Introduction: Prediabetes elevates the risk of type 2 diabetes (T2D), premature mortality, and complications like cardiovascular disease (CVD), driven by interacting risk factors, calling for risk stratification. High-risk individuals are prime candidates for targeted and cost-effective interventions. The challenge lies in identifying those at elevated risk of progressing to T2D and related complications.

Aims: To develop an electronic health record (EHR)-based machine-learning (ML) model to predict 5-year diabetes risk in adults with prediabetes, and long-term risks of complications and mortality.

Methods: A retrospective cohort of adults with prediabetes defined by fasting plasma glucose (FPG) and HbA1c (2002–2014, followed until December 31, 2019) was used to develop a ML model with routine measures (e.g., age, FPG, HbA1c, triglyceride-to-HDL-cholesterol ratio) from the Hong Kong territory-wide EHR system. Individuals were stratified into low/intermediate- and high-risk groups using validated model-derived risk scores. Incidence rates of diabetes and complications were calculated, and associations with all-cause mortality, CVD, and heart failure were assessed using Cox-proportional hazards regression, adjusted for age, sex, and clinical covariates (e.g., FPG).

Results: The EHR-based 5-year diabetes risk model included 545,054 adults with prediabetes (mean age: 61.3±13.1 years). The high-risk group (188,735; 34.6%) had a 5-year diabetes incidence rate of 92.5 events versus 24.3 events per 1,000 person-years in the low/intermediate group. During a mean duration follow-up of 9.4 years, the high-risk group had similar all-cause mortality risk (adjusted hazard ratio [aHR]=1.01, 95% CI 0.98–1.02) but higher risks of CVD (aHR=1.14, 95% CI 1.12–1.16) and heart failure (aHR=1.14, 95% CI 1.12–1.16) compared to the low/intermediate risk group.

Conclusions: We have developed an ML model based on EHR data which effectively identifies high-risk individuals for diabetes progression with cardiovascular complications. The ML model can be integrated within the EHR system to select high-risk individuals for early prevention.

Keywords: prediabetes; diabetes; prediction.

Introduction and aims: Target Trial Emulation (TTE) has ushered popularity because of its ability to improve the reliability of causal inference from observational data. This study aimed to comprehensively master related knowledge current use, potential challenges and insights of target trial in oncology.
Methods: Systematic literature review of global oncology TTE studies based on PubMed and Embase was conducted. Time trend, basic characterizes, bias assessment, and consistency with the results of emulated RCT were analyzed.
Results: A total of 60 TTE studies in cancer areas were identified, with the annual number in 2024 being equal to the sum of rest years. Among the 35 applications in cancer treatment, registry databases (51.4%) and overall survival (80%) were predominantly used as data sources and primary endpoint, respectively. 45.8% of included TTE suffered from immortal time bias, and 48.6% from prevalent user selection bias. Among the 18 trials from 17 studies aiming to calibrate the results from preexisting RCTs, only 44.4% trials met both statistical agreement and estimate agreement. The availability of fit-for-purpose data sources and the uncertainty of concordance in results were identified as the two main hurdles for limited quantity and quality of TTE in oncology areas considering its unique challenges.
Conclusions: The application of target trial emulation in oncology has seen a significant increase in 2024, while the overall quantity and quality are still limited, which could be largely constrained by the availability of fit-for-purpose data sources and the uncertainty of concordance in results. Potential solutions were recommended for improving the feasibility and quality of oncology trial emulation, including promoting regulatory acceptance, data integration of medical rerecords and linkage with insurance claims databases, as well as best practice in trial design, including modernizing eligibility criteria, using overall survival as primary endpoint.

Introduction: Methylphenidate (MPH) is widely used to treat attention-deficit/hyperactivity disorder (ADHD), including in patients with depression. However, concerns remain that MPH may increase the risk of manic episodes and transition to bipolar disorder (BD) in specific individuals.
Aims: We aim to estimate the treatment effect of administrating MPH on BD occurrence in patients with ADHD and depression using causal forest model with large-scale propensity scores.
Methods: We used the Health Insurance Review and Assessment Service-ADHD database (2016–2020), which contains nationwide claims data converted to OMOP-CDM. To evaluate the treatment effect of MPH on BD occurrence in adult patients with ADHD and depression, patients were divided into MPH-treated and untreated groups. Propensity scores were estimated via random forests after covariate screening using 10-fold cross-validation. A causal forest model with inverse-propensity weighting was used to estimate average and conditional treatment effects (ATE, CATE). To assess ATE heterogeneity, the validation set was divided into CATE-based quintiles, and ATEs were estimated using targeted minimum loss estimation. We identified key features of high versus low CATE groups by comparing distributions of the top 15 variables by importance.
Results: Among 28,939 patients, 19,939 were prescribed MPH and 1,881 were diagnosed with BD. A total of 4,608 baseline covariates were extracted and reduced to 4,477 after cross-validation. In the validation set, ATE was 0.018 [0.011–0.053]. ATEs by CATE quintiles (Q1–Q5) were 0.044, 0.036, 0.029, -0.002, and 0.015, respectively; Q1–Q3 showed significant effects. ATE heterogeneity among the groups was significant (χ²₄= 17.98, p = 0.0012); indicating treatment effects of MPH differed among these subgroups. All top 15 covariates differed significantly between high and low CATE groups, except lorazepam.
Conclusions: MPH may elevate BD risk in specific populations, and CATE-based heterogeneity supports individualized treatment strategies to guide safer MPH use.
Keywords: Machine learning, treatment effect, bipolar disorder

Introduction
Chronic Kidney Disease (CKD) is a growing global disease burden leading to a range of systemic complications. One of the major problems in dosage adjustment of renally excreted drugs in CKD patients is lack of sufficient data to calculate creatinine clearance. Availability of different formulas with different criteria are a solution for the same but agreement among these formula remains unclear.

