Abstract Background Tau PET tracers present distinct binding characteristics that might influence their trajectories and relationship with other biomarkers along the AD continuum. In a head‐to‐head study, we investigated the relationship between the emergence of PET tracers MK6240 and Flortaucipir, and plasma p ‐tau217 abnormalities as a function of Aβ PET deposition. We further assessed the concordance between tau PET and plasma p ‐tau217 positivity and its relationship with cognitive scores. Method We evaluated 352 individuals from the HEAD study (205 cognitively unimpaired and 147 cognitively impaired) with Aβ PET, MK6240 and Flortaucipir tau PET, and plasma p ‐tau217 (ALZpath) and GFAP (Quanterix) measures. Tau PET Braak regions, plasma p ‐tau217 trajectories were modeled as functions of Aβ burden (Centiloid scale) using the Lowess method. Biomarkers were z‐scored anchored on young individuals ( n = 19, < 28 years old) as anchors. Tau PET (Braak 1 region) and plasma p ‐tau217 were considered positive/abnormal when surpassing 2.5 z‐score. Result Among the tested markers, MK6240 was the earliest to show abnormality as a function of Centiloid, occurring at 22 Centiloids, followed by plasma p ‐tau217 at 38 Centiloid and Flortaucipir at Centiloid 56 (Figure 1). Tau PET and plasma p ‐tau217 positivity showed an overall high concordance (∼80% for both tracers; Figure 2). In MK6240 discordant cases, most individuals were MK6240+/ p ‐tau217‐ (13.8%), while 6.6% were MK6240‐/ p ‐tau217+ (Figure 2A). For Flortaucipir, 15% of discordant cases were Flortaucipir‐/ p ‐tau217+, while 4.8% were Flortaucipir+/ p ‐tau217‐ (Figure 2B). The discordant groups present higher Centiloid than the ptau217‐/Tau‐ group and overall decreased scores in cognitive tests, except for p ‐tau217+/MK6240‐ group (Figure 3). Conclusion MK6240 becomes abnormal at lower levels of Aβ burden compared to plasma p ‐tau217 and Flortaucipir. The relatively high prevalence of discordant tau PET positive or plasma p ‐tau217 positive and suggests that some individuals may show tau PET positivity first, while others may exhibit plasma p ‐tau217 positivity first. The distinct cognitive profiles of this groups suggest potential clinical relevance, warranting further investigation.
Head‐to‐Head evaluation of plasma
p
‐tau217 associations with MK6240, Flortaucipir, PI2620, and RO948 tau PET tracers
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 (Figures 1A‐D). Plasma p ‐tau217 explained a greater proportion of the variance in MK6240 than in FTP (Figures 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 (Figures 2A‐B). The magnitude of association was stronger for MK6240, followed by PI2620, RO948, and FTP (Figures 2C‐D). Plasma p ‐tau217 explained a greater proportion of the variance in MK6240, followed by FTP, RO948, and PI2620 (Figures 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.
