SCHWARTZ, G

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: Compared to those with natural nocturnal blood pressure (BP) decreases, those lacking nocturnal BP decreases (“non-dippers”) are at greater risk of cardiovascular disease (CVD). Substantial evidence also supports a role of chronic systemic inflammation in the pathogenesis of CVD. The purpose of this study was to examine associations of markers of inflammation with two common definitions of BP dipping status using 24-hour ambulatory blood pressure (AMBP) readings. Methods: AMBP monitoring and assays of C-reactive protein (CRP), interleukin-6 (IL-6), e-selectin, tissue inhibitor of metalloproteinase-2 (TIMP-2), and a selection of thirteen other inflammatory-pathway biomarkers were performed in 722 participants (mean age 64 [range 42-83] years, 36% men, 39% black, 83% hypertensive) in the Genetic Epidemiology Network of Arteriopathy (GENOA) Study (2001-2004). Non-dippers were classified using two common definitions: (1) nocturnal systolic and diastolic BP both decreased by <10% compared to daytime values [“dual non-dippers”, n=161 (22%)], and (2) nocturnal SBP decreased by <10% [“systolic non-dippers”, n=380 (53%)]. Logistic regressions, fit with generalized estimating equations to account for sibship clustering, yielded odds ratios (OR) of dipping status for log-transformed inflammatory biomarkers, adjusted for age, race, and sex. Results: IL-6 (range 0-37 pg/ml) was associated with dual non-dipping status (OR=1.55 [95%CI: 1.03, 2.32] p=0.035), translating into a 55% increase in the likelihood of being a dual non-dipper for every approximate doubling of IL-6. IL-6 (OR=1.63 [1.13, 2.35] p=0.009) and CRP (OR=1.35 [1.08,1.68] p=0.008) overall, as well as e-selectin (OR=0.49 [0.28, 0.87] p=0.014) and TIMP-2 (OR=0.54 [0.30,0.98] p=0.041) in hypertensive subjects were associated with systolic non-dipping status. Conclusions: These results suggest that non-dipping classification should be approached with prudence, as the two different criteria may influence conclusions about predictive marker associations. In addition, markers involved in direct versus inhibitory inflammation pathways may operate differentially across patient subgroups. Our findings support the link between inflammation and non-dipping circadian BP patterns and provide novel information on markers and definitions which may be useful for early detection and subsequent prevention of CVD.