Blinded Validation Research Articles

31P The IOpener study: Tyrosine kinase activity profiling to predict response to immune checkpoint inhibitors in patients with advanced stage non-small cell lung cancer

Treatment with immune checkpoint inhibitors (ICIs) is successful, but only in a subset of patients with advanced stage non-small cell lung cancer (NSCLC). Blood-based kinase profiling of peripheral blood mononuclear cells (PBMCs) has previously shown its predictive value for response to ICI treatment in a discovery cohort (Hurkmans et al., JITC 2020). The aim of the current prospective study was to perform a blind validation of the predictive value of kinase activity profiles in PBMCs for ICI response in patients with NSCLC in a standardized setting.

Codeveloping and Evaluating a Campaign to Reduce Dementia Misconceptions on Twitter: Machine Learning Study.

Dementia misconceptions on Twitter can have detrimental or harmful effects. Machine learning (ML) models codeveloped with carers provide a method to identify these and help in evaluating awareness campaigns. This study aimed to develop an ML model to distinguish between misconceptions and neutral tweets and to develop, deploy, and evaluate an awareness campaign to tackle dementia misconceptions. Taking 1414 tweets rated by carers from our previous work, we built 4 ML models. Using a 5-fold cross-validation, we evaluated them and performed a further blind validation with carers for the best 2 ML models; from this blind validation, we selected the best model overall. We codeveloped an awareness campaign and collected pre-post campaign tweets (N=4880), classifying them with our model as misconceptions or not. We analyzed dementia tweets from the United Kingdom across the campaign period (N=7124) to investigate how current events influenced misconception prevalence during this time. A random forest model best identified misconceptions with an accuracy of 82% from blind validation and found that 37% of the UK tweets (N=7124) about dementia across the campaign period were misconceptions. From this, we could track how the prevalence of misconceptions changed in response to top news stories in the United Kingdom. Misconceptions significantly rose around political topics and were highest (22/28, 79% of the dementia tweets) when there was controversy over the UK government allowing to continue hunting during the COVID-19 pandemic. After our campaign, there was no significant change in the prevalence of misconceptions. Through codevelopment with carers, we developed an accurate ML model to predict misconceptions in dementia tweets. Our awareness campaign was ineffective, but similar campaigns could be enhanced through ML to respond to current events that affect misconceptions in real time.

Adaptive Proxy-based Robust Production Optimization with Multilayer Perceptron

Machine learning (ML) has been a technique employed to build data-driven models that can map the relationship between the input and output data provided. ML-based data-driven models offer an alternative path to solving optimization problems, which are conventionally resolved by applying simulation models. Higher computational cost is induced if the simulation model is computationally intensive. Such a situation aptly applies to petroleum engineering, especially when different geological realizations of numerical reservoir simulation (NRS) models are considered for production optimization. Therefore, data-driven models are suggested as a substitute for NRS. In this work, we demonstrated how multilayer perceptron could be implemented to build data-driven models based on 10 realizations of the Egg Model. These models were then coupled with two nature-inspired algorithms, viz. particle swarm optimization and grey wolf optimizer to solve waterflooding optimization. These data-driven models were adaptively re-trained by applying a training database that was updated via the addition of extra samples retrieved from optimization with the proxy models. The details of the methodology will be divulged in the paper. According to the results obtained, we could deduce that the methodology generated reliable data-driven models to solve the optimization problem, as justified by the excellent performance of the ML-based proxy model (with a coefficient of determination, R2 exceeding 0.98 in training, testing, and blind validation) and accurate optimization result (less than 1% error between the Expected Net Present Values optimized using NRS and proxy models). This study aids in an enhanced understanding of implementing adaptive training in tandem with optimization in ML-based proxy modeling.

