validation across settings and age groups will support broader implementation.
Open AccessMultimodal Artificial Intelligence for Cause-of-Death Assignment From Verbal Autopsy Records: A National Benchmark Study in South Africa
Background: Timely cause-of-death (COD) information guides health policy, yet many deaths in low- and middle-income countries are not medically certified. In Africa, fewer than 10% of deaths are. Verbal autopsy (VA) yields a structured symptom questionnaire, and a free-text narrative. Mortality surveillance increasingly records the narrative alongside the questionnaire, but the algorithms used to produce VA-based statistics were designed for the structured responses only, so narratives inform physician COD assignment but do not contribute to algorithmic COD assignment. We evaluated whether supervised models could use both modalities to improve COD assignment. Methods: We analyzed 4,999 VA records for deaths at ages 12 years or older from a national sample of registered deaths in South Africa (RSA), using physician-assigned CODs as the reference standard. We trained unimodal and multimodal pipelines on the labelled records: pretrained language models (PLMs) fine-tuned on narratives alone, machine learning (ML) models on structured responses alone, and the two combined at the data, feature and decision levels. We also prompted a locally deployed generative large language model (LLM) on the same unimodal and multimodal inputs, without parameter updates. All were compared with InSilicoVA, the operational comparator representing current automated practice. We evaluated individual-level assignment, population-level agreement, calibration, and generalization across provinces in RSA, reporting these metrics rather than ranking pipelines on any one of them. Results: Every supervised pipeline improved on InSilicoVA by 8.5–15.9 percentage points in top-1 accuracy and 3.5–7.1 in cause-specific mortality fraction (CSMF) accuracy. The narrative and structured pipelines performed similarly in aggregate but differed by cause: 63.9% of deaths received identical predictions, 11.6% were correct only from the narrative pipeline and 13.6% only from the questionnaire pipeline, with the narrative pipeline more accurate for noncommunicable causes and the questionnaire pipeline for communicable causes. Against the unimodal approaches, the decision-fusion pipelines added 4.2 to 5.4 percentage points in top-1 accuracy, concentrated in the most common causes, whereas feature fusion with cross-modal attention improved the less common causes instead, adding 6.6 (1.7–11.5) points in balanced accuracy and 5.5 (0.6–10.3) in macro F1. Feature fusion reached the highest CSMF accuracy (95.7%). Paired differences from the better unimodal representative ranged from − 0.2 to +3.4 points, with all intervals including zero. The LLM exceeded InSilicoVA on individual-level metrics and reached the highest balanced accuracy of any pipeline (60.6%), while the supervised pipelines retained higher CSMF accuracy. Supervised pipelines had lower expected calibration error than InSilicoVA. Supervised gains held across all nine held-out provinces, including every top-1 comparison and nearly every macro F1 and CSMF comparison. Conclusions: PLMs using narratives alone substantially improved on the current best practice and matched ML pipelines using structured responses, while multiple multimodal strategies produced further gains in individual COD assignment. The zero-shot LLM showed particular strength on rare causes, whereas supervised pipelines better recovered the population cause distribution. These findings support incorporating narratives as a core input to automated VA systems and evaluating representative fusion strategies for the intended output. External validation across settings and age groups will support broader implementation.