Aim
To evaluate agreement across three standard equations to calculate creatinine clearance in hospitalized CKD patients.

Methods
A prospective observational study was carried out in Nephrology department after approval of institutional ethics committee. CKD patients admitted in Nephrology department with age ≥18 years were included in the study. Variables included age, gender, comorbidities, height, weight and serum creatinine values. Creatinine clearance or eGFR was calculated using Modification of Diet in Renal disease(MDRD),Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) and Cockcroft-Gault. Intraclass correlation coefficient (ICC) was used to assess agreement across three standard formulas.

Results
There was a male preponderance with 63.6 % out of 101 study sample. Diabetes mellitus was the predominant co-morbidity observed. CKD stage 5 patients comprised 83% study population. An intraclass correlation coefficient (ICC) of 0.917 was observed demonstrating excellent agreement between the three-creatinine clearance estimation formula. A strong agreement with ICC of 0.948 was observed in elderly patients also. There was no gender bias as both genders showed strong agreement.

Conclusion
As there exists a strong agreement between 3 commonly used formulas with different criteria to calculate creatinine clearance, anyone of these formulas can be used for dosage adjustment of renally impaired patients. Despite the strong statistical agreement, further research and reviews are needed to ensure safe and individualized treatment in such patients.

Keywords- MDRD,CKD-EPI,Cockcroft-Gault

Introduction: Diabetic kidney disease (DKD) is a major contributor to end-stage renal disease worldwide, particularly in Asia, and poses significant clinical and economic challenges. Early identification of biomarkers and repurposable drug targets is essential for timely intervention and improved patient outcomes. This study aimed to uncover key molecular signatures and therapeutic candidates in DKD through integrative transcriptomic and pharmacogenomic analysis.
Methods: A publicly available glomerular transcriptomic dataset of DKD patients and healthy controls was analyzed to identify differentially expressed genes (DEGs). Protein–protein interaction (PPI) networks were constructed using STRING and functionally annotated through Gene Ontology, KEGG, and Reactome. Upstream transcription factors were predicted via ChEA, and drug–gene interactions were identified using DGIdb, with a focus on drugs with existing clinical approval. Pathways and genes were prioritized based on relevance to glomerular filtration and podocyte function.
Results: Significant upregulated and downregulated DEGs formed a highly enriched PPI network (p < 1.0e-16). Enriched modules were associated with extracellular matrix remodeling, proteolysis regulation, and podocyte differentiation. Notably, genes such as SERPINA3, COL6A3, and AKR1B10 were upregulated, while NPHS1, PLCE1, and IGF1 were downregulated. Transcriptional regulators included RELA, ESR1, and CJUN (upregulated) and SMAD3, SUZ12 (downregulated). Drug–gene interaction analysis identified repositionable drugs with potential nephroprotective effects, including losartan (NPHS1), hydrochlorothiazide (PLCE1), tolrestat (AKR1B10), and fomepizole (ADH1B).
Conclusion: This integrative molecular epidemiology study highlights biomarker candidates and drug repurposing opportunities in DKD, offering a precision pharmacoepidemiologic approach. The findings support further validation to accelerate biomarker-guided therapeutic strategies and inform clinical decision-making, especially in resource-limited settings.
Keywords:
biomarkers, diabetic kidney disease, drug repositioning, molecular epidemiology, pharmacogenomics, transcriptomics.

Introduction: TreeScan-based approaches showed promise for drug safety monitoring in pregnancy. It is unknown whether TreeScan could be applied for surveillance of infant outcomes following paternal medication use.
Aims: To explore the performance of the TreeScan statistics on selected test cases using Norwegian linked health registries data.
Methods: We utilized the Norwegian linked health registries data, including the Medical Birth Registry of Norway, the Norwegian Prescribed Drug Registry, the Norwegian Patient Registry, and the Norwegian Control and Payment of Health Reimbursements. We included liveborn singleton children with identifiable fathers and mothers between 2010 and 2021. We excluded those with missing gestational age. We evaluated TreeScan using cohort design with propensity score adjustment via stratification. We used a test case that was previously investigated in several large observational studies: valproate and neurodevelopmental disorders (NDD). The unconditional Poisson TreeScan statistic was used to evaluate whether and which NDDs were associated with valproate exposure in the three months prior to conception (i.e., during spermatogenesis), compared to lamotrigine/levetiracetam exposure. Outcome nodes with adjusted p-value < 0.05 and at least two observed cases were considered statistical alerts.
Results: Of 850,493 offspring included, 854 and 1,808 children had fathers exposed to valproate and lamotrigine/levetiracetam in the three months prior to conception, respectively. We evaluated 42 nodes up to level 3 of the ICD-10 tree. None of them showed an increased risk of NDDs associated with paternal valproate exposure compared to lamotrigine/levetiracetam (highest relative risk = 0.96 for F84.0 - Autistic Disorder). These results are in line with previously published studies, suggesting that paternal valproate exposure around conception is unlikely to cause NDDs in offspring.
Conclusion: This evaluation shows preliminary results of TreeScan-based approaches for systematic surveillance of neurodevelopmental disorders following paternal exposure to medications during spermatogenesis. More test cases with other outcomes will be evaluated.