A colorimetric, sensitive, rapid, and simple diagnostic kit for the HLB putative causal agent detection

Huanglongbing (HLB) is one of the most devastating diseases in citrus worldwide. The Gram-negative bacterial plant pathogen “Candidatus Liberibacter spp.” is phloem-limited and vectored by citrus psyllids. The species “Candidatus Liberibacter asiaticus” (C.Las) has been detected in Argentina, and its vector has been found in at least nine provinces. Early detection of C.Las is critical for a successful management of HLB disease. Currently, HLB molecular diagnosis is carried out by PCR, nested PCR, real-time PCR, or another combination of these techniques, which require purification of genomic DNA, sophisticated equipment, and highly trained personnel. We have developed a prototype of a sensitive colorimetric kit to detect C.Las based on the specific DNA isothermal amplification of this microorganism. The reaction buffer contains hydroxynaphthol blue (HNB), an indicator dye that turns from violet to blue/light blue when the DNA amplification reaction is positive. Similar sensitivity to visualize a positive reaction was observed between HNB loop-mediated isothermal amplification and agarose gel electrophoresis analysis. The detection of C.Las-infected plants was up to 8 ng of total infected plant genomic DNA, similar to quantitative PCR. A blind validation test of the prototype kit was performed with purified DNA extracted from healthy or C.Las-infected midrib plants. Our kit showed 100% concordance with the results of a gold-standard quantitative PCR technique applied by the Laboratorio de Biología Molecular de EEA Montecarlo. The analysis of samples, without DNA purification to detect C.Las, showed a similar sensitivity to the analysis of the same samples in which C.Las DNA was previously purified.

Deep quantification of a refined adulterant blended into pure avocado oil

In this work, a new method has been developed to detect adulterations in avocado oil by combining optical images and their treatment with deep learning algorithms. For this purpose, samples of avocado oil adulterated with refined olive oil at concentrations from 1 % to 15 % (v/v) were prepared. Two groups of images of the different samples were obtained, one in conditions considered as bright and the other as dark, obtaining a total of 1,800 photographs. To obtain these images under both conditions, the exposure or shutter speed of the camera was modified (1/30 s for light conditions and 1/500 s for dark conditions). A residual neural network (ResNet34) was used to process and classify the images obtained. A different model was developed for each condition, and during blind validation of the models, ∼95 % of the images were correctly classified.

920P Blind validation of MSIntuit, an AI-based pre-screening tool for MSI detection from colorectal cancer H&E slides

MisMatch Repair deficiency (dMMR) / Microsatellite instability (MSI) is a crucial biomarker in colorectal cancer (CRC) predictive of response to immunotherapy and Lynch syndrome. Universal screening of newly diagnosed CRC for dMMR/MSI status is now recommended. With the growing numbers of biomarkers screened in clinical practice, dMMR/MSI testing, currently diagnosed by immunohistochemistry (IHC) and/or polymerase chain reaction (PCR), contributes to increasing pathologists workload and delaying therapeutic decisions.

147P Blind validation of an AI-based tool for predicting distant relapse from breast cancer HES stained slides

147P Blind validation of an AI-based tool for predicting distant relapse from breast cancer HES stained slides

Spatiotemporal analysis of interictal EEG for automated seizure detection and classification

ObjectiveSeizure type classification is important as therapy differs for different epilepsy subtypes.Currently, skilled neurologists classifyseizuresbased on visual analysis. However, manual EEG inspection is time-consuming, laborious,subjective, and prone to misclassification due to artifacts andEEGvariability. This work aims to address these limitations. MethodsIn this work, a quick, robust, and accurate spatiotemporal analytical algorithm is developed to classify epileptic seizures.TheEEG dataset is sampled at 125 Hz using aNicolet EEG system.Robust preprocessing, feature extraction, andoptimal classifiers captured IEDs (Interictal Epileptiform Discharges), includingspikes, sharps, slowwaves, and Spike-Wave Discharges (SWD). ResultsThe developed classifier resultsarevalidated againstclinicalimpressionsprovided by experienced epileptologists. Thealgorithm automatically classifies the EEG datainto four types: normal, focal, generalized, and absence, with 93.18 % accuracy (n = 88). ConclusionTheresultssuggest a novel way toscreen epileptic subjects without false positives (accuracy: 94.32 %, n = 88) and tentatively identify the seizure type. Blind validation further confirms the generalizability of the classifier (accuracy: 90.90 %, n = 11). SignificanceThe developed algorithm reduces the workload of neurologists for epilepsy screening and identifies seizure onset zone, temporal spread, and overall scalp distribution of epileptic activities.

