Miao, Jing

Background/Objective: Hypertension is the most prevalent chronic disease in primary care and a leading cause of cardiovascular morbidity and mortality. Despite existing guidelines, therapeutic inertia and suboptimal control persist. Large language models (LLMs) offer a potential valuable addition to augment clinical decision-making, yet their reliability for guideline-driven tasks remains unverified. This study evaluated the accuracy and safety of hypertension management recommendations generated by three LLMs compared to expert responses. Methods: Fifty-one clinical vignettes representing 17 core hypertension management concepts were constructed by hypertension experts. Each case was submitted to three LLMs (GPT-4, Gemini, MedLM) and a hypertension expert also wrote the “gold standard” answers. Three blinded expert reviewers rated each response on a 4-point accuracy scale, a binary safety (safe/unsafe) scale, and attempted to identify the source (LLM vs. expert) providing the response. Ratings were analyzed using mean scores, percentages of accurate and safe responses, and inter-rater agreement. Results: GPT-4 had the highest accuracy (83%) and safety (86%) scores among LLMs but remained inferior to expert responses (92% accuracy, 93% safety). Gemini and MedLM performed significantly worse (accuracy: 64% and 35%; safety: 73% and 39%, respectively). GPT-4 generated the most guideline-concordant responses (46%) among the three LLMs (Gemini 35%, MedLM 14%), but remains lower than experts’ responses (68%). Evaluators misidentified LLM responses as expert-written in 10 to 25% of cases, particularly with GPT-4. Inter-rater reliability for accuracy ratings was highest for expert-generated responses (ICC 0.81), with progressively lower agreement for GPT-4 (0.76), Gemini (0.70), and MedLM (0.68). A similar pattern was observed for safety and source discrimination ratings. The agreement was strongest for safety assessments and weakest for source discrimination. Conclusion: Among three tested LLMs, GPT-4 demonstrated closer agreement to expert decisions thereby showing greater potential for supporting hypertension management. However, current LLMs’ versions frequently produce inaccurate or unsafe recommendations and remain inferior to expert judgment. Human-in-the-loop supervision remains essential when deploying LLMs for clinical decision-making.

BACKGROUND: Hypertension, the leading cause of cardiovascular mortality, remains suboptimally controlled. Large language models (LLMs) could improve hypertension control by augmenting clinical decision-making, but their reliability for guideline-driven tasks is unverified. This study evaluated the accuracy and safety of hypertension management recommendations generated by 3 LLMs. METHODS: Fifty-one vignettes were constructed and submitted to the LLMs (GPT-4, Gemini, Medical Large Language Model [by Google; MedLM]) and a hypertension expert to generate the responses. Three blinded reviewers rated each response on a 4-point accuracy scale, a binary safety (safe/unsafe) scale, and attempted to identify the source (LLM versus expert) providing the response. RESULTS: GPT-4 had the highest accuracy (83%) and safety (86%) scores among LLMs but remained inferior to expert responses (92% accuracy, 93% safety). Gemini and MedLM performed significantly worse (accuracy: 64% and 35%; safety: 73% and 39%, respectively). GPT-4 generated the most guideline-concordant responses (46%) among the 3 LLMs (Gemini 35%, MedLM 14%) but was lower than expert responses (68%). Interrater reliability for accuracy ratings was higher for LLM-generated responses (GPT-4 [intraclass correlation coefficient, 0.30], Gemini [intraclass correlation coefficient, 0.61], and MedLM [intraclass correlation coefficient, 0.58]), with lower agreement for expert responses (intraclass correlation coefficient, 0.23). A similar pattern was observed for safety and source discrimination ratings. The agreement was strongest for safety assessments and weakest for source discrimination. CONCLUSIONS: Among the 3 tested LLMs, GPT-4 demonstrated closer agreement to expert decisions, thereby showing greater potential for supporting hypertension management. Despite their potential, current LLM versions are inferior to expert recommendations. Human-in-the-loop supervision remains essential when deploying LLMs for clinical decision-making.

