Daily Endocrinology Research Analysis
Three high-impact endocrinology papers stand out today: an AI system using retinal images accurately detects diabetic kidney disease and distinguishes isolated diabetic nephropathy from non-diabetic kidney disease across multi-ethnic cohorts; a meta-analysis of randomized trials signals a potential increase in deep-vein thrombosis with long-term GLP-1 receptor agonists; and 21-year follow-up from the DPP/DPPOS confirms durable reductions in type 2 diabetes incidence with lifestyle intervention a
Summary
Three high-impact endocrinology papers stand out today: an AI system using retinal images accurately detects diabetic kidney disease and distinguishes isolated diabetic nephropathy from non-diabetic kidney disease across multi-ethnic cohorts; a meta-analysis of randomized trials signals a potential increase in deep-vein thrombosis with long-term GLP-1 receptor agonists; and 21-year follow-up from the DPP/DPPOS confirms durable reductions in type 2 diabetes incidence with lifestyle intervention and metformin, with heterogeneity by baseline risk.
Research Themes
- AI-enabled noninvasive diagnostics in diabetic complications
- Long-term diabetes prevention and precision risk stratification
- Medication safety signals for GLP-1 receptor agonists
Selected Articles
1. Non-invasive biopsy diagnosis of diabetic kidney disease via deep learning applied to retinal images: a population-based study.
DeepDKD, a retinal image–based deep learning system, detected DKD with AUC 0.842 internally and 0.791–0.826 externally, and differentiated isolated diabetic nephropathy from NDKD with AUC up to 0.906 internal (0.733–0.844 external). In a prospective study, sensitivity surpassed a metadata model (89.8% vs 66.3%), and in a 4.6-year longitudinal cohort, AI-defined nephropathy classes showed divergent eGFR decline.
Impact: This work proposes a scalable, noninvasive pathway to screen for DKD and triage biopsy decisions across diverse populations, with prospective and longitudinal validation. It could transform access to nephropathy risk stratification in diabetes care.
Clinical Implications: Retinal-image AI could be deployed alongside albuminuria and eGFR to prioritize nephrology referral, intensify renoprotective therapy, and identify candidates for confirmatory workup when NDKD is suspected.
Key Findings
- Internal DKD detection AUC 0.842; external AUCs 0.791–0.826 across 10 multi-ethnic datasets.
- Isolated diabetic nephropathy vs NDKD differentiation: internal AUC 0.906; external AUCs 0.733–0.844.
- Prospective real-world study: sensitivity 89.8% vs 66.3% compared with a metadata model (p<0.0001).
- Longitudinal 4.6-year analysis: AI-identified groups differed in eGFR decline (27.45% vs 52.56%, p=0.0010).
Methodological Strengths
- Large-scale pretraining (734,084 images) with multi-ethnic external validation across five countries.
- Prospective and longitudinal proof-of-concept studies demonstrating clinical signal beyond cross-sectional accuracy.
Limitations
- Not a randomized implementation study; potential domain shift and spectrum bias across datasets.
- Black-box model interpretability and integration into clinical workflows remain to be addressed; image quality and device variability may affect performance.
Future Directions: Prospective implementation trials testing clinical outcomes and cost-effectiveness, fairness audits across demographics, model explainability, and integration with laboratory biomarkers and EHR for decision support.
2. Glucagon-Like Peptide-1 Receptor Agonists and Risk of Venous Thromboembolism: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.
Across 39 RCTs (70,499 participants), GLP-1 receptor agonists were associated with a significant increase in DVT risk (OR 1.64), particularly with treatment duration >1.5 years and in cardiovascular outcome trials, while overall VTE and pulmonary embolism were not significantly increased. This safety signal warrants vigilance in long-term users and high-risk patients.
Impact: Given the widespread and expanding use of GLP-1RAs for diabetes and obesity, identifying a time-dependent DVT risk has immediate implications for risk stratification and monitoring.
Clinical Implications: Consider baseline VTE risk assessment before initiating GLP-1RAs, counsel on DVT symptoms, reassess risk in prolonged therapy (>18 months), and individualize use in patients with prior VTE, thrombophilia, immobility, malignancy, or perioperative settings.
Key Findings
- Overall VTE showed a nonsignificant trend (OR 1.19; 95% CI 0.94–1.50), but DVT risk was significantly increased (OR 1.64; 95% CI 1.14–2.36).
- DVT risk increase was pronounced in trials with treatment duration >1.5 years (OR 2.32) and in cardiovascular outcome trials (OR 2.18).
- No significant association was found between GLP-1RAs and pulmonary embolism.
Methodological Strengths
- Meta-analysis restricted to randomized controlled trials with 70,499 participants.
- Inclusion of zero-event trials using continuity correction and predefined subgroup analyses by duration and trial type.
Limitations
- Fixed-effects models may underrepresent between-study heterogeneity; lack of patient-level data limits adjustment for confounders.
- Event adjudication and reporting may vary across trials; drug-, dose-, and indication-specific risks could not be fully disentangled.
Future Directions: Individual patient data meta-analyses, mechanistic studies on prothrombotic pathways, and prospective pharmacovigilance to refine risk stratification by drug, dose, duration, and patient phenotype.
3. Long-term effects and effect heterogeneity of lifestyle and metformin interventions on type 2 diabetes incidence over 21 years in the US Diabetes Prevention Program randomised clinical trial.
Over ~21 years, the original ILS and metformin groups maintained lower diabetes incidence than placebo (HR 0.76 and 0.83), increasing diabetes-free survival by 3.5 and 2.5 years. Effects were strongest early and showed larger absolute benefits among participants with higher baseline glycemic risk.
Impact: This definitive long-term analysis confirms durable prevention benefits of lifestyle and metformin, informing precision prevention policies and long-horizon health economic models.
Clinical Implications: Prioritize intensive lifestyle programs for high-risk prediabetes; consider metformin particularly for younger, heavier, or higher-glycemia individuals; monitor heterogeneity to tailor prevention intensity.
Key Findings
- ILS reduced diabetes incidence (HR 0.76; RD −1.59 per 100 person-years) and metformin reduced incidence (HR 0.83; RD −1.17).
- Median diabetes-free survival increased by 3.5 years (ILS) and 2.5 years (metformin); mean increases 2.0 and 1.2 years, respectively.
- Treatment curves diverged early (first 3 years) with convergence over time; absolute benefits were larger in participants with higher baseline glycemic risk.
Methodological Strengths
- Large randomized cohort with intention-to-treat analysis and long-term follow-up.
- Pre-specified outcomes and robust survival analyses across decades.
Limitations
- Post-DPP protocol changes (placebo discontinuation, unmasked metformin continuation, lifestyle offered to all) likely diluted between-group differences.
- Administrative censoring and varying follow-up durations; effect heterogeneity not fully characterized for all subgroups.
Future Directions: Targeted prevention trials leveraging baseline glycemic and metabolic risk to maximize absolute benefit; implementation research to scale intensive lifestyle programs.