Superior reproducibility and repeatability in automated quantitative pupillometry compared to standard manual assessment, and quantitative pupillary response parameters present high reliability in critically ill cardiac patients.

Quantitative pupillometry is part of multimodal neuroprognostication of comatose patients after out-of-hospital cardiac arrest (OHCA). However, the reproducibility, repeatability, and reliability of quantitative pupillometry in this setting have not been investigated. In a prospective blinded validation study, we compared manual and quantitative measurements of pupil size. Observer and device variability for all available parameters are expressed as mean difference (bias), limits of agreement (LoA), and reliability expressed as intraclass correlation coefficients (ICC) with a 95% confidence interval. Fifty-six unique quadrupled sets of measurement derived from 14 sedated and comatose patients (mean age 70±12 years) were included. For manually measured pupil size, inter-observer bias was -0.14±0.44 mm, LoA of -1.00 to 0.71 mm, and ICC at 0.92 (0.86-0.95). For quantitative pupillometry, we found bias at 0.03±0.17 mm, LoA of -0.31 to 0.36 mm and ICCs at 0.99. Quantitative pupillometry also yielded lower bias and LoA and higher ICC for intra-observer and inter-device measurements. Correlation between manual and automated pupillometry was better in larger pupils, and quantitative pupillometry had less variability and higher ICC, when assessing small pupils. Further, observers failed to detect 26% of the quantitatively estimated abnormal reactivity with manual assessment. We found ICC >0.91 for all quantitative pupillary response parameters (except for latency with ICC 0.81-0.91). Automated quantitative pupillometry has excellent reliability and twice the reproducibility and repeatability than manual pupillometry. This study further presents novel estimates of variability for all quantitative pupillary response parameters with excellent reliability.

Diagnostic value of striatal 18F-FP-DTBZ PET in Parkinson's disease.

Background18F-FP-DTBZ has been proven as a biomarker for quantifying the concentration of presynaptic vesicular monoamine transporter 2 (VMAT2). However, its clinical application is still limited.ObjectivesTo evaluate the difference in dopaminergic integrity between patients with Parkinson’s disease (PD) and healthy controls (HC) using 18F-FP-DTBZ PET in vivo and to determine the diagnostic value of standardized uptake value ratios (SUVRs) using the Receiver Operating Characteristic (ROC) curve.MethodsA total of 34 PD and 31 HC participants were enrolled in the PET/MR derivation cohort, while 89 PD and 18 HC participants were recruited in the PET/CT validation cohort. The Hoehn–Yahr Scale and the third part of the MDS-Unified Parkinson’s Disease Rating Scale (MDSUPDRS-III) were used to evaluate the disease staging and severity. All assessments and PET scanning were performed in drug-off states. The striatum was segmented into five subregions as follows: caudate, anterior dorsal putamen (ADP), anterior ventral putamen (AVP), posterior dorsal putamen (PDP), and posterior ventral putamen (PVP) using automatic pipeline built with the PMOD software (version 4.105). The SUVRs of the targeted subregions were calculated using the bilateral occipital cortex as the reference region.ResultsRegarding the diagnostic value, ROC curve and blind validation showed that the contralateral PDP (SUVR = 3.43) had the best diagnostic accuracy (AUC = 0.973; P < 0.05), with a sensitivity of 97.1% (95% CI: 82.9–99.8%), specificity of 100% (95% CI: 86.3–100%), positive predictive value (PPV) of 100% (95% CI: 87.0–100%), negative predictive value (NPV) of 96.9% (95% CI: 82.0–99.8%), and an accuracy of 98.5% for the diagnosis of PD in the derivation cohort. Blind validation of 18F-FP-DTBZ PET imaging diagnosis was done using the PET/CT cohort, where participants with a SUVR of the PDP <3.43 were defined as PD. Kappa test showed a consistency of 0.933 (P < 0.05) between clinical diagnosis and imaging diagnosis, with a sensitivity of 98.9% (95% CI: 93.0–99.9%), specificity of 94.4% (95% CI: 70.6–99.7%), PPV of 98.9% (95% CI: 93.0–99.9%), NPV of 94.4% (95% CI: 70.6–99.7%), and a diagnostic accuracy of 98.1%.ConclusionsOur results showed that an SUVR threshold of 3.43 in the PDP could effectively distinguish patients with PD from HC.

