Evaluating Diagnostic Accuracy and Determining Optimal Diagnostic Thresholds of Three Different Approaches to [68Ga]-DOTATATE PET/MRI Analysis in Patients with Meningioma

PURPOSEMultiple approaches with [Ga68]-DOTATATE, a somatostatin analog PET radiotracer, have demonstrated clinical utility in evaluation of meningioma but have not been compared directly. Our purpose was to compare diagnostic performance of three approaches to quantitative brain [68Ga]-DOTATATE PET/MRI analysis in patients with suspected meningioma recurrence and to establish the optimal diagnostic threshold for each method.METHODSPatients with suspected meningioma were imaged prospectively with [68Ga]-DOTATATE brain PET/MRI. Lesions were classified as meningiomas and post-treatment change (PTC), based on pathology findings and follow up MRI appearance. Lesions were reclassified using the following methods: absolute SUV threshold (SUV), SUV ratio (SUVR) to superior sagittal sinus (SSS) (SUVRsss), and SUVR to the pituitary gland (SUVRpit). Diagnostic performance of the three methods was compared using contingency tables and McNemar’s test. Previously published pre-determined thresholds were assessed where applicable. The optimal thresholds for each method were identified using Youden’s J statistics.RESULTS166 meningiomas and 41 PTC lesions were identified across 62 patients. SUV, SUVRsss, and SUVRpit of meningioma were significantly higher than those of PTC (P<0.0001). The optimal thresholds for SUV, SUVRsss, and SUVRpit were 4.65, 3.23, and 0.260, respectively. At the optimal thresholds, SUV had the highest specificity (97.6%) and SUVRsss had the highest sensitivity (86.1%). An ROC analysis of SUV, SUVRsss, and SUVRpit revealed AUC of 0.932, 0.910, and 0.915, respectively (P<0.0001).CONCLUSIONWe found that the SUVRsss method may have the most robust combination of sensitivity and specificity in the diagnosis of meningioma in the post-treatment setting, with the optimal threshold of 3.23. Future studies validating our findings in different patient populations are needed to continue optimizing the diagnostic performance of [68Ga]-DOTATATE PET/MRI in meningioma patients. Clinical Trial Registration: ClinicalTrials.gov Identifier: NCT04081701. Registered 9 September 2019. https://clinicaltrials.gov/ct2/show/NCT04081701

Application of Association Rules Analysis in Mining Adverse Drug Reaction Signals

Adverse drug reactions (ADRs) are increasingly becoming a serious public health problem. Spontaneous reporting systems (SRSs) are an important way for many countries to monitor ADRs produced in the clinical use of drugs, and they are the main data source for ADR signal detection. The traditional signal detection methods are based on disproportionality analysis (DPA) and lack the application of data mining technology. In this paper, we selected the spontaneous reports from 2011 to 2018 in Jiangsu Province of China as the research data and used association rules analysis (ARA) to mine signals. We defined some important metrics of the ARA according to the two-dimensional contingency table of ADRs, such as Confidence and Lift, and constructed performance evaluation indicators such as Precision, Recall, and F1 as objective standards. We used experimental methods based on data to objectively determine the optimal thresholds of the corresponding metrics, which, in the best case, are Confidence = 0.007 and Lift = 1. We obtained the average performance of the method through 10-fold cross-validation. The experimental results showed that F1 increased from 31.43% in the MHRA method to 40.38% in the ARA method; this was a significant improvement. To reduce drug risk and provide decision making for drug safety, more data mining methods need to be introduced and applied to ADR signal detection.

RTID-08. MACHINE LEARNING TO UNCOVER SIGNATURES OF VULNERABILITY IN GLIOBLASTOMA UMBRELLA SIGNATURE TRIAL (GUST)