- Open Access
- 11 September 2026
- Published by Elsevier in Environmental Research
- Vol. 308
Aflatoxin exposure and cognitive performance in Mexican adolescents
- 11 September 2026
- Published by Elsevier in Environmental Research
- Vol. 308
Open AccessAflatoxin exposure and cognitive performance in Mexican adolescents
- Open Access
- 31 August 2026
- Published by The Open Journal in Journal of Open Source Software
- Vol. 11 (124) , 9689
PyCFRL: A Python library for counterfactually fair offline reinforcement learning via sequential data preprocessing
- 31 August 2026
- Published by The Open Journal in Journal of Open Source Software
- Vol. 11 (124) , 9689
Open AccessPyCFRL: A Python library for counterfactually fair offline reinforcement learning via sequential data preprocessing
- Open Access
A Multimodal Benchmark for Evaluating Cause-of-Death Inference Using Child Health and Mortality Data
A bs t r act: Accurately attributing causes of death is vital for global health, yet fewer than 5% of deaths in resource-constrained regions are medically certified. To assign causes to these unlabeled deaths at scale, practitioners traditionally rely on verbal autopsy, using supervised statistical models to classify based on structured survey data. However, modern mortality surveillance increasingly collects rich, unstructured multimodal data, such as free-text caregiver narratives and postmortem diagnostics, which traditional supervised statistical models struggle to seamlessly integrate. In this paper, we present a comprehensive, multimodal benchmark for cause-of-death classification using data from the Child Health and Mortality Prevention Surveillance (CHAMPS) network, a unique surveillance platform spanning nine countries across South Asia and Sub-Saharan Africa. Using this dataset, we introduce an evaluation framework designed to rigorously assess diagnostic reasoning, moving beyond traditional metrics that fail to capture complex clinical realities. We demonstrate the utility of this benchmark by evaluating zero-shot large language models against supervised baselines across various data modalities. Our results reveal distinct differences in how these modeling approaches synthesize unstructured medical evidence. This benchmark provide a rigorously defined resource for assessing clinical reasoning in next-generation mortality surveillance.
Open AccessA Multimodal Benchmark for Evaluating Cause-of-Death Inference Using Child Health and Mortality Data
A bs t r act: Accurately attributing causes of death is vital for global health, yet fewer than 5% of deaths in resource-constrained regions are medically certified. To assign causes to these unlabeled deaths at scale, practitioners traditionally rely on verbal autopsy, using supervised statistical models to classify based on structured survey data. However, modern mortality surveillance increasingly collects rich, unstructured multimodal data, such as free-text caregiver narratives and postmortem diagnostics, which traditional supervised statistical models struggle to seamlessly integrate. In this paper, we present a comprehensive, multimodal benchmark for cause-of-death classification using data from the Child Health and Mortality Prevention Surveillance (CHAMPS) network, a unique surveillance platform spanning nine countries across South Asia and Sub-Saharan Africa. Using this dataset, we introduce an evaluation framework designed to rigorously assess diagnostic reasoning, moving beyond traditional metrics that fail to capture complex clinical realities. We demonstrate the utility of this benchmark by evaluating zero-shot large language models against supervised baselines across various data modalities. Our results reveal distinct differences in how these modeling approaches synthesize unstructured medical evidence. This benchmark provide a rigorously defined resource for assessing clinical reasoning in next-generation mortality surveillance.
- Open Access
- 12 June 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14 (1) , e77761
Feasibility and Acceptability of a Mobile App and Wearable Device for Collecting Mental Health Survey and Passively Sensed Data Among Health Care Workers in Kenya: Mixed Methods Pilot Study
Background: Mobile apps and wearable devices may help to facilitate early detection of mental health conditions by providing objective, real-time data to supplement other forms of feedback and diagnoses. Few studies have investigated the acceptability and feasibility of using a mobile app to track survey- and wearable-based data in mental health research in Sub-Saharan Africa. Objective: This pilot study evaluated the feasibility and acceptability of using a mobile app and wearables to capture mental health–based survey data and passively sensed data among Kenyan health care workers. Methods: A mixed methods study was conducted among health care workers employed at 4 hospitals in Nairobi, Kenya, over 30 days. A mobile app was used to collect and integrate active (baseline questionnaire and daily mood) and passive (wearable) data. The baseline questionnaire gathered information on sociodemographics, work environment, and mental health assessments on depression, anxiety, personality, early family environment, posttraumatic stress disorder, and substance use. A wearable device was used to gather data on steps, heart rate, and sleep. Qualitative interviews were conducted post trial to gain in-depth insights into participants’ experiences during the study. Results: Fifty-one participants enrolled in the pilot study. They were primarily nurses (47%) and female (70%), with a median (IQR) age of 32 (29-36) years. Attrition over 30 days was low, with only one participant dropping out due to device malfunction, which was a broken screen. Completeness of the baseline survey was high, with participants completing 96.1% of the questions. Further, 58% of the daily mood ratings were completed over the 30 days. For the wearable measures, participants submitted steps, heart rate, and sleep data on 93%, 73%, and 51% of study days, respectively. The proportion of days the wearable was worn for over 10 hours was 63%. Interviews revealed 2 primary themes. The first was intrinsic and extrinsic motivation; participants indicated that they liked having their health metrics tracked and receiving congratulatory messages from the app, encouraging increased step counts. The second theme was technical and usability challenges; 48% (10/21) of the participants reported discomfort wearing the watch while sleeping and challenges with synchronization of data due to the nonautomated nature of the process. Participants suggested additional prompts to remind them to complete the daily mood question. Conclusions: This pilot study demonstrates the feasibility of deploying mental health surveys, collecting data through wearable devices, and integrating such data within a single mobile platform under real-world infrastructure constraints. Health care workers in Kenya were willing to provide sensitive information through mental health assessments using a mobile app. To improve adherence, future studies should consider addressing some contextual factors such as daily prompts, enhanced data synchronization methods, and comfort concerns to improve adherence, especially during sleep.
