Introduction: Head-to-head comparison of atezolizumab- bevacizumab and lenvatinib in the patients with unresectable hepatocellular carcinoma (HCC) remained limited in the Asian population.

Aims: We aimed to compare atezolizumab-bevacizumab and lenvatinib among HCC patients in viral and non-viral hepatitis groups using a target trial emulation framework.

Methods: We emulated a target trial that compared two first-line drugs using a multi-institutional hospital electronic medical record database. We included HCC patients who were newly treated with atezolizumab-bevacizumab or lenvatinib between 2023 and 2024. To confirm the HCC stage, we used an on-premises LLM, the GPT-OSS 20B model, to extract unstructured imaging text reports within 3 months of the first treatment date. Clinical outcomes were progression-free survival (PFS) and overall survival (OS). We used inverse probability of treatment weighting (IPTW) as a proxy randomization method. Pure subgroup analyses were conducted stratified by viral hepatitis, non-viral hepatitis, and liver function. Cox regression model was applied to analyze the relative effect between the two groups.

Results: A total of 397 patients were included with a median age of 67.8 years and 79.1% male. 52.5% and 24.2% patients experienced HBV and HCV infection, respectively. 34.3% and 23.6% patients had extrahepatic spread and macrovascular invasion, respectively. After IPTW, atezolizumab-bevacizumab had similar clinical outcomes with lenvatinib [PFS: 1.19 (95% CI: 0.86 - 1.65); OS: 1.14 (0.81 - 1.60)] among the entire cohort and the viral hepatitis subgroup. However, lenvatinib had better PFS and OS than atezolizumab-bevacizumab among the non-viral hepatitis group [PFS: 2.11 (95% CI: 1.15 - 3.89); OS: 1.90 (1.01 - 3.57)].

Conclusion: Our study sugguested that immunotherapy and tyrosine kinase inhibitors might have different clinical benefits for different etiologies of HCC. Futher study is suggested to confirm our finding.

Key word: target trial emulation, HCC, atezolizumab-bevacizumab, lenvatinib

INTRODUCTION: In January 2022, sacubitril-valsartan was included on the funding list for patients with heart failure with reduced ejection fraction (HFrEF). A local real-world study was conducted to examine how pivotal PARADIGM-HF trial evidence translates to the local context.

AIMS: This study assessed real-world outcomes of sacubitril-valsartan compared to angiotensin-converting enzyme inhibitor (ACEi) or angiotensin II receptor blocker (ARB), among patients with heart failure (HF).

METHODS: A retrospective cohort study was conducted using linked national health record databases to identify patients with HF who initiated sacubitril-valsartan or ACEi/ARB between 2017 and 2023. Prevalent new user design was adopted for cohort treatment assignment. Patient characteristics such as demographics, comorbidities, previous HF hospitalisation, baseline biomarker values and use of concomitant drugs were balanced using Lasso techniques. Poisson regression estimated incidence rate ratios (IRR) with 95% confidence intervals (CIs) for cardiovascular (CV) deaths, HF hospitalisation, composite outcomes, and secondary outcomes during first year follow-up.

RESULTS: The final cohort included 3,044 patients (sacubitril-valsartan, n=1,667; ACEi/ARB, n=1,377). Patients on sacubitril-valsartan had lower but non-significant CV deaths (IRR: 0.77; 95% CI 0.58, 1.02), HF hospitalisation (IRR: 0.86; 95% CI 0.65, 1.14) and composite outcomes (IRR: 0.85; 95% CI 0.68, 1.06). However, sacubitril-valsartan was associated with significant reductions in all-cause mortality (IRR: 0.77; 95% CI 0.64, 0.93) and hospitalisations (IRR: 0.79; 95% CI 0.70, 0.90).

CONCLUSION: Sacubitril-valsartan was associated with reduced HF hospitalisation and CV deaths, although these primary outcomes did not reach statistical significance. The findings align with other published real-world studies where effect estimates have shown variable statistical significance. A key limitation was the absence of left ventricular ejection fraction data, which may have impacted patient selection and confounding adjustment. Despite these limitations, sacubitril-valsartan demonstrated significant reductions in all-cause hospitalisations and mortality, suggesting clinical benefit in the local population.

KEYWORDS: Heart failure, sacubitril-valsartan, real-world outcomes

Introduction: Preventing vascular complications is central to type 2 diabetes (T2D) management. Sulfonylureas remain widely used glucose-lowering agents worldwide. Previous studies indicated possible cardiovascular harms from sulfonylureas, but findings are inconsistent. Evidence simultaneously assessing cardiovascular, peripheral vascular, and microvascular complications within the same population is limited.

Aims: To systematically evaluate associations between sulfonylurea use and vascular complications among T2D patients and assess potential causality using Mendelian randomization.