Harmonization of Flortaucipir, MK6240, PI2620 and RO948 with the Uniτ scale
Abstract Background Precise cross‐tracer harmonization of tau PET imaging is essential for comparing tau burden across different tracers and studies. Here, we evaluated the performance of the universal tau PET scale (Uniτ) in harmonizing Flortaucipir, MK6240, PI2620, and RO948 tau PET images into a universal scale. Method We assessed 485 individuals across aging and AD spectrum scanned head‐to‐head with Flortaucipir and MK6240, and a subset of 90 individuals with additional PI2620 and RO948. We generated tau PET SUVRs using the inferior cerebellar gray matter as reference with a common 8mm FWHM. We estimated Uniτ parameters on a training set ( n = 200) by fitting a smoothed hyperbolic tangent equation to the Meta‐Temporal ROI anchored in the mean SUVR of young participants and the 90th percentile from cognitively impaired individuals. We compared tau positivity on the Uniτ scale with classifications from SUVRs (mean + 3 SD from Youngs). We applied the same equation to all brain voxels to generate Uniτ 3D images. Then, we extracted mean Uniτ values for key ROIs and correlated them with ROI‐based Uniτ values to evaluate voxel‐wise estimates. Result Using tracer‐specific parameters, Uniτ harmonized Flortaucipir, MK6240, PI2620, and RO948 to the same scale, aligning values near the identity line and confirming its applicability across tau PET tracers (Figure 1). For Flortaucipir, Uniτ tau positivity matched original SUVR classifications (ground truth), with only one mismatch for MK6240 (Figure 1). Applying the Uniτ transformation to all brain voxels effectively harmonized 3D images to a common scale, reducing visual variabilities (Figure 2). Furthermore, ROI‐based estimates and those extracted from the 3D Uniτ images were identical (Figure 3). Conclusion Our results indicate that tau PET tracers can be harmonized to a common scale using a large head‐to‐head dataset. The Uniτ scale can harmonize entire 3D tau PET images, suggesting that there is no need to use pre‐established ROIs, which would constrain the analysis to a few brain regions. Uniτ is freely accessible across platforms ( www.unitau.app ) for ROI and 3D tau PET harmonization.
Effects of AD modifiable risk factors to tau‐PET tracer uptake and its association with cognitiion in early Braak stages
Abstract Background Many risk factors can contribute to the occurrence of Alzheimer`s Disease (AD). However, little is known about the impact of dementia risk factors to the uptake of tau‐PET tracers. Therefore, in this work we aim to investigate the influence of dementia risk factors and comorbidities on 18 F‐Flortaucipir (FTP) and 18 F‐MK6240 (MK) tau‐PET tracers’ uptake. Additionally, we will assess how these factors impact the association of tau‐PET and cognition. Method We accessed 436 individuals across the aging and AD spectrum (251 amyloid negative and 185 amyloid positive) from the HEAD study, with available Aβ‐PET, FTP, MK, and clinical assessments. Linear regression models corrected for age, sex, clinical diagnosis, and study site tested the association of factors with tau‐PET tracers in the medial temporal lobe (MTL). A tau‐PET × risk factor term was added to test the influence of risk factors to the association of tau with cognition. Result In amyloid‐β negative individuals, high BMI were positively associated with the uptake of both FTP and MK, whereas hearing loss were positively associated only with MK in the MTL (Figure 1A). In amyloid‐β positive individuals, high body mass index (BMI), hearing loss and sleep disorders were negatively associated with the uptake of both tau‐PET tracers in the MTL. On the other hand, hypertension showed negative association only with MK uptake (Figure 1B). Using Mini‐Mental State Examination (MMSE) scores as outcome, we observed that amyloid‐β negative individuals with high BMI showed worse cognitive performance as a function of both MK and FTP in the MTL, whereas individuals with vision impairment and hearing loss showed worse cognitive performance as a function of MK only (Figure 2A). Amyloid‐β positive individuals with hypercholesterolemia and hypertension presented worse cognitive performance as a function of both MK and FTP in the MTL (Figure 2B). Conclusion In this preliminary analysis, sleep disorders, hypertension, and high BMI were independently associated with tau‐PET tracer uptake, with the effects varying according to amyloid‐β pathology. These prevalent factors in the elderly also changed the association between tau‐PET and cognition, underscoring the need for further studies to better understand their role in modulating this relationship.