Abstract Background and Aims Malignancy is commonly seen in patients with membranous Nephropathy (MN) with a prevalence up to 22%. M-type phospholipase A2 receptor (PLA2R) antigen and its associated antibody account for up to 80% of the cases but its association with malignancy is less clear. To date, no study has specifically evaluated cancer incidence and risk factors in patients with PLA2R associated MN as most studies were conducted in an era before many MN antigens were identified. Here, we report cancer incidence, types of cancer, and effect of immunosuppressive therapy and clinical parameters in a cohort of patients with PLA2R-associated MN. Method A single center retrospective cohort study in adults with diagnosis of PLA2R associated MN from 1/1/1996 to 12/31/2021. Follow-up was defined as the time from MN diagnosis to the last follow-up or if the patient initiated dialysis or underwent transplantation. The primary outcome in this study was the incidence of cancer events reported as cancer incidence/1000 patients/year. The cancer events were compared to the age-adjusted United States (U.S.) general population using Centers for Disease Control (C.D.C) data. The standardized incidence ratio (SIR) was calculated by dividing the cancer rates in our cohort to the cancer rates in general population. Time-to-event analysis was conducted using the Kaplan-Meier method. The log-rank test compared age, sex, smoking habits, hypertension, diabetes, body mass index (BMI), previous immunosuppression (IS), remission of the MN, PLA2R titers at the time of diagnosis. Cox proportional hazards regression was used to evaluate the associations between baseline characteristics and cancer incidence. For the multivariate model of cancer incidence, we evaluated variables clinically relevant for the outcome. Results We identified 239 patients with PLA2R-associated MN, with mean (±SD) age of 52 (±13) years, with 72.8% male (n = 174), median [IQR] BMI of 29.8 [26.3, 33.6] kg/m2, median eGFR of 68 [45, 94] ml/min, median proteinuria of 8.2 [5.1, 11.2] g/24h, and median PLA2R titers of 38 [9, 122] RU/ml. Common comorbidities included hypertension (60.2%), and diabetes (17.1%). Majority of patients drank alcohol (90.7%) and 32.6% were smokers (active or former). Of 239 patients, 15 had a history of cancer before the MN diagnosis. But in all these patients, cancer had been treated and was considered to be in full remission. A diagnosis of cancer was made in 37 patients (15.4%) during a median follow-up of 60 [27.5, 145.5] months. Incidence of cancer in our cohort was 22.6 [95% CI, 15.9 to 31.1] per 1000 persons/year. In comparison, age adjusted U.S. general population has a lower cancer incidence of 15.9 [95% CI, 15.9 to 16.1] cases per 1000 person/year. The resulting SIR of cancer in our cohort was 1.42. This SIR is lower than the one reported by Bjørneklett and al. (2007) in a Norwegian cohort that included all patients with MN and was not limited to PLA2R-positive MN. The cumulative incidence of cancer diagnosis is shown in Fig. 1. Mean age at cancer diagnosis was 66 (±10.2) years old. Prostate cancer was the most common (n = 11) type of cancer, followed by skin cancer (n = 9), followed by cancer of the urinary tract (n = 5). In univariate and multivariate analysis (Table 1), only older age (>65 years) and tobacco use were associated with statistically significant increased risk of cancer. Other clinical parameters such as sex, eGFR, baseline proteinuria, BMI, hypertension, diabetes, alcohol use, anti-PLA2R antibody titer, type of immunosuppression, remission status (partial, complete, no response) were not associated with a significant increased risk of cancer. Conclusion Patients with PLA2R associated MN are at increased risk of developing cancer following diagnosis compared to general population and routine age-appropriate cancer screening in this population is essential. The increased incidence of cancer is independent of IS therapy, response to therapy or other comorbidities. These data suggests that there are likely mechanisms beyond the traditional risk factors at play that may be unique to the presence of PLA2R antibodies.

