Citations

Abstract Background Measurement of cardiac troponin (cTn) using high-sensitivity assays is recommended for the diagnosis of myocardial infarction. We determined sex-specific and uniform 99th percentile upper reference limits (URLs) using the new Elecsys® Troponin T high-sensitivity Gen 6 assay in a global, healthy reference range cohort. Methods Lithium-heparin plasma and serum samples were prospectively collected from apparently healthy individuals aged ≥20 years across 34 global sites in the United States, Europe, China, and Japan. cTnT was measured using the Troponin T high-sensitivity Gen 6 assay on the Cobas® e 801 analyzer. Exclusion criteria were defined according to the 2022 IFCC guidance. Uniform and sex-specific 99th percentile URLs and nonparametric 95% CIs were determined in plasma and serum separately and combined. Results The final study population comprised 4147 participants (52.5% female) with a median (25th–75th percentiles) age of 48.0 (33.0–59.0) years; 45.8%, 47.9%, 5.2%, and 1.1% were White, Asian, Black, and other/unknown, respectively. For sample matrices combined (n = 8294), 81.0% and 99.2% of cTnT values were above the limit of detection in females and males, respectively. Sex-specific 99th percentile URLs (95% CI) were 18 (16–23) ng/L for females and 32 (28–35) ng/L for males; the uniform 99th percentile URL was 27 (24–31) ng/L. URLs were comparable in plasma and serum samples. Conclusions This study determined sex-specific and uniform 99th percentile URLs for the Troponin T high-sensitivity Gen 6 assay that were comparable irrespective of the matrix used in a large, global, healthy reference population.

Abstract Background The accuracy and safety of generating medication orders by large language models (LLMs) must be demonstrated. Without standardization, performance evaluation is limited to time and resource-intensive clinician grading. This evaluation aimed to develop a standardized medication format that supports automated performance evaluation (MedMatch). Methods First, a survey of 40 medication prompts was given to clinicians to assess agreement in medication order communication. Second, a clinician panel developed a standardized medication format (MedMatch) for oral and intravenous medications. Third, a clinician-annotated dataset of medication prompts and standardized answers in the MedMatch format was developed for LLM testing. Finally, LLMs were retested with the same dataset, adjusted to exclude route information, to evaluate the appropriate categorization of medication route. Results The formal medication orders consistently showed low omission rates and high overlap for all entities, compared to the verbal and brief written communication types. Lexical overlap results demonstrated pattern norms amongst clinicians with entities appearing most commonly in positions 1-5 in the order of drug name, dose, unit, route, and frequency. In the second survey, the formal written group performed the highest with 78.3% of prompts considered appropriate as a computer-generated response. LLM accuracy on MedMatch order standardization was highest in oral solid (64.2-72.5%), intravenous intermittent (72.5-84.3%), and intravenous push (62.7-74.5%) categories. LLMs performed the worst at categorizing medication orders accurately into intravenous push (18-61%) and intravenous intermittent (51-100%) routes. Conclusions Standardized format for computer-based outputs may support automated performance analysis and enhance the clarity of medication communication.

Abstract Background Large language models (LLMs) are increasingly used in medication-related tasks despite limited evidence supporting their accuracy and safety. LLMs are vulnerable to adversarial attacks and may confabulate in response to fabricated information in the prompt. Given the substantial risk of harm from medication errors, the purpose of this study was to establish a benchmark task to evaluate LLM performance in distinguishing fabricated from real medications. Methods Two datasets (brand and generic), each consisting of 250 medication lists were developed, each including 4-6 medications and one fabricated medication (specifically a Pokémon character), with complete dosing information provided. Six LLMs were evaluated (GPT-5-Chat, GPT-4o-mini, DrugGPT, Gemma-3-27B-IT, Llama-3.3-70B-Instruct, and Qwen3-32B). For each medication list, models were queried for dosing information and disease indication across three experimental conditions: standard decoding, standard decoding with mitigation prompting, and deterministic decoding (temperature = 0). Each experiment was performed in triplicate. Outputs were deemed confabulations if the LLM failed to recognize and reject the fabricated medication. The primary outcome was the rate of confabulations. Exact paired-permutation tests were used to compare confabulation rates across LLMs and prompting approaches. Results Confabulation rates under standard conditions ranged from 2.7% to 99.6%. DrugGPT demonstrated the lowest baseline confabulation rates (2.7-6.4%), whereas other models demonstrated substantially higher rates (47.2-99.6%). Mitigation prompting significantly reduced confabulation rates compared to standard and deterministic decoding across all tasks and datasets (p < 0.001): GPT-5-Chat achieved confabulation rates of 0% with mitigation prompting from baseline rates of 66.4-88.5%. Deterministic decoding did not substantially impact confabulation rates. Conclusions LLMs performing medication-related tasks are susceptible to confabulation in response to invalid or fabricated inputs, raising important safety concerns for clinical use. While mitigation prompting reduced confabulation rates, this study underscores the importance of robust safeguards when deploying LLMs in medication-related decision-making.