Highly Sensitive EGFRvIII Detection in Circulating Extracellular Vesicle RNA of Glioma Patients.

Liquid biopsy offers an attractive platform for noninvasive tumor diagnosis, prognostication, and prediction of glioblastoma clinical outcomes. Prior studies report that 30% to 50% of GBM lesions characterized by EGFR amplification also harbor the EGFRvIII mutation. A novel digital droplet PCR (ddPCR) assay for high GC content amplicons was developed and optimized for sensitive detection of EGFRvIII in tumor tissue and circulating extracellular vesicle RNA (EV RNA) isolated from the plasma of patients with glioma. Our optimized qPCR assay detected EGFRvIII mRNA in 81% [95% confidence interval (CI), 68%-94%] of EGFR-amplified glioma tumor tissue, indicating a higher than previously reported prevalence of EGFRvIII in glioma. Using the optimized ddPCR assay in discovery and blinded validation cohorts, we detected EGFRvIII mutation in 73% (95% CI, 64%-82%) of patients with a specificity of 98% (95% CI, 87%-100%), compared with qPCR tumor tissue analysis. In addition, upon longitudinal monitoring in 4 patients, we report detection of EGFRvIII in the plasma of patients with different clinical outcomes, rising with tumor progression, and decreasing in response to treatment. This study demonstrates the feasibility of detecting EGFRvIII mutation in plasma using a highly sensitive and specific ddPCR assay. We also show a higher than previously reported EGFRvIII prevalence in glioma tumor tissue. Several features of the assay are favorable for clinical implementation for detection and monitoring of EGFRvIII-positive tumors.

Characterization of new psychoactive substances by integrating benchtop NMR to multi-technique databases.

New psychoactive substances (NPS) have become a serious threat for public health due to their ability to be sold in the street or on internet. NPS are either derived from commercial drugs which are misused (recreational rather than medical use) or whose structure is slightly modified. To regulate NPS, it is essential to accurately characterize them, either to recognize molecules that were previously identified or to quickly elucidate the structure of unknown ones. Most approaches rely on the determination of the exact mass obtained by high‐resolution mass spectrometry requiring expensive equipment. This motivated us to develop a workflow in which the elucidation is assisted with databases and does not need the exact mass. This workflow combines 1D and 2D NMR measurements performed on a benchtop spectrometer with IR spectroscopy, for creating a multi‐technique database to characterize pure and mixed NPS. The experimental database was created with 57 entries mostly coming from seizures, mainly cathinones, cannabinoids, amphetamines, arylcyclohexylamines, and fentanyl. A blind validation of the workflow was carried out on a set of six unknown seizures. In the first three cases, AF, AB‐FUBINACA, and a mixture of 2C‐I and 2C‐E could be straightforwardly identified with the help of their reference spectra in the database. The two next samples were elucidated for the first time with the help of the database to reveal NEK and MPHP substances. Finally, a precise quantification of each characterized NPS was obtained in order to track NPS trafficking networks.

Multi-strut and empirical formula-based macro modeling for masonry infilled RC frames

It is widely accepted that infill walls play an important role in the structural response of buildings under seismic excitations. Disregarding their influences on the seismic design and evaluation of structures could be potentially dangerous. Therefore, a reliable numerical model is necessary. Because of its simplicity and efficiency, the equivalent strut model is the most widely used to represent infilled frames. In this study, a novel multi-strut macro model and associated empirical formulas for modeling infilled RC frames are proposed. The model was developed based on previous experimental observations. Fiber-section truss elements are used as the struts. All potential failure modes of the infill wall were considered by means of the equivalent stress–strain parameters of the struts. Empirical formulas for predicting these equivalent stress–strain parameters were developed and calibrated based on a comprehensive experimental database and regression analysis. Finally, blind validation tests of the proposed model and empirical formulas were performed to assess their reliability. Monotonic and cyclic load simulations of the infilled RC frames were carried out and compared with experimental results. The application of the proposed approach for both single-bay and multi-bay infilled frame simulations was examined.