- 12 June 2026
- Published by JMIR Publications Inc. in JMIR mHealth and uHealth
- Vol. 14 (1) , e77761
Open AccessFeasibility and Acceptability of a Mobile App and Wearable Device for Collecting Mental Health Survey and Passively Sensed Data Among Health Care Workers in Kenya: Mixed Methods Pilot Study
Background: Mobile apps and wearable devices may help to facilitate early detection of mental health conditions by providing objective, real-time data to supplement other forms of feedback and diagnoses. Few studies have investigated the acceptability and feasibility of using a mobile app to track survey- and wearable-based data in mental health research in Sub-Saharan Africa. Objective: This pilot study evaluated the feasibility and acceptability of using a mobile app and wearables to capture mental health–based survey data and passively sensed data among Kenyan health care workers. Methods: A mixed methods study was conducted among health care workers employed at 4 hospitals in Nairobi, Kenya, over 30 days. A mobile app was used to collect and integrate active (baseline questionnaire and daily mood) and passive (wearable) data. The baseline questionnaire gathered information on sociodemographics, work environment, and mental health assessments on depression, anxiety, personality, early family environment, posttraumatic stress disorder, and substance use. A wearable device was used to gather data on steps, heart rate, and sleep. Qualitative interviews were conducted post trial to gain in-depth insights into participants’ experiences during the study. Results: Fifty-one participants enrolled in the pilot study. They were primarily nurses (47%) and female (70%), with a median (IQR) age of 32 (29-36) years. Attrition over 30 days was low, with only one participant dropping out due to device malfunction, which was a broken screen. Completeness of the baseline survey was high, with participants completing 96.1% of the questions. Further, 58% of the daily mood ratings were completed over the 30 days. For the wearable measures, participants submitted steps, heart rate, and sleep data on 93%, 73%, and 51% of study days, respectively. The proportion of days the wearable was worn for over 10 hours was 63%. Interviews revealed 2 primary themes. The first was intrinsic and extrinsic motivation; participants indicated that they liked having their health metrics tracked and receiving congratulatory messages from the app, encouraging increased step counts. The second theme was technical and usability challenges; 48% (10/21) of the participants reported discomfort wearing the watch while sleeping and challenges with synchronization of data due to the nonautomated nature of the process. Participants suggested additional prompts to remind them to complete the daily mood question. Conclusions: This pilot study demonstrates the feasibility of deploying mental health surveys, collecting data through wearable devices, and integrating such data within a single mobile platform under real-world infrastructure constraints. Health care workers in Kenya were willing to provide sensitive information through mental health assessments using a mobile app. To improve adherence, future studies should consider addressing some contextual factors such as daily prompts, enhanced data synchronization methods, and comfort concerns to improve adherence, especially during sleep.
- 4 May 2026
- Published by American Psychological Association (APA) in Rehabilitation Psychology
Understanding the impact of perceived app usability on the efficacy of mobile health intervention for traumatic brain injury caregivers.
