Deqiang Zheng

Background: Polygenic risk scores (PRS) and lifestyle factors are associated with colorectal cancer (CRC), but whether their associations and joint effects vary according to age at CRC onset remains unclear. We evaluated age‐related heterogeneity in the associations of genetic risk and lifestyle with CRC and examined their additive interaction in early‐ and late‐onset CRC. Methods: We analyzed data from 405,453 cancer‐free participants aged 38–73 years at baseline in the UK Biobank. PRS and a healthy lifestyle score (based on eight modifiable factors) were assessed. Age‐scale Cox proportional hazards models estimated hazard ratios (HRs) for CRC risk by age at onset. Heterogeneity across age intervals was assessed using likelihood ratio tests. Additive interaction was assessed using the relative excess risk due to interaction (RERI), and differences in standardized incidence rates were evaluated. Results: During a median follow‐up of 13.65 years, 6620 CRC cases were identified. Compared with low genetic risk, high genetic risk was associated with HRs of 3.72, 3.32, 2.80, and 2.01 for CRC occurring at ages < 55, 55–64, 65–74, and ≥ 75 years, respectively, with significant heterogeneity across age intervals ( p for heterogeneity < 0.001). Compared with a favorable lifestyle, the corresponding HRs for an unfavorable lifestyle were 1.10, 1.50, 1.58, and 1.28, respectively, with evidence of overall heterogeneity ( p for heterogeneity = 0.015). For late‐onset CRC, positive additive interactions were observed for an unfavorable lifestyle combined with intermediate or high genetic risk, whereas no such interaction was detected for early‐onset CRC. Conclusions: The associations of genetic susceptibility and baseline lifestyle with CRC varied according to age at onset. Positive additive interaction between genetic risk and an unfavorable lifestyle was observed for late‐onset but not early‐onset CRC. These findings provide further evidence that the relative and joint associations of genetic and lifestyle factors with CRC vary across the life course.

While current colorectal cancer (CRC) screening guidelines consider family history to determine starting age for screening, they inadequately incorporate other important unmodifiable risk factors, including polygenic risk score (PRS) and biological sex. In 353,118 average‐risk UK Biobank participants, we used family history‐stratified Cox models to estimate associations of sex and PRS with CRC incidence, metastasis, and CRC‐specific mortality. We translated risk differences into risk‐adapted initiation screening ages using both risk advancement period (RAP) and absolute 10‐year cumulative risk. Men had a higher risk of CRC compared to women, with HRs of 1.41 (95% CI, 1.33–1.50) for those without family history and 1.42 (95% CI, 1.26–1.61) for those with family history. For each one‐standard deviation (SD) increase in PRS, the risk was 1.49 times higher (95% CI, 1.45–1.53) in participants without a family history and 1.42 times higher (95% CI, 1.34–1.51) in those with a family history. The increase in PRS was associated with elevated risks of both metastatic CRC at diagnosis and CRC‐specific mortality, indicating delayed diagnosis in high‐PRS individuals and highlighting the potential benefit of earlier screening. Consequently, risk‐adapted initiation ages ranged from before 40 years for men with a family history in the high‐PRS group to after 60 years for women without a family history in the low‐PRS group. Our findings demonstrate that the integration of PRS, biological sex, and family history identifies substantial heterogeneity in CRC risk and risk‐adapted screening initiation ages.

Aims: The long-term depressive symptom trajectories across the transition to incident cardiovascular disease (CVD) remain poorly characterized. We aimed to investigate trajectories of depressive symptoms before and after incident CVD in multinational ageing populations. Methods and results: In this multicohort study, we analysed longitudinal data from the Health and Retirement Study (HRS, 2002–2018), English Longitudinal Study of Ageing (ELSA, 2002–2019), and China Health and Retirement Longitudinal Study (CHARLS, 2011–2018). Depressive symptoms were measured biennially by the 8 item (HRS and ELSA) and 10 item (CHARLS) Center for Epidemiologic Studies Depression Scale and standardized to Z-scores. Incident CVD was ascertained by medical history. Data were analysed using discontinuous growth curve modelling and pooled using meta-analysis. Among 28 704 included participants (mean [SD] age, 62.3 [9.8] years; 43.4% male), 7363 developed CVD during follow-up. Participants who developed CVD had higher depressive symptom scores than those without CVD (pooled β [95% confidence interval]: 0.161 SD [0.086–0.236]). A significant increase in depressive symptom scores was observed around the time of CVD diagnosis (pooled β: 0.089 SD [0.060–0.118]). Additionally, depressive symptom scores increased both in the years preceding CVD onset (pooled β: 0.007 SD/year [0.004–0.010]) and following diagnosis (pooled β: 0.009 SD/year [0.004–0.013]). Conclusion: Depressive symptoms show a gradual increase before CVD diagnosis, elevate further around the time of diagnosis, and continue to worsen thereafter. These findings call for the integration of mental health screening and management into CVD care at all stages, from primary prevention to long-term disease management.