Methods: UK Biobank participants with T2D receiving oral antidiabetics (sulfonylurea or others) and free of targeted vascular complications were included. Medication use was self-reported. Incident complications were identified using diagnostic codes from linked self-report, hospital, primary care, and death records. Multivariable Cox proportional hazards models were used to estimate adjusted hazard ratios. Two-sample Mendelian randomization was performed to assess causal relationships.

Results: A total of 6,144 T2D patients (2,153 sulfonylurea users) aged 59.2 ± 7.1 years with mean HbA1c of 7.0% and mean diabetes duration of 6.6 years were included. After median follow-up of 12.8–13.5 years, compared with other oral antidiabetics, sulfonylurea use was associated with significantly higher risks of peripheral artery disease (HR 1.29, 1.01–1.65) and microvascular disease (HR 1.27, 1.16–1.39). Subtype analyses for microvascular disease showed increased risks for diabetic retinopathy (HR 1.33, 1.19-1.48), neuropathy (HR 1.37, 1.10-1.70), and nephropathy (HR 1.18, 1.03-1.35). No significant associations were observed for cardiovascular disease, ischemic heart disease, or stroke. Mendelian randomization suggested causal effects of sulfonylureas on diabetic retinopathy (OR 4.85, 2.59–9.08) and coronary heart disease (OR 1.91, 1.29–2.84). No causal associations were found for the other outcomes.

Conclusions: Sulfonylureas increase the risk of microvascular disease, particularly retinopathy, among T2D patients. These findings were partially supported by Mendelian randomization analyses and highlight the need for further studies to confirm.

Keywords: Sulfonylureas; Type 2 diabetes; Diabetic vascular complications

Introduction: Patients with drug-resistant epilepsy are at increased risk of mortality, disability, and poor quality of life. Epilepsy surgery is an important treatment option for these patients; however, accurate localization of the epileptogenic zone is essential prior to surgery. In some patients, non-invasive investigation methods fail to precisely identify the seizure onset zone, necessitating invasive intracranial monitoring. Two principal invasive investigations, subdural electrode (SDE) and stereo-electroencephalography (SEEG) monitoring, are being considered for inclusion in Thailand’s Universal Health Coverage benefit package.

Aims: To evaluate the cost–utility of SEEG and SDE in patients with drug-resistant epilepsy whose epileptogenic zone could not be identified using non-invasive pre-surgical investigations.

Methods: A cost–utility analysis was conducted using a decision tree combined with a Markov model to estimate lifetime costs and quality-adjusted life years (QALYs) associated with SEEG and SDE compared with standard anti-epileptic drug treatment. Clinical effectiveness and transition probabilities were derived from patient data obtained from the Neurological Institute of Thailand. Direct medical costs were collected from the institute’s financial database, while direct non-medical costs and utility data were obtained through patient interviews. Additional model parameters were identified through a review of national and international literature. Outcomes were presented as incremental cost-effectiveness ratios. One-way and probabilistic sensitivity analyses were conducted to assess the robustness of the model parameters.

Results: Both SEEG and SDE were found to be cost-effective compared with standard anti-epileptic drug treatment, as they resulted in lower overall costs and greater QALYs. SEEG reduced treatment costs by 300,292 THB and increased QALYs by 1.297 years, while SDE reduced costs by 346,577 THB and increased QALYs by 0.247 years.

Conclusions: SEEG and SDE should be included in Thailand’s health benefit package to improve access to epilepsy surgery evaluation while generating cost savings from a societal perspective.

Keywords: Drug-resistant epilepsy; stereoelectroencephalography (SEEG); cost–utility analysis

Introduction: Health technology assessment (HTA) for cardiovascular disease (CVD) primary prevention relies on individual-level microsimulation, yet the nested-loop structure of probabilistic sensitivity analysis (PSA) creates a well-recognised computational bottleneck—published runtimes range from hours to days, leading to the view that PSA is “impractical for patient-level models.” Existing remedies use statistical variance reduction or cloud-based horizontal scaling, but neither restructures the simulation algorithm itself; current health economic modelling software remains CPU-bound with no GPU support.

Aims: To develop a precision medicine model for health technology assessment

Methods: We developed PPCVD-HEM+, a high-performance extension of the Primary Prevention of Cardiovascular Disease Health Economics Model that re-encodes the Markov microsimulation as tensorised array operations amenable to hardware acceleration. Three advances are introduced: (1) accelerated backends—a Numba JIT engine for CPU-vectorised execution and a PyTorch engine for GPU-batched execution (MPS/CUDA)—replacing sequential loops with parallel tensor algebra; (2) a Monte Carlo Tree Search (MCTS) framework for automated treatment optimisation across 136 PBS-listed drug-dose entries; and (3) a web platform (https://ppcvdhem.com) for browser-based cost-effectiveness analysis. All engines preserve the original sum-then-split event logic. To our knowledge, this is the first application of GPU-accelerated tensor computation to health economic microsimulation.