Glioblastoma is characterized by intra- and inter-tumoral heterogeneity. A glioblastoma umbrella signature trial (GUST) posits multiple investigational treatment arms based on corresponding biomarker signatures. A contingency of an efficient umbrella trial is a suite of orthogonal signatures to classify patients into the likely-most-beneficial arm. Assigning optimal thresholds of vulnerability signatures to classify patients as “most-likely responders” for each specific treatment arm is a crucial task. We utilized semi-supervised machine learning, Entropy-Regularized Logistic Regression, to predict vulnerability classification. By applying semi-supervised algorithms to the TCGA GBM cohort, we were able to transform the samples with the highest certainty of predicted response into a self-labeled dataset and thus augment the training data. In this case, we developed a predictive model with a larger sample size and potential better performance. Our GUST design currently includes four treatment arms for GBM patients: Arsenic Trioxide, Methoxyamine, Selinexor and Pevonedistat. Each treatment arm manifests its own signature developed by the customized machine learning pipelines based on selected gene mutation status and whole transcriptome data. In order to increase the robustness and scalability, we also developed a multi-class/label classification ensemble model that’s capable of predicting a probability of “fitness” of each novel therapeutic agent for each patient. Such a multi-class model would also enable us to rank each arm and provide sequential treatment planning. By expansion to four independent treatment arms within a single umbrella trial, a “mock” stratification of TCGA GBM patients labeled 56% of all cases into at least one “high likelihood of response” arm. Predicted vulnerability using genomic data from preclinical PDX models correctly placed 4 out of 6 models into the “responder” group. Our utilization of multiple vulnerability signatures in a GUST trial demonstrates how a precision medicine model can support an efficient clinical trial for heterogeneous diseases such as GBM. Surgical Therapies

SARS-COV-2 Neutralizing Activity in Serum Collected from Recovered and Vaccinated Health Care Workers

Vaccination against the ongoing COVID-19 is the key point in fight against the pandemic. The Spike (S) glycoprotein of SARS-CoV-2 is the major target of the neutralizing humoral response. We evaluated analytical and clinical performances of a surrogate virus neutralization test (sVNT) (iFlash-2019-nCoV Nab assay, Ylho, China) compared to the conventional neutralization tests (cVNT) and anti-S eCLIA assays (Roche Diagnostics, Switzerland) in recovered and/or vaccinated health care workers. Our results indicate that sVNT displayed high specificity and no cross reactivity. Both Roche and iFlash immunoassays were good in identifying cVNT serum dilution > 1:16. Optimal thresholds in identifying cVNT titers ≥ 1:16, were 74.5 U/ml and 49.4 IU/ml for anti-S eCLIA and sVNT, respectively. Our data show that Nab neutralizing antibodies titers depend on individuals and may abate over time. Specific assays such as sVNT may be a reliable complementary tool to routine anti-S serology assays.

Interpreting change on the Symbol Digit Modalities Test in people with relapsing multiple sclerosis using the reliable change methodology

Background: The Symbol Digit Modalities Test (SDMT) is increasingly utilized in clinical trials. A SDMT score change of 4 points is considered clinically important, based on association with employment anchors. Optimal thresholds for statistically reliable SDMT changes, accounting for test reliability and measurement error, are yet to be applied to individual cases. Objective: The aim of this study was to derive a statistically reliable marker of individual change on the SDMT. Methods: This prospective, case–control study enrolled 166 patients with multiple sclerosis (MS). SDMT scores at baseline, relapse, and 3-month follow-up were compared between relapsing and stable patient groups. Using data from the stable group and three previously published studies, candidate thresholds for reliable decline were calculated and validated against other tests and a clinically meaningful anchor—cognitive relapse. Results: Candidate thresholds for reliable decline at the 80% confidence level varied between 6 and 11 points. An SDMT change of 8 or more raw score points was deemed to offer the best balance of discriminatory power and external validity for estimating cognitive decline. Conclusion: This study illustrates the feasibility and usefulness of reliable change methodology for identifying statistically meaningful cognitive decline that could be implemented to identify change in individual patients, for both clinical management and clinical trial outcomes.

On optimal thresholds for pairs trading in a one-dimensional diffusion model

We study the static maximization of long-term averaged profit, when optimal preset thresholds are determined to describe a pairs trading strategy in a general one-dimensional ergodic diffusion model of a stochastic spread process. An explicit formula for the expected value of a certain first passage time is given, which is used to derive a simple equation for determining the optimal thresholds. Asymptotic arbitrage in the long run of the threshold strategy is observed. doi:10.1017/S1446181121000298

Grading mitral regurgitation using 4D flow CMR: comparison to transthoracic echocardiography

