Shi-Heng Wang

Background and ObjectivesEvidence remains inconclusive on whether paternal valproate exposure during spermatogenesis is associated with adverse offspring outcomes, with findings lacking beyond Nord...

Objective: This study examined employment trajectories following the diagnosis and antipsychotic treatment of schizophrenia, with particular focus on transitions to long-acting injectable antipsychotic treatment. Methods: We conducted a nationwide matched cohort study using Taiwan’s National Health Insurance Research Database (2001–2021). Individuals aged 18–64 years with newly diagnosed schizophrenia spectrum disorders were matched 1:1 with individuals without schizophrenia by birth year, sex, and index date. Employment trajectories were analyzed using controlled interrupted time-series models. To evaluate treatment-related transitions, individuals initiating long-acting injectable antipsychotics were compared with matched oral antipsychotic users. Subgroup analyses were conducted by illness duration and baseline medication adherence. Results: The schizophrenia cohort (n = 93,041) showed substantially lower employment odds than matched comparisons (OR = 0.31), with a marked decline around diagnosis (OR = 0.90) and only modest recovery thereafter (post-diagnosis trend OR = 1.09 per year). In the treatment analysis (18,488 matched pairs), long-acting injectable users had lower baseline employment odds than oral antipsychotic users (OR = 0.85) and an immediate decline around long-acting injectable initiation (OR = 0.92), followed by a modestly more favorable post-initiation employment trend (OR = 1.03). More favorable post-initiation trends were observed among individuals with longer illness duration and lower baseline adherence. No meaningful differences in employment trajectories were observed across different long-acting injectable formulations, including flupentixol, risperidone, haloperidol, paliperidone, and fluphenazine. Conclusions: Employment outcomes in schizophrenia appear shaped by diagnostic and treatment-related transitions rather than static disease status. Stabilization after long-acting injectable initiation may be linked to incremental improvements in employment trajectories for some subgroups, but optimizing long-term vocational outcomes likely requires integrated psychosocial and vocational support.

Background: Clinical course information, including disease onset, episode recurrence, and hospitalization history, is essential for psychiatric care and research. However, these data are embedded in unstructured clinical narratives with substantial linguistic variability, making manual extraction labor-intensive and rule-based extraction difficult to scale. Fine-tuned large language models (LLMs) may flexibly extract such information from privacy-sensitive psychiatric records. Objective: This study aimed to evaluate the performance of LLMs for automatically extracting temporal and clinical course information from psychiatric discharge summaries. Methods: We analyzed 500 psychiatric discharge summaries from the Integrated Medical Database of National Taiwan University Hospital. A psychiatrist and a natural language processing researcher manually annotated clinical events and temporal information. Four open-source LLMs (LLaMA, MentaLLaMA, OpenBioLLM, and Mistral) were fine-tuned using low-rank adaptation and evaluated using 10-fold cross-validation. A 2-stage framework was developed in which sentence-level extraction of clinical events and temporal information was followed by chart-level prediction of 4 clinical course features: first-episode onset time, episode count, number of psychiatric hospitalizations, and most recent hospitalization. Performance was assessed using precision, recall, F ₁ -score, accuracy, mean absolute error (MAE), and bootstrap significance testing. Results: A total of 12,947 sentences were extracted from 500 discharge summaries, yielding 7177 clinical event annotations and 4842 temporal annotations. At the sentence level, Mistral achieved the highest F ₁ -scores for clinical event extraction, including symptom/episode detection (0.925, 95% CI 0.920‐0.930), hospitalization detection (0.944, 95% CI 0.934‐0.952), and remission/response detection (0.867, 95% CI 0.851‐0.883). Mistral also achieved the highest F ₁ -scores for most temporal information categories, including age expressions (0.983, 95% CI 0.973‐0.993), relative time expressions (0.953, 95% CI 0.944‐0.963), duration expressions (0.976, 95% CI 0.962‐0.987), and vague temporal expressions (0.888, 95% CI 0.871‐0.905). At the chart level, the proposed 2-stage framework showed the clearest benefit for first-episode onset prediction. Using the 2-stage framework, Mistral achieved the highest point estimate for onset accuracy (0.772), although its performance did not significantly differ from that of LLaMA using the same framework (0.744; bootstrap P =.26). For onset prediction, the 2-stage framework significantly outperformed the direct and joint extraction approaches across all evaluated models (all bootstrap P ≤.002). For other chart-level features, the best-performing approach varied: using the 2-stage framework, Mistral achieved the highest episode count accuracy (0.624; MAE=0.428); using the direct approach, LLaMA achieved the highest hospitalization count accuracy (0.692; MAE=0.379); and using the 2-stage framework, OpenBioLLM achieved the highest most recent hospitalization accuracy (0.868; F ₁ -score=0.617). Conclusions: Fine-tuned, locally deployable, open-source LLMs can extract temporal and longitudinal disease course information from psychiatric discharge summaries. The 2-stage framework was most beneficial for first-episode onset prediction and performed competitively across other chart-level features, supporting the use of LLMs to transform heterogeneous psychiatric narratives into structured data for research and decision support.

