Bowen Liu

A bstract: Breast cancer survivors frequently experience upper-limb impairments, making continuous monitoring essential for effective rehabilitation. We propose REINA (Recognize-Then-Infer Wearable-to-App A I Framework), a two-stage deep-learning approach for remote monitoring of motor function during breast cancer rehabilitation using wearable-device data. Inertial measurement unit (IMU) signals from wearable devices are first used to recognize physical activities via supervised learning, followed by an activity-specific recurrent neural network (RNN) to infer corresponding electromyography (EMG) signals. REINA establishes reliable inference of neuromuscular activity from wearable IMU data, enabling real-time, cost-effective assessment of motor function recovery in real-world settings.

Trigger–outcome and risk-factor analyses in substance use disorder (SUD) care are sensitive to modeling choices: alternative trigger definitions, adjustment sets, study periods, and lag windows can yield different risk contrasts. We introduce risk-envelope trigger–outcome analysis, a robustness-first framework that reports adjusted risk differences (ARDs) across a prespecified family of clinically plausible specifications rather than relying on a single analytic route. The goal is screening and prioritization, not point-identified causal inference. We apply the framework to a regional SUD patient-quarter panel from 2017Q1 through 2022Q4 that is linked to tract/county context indicators. The analysis identifies stable diagnosis-side signals, especially suicidal ideation followed by next-quarter SubA (polysubstance use excluding alcohol and nicotine) evidence, while showing that some clinically plausible pairs are specification-sensitive. Short-term recurrence, specification-adjusted transition checks, and SubA–opioid cross-transitions provide operational markers for follow-up, and multiplicity-controlled context screening identifies community indicators that co-move with documented SUD evidence. Risk-envelope analysis provides a transparent basis for SUD trigger screening under observational and specification uncertainty.

Transformed gamma distributions are widely used in various fields such as survival analysis, reliability engineering, and environmental studies. Statistical inference for the quantities of the transformed gamma distributions is of great importance in these applications. However, traditional inference methods often rely on large sample approximations, which is not accurate for small sample sizes. To address this issue, we introduced an exact statistical inference framework for quantities of the transformed gamma distributions and applied the framework to loggamma distribution. The performance of the proposed framework was extensively evaluated via comprehensive simulation studies. The results showed that the proposed framework outperforms the parametric bootstrapping method in terms of coverage probability and type 1 error rate. Two real data applications are provided for illustrative purposes.

Areas at low land elevations exhibit increased exposures to flooding. These exposures are exacerbated by the presence of community social vulnerability. This study examines the relationship between mean land elevations and social vulnerability in Metropolitan Kansas City. A database with socio‐demographic information, anthropogenic hazards, and mean elevations was created at the census tract level ( N = 556). The findings suggest that the population aged 25 and over with a bachelor's degree (%), the population density (sq. mi.), and the population under the age of 18 (%) are positive predictors of mean land elevations (ft.). Families living in poverty (%), the minority population (%), and TRI facilities are negative predictors of mean land elevations (ft.). These findings suggest that socially vulnerable populations are at higher risk of flooding and experiencing compound disasters from the disproportionate presence of TRI facilities at lower land elevations. Owner‐occupied housing (%), single family housing (%), median home value (USDs), group quarters, and RCRA hazardous waste sites are not predictors of mean land elevations. Policy implications call for increased community participation in the flood mitigation process through programs like Community Emergency Response Teams and the Community Rating System.

Modeling excesses remains to be an important topic in insurance data modeling. Among the alternatives to modeling excess, the Peaks Over Threshold (POT) framework with Generalized Pareto distribution (GPD) is regarded as an efficient approach due to its flexibility. However, selecting an appropriate threshold for such a framework is a major difficulty. To address such difficulty, we applied several accumulation tests along with Anderson-Darling test to determine an optimal threshold. Limited simulations were conducted to assess the performance of accumulation tests. Based on the selected thresholds, the fitted GPD with the estimated quantiles can be found. We applied the procedure to the wellknown Norwegian fire insurance data and AON Re Belgian fire loss data. With the selected thresholds, confidence intervals for the Value-at-Risks (VaR) were constructed correspondingly. The accumulation test approach provides satisfactory performance in modeling the VaR of both data sets compared to the previous graphical methods.

Aim: Dengue cases are on the rise in Thailand, coinciding with increases in flooding events. Studies pertaining to public knowledge, attitudes, and practices (KAP) of dengue prevention have frequently been used to better understand the public’s needs towards dengue. While these studies were conducted in different settings, it is necessary to initiate a systematic review and meta-analysis of relevant studies. Subjects and method: We conducted a systematic review and meta-analysis of prior studies in Thailand that assessed the KAP towards dengue. Eligibility criteria were established and independently used by reviewers to select nine studies for the systematic review and three for the meta-analysis. Collectively, the nine studies included 3,058 individuals and 2,519 households. Results: The overall estimate of the proportion of participants with good knowledge of dengue prevention is 35% (95% CI: 14-59%), suggesting the majority of the population in Thailand had low levels of knowledge towards dengue. The poor levels of practice in Thailand were also observed and confirmed for the majority of the population by meta-analysis, with the pooled estimate of the proportion of participants with good practice of dengue prevention being 25% (95% CI: 22-27%). In contrast, most of the studies included in the systematic review reported positive attitudes towards dengue prevention, and this finding was also affirmed by the meta-analysis, as the pooled estimate of the proportion of positive attitudes towards dengue prevention is 61% (95% CI: 43-77%). Conclusion: Despite good attitudes towards dengue prevention, poor knowledge and poor practices predominate, highlighting the need for enhanced public health campaigns to educate the public on dengue risks and prevention methods.

As human complex diseases are influenced by the interaction between genetics and the environment, identifying gene-environment interactions (G × E) is crucial for understanding disease mechanisms and predicting risk. Developing robust quantitative tools for G × E analysis can enhance the study of complex diseases. However, many existing methods that explore G × E focus on the interplay between an environmental factor and genetic variants, exclusively for common or rare variants. In this study, we developed MAGEIT_RAN and MAGEIT_FIX to identify interactions between an environmental factor and a set of genetic markers, including both rare and common variants, based on the MinQue for Summary statistics. The genetic main effects in MAGEIT_RAN and MAGEIT_FIX are modeled as random and fixed effects, respectively. Simulation studies showed that both tests had type I error under control, with MAGEIT_RAN being the most powerful test. Applying MAGEIT to a genome-wide analysis of gene-alcohol interactions on hypertension and seated systolic blood pressure in the Multi-Ethnic Study of Atherosclerosis revealed genes like EIF2AK2, CCNDBP1 and EPB42 influencing blood pressure through alcohol interaction. Pathway analysis identified one apoptosis and survival pathway involving PKR and two signal transduction pathways associated with hypertension and alcohol intake, demonstrating MAGEIT_RAN's ability to detect biologically relevant gene-environment interactions.