Abstract As artificial intelligence (AI) Large Language Models (LLM) like ChatGPT become more common in healthcare, patients increasingly use them to find medical information. While ChatGPT may support health literacy, its readability and accuracy compared to established resources remain unclear. The purpose of this study is to evaluate ChatGPT-4 responses on femoroacetabular impingement (FAI) and its surgical management compared to AAOS OrthoInfo content. We hypothesize that ChatGPT may deliver medically accurate information, but its readability may fall short when compared to OrthoInfo. Nine questions based on the OrthoInfo FAI page were submitted to ChatGPT-4, with and without a readability prompt. Topics included anatomy, pathology, cause, symptoms, workup, imaging, treatment, role of hip arthroscopy, and outcomes. Readability was assessed using validated indices. Accuracy was independently rated using a 4-point scale. Statistical comparisons were made using t-tests and ANOVA (P &lt; 0.01 threshold). OrthoInfo content had a mean reading grade level of 8.0 and a Flesch Reading Ease score of 60.7. Unprompted ChatGPT responses were significantly less readable (grade level 16.2; Flesch score 21.7; P &lt; 0.001). Prompting for readability improved ChatGPT outputs (grade level 10.6; Flesch score 58.2), making them comparable to OrthoInfo (P &lt; 0.09). Accuracy was high across all sources, though OrthoInfo's response on FAI causes scored slightly lower (3.5 versus 4.0; P &lt; 0.46). ChatGPT provides accurate information on FAI. However, without prompting, its complexity may hinder patient understanding. Prompting for readability enables ChatGPT to match established resources like OrthoInfo. Tailored prompting is key to using LLM effectively in patient education and promoting health literacy.

Abstract Background Catheter related bloodstream infections (CRBSI) are associated with increased resource use, treatment costs, and mortality among patients with cancer. Coagulase-negative staphylococci (CoNS) are the most common cause of central venous catheter (CVC) infection in this population. While CVC removal remains the gold standard, CRBSI with CoNS is sometimes managed with CVC salvage, despite incomplete safety and efficacy data in patients with hematological malignancy.Table 1.Baseline Demographic and Clinical DataTable 2.Baseline Characteristics by Treatment Strategy Methods A retrospective review was conducted of adult patients at Mayo Clinic Rochester from 2013 to 2024 diagnosed with hematological malignancy and CRBSI with CoNS. Patients were included in a “CVC removed” group if the line was removed within 5 days of bacteremia onset, and all others were included in a “CVC retained” group. Electronic medical records were evaluated for relevant clinical comorbidities, risk factors, and management. The primary outcomes of interest were 90-day mortality and relapse. To control for baseline differences between groups, we employed a propensity weighted analysis based on inverse probability of treatment weighting using the propensity score.Weighted cumulative event rates for Relapse and Mortality Results A total of 140 cases were included, of whom 95 (67.9%) had undergone BMT and 111 (80.4%) were neutropenic at time of CRBSI. The predominant CoNS identified was Staphylococcus epidermidis (88.6%). 79 (56.4%) patients had the CVC retained and 61 (43.6%) underwent CVC removal. Those in whom the CVC was retained were more likely to be BMT recipients (p=0.007). However, line removal was more common with PICC lines (p&lt; 0.001) and in those hospitalized for management of line-associated bacteremia (p=0.022). Within 90 days, there were 10 patients (cumulative incidence 8.2%) who had microbiological relapse (2 [4.3%] in the CVC removed group and 8 [11.1%] in the CVC retained group). Propensity weighted analysis demonstrated significantly higher rate of microbiologic relapse with CVC retention (p=0.043) but no difference in mortality (p=0.128), as shown by weighted cumulative event rate curves in Figure 1. Conclusion Despite the finding of increased risk of relapse associated with CVC retention, there was no significant difference in mortality between CVC removal and retention among individuals with CoNS CRBSI and hematological malignancy. Disclosures All Authors: No reported disclosures