Background: The high comorbidity between mental and cardiovascular diseases is widely recognized, but the underlying mechanisms are not yet well understood. We investigated the familial coaggregation and shared genetic loading between these diseases. Methods: Using claims data from Taiwan’s universal health insurance program, we identified cohorts of 4,513,509 individuals with parental information and 3,330,181 with full-sibling information born between 1970 and 1999, and followed up until 2020. Genotyping was available for 106,796 unrelated participants from the Taiwan Biobank. Multiple logistic regression models were used to estimate the association of parental history, full-sibling history, and polygenic risk scores (PRSs) for severe mental illnesses (schizophrenia, bipolar disorder, major depressive disorder [MDD]) with the risk of myocardial infarction, stroke, peripheral arterial disease (PAD), and heart failure. Results: Individuals with a parental history of schizophrenia, bipolar disorder, or MDD were more likely to have stroke and PAD. Coaggregation in full-siblings was observed for MDD with myocardial infarction and heart failure, bipolar disorders with stroke, and schizophrenia with PAD. The PAD risk increased with more relatives affected by MDD; the magnitude of association was larger when both parents were affected than when either parent was affected, and when there were two affected siblings than in those with only one. MDD PRS was positively associated with the risk of myocardial infarction and PAD. Conclusions: This study revealed familial coaggregation between mental and cardiovascular diseases, and shared polygenic liability of MDD with cardiovascular diseases. Our findings suggest the potential benefit of family-based integrated screening and preventive strategies.

Observational studies found that individuals with schizophrenia (SCZ) exhibit higher prevalence of obesity-related traits, posing significant health risks. However, the neuropsychiatric mechanism remains unclear. This study aimed to investigate the genetic connections between schizophrenia and obesity. We examined the effect of genetic liability for SCZ on obesity in Taiwan Biobank samples of East Asian populations (N = 106,614) using linear regression models. We also performed linkage disequilibrium score regression to estimate genetic correlation in East Asian and European populations, respectively, and conducted two-sample Mendelian randomization (MR) to infer their causality in European populations. Our results indicated that higher genetic predisposition for SCZ was associated with lower body mass index (BMI) (beta±SE = -0.11 ± 0.01, per SD of genetic predisposition for SCZ), body fat percentage (BFP) (beta±SE = -0.16 ± 0.02), waist circumference (WC) (beta±SE = -0.20 ± 0.03), hip circumference (HC), (beta±SE = -0.17 ± 0.02) and waist-to-hip ratio (WHR) (beta±SE = -0.06 ± 0.02). SCZ showed negative genetic correlations with BMI (rg±SE = -0.15 ± 0.03 and -0.11 ± 0.01), WC (rg±SE = -0.10 ± 0.03 and -0.08 ± 0.02), and HC (rg±SE = -0.12 ± 0.03 and -0.08 ± 0.03) in both East Asian and European populations. SCZ showed negative genetic correlation with BFP in East Asian populations (rg±SE = -0.13 ± 0.03, p-value = 0.0001) but not in European populations (rg±SE = -0.02 ± 0.04, p-value = 0.62). MR analyses revealed negative causal relationships from SCZ to BMI, BFP, WC, and HC. This study clarified that the increased obesity prevalence in individuals with SCZ is not due to the genetic effect of SCZ itself. Rather, genetic variants associated with SCZ have a negative impact on both causality and correlations between SCZ and obesity.