Despite adequate glycemic control, a proportion of patients develop diabetic retinopathy (DR), suggesting the contribution of other mechanisms. To explore genetic associations with DR, we conducted a whole-exome sequencing (WES) study of DR using a dual-control design in individuals of European ancestry. Leveraging UK Biobank WES data, we implemented a dual-control design to identify candidate genes associated with DR, comparing cases with both diabetes controls and the general population. Rare variant associations were tested at gene-level using SAIGE-GENE+, and common variants were analyzed via PLINK2. Significant findings were assessed through the exclusion of self-reported cases, leave-one-variant-out (LOVO) analysis, time-to-event analysis, transcriptomics analysis, virtual knockout experiments, proteomics, and protein-protein interaction (PPI) analysis. Exome-wide gene-based association analysis identified FRZB (frizzled-related protein) as a candidate gene potentially associated with DR in comparisons with diabetic controls (odds ratio [OR] = 1.09; 95% confidence interval [CI], 1.04-1.15; P = 2.55 × 10-7). The observed association showed generally consistent patterns across several sensitivity analyses, including exclusion of self-reported cases, LOVO analyses, and Cox proportional hazards models. In addition, transcriptomic, proteomic, virtual knockout, and PPI analyses provided complementary exploratory evidence supporting a possible involvement of FRZB in DR-related biological processes. Additionally, single-variant analysis revealed two loci, FAM160A1 (4:151662612:C:G) and HK1 (10:69300854:A:G), associated with DR in the general population (OR = 1.82 and OR = 1.50, respectively). However, multiomics support for these signals was limited. This multilayered genetic study identified FRZB as a candidate gene potentially associated with DR. However, this finding should be interpreted cautiously, and further experimental studies and independent replication are needed to clarify the potential relevance of FRZB to DR.

Background: Evidence on whether modifiable behaviours, leisure engagement and subjective well-being jointly shape longevity in the oldest-old (≥80 years) remains limited, especially in China. We aimed to examine the associations between modifiable factors and survival after age 80 in China. Methods: This study comprised 30 398 participants aged 80 years and over, drawn from the Chinese Longitudinal Healthy Longevity Survey (CLHLS 1998–2018). Lifestyles (smoking status, drinking status, physical activity and plant-based diet index), leisure activities (mental, physical, productive and social activity) and subjective well-being were assessed using structured interviews. Median age at death after age 80 was estimated using Laplace regression. Results: During follow-up, 21 648 (71.2%) participants died. Having healthy lifestyles (never smoking [4.2 months], never drinking [1.6 months], engaging in physical activity [3.4 months], eating healthful diet [2.2 months]), engaging in leisure activities (mental: 3.2 months; physical: 8.7 months; productive: 2.0 months; social: 2.8 months) and reporting high subjective well-being (5.4 months) were found to have significantly longer survival times. The median survival of people with a low-risk profile (maintained at least three healthy lifestyles, engaged in at least one leisure activity and reported medium or high subjective well-being) was 18 months (95% CI 15.5–19.8) longer than that of those with a high-risk profile (unhealthy lifestyle, engaged in no leisure activities and reported low subjective well-being). Conclusions: The study findings have demonstrated that adopting a healthy lifestyle, engaging in leisure activities and fostering subjective well-being are associated with longer life expectancy among the oldest-old in China.

Aims: Drug–gene interactions (DGIs) modify drug response and safety, yet their influence on diabetic microvascular complications remains unclear. This study aimed to elucidate the role of DGIs in these complications. Materials and Methods: Using UK Biobank (UKB) data, we identified medications frequently prescribed to individuals with diabetes and defined DGIs based on the Food and Drug Administration (FDA) and the Clinical Pharmacogenetics Implementation Consortium (CPIC) guidelines. Associations between DGIs and diabetic microvascular complications were evaluated using Cox proportional hazards models, which are suited for longitudinal time‐to‐event data. Two complementary analyses were performed: (1) a therapeutic class‐level analysis among medication users, and (2) a genotype‐level analysis among individuals with non‐normal metabolizer phenotypes who used the corresponding medications. Results: We identified 368 medications preferentially used among participants with diabetes, primarily cardiovascular agents and detected 55 clinically relevant DGIs implicating 30 medications and 7 genes. Among users of antithrombotic agents, the presence of DGIs was associated with diabetic kidney disease (DKD) (hazard ratio [HR]: 1.44, 95% confidence interval [CI]: 1.12–1.86) and diabetic neuropathy (DN) (HR: 2.13, 95% CI: 1.39–3.28). Likewise, among individuals with non‐normal metabolizer status for CYP2C19 or CYP2D6 , DGIs conferred elevated risks for DKD and DN (HR range: 1.26–2.11). However, no significant association was found between DGI and DR. Conclusion: This study provides the first comprehensive assessment of DGIs and diabetic microvascular complications. DGIs involving antithrombotic agents and non‐normal CYP2C19 or CYP2D6 metabolizers were significantly linked to higher risks of DKD and DN. These findings underscore the potential of pharmacogenomic‐guided prescribing to enhance drug safety.

Reliable feature selection is critical for extracting small, interpretable biomarker panels from high-dimensional omics, yet deep learning–based feature selection methods have rarely been compared systematically. We present a unified and reproducible benchmark of post-hoc explainers—attribution (GradientSHAP, and DeepLIFT), perturbation (LIME, Feature Ablation, and Occlusion) and embedded selectors (CancelOut, EAR-FS, and GRACES). Using standardized preprocessing, a shared neural network, and Optuna tuning, we evaluate feature selection performance and computational efficiency in simulations and downstream predictive accuracy on diverse real datasets, including gene expression datasets with binary outcomes, five TCGA projects spanning mRNA, methylation, and SNP modalities with multi-class outcomes, and ADNI gene expression with continuous neuroimaging phenotypes. EAR-FS and GRACES consistently perform best: GRACES is most robust but computationally intensive, whereas EAR-FS achieves similar accuracy with much lower computational cost. Classical methods remain competitive when signals are sparse and near-linear. Post-hoc explainers contribute most as interpretability tools and model auditors rather than as primary subset selectors. To enable reproducibility and broad adoption, we provide a software hub and website implementing all these methods with standardized pipelines and evaluation routines, facilitating efficient feature selection under typical constraints of budget, sample size, and turnaround time.