Abstract Background Native vertebral osteomyelitis (NVO) is a life-threatening spinal infection with considerable clinical burden. Although increasing literature has described its features and associated treatment failure, large-scale cohorts with long-term follow-up remain limited. Herein, we evaluated a large cohort of patients with NVO to characterize the clinical manifestations, management, and long-term outcomes, and to further explore factors associated with treatment failure.Table 1.Baseline characteristics of patients with native vertebral osteomyelitisTable 2.Treatment, failure rates, and mortality in patients with native vertebral osteomyelitis Methods We conducted a retrospective multicenter cohort study of adults (≥18 years) with NVO across Mayo Clinic sites between 1999 and 2024. Data on demographics, clinical manifestations, antimicrobial therapy, and surgical intervention were collected. Cumulative incidence of treatment failure was estimated as a function of follow-up accounting for the competing risk of death. Independent predictors of treatment failure were evaluated in a multivariable Cox regression model with time-dependent covariates.Table 3.Multivariable predictors of treatment failure in patients with native vertebral osteomyelitisFigure 1.Cumulative incidence of death and treatment failure following diagnosis of native vertebral osteomyelitis(A) Short-term (12-month) cumulative incidence of death and treatment failure (magnified view of the first year from panel B).(B) Long-term (up to 7 years) cumulative incidence of death (dotted orange line) and treatment failure (solid blue line). Results Among 1,255 patients included (median age 67 years; 66% male), lumbosacral involvement was most common (65%), with multilevel vertebral involvement ( &gt;2 levels) in 21%. Epidural abscesses were present in 43%, paravertebral abscesses in 42%, and psoas abscesses in 20%. Pathogens were identified in 77%, most commonly S. aureus (50%; MSSA 37%, MRSA 13%), followed by Streptococcus species (22%), and Gram-negative bacilli (11%) (Table 1). The median total antibiotic duration was 59 days (IQR, 42–132). Surgical intervention was performed in 13% after a median of 2 days (IQR, 1-5). Treatment failure occurred in 162 patients during a median follow-up of 2.8 years (6-month and 5-year cumulative incidence were 11% and 14%, respectively), with the median time to failure of 3.1 months. Cumulative mortality was 12% at 6 months and 33% at 5 years (Table 2, Figure 1). Diabetes (HR 1.90, 95% CI 1.16–3.12), multilevel vertebral involvement (HR 1.75, 95% CI 1.22–2.51), and MRSA infection (HR 2.11, 95% CI 1.06–4.18) were independently associated with increased risk of treatment failure (Table 3). Conclusion This 25-year study highlights the persistent burden of treatment failure in NVO and associated risk factors, underscoring the importance of early recognition of high-risk features to optimize outcomes. Disclosures All Authors: No reported disclosures

Abstract Background Native joint septic arthritis (NJSA) is a serious musculoskeletal infection requiring timely surgical intervention to prevent long-term morbidity. Although current guidelines recommend early surgical debridement, supporting data, particularly on long-term outcomes, remain scarce. Given this, we evaluated the association between surgical timing and 1-year clinical outcomes using the Desirability of Outcome Ranking (DOOR) framework, applied for the first time in NJSA.Table 1.Desirability of Outcome Ranking (DOOR) ordinal scaleTable 2.Baseline demographic, clinical characteristics, and treatment outcomes Methods We conducted a retrospective multicenter cohort study of adults (≥18 years) with NJSA who underwent surgery across Mayo Clinic campuses between 2012 and 2021. Clinical outcomes at one year were assessed using a 9-level DOOR scale, incorporating survival, treatment failure (relapse, reinfection, readmission, or significant surgical events), and joint recovery (full vs qualified) (Table 1). Time to surgery from hospital admission was analyzed both as a continuous variable and as categories: &lt; 1, 1–2, or ≥3 days.Table 3.Associations between duration and outcomesFigure 1.Association Between Time to Surgery and DOOR Scores in Native Joint Septic Arthritis.A) Distribution of Modified DOOR Scores by Surgical Timing. B) Modeled Probability of DOOR Scores by Time to Surgery. Results Among 268 patients (median age 62 years, 61% male), 30% underwent surgery &lt; 1 day from admission, 47% within 1–2 days, and 24% at ≥3 days. At 1 year, 57% achieved full recovery without unfavorable events (DOOR score 1), while treatment failure occurred in 34% (Table 2). In unadjusted analyses, longer surgical delay was significantly associated with higher DOOR scores (per IQR increase [from 0 to 2 days], OR: 1.5; 95% CI: 1.0–2.1; p = 0.026) and with increased 1-year mortality (HR: 1.7; 95% CI: 1.1–2.6; p = 0.019) and treatment failure (HR: 1.5; 95% CI: 1.1–2.0; p = 0.007). Similar results were obtained when delay was used as a categorical 3-level variable, with the highest risk observed in patients undergoing surgery after ≥3 days. After adjusting for age and Charlson index, only the association between surgical delay and treatment failure remained statistically significant (HR: 1.5; 95% CI, 1.1–2.0; p=0.016) (Table 3, Figure 1). Conclusion Delayed surgical intervention for management of NJSA was independently associated with higher risk of treatment failure at 1 year. These findings reinforce current guideline recommendations for timely operative management and further underscore the potential of the DOOR framework to capture clinically meaningful, multidimensional outcomes in NJSA. Disclosures All Authors: No reported disclosures