Caregiving for individuals with traumatic brain injury (TBI) is often highly stressful, and traditional in-person interventions can be inaccessible given the demands of their caregiver role. Mobile health (mHealth) interventions offer a low-burden, scalable alternative by delivering personalized, real-time support. However, how effective these interventions depends on whether people perceive them as useful and engage with them. The role of perceived usability in mHealth efficacy remains underexplored in the TBI caregiver population. This study primarily evaluates the relationship between self-reported app usability and the efficacy of a fully automated mHealth intervention, delivered via personalized mobile-app messages, in reducing caregiver strain, anxiety, and depression in a TBI caregiver population. The secondary aim is to identify variables that moderate the intervention efficacy among caregivers reporting high app usability. We analyzed data from 122 TBI caregivers assigned to receive self-monitoring plus push notifications for self-care as part of a larger randomized controlled trial. Perceived app usability was assessed via caregiver responses to a questionnaire item. We applied multivariate linear models with a weighted and centered least squares estimator to assess the moderating effects of perceived app usability and other variables on message efficacy. Messages were more effective in reducing depression among TBI caregivers who reported high app usability. Moreover, shorter caregiving duration, higher Fitbit step count, and lower prior-week anxiety were significantly associated with improved message efficacy among this high usability group. Perceived app usability plays a critical role in the mHealth message efficacy among TBI caregivers. Tailoring interventions based on perceived app usability and identified moderators may optimize health outcomes and support more personalized care for this population. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
- 4 May 2026
- Published by American Psychological Association (APA) in Rehabilitation Psychology
Understanding the impact of perceived app usability on the efficacy of mobile health intervention for traumatic brain injury caregivers.
Caregiving for individuals with traumatic brain injury (TBI) is often highly stressful, and traditional in-person interventions can be inaccessible given the demands of their caregiver role. Mobile health (mHealth) interventions offer a low-burden, scalable alternative by delivering personalized, real-time support. However, how effective these interventions depends on whether people perceive them as useful and engage with them. The role of perceived usability in mHealth efficacy remains underexplored in the TBI caregiver population. This study primarily evaluates the relationship between self-reported app usability and the efficacy of a fully automated mHealth intervention, delivered via personalized mobile-app messages, in reducing caregiver strain, anxiety, and depression in a TBI caregiver population. The secondary aim is to identify variables that moderate the intervention efficacy among caregivers reporting high app usability. We analyzed data from 122 TBI caregivers assigned to receive self-monitoring plus push notifications for self-care as part of a larger randomized controlled trial. Perceived app usability was assessed via caregiver responses to a questionnaire item. We applied multivariate linear models with a weighted and centered least squares estimator to assess the moderating effects of perceived app usability and other variables on message efficacy. Messages were more effective in reducing depression among TBI caregivers who reported high app usability. Moreover, shorter caregiving duration, higher Fitbit step count, and lower prior-week anxiety were significantly associated with improved message efficacy among this high usability group. Perceived app usability plays a critical role in the mHealth message efficacy among TBI caregivers. Tailoring interventions based on perceived app usability and identified moderators may optimize health outcomes and support more personalized care for this population. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
- 9 April 2026
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 82 (2)
Joint modeling of multiple longitudinal biomarkers and survival outcomes via threshold regression: variability as a predictor
Longitudinal biomarker data and health outcomes are routinely collected in many studies to assess how biomarker trajectories predict health outcomes. Existing methods primarily focus on mean biomarker profiles, treating variability as a nuisance. However, excess variability may indicate system dysregulations that may be associated with poor outcomes. In this paper, we address the long-standing problem of using variability information of multiple longitudinal biomarkers in time-to-event analyses by formulating and studying a Bayesian joint model. We first model multiple longitudinal biomarkers, some of which are subject to limit-of-detection censoring. We then model the survival times by incorporating random effects and variances from the longitudinal component as predictors through threshold regression that admits nonproportional hazards. We demonstrate the operating characteristics of the proposed joint model through simulations and apply it to data from the Study of Women’s Health Across the Nation to investigate the impact of the mean and variability of follicle-stimulating hormone and anti-Müllerian hormone on age at the final menstrual period.
