Hommos, Musab S

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.

Aging is associated with significant changes in structure and function of the kidney, even in the absence of age-related comorbidities. On the macrostructural level, kidney cortical volume decreases, surface roughness increases, and the number and size of simple renal cysts increase with age. On the microstructural level, the histologic signs of nephrosclerosis (arteriosclerosis/arteriolosclerosis, global glomerulosclerosis, interstitial fibrosis, and tubular atrophy) all increase with age. The decline of nephron number is accompanied by a comparable reduction in measured whole-kidney GFR. However, single-nephron GFR remains relatively constant with healthy aging as does glomerular volume. Only when glomerulosclerosis and arteriosclerosis exceed that expected for age is there an increase in single-nephron GFR. In the absence of albuminuria, age-related reduction in GFR with the corresponding increase in CKD (defined by an eGFR<60 ml/min per 1.73 m 2 ) has been shown to associate with a very modest to no increase in age-standardized mortality risk or ESRD. These findings raise the question of whether disease labeling of an age-related decline in GFR is appropriate. These findings also emphasize the need for a different management approach for many elderly individuals considered to have CKD by current criteria.

Purpose To develop and determine the reliability of a novel measurement instrument assessing the quality of residents’ discharge summaries. Method In 2014, the authors created a discharge summary evaluation instrument based on consensus recommendations from national regulatory bodies and input from primary care providers at their institution. After a brief pilot, they used the instrument to evaluate discharge summaries written by first-year internal medicine residents (n = 24) at a single U.S. teaching hospital during the 2013–2014 academic year. They conducted a generalizability study to determine the reliability of the instrument and a series of decision studies to determine the number of discharge summaries and raters needed to achieve a reliable evaluation score. Results The generalizability study demonstrated that 37% of the variance reflected residents’ ability to generate an adequate discharge summary (true score variance). The decision studies estimated that the mean score from six discharge summary reviews completed by a unique rater for each review would yield a reliability coefficient of 0.75. Because of high interrater reliability, multiple raters per discharge summary would not significantly enhance the reliability of the mean rating. Conclusions This evaluation instrument reliably measured residents’ performance writing discharge summaries. A single rating of six discharge summaries can achieve a reliable mean evaluation score. Using this instrument is feasible even for programs with a limited number of inpatient encounters and a small pool of faculty preceptors.