Defining tau PET positivity grey zones for MK6240 and Flortaucipir quantification
Abstract Background Tau PET measures are inherently continuous and applying dichotomized thresholds introduces conceptual and analytical idiosyncrasies. Understanding the limitations in the transition from tau‐negative to tau‐positive classifications is crucial for the effective use of these thresholds. This study aims to determine the confidence levels of tau PET thresholds of abnormality for different tau PET tracers by characterizing their "gray zone" using the universal tau PET scale (Uniτ, www.unitau.app ). Method We evaluated 485 individuals across the aging and AD spectrum from the HEAD study, with head‐to‐head scans for MK6240 and Flortaucipir. Uniτ estimates were derived from the Meta‐Temporal ROI. Tau positivity (T+) was defined as Uniτ values exceeding the mean plus 3 SD of individuals younger than 28 years ( n = 24). Two physicians independently performed a visual assessment of tau positivity for each tracer, with agreement indicating clear tau pathology (TVR+). Logistic regression was used to estimate the probability of TVR+ across Uniτ values for each tracer ( n = 189, CU elderly Aβ‐ and CI Aβ+). Individuals were classified as negative, positive, or within a "gray zone" between the most liberal T+ threshold and varying TVR+ probability thresholds (50%, 75%, 90%, 95%, and 99%). Result The Uniτ gray zone, defined as a function of TVR+ probabilities, demonstrated consistency between the two tracers, with differences observed only in the decimal range. The most liberal Uniτ threshold for T+ was 11.0 for both MK6240 and Flortaucipir, closely matching the 50% TVR+ probability threshold (11.1 for MK6240 and 11.7 for Flortaucipir). Higher TVR+ probabilities corresponded to increased Uniτ values with MK6240 and Flortaucipir showing near‐identical values between the tracers: 14.2 and 14.8 for 75%, 17.3 and 17.9 for 90%, 19.4 and 19.9 for 95%, and 24.1 and 24.6 for 99%, respectively (Figure 1). In total, 26 participants fell within the gray zone for MK6240, compared to 47 participants for Flortaucipir. Conclusion These findings highlight the potential of Uniτ to provide a standardized approach for assessing tau positivity across tracers. By combining quantitative Uniτ measures with visual assessments, we enhance the understanding of tau positivity certainty, particularly in the transition between negative and positive classifications.
Assessing Braak stage agreement between four Tau PET tracers
Abstract Background The Alzheimer's disease (AD) Braak staging is a key framework for classifying tau pathology progression in AD based on histopathological post‐mortem brain examinations. However, adapting it to PET imaging can be challenging due to differences in tracer uptake patterns and binding properties, which affect sensitivity, specificity, and regional staging. This study compares Braak staging across four tau PET tracers: Flortaucipir, MK6240, PI2620, and RO948. Methods We assessed 90 participants across the AD spectrum (46 CU, 31 MCI, 13 dementia; mean age 66.1 ± 7.8) using Aβ PET and four tau PET tracers: (Flortaucipir, MK6240, PI2620, and RO948). Braak positivity was defined based on Aβ− CU individuals (mean +2.5 SD, SUVR). To evaluate systematic bias and agreement between tracers, we computed pairwise differences at corresponding Braak stage estimates and applied the Bland‐Altman method to assess mean bias and limits of agreement. Additionally, Tau PET Braak region trajectories were modeled as functions of Aβ burden (Centiloid scale) using the Lowess method. Results Braak stage trajectories as a function of Aβ differ depending on the tracer and the sequential order of abnormality is highly variable. For instance, while for MK6240, RO948 and PI2620, Braak I is the first region to became abnormal, for Flortaucipir the earliest region to became abnormal is Braak IV (Figure 1). This variable pattern of abnormality impacts in the concordance of Braak staging between tracers with the highest Braak staging agreement resulting in concordance levels of approximately 70%. The highest levels of agreement between tracers usually happen at Braak 0 or Braak IV‐V, with intermediate stages showing very low concordance (Figure 2). The Bland‐Altman analysis identified wide limits of agreement, suggesting high variability and high tracer‐specific differences (Figure 3). On the other hand, it also identified that mean differences between tracers were small, indicating minimal systematic bias. Conclusion These preliminary findings reveal discrepancies in Braak staging when comparing Flortaucipir, MK6240, PI2620 and RO948. These findings suggest that while the tracers provide comparable stages on average, they may not be fully interchangeable in individual cases.