Background We aimed to assess the longitudinal changes of glycocalyx, a proteoglycan/glycoprotein layer that lines and protects endothelial cells in normotensive pregnancies (NPs); to correlate glycocalyx parameters with kidney injury markers (podocin‐positive to nephrin‐positive [P/N] ratio); and to explore changes in P/N ratio in preeclampsia. Methods Plasma and urine samples and glycocalyx measurements using side‐stream darkfield imaging were collected at each trimester from pregnant women (n=31). Plasma glycocalyx components were measured by ELISA, and urinary P/N‐positive extracellular vesicles by flow cytometry. Results In NPs (n=26), glycocalyx degradation increased from the first to the second trimester ( P =0.008) and returned to first‐trimester values in the third trimester ( P =0.008). Microvascular perfusion decreased from the first to the second trimester ( P =0.006) and increased back to first‐trimester values by the third trimester ( P =0.039). Plasma syndecan 1 level showed a steady increase from the first to the second trimester ( P <0.001) and further in the third trimester ( P <0.001); hyaluronic acid increased from the first to the second trimester ( P =0.016) without further changes. Glycocalyx degradation was associated with an increase in P/N ratio from the first to the second trimester of NPs and remained stable through the third trimester. P/N ratio strongly correlated with elevated syndecan 1 and hyaluronic acid levels in the third trimester. Comparison of P/N ratios between NP and preeclampsia (n=5) showed higher values in preeclampsia across all trimesters ( P =0.028). Conclusions Glycocalyx changes during NPs are associated with elevated urinary podocyte markers. Increase in P/N ratio across NPs, and further in preeclampsia, suggests that podocyte injury in preeclampsia is extension of physiological changes in pregnancy, which may predate its clinical presentation.

ABSTRACT Background Membranous nephropathy (MN) associated with phospholipase A2 receptor (PLA2R) antibodies is the most common cause of nephrotic syndrome in non-diabetic adult patients. This study investigated the relationship between clinical phenotypes and renal outcomes in this population, emphasizing the potential for phenotype-based treatment stratification. Methods We conducted a retrospective, multi-center cohort study of PLA2R-positive MN. Unsupervised cluster analysis grouped patients based on clinicopathological characteristics. Primary outcomes included complete or partial remission within 2 years of biopsy and end-stage kidney disease (ESKD) or death during follow-up. Results Among 178 patients, three distinct clusters emerged (n = 89, 70, and 19). Within 2 years of biopsy, 102 patients (57%) achieved complete or partial remission. Cluster 1, characterized by the mildest disease markers, including the lowest body mass index, serum anti-PLA2R titer, serum creatinine, proteinuria, and triglycerides, had the highest remission rate of 72%, compared with 50% in cluster 2 and 54% in cluster 3. Cluster 2 had significantly lower remission compared to cluster 1 (HR 0.64, 95% CI 0.42–0.96, P = .03); results were similar in cluster 3, albeit not statistically significant (HR 0.50, 95% CI 0.23–1.09, P = .08). Elevated anti-PLA2R levels (&gt;100 U/ml) and proteinuria (&gt;8 g/24-hour) predicted reduced remission (HR 0.56, 95% CI 0.36–0.88, P = .01; HR 0.43, 95% CI 0.26–0.73, P = .002, respectively), while high high-density lipoprotein levels were protective (HR 1.01, 95% CI 1.00–1.02, P = .048). Overall, 21 patients (12%) developed ESKD or died. ESKD-free survival differed significantly across clusters (log-rank test P = .04). Conclusions Unsupervised clustering identified distinct clinical phenotypes in PLA2R-positive MN, each associated with different renal prognoses. Phenotype-based risk stratification could enhance treatment precision, improve patient outcomes, and potentially reduce treatment-related adverse effects.