- 9 April 2026
- Published by Oxford University Press (OUP) in Biometrics
- Vol. 82 (2)
Joint modeling of multiple longitudinal biomarkers and survival outcomes via threshold regression: variability as a predictor
Longitudinal biomarker data and health outcomes are routinely collected in many studies to assess how biomarker trajectories predict health outcomes. Existing methods primarily focus on mean biomarker profiles, treating variability as a nuisance. However, excess variability may indicate system dysregulations that may be associated with poor outcomes. In this paper, we address the long-standing problem of using variability information of multiple longitudinal biomarkers in time-to-event analyses by formulating and studying a Bayesian joint model. We first model multiple longitudinal biomarkers, some of which are subject to limit-of-detection censoring. We then model the survival times by incorporating random effects and variances from the longitudinal component as predictors through threshold regression that admits nonproportional hazards. We demonstrate the operating characteristics of the proposed joint model through simulations and apply it to data from the Study of Women’s Health Across the Nation to investigate the impact of the mean and variability of follicle-stimulating hormone and anti-Müllerian hormone on age at the final menstrual period.
- Open Access
- 26 March 2026
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 5 (3) , e0001281
Patterns of smartphone typing performance by time awake: implications for unobtrusive ambulatory mental fatigue assessment
Mental fatigue undermines workplace safety and productivity. Early detection of subtle declines in objective alertness and cognitive performance in workers can enable timely interventions to prevent costly errors and safeguard employee health. However, conventional assessments often require controlled laboratory conditions and prolonged testing, limiting their real-world applicability. Here, we demonstrate an approach that utilizes smartphone keyboard metrics in everyday use to provide a more scalable, continuous, and ambulatory method for evaluating mental fatigue. We examined the adjusted association between novel yet widely available SensorKit typing performance metrics and wearable-derived time since waking from a most recent major sleep episode or napping among 366 first-year training physicians in the United States who generated 45,042 typing sessions over a two-month period. Typing performance, especially typing speed, has a significant non-linear adjusted relationship with time awake. At the population-level, typing speed increases and peaks around 7.5 hours since awake and has a substantial decrease around 15.3 hours of time awake, highly consistent with classical lab-based active task Psychomotor Vigilance Test findings that showed increased response lapses beyond 15.8 hours of wake period. Our findings are relevant for developing a ubiquitous and unobtrusive tool to assess, monitor, and manage mental fatigue on a continuous basis in everyday life, especially for populations in high-risk and high-stake settings.
- 26 March 2026
- Published by Public Library of Science (PLoS) in PLOS Digital Health
- Vol. 5 (3) , e0001281
Open AccessPatterns of smartphone typing performance by time awake: implications for unobtrusive ambulatory mental fatigue assessment
Mental fatigue undermines workplace safety and productivity. Early detection of subtle declines in objective alertness and cognitive performance in workers can enable timely interventions to prevent costly errors and safeguard employee health. However, conventional assessments often require controlled laboratory conditions and prolonged testing, limiting their real-world applicability. Here, we demonstrate an approach that utilizes smartphone keyboard metrics in everyday use to provide a more scalable, continuous, and ambulatory method for evaluating mental fatigue. We examined the adjusted association between novel yet widely available SensorKit typing performance metrics and wearable-derived time since waking from a most recent major sleep episode or napping among 366 first-year training physicians in the United States who generated 45,042 typing sessions over a two-month period. Typing performance, especially typing speed, has a significant non-linear adjusted relationship with time awake. At the population-level, typing speed increases and peaks around 7.5 hours since awake and has a substantial decrease around 15.3 hours of time awake, highly consistent with classical lab-based active task Psychomotor Vigilance Test findings that showed increased response lapses beyond 15.8 hours of wake period. Our findings are relevant for developing a ubiquitous and unobtrusive tool to assess, monitor, and manage mental fatigue on a continuous basis in everyday life, especially for populations in high-risk and high-stake settings.