Longitudinal progression and harmonization of tau‐PET tracers
Abstract Background Tau‐PET tracers have been used to monitor the progression of Alzheimer’s disease (AD). However, different tracers present distinct patterns of binding throughout the brain, challenging the harmonization of their findings. Leveraging the HEAD Study, the largest head‐to‐head study of tau‐PET tracers, we recently developed the Uniτ scale, which cross‐sectionally harmonizes Flortaucipir and MK6240 onto a universal tau‐PET measurement. Here, we provide a preliminary evaluation of the Uniτ scale’s longitudinal performance in HEAD and two independent cohorts. Methods We assessed 422 individuals across the AD spectrum with longitudinal tau‐PET from three cohorts: HEAD [13 cognitively unimpaired (CU), 9 cognitively impaired (CI) individuals, scanned head‐to‐head with Flortaucipir and MK6240], ADNI [74 CU, 208 CI, tracer: Flortaucipir], and TRIAD [72 CU, 46 CI, tracer: MK6240]. Standardized uptake ratios (SUVRs) were harmonized to Uniτ using the Uniτ Ecosystem ( unitau.app ). Braak I‐II and the Meta‐Temporal regions were used as regions of interest. Annual tau‐PET uptake change was calculated, and the effect size was defined as the mean annual change divided by its standard deviation. Results In HEAD, annual change in tau‐PET uptake in Braak I‐II across CU and CI was only detectable in MK6240 (Figure 1A). However, both tracers detected significant annual changes in the Meta‐Temporal ROI for CI (Figure 1B). In both ADNI (Flortaucipir) and TRIAD (MK6240), changes in tau‐PET uptake in Braak I‐II regions were not significant (Figures 2‐3A). However, across all cohorts, the Meta‐Temporal ROI consistently showed detectable annual tau‐PET increases—most pronounced among CI—leading to larger effect sizes than in Braak I‐II (Figures 1‐3B). The Uniτ harmonization did not fundamentally alter the pattern of the findings. However, in ADNI (Flortaucipir), Uniτ yielded a slightly higher effect size than SUVR in CU for both Braak I‐II and Meta‐Temporal regions. Conclusion In this large longitudinal sample, our findings confirm that Flortaucipir and MK6240 can detect tau PET changes over time likely associated with the progression of tau tangle pathology. While MK6240 appears to show greater progression in CU, both tracers progress similarly in CI. Our data also suggested that Uniτ harmonized tau PET measurements maintain the longitudinal characteristics of each tau PET tracer for use in clinical trials.
Definitions for Vasoplegia Associated With On-Pump Cardiac Surgery: A Systematic Review With Meta-Analysis
OBJECTIVES: Vasoplegia is a common complication of cardiac surgery that uses cardiopulmonary bypass and contributes to morbidity and mortality, yet a consensus definition does not exist. The objective of this study was to evaluate the diagnostic criteria and definitions used to characterize vasoplegia and how different criteria influence incidence estimates. DATA SOURCES: Ovid Embase, Ovid MEDLINE, Scopus, Web of Science Core Collection, ClinicalTrials.gov, Ovid Cochrane Central Register of Controlled Trials, and the World Health Organization’s International Clinical Trials Registry Platform clinical trials registry. STUDY SELECTION: Randomized clinical trials and observational studies reporting on vasoplegia in adults undergoing any type of cardiac surgery that used cardiopulmonary bypass. DATA EXTRACTION: Proportional meta-analysis using a random-effects model and the inverse variance method was used to calculate the pooled incidence of vasoplegia and its clinical outcomes. DATA SYNTHESIS: A total of 68 studies encompassing 56,580 patients were identified, from which 63 unique vasoplegia definitions were used. Blood pressure ( n = 57 studies, 84%) and cardiac output ( n = 50 studies, 74%) were among the most common criteria used in vasoplegia definitions; however, there was a vast variety of threshold values applied within these criteria and all other criteria comprising the definitions. The pooled incidence of vasoplegia was 21% (95% CI, 17–25%), acute kidney injury was 32% (95% CI, 21–45%), and mortality was 12% (95% CI, 9–16%). Subgroup analysis revealed that transplantation and left ventricular assist device implantation surgeries, and those with baseline left ventricular ejection fraction less than 40% had a significantly greater incidence of vasoplegia. CONCLUSIONS: The published literature varies greatly in the criteria used to define vasoplegia associated with on-pump cardiac surgery. Generation and adoption of a unified definition for vasoplegia must be an international priority.