Key Points Preexisting CKD, elevated serum creatinine, and severe interstitial fibrosis and tubular atrophy were linked to poor renal prognosis in renal sarcoidosis. Steroid therapy was strongly associated with renal recovery and improved long-term outcomes in renal sarcoidosis. Early detection and management of renal involvement in sarcoidosis are critical to preventing irreversible kidney damage. Background Renal sarcoidosis (RS) is a rare but potentially severe manifestation of sarcoidosis, primarily affecting the kidneys through granulomatous interstitial nephritis and calcium metabolic disturbances. This study evaluates the clinicopathologic features and renal outcomes of biopsy-proven RS, focusing on identifying predictors of renal recovery and disease progression. Methods This retrospective study included 43 biopsy-proven RS at Mayo Clinic (2012–2024). Demographic, clinical, laboratory, and histopathologic data were analyzed. Renal recovery within 6 months was classified as complete, partial, or no recovery based on serum creatinine (sCr) changes. Long-term adverse composite outcomes included the CKD onset, CKD progression, ESKD, dialysis initiation, or death. Logistic regression and Cox proportional hazards regression were used to assess predictors of renal outcomes. Results Among the cohort, 84% exhibited interstitial nephritis, with granulomatous interstitial nephritis in 33%. Calcium phosphate deposits were found in 58% of cases. Hypercalciuria (90%) was more common than hypercalcemia (74%). Within 6 months, 86% achieved recovery (74% complete, 12% partial), while 40% experienced long-term adverse composite outcomes. Higher sCr at biopsy (≥2.4 mg/dl) and moderate-to-severe interstitial fibrosis and tubular atrophy were associated with reduced complete recovery and worse long-term outcomes, while preexisting CKD significantly increased the risk of adverse long-term outcomes. Steroid therapy significantly increased the likelihood of complete recovery and reduced the risk of adverse long-term outcomes. Conclusions Preexisting CKD, elevated sCr at biopsy, interstitial fibrosis and tubular atrophy, and steroid therapy are key predictors of renal outcomes in RS. Early diagnosis and intervention are crucial to preventing irreversible kidney damage.

Background The digital transformation in medical education is reshaping how clinical skills, such as point-of-care ultrasound (POCUS), are taught. In nephrology fellowship programs, POCUS is essential for enhancing diagnostic accuracy, guiding procedures, and optimizing patient management. To address these evolving demands, we developed an artificial intelligence (AI)-driven POCUS curriculum using a multiagent approach that integrates human expertise with advanced AI models, thereby elevating educational standards and better preparing fellows for contemporary clinical practice. Methods In April 2024, the Mayo Clinic Minnesota Nephrology Fellowship Program initiated a novel AI-assisted process to design a comprehensive POCUS curriculum. This process integrated multiple advanced AI models—including GPT-4.0, Claude 3.0 Opus, Gemini Advanced, and Meta AI with Llama 3—to generate initial drafts and iteratively refine content. A panel of blinded nephrology POCUS experts subsequently reviewed and modified the AI-generated material to ensure both clinical relevance and educational rigor. Results The curriculum underwent 12 iterative revisions, incorporating feedback from 29 communications across AI models. Key features of the final curriculum included expanded core topics, diversified teaching methods, enhanced assessment tools, and integration into inpatient and outpatient nephrology rotations. The curriculum emphasized quality assurance, POCUS limitations, and essential clinical applications, such as fistula/graft evaluation and software integration. Alignment with certification standards further strengthened its utility. AI models contributed significantly to the curriculum's foundational structure, while human experts provided critical clinical insights. Conclusion This curriculum, enhanced through a multiagent approach that combines AI and human collaboration, exemplifies the transformative potential of digital tools in nephrology education. The innovative framework seamlessly integrates advanced AI models with expert clinical insights, providing a scalable model for medical curriculum development that is responsive to evolving educational demands. The synergy between technological innovation and human expertise holds promising implications for advancing fellowship training. Future studies should evaluate its impact on clinical competencies and patient outcomes across diverse practice environments.