- Open Access
Third places visits and well-being: insights from longitudinal passive sensing data
Spending time in locations outside the home and workplace (termed “third places”), has been linked to better mental health. However, studies to date have typically been cross-sectional, based on self-reported location data and employed small sample sizes, limiting their ability to assess the presence and nature of the association between third places and mental health. To overcome these limitations, we collected 18,795 person-days of objective SensorKit location data passively from a national cohort of 410 first-year medical residents across the United States, to assess visits to third places and their associations with mood and depression over the course of one year. On average, participants visited 3.3 unique locations per day (SD=1.7) and spent 17.9% of their time at third places (SD = 26.5%). Within individuals, both a higher percentage of time spent at third places (B=0.013 [per 10% increase], pp p p p <0.001) compared to non-depressed subjects. These findings demonstrate the relationship between visiting third places and well-being, and suggest that interventions and policies aimed at encouraging third places visits have the potential to improve mental health.
Open AccessThird places visits and well-being: insights from longitudinal passive sensing data
Spending time in locations outside the home and workplace (termed “third places”), has been linked to better mental health. However, studies to date have typically been cross-sectional, based on self-reported location data and employed small sample sizes, limiting their ability to assess the presence and nature of the association between third places and mental health. To overcome these limitations, we collected 18,795 person-days of objective SensorKit location data passively from a national cohort of 410 first-year medical residents across the United States, to assess visits to third places and their associations with mood and depression over the course of one year. On average, participants visited 3.3 unique locations per day (SD=1.7) and spent 17.9% of their time at third places (SD = 26.5%). Within individuals, both a higher percentage of time spent at third places (B=0.013 [per 10% increase], pp p p p <0.001) compared to non-depressed subjects. These findings demonstrate the relationship between visiting third places and well-being, and suggest that interventions and policies aimed at encouraging third places visits have the potential to improve mental health.
- 1 December 2025
- Published by Institute of Mathematical Statistics in The Annals of Applied Statistics
- Vol. 19 (4) , 3003-3022
Tree-regularized Bayesian latent class analysis for improving weakly separated dietary pattern subtyping in small-sized subpopulations
Dietary patterns synthesize multiple related diet components, which can be used by nutrition researchers to examine diet-disease relationships. Latent class models (LCMs) have been used to derive dietary patterns from dietary intake assessment, where each class profile represents the probabilities of exposure to a set of diet components. However, LCM-derived dietary patterns can exhibit strong similarities, or weak separation, resulting in numerical and inferential instabilities that challenge scientific interpretation. This issue is exacerbated in small-sized subpopulations. To address these issues, we provide a simple solution that empowers LCMs to improve dietary pattern estimation. We develop a tree-regularized Bayesian LCM that shares statistical strength between dietary patterns to make better estimates using limited data. This is achieved via a Dirichlet diffusion tree process that specifies a prior distribution for the unknown tree over classes. Dietary patterns that share proximity to one another in the tree are shrunk toward ancestral dietary patterns a priori, with the degree of shrinkage varying across prespecified food groups. Using dietary intake data from the Hispanic Community Health Study/Study of Latinos, we apply the proposed approach to a sample of 496 U.S. adults of South American ethnic background to identify and compare dietary patterns.
- 1 December 2025
- Published by Institute of Mathematical Statistics in The Annals of Applied Statistics
- Vol. 19 (4) , 3003-3022
Tree-regularized Bayesian latent class analysis for improving weakly separated dietary pattern subtyping in small-sized subpopulations
Dietary patterns synthesize multiple related diet components, which can be used by nutrition researchers to examine diet-disease relationships. Latent class models (LCMs) have been used to derive dietary patterns from dietary intake assessment, where each class profile represents the probabilities of exposure to a set of diet components. However, LCM-derived dietary patterns can exhibit strong similarities, or weak separation, resulting in numerical and inferential instabilities that challenge scientific interpretation. This issue is exacerbated in small-sized subpopulations. To address these issues, we provide a simple solution that empowers LCMs to improve dietary pattern estimation. We develop a tree-regularized Bayesian LCM that shares statistical strength between dietary patterns to make better estimates using limited data. This is achieved via a Dirichlet diffusion tree process that specifies a prior distribution for the unknown tree over classes. Dietary patterns that share proximity to one another in the tree are shrunk toward ancestral dietary patterns a priori, with the degree of shrinkage varying across prespecified food groups. Using dietary intake data from the Hispanic Community Health Study/Study of Latinos, we apply the proposed approach to a sample of 496 U.S. adults of South American ethnic background to identify and compare dietary patterns.