Rx-LLM: a benchmarking suite to evaluate safe large language model performance for medication-related tasks
Abstract Background: For large language models (LLMs) to reach their potential as information technology tools that make medication use safer, clinically relevant benchmarks capable of automated grading and designed specifically to measure the performance of LLMs for medication tasks are required. The purpose of this study was to design a suite of benchmarking tests reflective of Comprehensive Medication Management (CMM; the standard of care for medication optimization) and quantify the baseline performance of the latest LLMs. Methods: We established six benchmarks representing critical stages of the CMM process: drug formulation matching, drug order (sig) generation, drug route matching, drug-drug interaction identification, renal dose identification, and drug-indication matching. For each benchmark, we curated a clinician-annotated dataset comprising 250 standardized input-output pairs including both inpatient and outpatient medications. We evaluated the clinical knowledge retrieval capabilities of three LLMs: GPT-4o-mini, MedGemma-27B, and LLaMA3-70B. We employed a zero-shot prompting strategy, excluding in-context examples, to assess the models’ internal clinical knowledge rather than their few-shot learning potential. To check reliability, each model was run three times using a temperature of 0.7 (a mid-range value of an LLM setting controlling text generation randomness). Performance was assessed using task-specific evaluation metrics including precision (positive predictive value), recall (sensitivity), F1-score, accuracy, and correctness consistency across trials. Results: Across six benchmarks, LLaMA3-70B demonstrated the highest performance in four tasks: drug-formulation matching (F1, 54.0% [95 CI: 50.1-58]), drug-order generation (accuracy, 88.0%), drug-route identification (F1, 74.3% [95 CI: 71-78]), and drug-indication identification (accuracy, 97.6% [95 CI: 95.6-99.2]). In the drug–drug interaction task, GPT-4o-mini achieved the highest overall accuracy (70.4% [95 CI: 64.8-75.7]). For renal dose–adjustment identification, GPT-4o-mini demonstrated the highest F1 score (83.3% [95 CI: 77.6-88]). Correctness-consistency scores ranged from 8.0% to 97.6% across benchmarks, with no model exhibiting uniformly superior consistency. Conclusions: Model performance varied substantially across medication-related tasks. LLaMA3-70B demonstrated promising baseline performance in tasks involving formulation, ordering, route, and indication. GPT-4o-mini showed potential advantages in drug–drug interaction detection and renal dose adjustment. These findings underscore the need for task-specific evaluation when deploying models for medication-focused clinical decision support.
Drug-drug interaction identification using large language models
Abstract Background The purpose of this study was to develop a series of benchmarks specifically for large language model (LLM) performance in the identification of drug-drug interactions (DDIs). Methods We evaluated six LLMs using a clinician-designed dataset of 750 DDI scenarios. Tasks included: (1) a pointwise two-drug classification task (2) a pairwise three-drug discrimination task and (3) a listwise 4–6 drug selection task. Zero-shot prompting was applied and performance was assessed using precision, recall, F1 score, accuracy, and self-consistency. Results Model performance varied by task structure and interaction severity. GPT-5-Chat demonstrated the highest recall and F1 score in the pointwise task (80.7% and 80.2%), whereas DrugGPT achieved superior accuracy in the pairwise (93.7%) and listwise task (92.6%). All models had improved performance with more severe compared to less severe DDIs (accuracy 72.0-99.2% vs. 58.3-86.7%). All models tended to have higher self-consistency in the pairwise and listwise experiments compared to the pointwise experiment (68.8-93.6% vs. 31.0-74.2%). No model exhibited uniformly high reliability across all tasks. Conclusions Current LLMs show inconsistent capabilities in identifying DDIs. Performance degrades as reasoning space expands with limited stability. These findings emphasize the need for multi-format evaluation frameworks and reliability-aware assessment when considering LLMs for medication-safety applications.