Results: The CPU engine achieved 29.7-fold speedup with negligible outcome differences (standardised mean difference < 0.05). The GPU engine achieved 208-fold speedup, completing 1,000 PSA iterations in 53 seconds versus ∼3.1 hours for STATA. Multi-arm GPU-batched PSA completed 1,000 iterations in ∼90 seconds (60-fold improvement). MCTS identified early intensive combination therapy as dominant across risk profiles, with net monetary benefits of AU$19,252– 143,953 at a willingness-to-pay threshold of AU$28,000/QALY.

Conclusions: By replacing traditional microsimulation bottlenecks with tensorised GPU-batched computation, PPCVD-HEM+ achieves orders-of-magnitude speedups that enable real-time HTA workflows, automated treatment optimisation, and accessible web-based cost-effectiveness analysis.

Keywords: Health Economics

Introduction: Type 2 diabetes mellitus (T2DM) significantly accelerates cardiovascular risk through emerging comorbidities.

Aims: This study utilizes a multi-state illness-death model to map the precise velocities, sequential trajectories, and direct mortality risks associated with the development of isolated or concurrent hypertension (HT) and dyslipidemia (DLP) following a T2DM diagnosis.

Methods: A dynamic cohort of 56,584 adult patients diagnosed with baseline T2DM between 2015 and 2017 was tracked over a strict 4-year fixed administrative horizon. Patients and clinical transitions were identified using validated ICD-10 diagnostic codes. Continuous-time state changes were dual-validated using non-parametric estimation across a 6-state matrix spanning baseline T2DM, intermediate comorbidities, compound sequential multimorbidity, and all-cause mortality.

Results: Out of baseline T2DM, 32.76% of patients remained stable without developing secondary complications over 4 years. Transitioning to an isolated comorbidity was highly prevalent, with 30.85% developing isolated HT first and 24.04% developing isolated DLP first, while 6.31% experienced simultaneous onset. Crucially, intermediate pathways exhibited highly divergent survival windows. Patients occupying the isolated HT state faced a devastating 10.79% probability of direct all-cause mortality before developing further complications. Conversely, patients occupying the isolated DLP state experienced a direct mortality probability of only 3.15%, revealing a 3.4-fold crude risk increase for the hypertension pathway. Once patients progressed to the dual-comorbidity compound state, 4-year mortality risks converged and stabilized at 10.33% for simultaneous onset and 10.03% for sequential onset.

Conclusions: Entering the isolated HT phase represents a highly volatile clinical inflection point carrying a disproportionately elevated mortality toll compared to isolated lipid progression. These findings mandate aggressive, immediate blood pressure stabilization strategies at the earliest presentation of isolated HT, long before dual multi-morbidity manifests, to mitigate the steep velocity of early mortality in T2DM patients.

Keyword : Competing risks, Type 2 diabetes mellitus, Real-world evidence

Introduction: Combination therapy with glucagon-like peptide-1 receptor agonists and sodium-glucose cotransporter-2 inhibitors is increasingly used in type 2 diabetes (T2D). However, the joint effects of these therapeutic targets on kidney function remain incompletely understood. Drug-target Mendelian randomization may help evaluate the long-term effects of genetically proxied target perturbation on renal outcomes.
Aims: To investigate the independent associations of GLP1R and SLC5A2 variants with kidney function. To assess potential interaction effects between these therapeutic targets on estimated glomerular filtration rate (eGFR).
Methods: Target-specific genetic scores for GLP1R and SLC5A2 were constructed using cis-acting functional variants within each gene region. Multiplicative interaction terms between genetic scores were evaluated using linear regression models adjusted for age and sex. Analyses were conducted in 464,045 UK Biobank participants with available genotype and eGFR data. Associations were estimated using beta coefficients, 95% confidence intervals (CI), and p-values.
Results: Both GLP1R and SLC5A2 genetic scores were associated with lower eGFR. The association for SLC5A2 was β = −0.101 (95% CI: −0.135 to −0.067; p = 6.78 × 10^−9). The association for GLP1R was β = −0.082 (95% CI: −0.116 to −0.048; p = 2.29 × 10^−6). A significant interaction between GLP1R and SLC5A2 genetic scores was observed (β = −0.052; p = 0.0028). The association between GLP1R scores and eGFR was stronger among individuals with higher SLC5A2 genetic scores (75th percentile: β = −0.084, 95% CI: −0.118 to −0.050; p = 1.20 × 10^−6). These findings suggest potential joint effects of GLP1R and SLC5A2 pathways on kidney function.
Conclusions: Genetic variation in GLP1R and SLC5A2 was associated with kidney function. Evidence of interaction between these pathways may inform future studies evaluating combination therapies and renal outcomes in T2D.
Keywords: Mendelian randomization; GLP1R; SLC5A2; eGFR; kidney function