Background Mitral Regurgitation (MR) is currently primarily assessed using multiple transthoracic echocardiography (TTE) parameters. Two-dimensional Cardiac Magnetic Resonance (CMR) can be used in difficult cases but has limited agreement with TTE for quantifying MR. We hypothesized that 4D Flow CMR may help to quantify MR. Purpose To determine the 4D Flow CMR thresholds that achieve the best agreement with TTE for grading MR. Methods We conducted a single-center prospective study of patients evaluated for chronic primary MR in 2016–2020. MR was evaluated blindly by TTE and 4D Flow CMR respectively by two cardiologists and two radiologists with decades of experience. MR was graded with both methods as mild, moderate or severe. 4D Flow CMR measurements included MR regurgitant volume per beat (RV) and mitral anterograde flow per beat (MF). RF was obtained as the ratio RV/MF. Additionally, MF was compared to left ventricular stroke volume (LVSV) by cine-CMR. Results We included 33 patients in the initial cohort and 33 in the validation cohort. Inter-observer agreement was good for TTE and excellent for 4D Flow CMR. Agreement between MF and LVSV was excellent. Using recommended TTE thresholds (30 mL, 60 mL, 30%, 50%), agreement was moderate for RV and RF. The best agreement between 4D Flow CMR and TTE was obtained with CMR thresholds of 20 mL and 40 mL for RV (κ=0.93; 95% CI, 0.8–1) and 20% and 37% for RF (κ=0.90; 95% CI, 0.7–0.9). In the validation cohort, agreement between TTE and 4D Flow CMR was good with the optimal thresholds (κ= 0.78; 95% CI, 0.61–0.94). Conclusion We propose CMR thresholds that provide a good agreement between TTE and CMR for grading MR. Further studies are needed to fully validate 4D-Flow CMR accuracy for primary MR quantification. FUNDunding Acknowledgement Type of funding sources: None. Quantification of MR using 4D Flow CMR

CLRM-01. MACHINE LEARNING TO UNCOVER SIGNATURES OF VULNERABILITY IN GLIOBLASTOMA UMBRELLA SIGNATURE TRIAL (GUST)

Glioblastoma is characterized by intra- and inter-tumoral heterogeneity. A glioblastoma umbrella signature trial (GUST) posits multiple investigational treatment arms based on corresponding biomarker signatures. A contingency of an efficient umbrella trial is a suite of orthogonal signatures to classify patients into the likely-most-beneficial arm. Assigning optimal thresholds of vulnerability signatures to classify patients as “most-likely responders” for each specific treatment arm is a crucial task. We utilized semi-supervised machine learning, Entropy-Regularized Logistic Regression, to predict vulnerability classification. By applying semi-supervised algorithms to the TCGA GBM cohort, we were able to transform the samples with the highest certainty of predicted response into a self-labeled dataset and thus augment the training data. In this case, we developed a predictive model with a larger sample size and potential better performance. Our GUST design currently includes four treatment arms for GBM patients: Arsenic Trioxide, Methoxyamine, Selinexor and Pevonedistat. Each treatment arm manifests its own signature developed by the customized machine learning pipelines based on selected gene mutation status and whole transcriptome data. In order to increase the robustness and scalability, we also developed a multi-class/label classification ensemble model that’s capable of predicting a probability of “fitness” of each novel therapeutic agent for each patient. Such a multi-class model would also enable us to rank each arm and provide sequential treatment planning. By expansion to four independent treatment arms within a single umbrella trial, a “mock” stratification of TCGA GBM patients labeled 56% of all cases into at least one “high likelihood of response” arm. Predicted vulnerability using genomic data from preclinical PDX models correctly placed 4 out of 6 models into the “responder” group. Our utilization of multiple vulnerability signatures in a GUST trial demonstrates how a precision medicine model can support an efficient clinical trial for heterogeneous diseases such as GBM.

ON OPTIMAL THRESHOLDS FOR PAIRS TRADING IN A ONE-DIMENSIONAL DIFFUSION MODEL

We study the static maximization of long-term averaged profit, when optimal preset thresholds are determined to describe a pairs trading strategy in a general one-dimensional ergodic diffusion model of a stochastic spread process. An explicit formula for the expected value of a certain first passage time is given, which is used to derive a simple equation for determining the optimal thresholds. Asymptotic arbitrage in the long run of the threshold strategy is observed.