Background Sleep traits have been linked to cognitive aging, but observational associations are susceptible to confounding and reverse causation. Using genetically informed polygenic risk scores (PRSs), this study examined associations between genetic liability to sleep traits and cognitive function in an East Asian population. We further assessed whether educational attainment, as a marker of cognitive reserve, modifies these associations. Methods Participants aged ≥ 60 years from the Taiwan Biobank underwent cognitive assessment using the Mini-Mental State Examination (MMSE) at baseline (n = 27,343) and follow-up (n = 6,273; mean follow-up = 3.9 years). PRSs for chronotype, sleep duration, long sleep, short sleep, insomnia, and daytime napping were constructed. Associations with baseline MMSE and longitudinal MMSE change were evaluated, and PRS × education interactions were tested. Results Higher PRSs for early chronotype, longer sleep duration, and long sleep (> 8 h) were associated with lower baseline MMSE scores (β range per SD increase: −0.05 to − 0.03), but were not associated with MMSE change over follow-up. A significant interaction was observed between long-sleep PRS and education for MMSE change (p for interaction = 0.0008), with greater decline among individuals with lower educational attainment. In contrast, no association was observed in higher-education groups. Conclusions Genetic liability to early chronotype and long sleep duration is associated with lower late-life cognitive performance, primarily reflecting differences in baseline cognitive status rather than short-term cognitive decline. Higher educational attainment may mitigate cognitive vulnerability associated with genetic predisposition to long sleep, consistent with a cognitive-reserve framework.

BACKGROUND Understanding the clinical course of psychiatric disorders is vital for informed decision-making. Extracting details such as onset time, episode count, and hospitalization history from unstructured psychiatric notes remains challenging. OBJECTIVE This study evaluates the performance of large language models (LLMs)—LLaMA, MentaLLaMA, OpenBioLLM, and Mistral—in extracting temporal and event-based clinical information from psychiatric discharge summaries. METHODS We used 500 annotated discharge summaries from the NTUH-iMD, covering psychiatric diagnoses (ICD-9-CM: 290–319, ICD-10-CM: F00–F99). Key temporal and clinical course features were annotated by a psychiatrist and an NLP researcher. A two-stage extraction process was implemented: first, sentence-level models identified clinical events and temporal cues; then, a chart-level model predicted four clinical course features: onset time, episode count, number of hospitalizations, and most recent hospitalization time. Performance was evaluated using F1-scores. RESULTS Among 12,947 analyzed sentences, 7,177 included clinical events and 4,842 contained temporal information. Mistral achieved the best performance in event extraction (episodes: 0.968; hospitalizations: 0.933; remission/response: 0.901) and temporal information (age: 0.968; time expressions: 0.967; duration: 0.901). In chart-level extraction, F1-scores were highest for Mistral in onset time (0.714), episode count (0.624), number of hospitalizations (0.676), and last hospitalization time (0.842). CONCLUSIONS Fine-tuned LLMs, especially Mistral, can accurately extract structured clinical course information from psychiatric notes, offering a scalable alternative to manual review. Future work should address vague temporal expressions, expand feature sets, and validate generalizability across diverse settings. CLINICALTRIAL None

Type 2 diabetes is a complex disease influenced by both genetic and environmental factors. Although polygenic risk scores (PRS) and family history (FH) are established predictors, their combined utility in East Asian populations remains unclear. In this study, we analyzed 111,899 unrelated participants from the Taiwan Biobank with linkage to nationwide health registry data. We constructed transancestry, East Asian–specific, and European-derived PRS and assessed their associations with type 2 diabetes. The transancestry PRS improved the explained variance in disease risk compared with ancestry-specific PRS and FH. FH, particularly sibling history, also showed a strong association with type 2 diabetes. When combined, PRS and FH contributed independently and additively, improving predictive performance. These findings demonstrate that PRS and FH are independent yet complementary risk indicators and support the application of transancestry PRS models in East Asian populations. Incorporating both genetic and familial risk factors may enhance early detection and facilitate more personalized prevention strategies for type 2 diabetes. Article Highlights: Type 2 diabetes is influenced by genetic and environmental factors. Although polygenic risk scores (PRS) and family history (FH) are established predictors, their combined utility in East Asian populations is unclear. We evaluated the predictive performance of PRS and the independent and joint effects of PRS and FH on diabetes risk in an East Asian population. The transancestry PRS explained more variance than ancestry-specific PRS and FH. PRS and FH were independently predictive, and their combination improved prediction, with stronger effects in men. Sibling history showed the strongest familial association. Integrating PRS and FH may enhance diabetes risk prediction in East Asian populations.