Abstract Background Older adults with brain tau pathology but no evident amyloid‐beta (A) pathology can be referred to as primary age‐related tauopathy (PART). However, in vivo characterization of PART using biomarkers remains unclear, varying based on the method used to define tau pathology (e.g. different tau‐PET tracers). Here, we conducted a head‐to‐head characterization of PART using two different tau‐PET tracers. Method We studied 433 individuals from the HEAD cohort (244 CU, 138 MCI, and 51 with dementia). Participants underwent clinical assessments, MRI, A‐PET, both MK‐6240 and Flortaucipir scans, and a subset with plasma biomarkers ( n = 332). A‐PET positivity was defined as Centiloid>24. Tau‐positivity(T) was determined by MK‐6240 and Flortaucipir SUVR values exceeding the mean+2SD in at least one Braak region anchored in CU A‐ individuals followed by visual confirmation in A‐/T+ cases. Concordance across biomarkers was assessed among PART (A‐/T+), A‐/T‐ and A+/T+ groups. Plasma GFAP, p ‐tau217, NFL levels, and Centiloids were compared across A/T groups using linear regression. Result We found a total of 25 (5.8 %) cases classified as A‐/T+ defined by at least one tau tracer (74±6.3 years, 52% females, 17 CU, 3 MCI, 5 with dementia). Detection of A‐/T+ cases varied by tracer: MK‐6240( n = 20), Flortaucipir( n = 16) (Figure 1A). Among A‐/T+ individuals, MK‐6240 identified 13 CU, 4 MCI, and 3 dementia cases, while Flortaucipir identified 11 CU, 3 MCI, and 2 dementia cases(Figure 1B). Concordance between MK‐6240 and Flortaucipir in A‐/T+ cases was low ( n = 11, 44%)(Figure 1C). A‐/T+ individuals showed no significant biomarker differences from A‐/T‐ but had lower biomarker levels than A+/T+: Centiloids (MK‐6240: T=15.08, p <0.001; Flortaucipir: T=14.36, p <0.001), GFAP (MK‐6240: T=5.04, p < 0.01; Flortaucipir: T=2.92, p <0.001), p ‐tau217 (MK‐6240: T=7.77, p <0.001; Flortaucipir: T=6.34, p <0.001), except for NFL (MK‐6240: T=1.35, p = 0.18; Flortaucipir: T=1.14, p = 0.25)(Figure 2). Representative A and tau biomarker cases (Figure 3A), along with discordant cases classified as T+ by only one tracer (Figure 3B), are illustrated. Conclusion Our preliminary analysis revealed a higher prevalence of A‐/T+ cases with MK‐6240 than Flortaucipir, low concordance between tracers and consistently lower biomarker levels in A‐/T+ compared to A+/T+. Larger studies and longitudinal analyses are necessary to ascertain the pathological trajectories of A‐/T+ individuals with varying tau biomarker profiles.