Personalized approach for response prediction and treatment management for non-small cell lung cancer patients based on a liquid biopsy.

e21132 Background: To date, predicting response to immune checkpoint blockade (ICB) therapy in non-small cell lung cancer (NSCLC) patients is based on assessing PD-L1 levels in tumor biopsies. However, such assays are only moderately predictive. In addition, the assays require a tumor biopsy and do not aid in identifying patient-specific resistance mechanisms beyond PD-L1. To overcome these issues, we developed a novel computational approach for predicting response to ICB based on pre-treatment proteomic measurements in liquid biopsies. Methods: Plasma samples were collected from 184 NSCLC patients prior to treatment, along with comprehensive clinical data, as part of an ongoing multi-center clinical trial (PROPHETIC; NCT04056247). Overall response rate (ORR) was assessed 3- and 6-months following treatment initiation. A deep proteomic profiling of each plasma sample was performed, measuring the expression levels of approximately 7000 proteins. A novel proprietary machine learning approach was developed on a subset of samples (training set; n = 110) and then was tested on a blind independent validation set (n = 74). Results: A computational model was developed on the proteomic data by identifying patient-specific Resistance Associated Proteins (RAPs). Focusing on differentially expressed proteins between responders and non-responders, a protein was defined as a RAP in a given patient based on its expression level in the patient relative to the expression distribution of the RAP in responders and non-responders. The probability of response to ICB treatment was determined based on the patient’s RAP profile together with 4 clinical parameters. The RAP-based machine learning model successfully stratified between patients with prolonged and limited benefit with a hazard ratio (HR) of 4.5 (confidence interval 2.07-9.77; p-value &lt; 0.0001) and 2.27 (confidence interval 1.7-4.03; p-value = 0.004) for overall survival and progression free survival, respectively. Each patient displayed a resistance map comprised of a unique combination of RAPs, suggesting a new approach for personalized medicine based on patient-specific pathway blockade. For example, a patient with KDR and IL-6 defined as RAPs may benefit from a clinical trial that targets any of these RAPS in combination with ICB. Last, an exploration into the biological functions of the identified RAPs revealed specific biological processes in each response group, including splicing, complement system, coagulation and signaling. Conclusions: We have developed a novel computational approach based on proteomic profiling of liquid biopsies for predicting response to ICB treatment in NSCLC patients. Our approach also sheds light on patient-specific resistance mechanisms, potentially enabling personalized treatment options and patient monitoring over time.

Methylated DNA markers for plasma detection of ovarian cancer: Discovery, validation, and clinical feasibility

ObjectiveAberrant DNA methylation is an early event in carcinogenesis which could be leveraged to detect ovarian cancer (OC) in plasma. MethodsDNA from frozen OC tissues, benign fallopian tube epithelium (FTE), and buffy coats from cancer-free women underwent reduced representation bisulfite sequencing (RRBS) to identify OC MDMs. Candidate MDM selection was based on receiver operating characteristic (ROC) discrimination, methylation fold change, and low background methylation among controls. Blinded biological validation was performed using methylated specific PCR on DNA extracted from independent OC and FTE FFPE tissues. MDMs were tested using Target Enrichment Long-probe Quantitative Amplified Signal (TELQAS) assays in pre-treatment plasma from women newly diagnosed with OC and population-sampled healthy women. A random forest modeling analysis was performed to generate predictive probability of disease; results were 500-fold in silico cross-validated. ResultsThirty-three MDMs showed marked methylation fold changes (10 to >1000) across all OC subtypes vs FTE. Eleven MDMs (GPRIN1, CDO1, SRC, SIM2, AGRN, FAIM2, CELF2, RIPPLY3, GYPC, CAPN2, BCAT1) were tested on plasma from 91 women with OC (73 (80%) high-grade serous (HGS)) and 91 without OC; the cross-validated 11-MDM panel highly discriminated OC from controls (96% (95% CI, 89–99%) specificity; 79% (69–87%) sensitivity, and AUC 0.91 (0.86–0.96)). Among the 5 stage I/II HGS OCs included, all were correctly identified. ConclusionsWhole methylome sequencing, stringent filtering criteria, and biological validation yielded candidate MDMs for OC that performed with high sensitivity and specificity in plasma. Larger plasma-based OC MDM studies, including testing of pre-diagnostic specimens, are warranted.