Abstract Background Tau phosphorylation and aggregation are hallmark features of Alzheimer's disease pathology(AD). Previous studies have shown that plasma phosphorylated tau( p ‐tau) at threonine 217 can detect AD‐related tau tangle pathology. However, the head‐to‐head association between p ‐tau217 and different tau PET tracers remains unexplored. Here, we conducted a head‐to‐head comparison of the association between plasma p ‐tau217 and four distinct tau PET tracers[MK6240, Flortaucipir(FTP), PI2620, RO948]. Method We studied 338 individuals across the AD continuum from the HEAD study[190 cognitively unimpaired elderly(CU), 109 with mild cognitive impairment(MCI), and 39 with dementia] with available FTP, MK6240, and plasma p ‐tau217 measured by ALZpath assay. A subset of 64 individuals(28 CU, 25 MCI, and 11 dementia) also had PI2620 and RO948, and p ‐tau217 measured by Lumipulse. The strength of the association was determined using t‐value maps from regression models testing the association between each tau PET tracer and plasma p ‐tau217, and the extent was measured by the percentage of significant voxels(t‐value>3.2). R‐square metric was used to determine how much of the variance in tau PET tracer estimates was explained by plasma p ‐tau217 concentrations. Associations between p ‐tau217 with tau PET were evaluated using the Spearman test. Result We found that in CU, both the extent and magnitude of the association with plasma p ‐tau217 were higher for MK6240 compared to FTP (Figure 1A‐D). Plasma p ‐tau217 explained a greater proportion of the variance in MK6240 than in FTP (Figure 1E‐F). In CI, the extent and magnitude of the association were similar for MK6240 and FTP. Additionally, plasma p ‐tau217 explained a similar proportion of the variance in MK6240 and FTP. Using the subset of individuals who had all tau tracers, MK6240 showed a slightly higher extent of association with p ‐tau217, followed by FTP, PI2620, and RO948 (Figure 2A‐B). The magnitude of association was stronger for MK6240, followed by PI2620, RO948, and FTP (Figure 2C‐D). Plasma p ‐tau217 explained a greater proportion of the variance in MK6240, followed by FTP, RO948, and PI2620 (Figure 2E‐F). We compared the associations between different p ‐tau217 assays, and both assays presented similar associations with all tau PET tracers (Figure 3). Conclusion In summary, our results indicate that the four tau PET tracers exhibit robust associations with plasma p ‐tau217 in similar topographical regions.

Abstract Background The overlap index (OI), previously introduced by Lee et al. (2022), has proven to be a reliable method for detecting tau accumulation via tau‐PET imaging with flortaucipir ([ 18 F]AV‐1451). This technique identifies voxel‐wise increases in standardized uptake value ratio (SUVr) across serial scans. However, the relationship between tau‐PET measurements obtained with [ 18 F]MK‐6240 and [ 18 F]AV‐1451 using the OI remains unclear. This study aims to investigate the relationship between [ 18 F]MK‐6240 and [ 18 F]AV‐1451 tau‐PET using the tau‐PET OI. Method The study included 27 participants from the HEAD project, all of whom underwent two serial tau‐PET scans (each [ 18 F]MK‐6240 and [ 18 F]AV‐1451) along with 3T T1‐weighted MRI, acquired within an average interval of 18 months. SUVr maps for each tau‐PET tracer were normalized to the cerebellar crus grey matter, and MR images were co‐registered to the MCALT T1 template. Tau‐PET images were spatially resampled to the template space, and OI was computed within a predefined META‐ROI using an intensity threshold of 1.3 and a cutoff of 0.5. Clusters containing fewer than 20 contiguous voxels were excluded from analysis. The META‐ROI SUVr was calculated as the average SUVr across the selected region's SUVr. Additionally, a visual comparison was performed to assess the spatial overlap of OI‐identified voxels between the two tau‐PET tracers. Result OI values within the META‐ROI exhibited a broad distribution between [ 18 F]MK‐6240 and [ 18 F]AV‐1451 tau‐PET, particularly in cognitively unimpaired (CU) and younger individuals (Figure 1A). Unlike the OI distribution, the relationship between META‐ROI SUVr values of both tracers followed an exponential trend (Figure 1B). Visual inspection revealed that OI‐detected voxels in [ 18 F]MK‐6240 tau‐PET were frequently localized to the meninges, an established site of off‐target binding, which was not observed in [ 18 F]AV‐1451 tau‐PET (Figure 2). Conclusion The association between [ 18 F]MK‐6240 and [ 18 F]AV‐1451 tau‐PET, as assessed using the OI, exhibited considerable variability in CU and younger participants. Additionally, the presence of overlapping voxels in the meninges in [ 18 F]MK‐6240 tau‐PET highlights the need to account for and potentially exclude off‐target binding effects in these populations. The influence of meningeal signal and other off‐target binding on the OI in the temporal META‐ROI needs to be further evaluated.