An Office-Based, Point-of-Care Test Predicts Treatment Outcomes With Community-Based Pelvic Floor Physical Therapy in Patients With Chronic Constipation

Rectal evacuation disorders are common among constipated patients. We aimed to evaluate the accuracy of an investigational point-of-care test (rectal expulsion device [RED]) to predict outcomes with community-based pelvic floor physical therapy. We enrolled patients meeting Rome IV criteria for functional constipation failing fiber/laxatives for more than 2 weeks. RED was inserted and self-inflated, and then time-to-expel was measured in a left lateral position. All patients underwent empiric community-based pelvic floor physical therapy in routine care with outcomes measured at 12 weeks. The primary end point was global clinical response (Patient Assessment of Constipation Symptoms score reduction, >0.75 vs baseline). Secondary end points included improvement in health-related quality-of-life (Patient Assessment of Constipation Quality of Life score reduction, >1.0) and complete spontaneous bowel movement frequency (Food and Drug Administration complete spontaneous bowel movement responder definition). Thirty-nine patients enrolled in a feasibility phase to develop the use-case protocol. Sixty patients enrolled in a blinded validation phase; 52 patients (mean, 46.9 y; 94.2% women) were included in the intention-to-treat analysis. In the left lateral position, RED predicted global clinical response (generalized area under the curve [gAUC], 0.67; 95% CI, 0.58-0.76]), health-related quality-of-life response (gAUC, 0.67; 95% CI, 0.58-0.77; P < .001), and complete spontaneous bowel movement response (gAUC, 0.63; 95% CI, 0.57-0.71; P < .001). As a screening test, a normal RED effectively rules out evacuation disorders (expected clinical response, 8.9%; P= .042). Abnormal RED in the left lateral position (defined as expulsion within 5 seconds or >120 seconds) predicted 48.9% clinical response to physical therapy. A seated maneuver enhanced the likelihood of clinical response (71.1% response with seated RED retained >13 seconds) but likely is unnecessary in most settings. RED offers an opportunity to disrupt the paradigm by offering a personalized approach to managing chronic constipation in the community (Clinicaltrials.gov: NCT04159350).

Screening for delirium and dementia in older hospitalised adults in Zambia

Delirium prevalence and aetiology in older people in hospital or community settings in sub-Saharan Africa (SSA) is largely unknown. Cognitive screening tools designed for high-income countries (HICs) may be inappropriate due to cultural and educational differences, and delirium-specific measures lack validation in this context. The ‘Identification and Intervention for Dementia in Elderly Africans’ (IDEA) screen is a low-literacy tool developed and validated for dementia and delirium screening in Tanzania and Nigeria.This study aims to determine the prevalence and aetiology of delirium and dementia in older hospitalised patients in Zambia and to assess the utility of the IDEA screen for identification of major cognitive impairment in this setting.This was a blinded 4-month validation study which took place February–June 2015. Consecutive inpatient admissions of a rural mission hospital aged ≥60 years were administered the IDEA screen onadmission. Individuals were evaluated for dementia or delirium based on clinical examination, notes review and the Confusion Assessment Method. Delirium aetiological factors were recorded and classified (infectious/non-infectious).Of 136 patients recruited, dementia, delirium and major cognitive impairment were identified in 37 (27.2%), 45 (33.1%) and 62 (45.6%) respectively. Diagnostic accuracy of the IDEA screen for dementia and delirium was 0.661–0.795 (AUROC). Of those with delirium, 18 (40%) were classified infectious and 26 (57.8%) were classified non-infectious aetiologies.Dementia and delirium prevalence in older Zambian inpatients is comparable tohigh-income countries. The IDEA screen ispotentially clinically useful in this setting though diagnostic accuracy was lower than in initial validation studies. Non-infectious diseases are more highly represented amongst delirium precipitants than anticipated.

A Clinically Translatable, Extensively Validated Immune-based Classification of Human Papillomavirus-Associated Head and Neck Cancer With Implications for Treatment Deintensification and Immunotherapy

<h3>Purpose/Objective(s)</h3> Human papillomavirus-associated (HPV+) head and neck squamous cell carcinoma (HNSCC) is the fastest rising cancer in North America, with generally favourable prognosis. There is significant interest in treatment de-escalation for these patients, but 15-30% of patients recur after primary treatment, reflecting a need for improved risk-stratification tools. We sought to develop a molecular test to predict survival of patients with newly diagnosed HPV+ HNSCC. <h3>Materials/Methods</h3> We created an RNA-based prognostic score which was successfully validated in six cohorts comprising a total of 906 patients, including blinded external validations in a large retrospective cohort (n = 269) and in a prospective cohort (BD2Decide, NCT02832102, n = 286). We assessed the feasibility of using IHC to calculate the UWO3 score using an immunohistochemistry-based tissue microarray cohort (n = 197). Transcriptomic data from two aggressive treatment de-escalation cohorts were used to assess ability of UWO3 to identify patients who recur. Multivariate Cox models were used to assess the associations between the UWO3 score and outcomes, adjusting for potential confounders. <h3>Results</h3> Analysis of HPV+ HNSCC tumours revealed three distinct tumor microenvironment subtypes: immune rich, immune desert, and mixed. A three gene immune score (UWO3) classified patients into the three immune groups and was strongly associated with disease-free survival in six datasets, including blinded validation in a retrospective and a prospective dataset. Pooled analysis (n = 906) demonstrated that the immune rich group had superior disease-free survival at 5 years. The association between UWO3 immune class and time to recurrence was independent of patient age, sex, smoking status, alcohol intake status, and AJCC 8th edition clinical stage. Finally, UWO3 was able to identify patients who respond to aggressive treatment de-escalation to 30 Gy radiation in two treatment de-escalation cohorts. <h3>Conclusion</h3> The UWO3 immune score could inform clinical decision making for patients with HPV+ HNSCC based on robust outcome prediction across six independent cohorts. The superior survival of immune rich patients supports de-intensification strategies for this cohort, while the inferior outcomes of the immune desert patients suggest potential for intensification and/or immunotherapy. Prospective de-escalation and intensification clinical trials are currently being planned.

155 TERT Promoter Mutation Analysis for Blood-based Diagnosis and Monitoring of Gliomas

INTRODUCTION: Liquid biopsy offers a minimally invasive tool to diagnose and monitor the heterogeneous molecular landscape of tumors over time and therapy. Detection of TERT promoter mutations (C228T, C250T) in cfDNA has been successful for some systemic cancers but has yet to be demonstrated in gliomas, despite the high prevalence of these mutations in glioma tissue (&gt;60% of all tumors). METHODS: Here, we developed a novel digital droplet PCR (ddPCR) assay that incorporates features to improve sensitivity and allows for the simultaneous detection and longitudinal monitoring of two TERT promoter mutations (C228T and C250T) in cfDNA from the plasma of patients with glioma. RESULTS: In baseline performance in tumor tissue, the assay had perfect concordance with an independently performed clinical pathology laboratory assessment of TERT promoter mutations in the same tumor samples [95% confidence interval (CI), 94%-100%]. Extending to matched plasma samples, we detected TERT mutations in both discovery and blinded multi-institution validation cohorts with an overall sensitivity of 62.5% (95% CI, 52%-73%) and a specificity of 90% (95% CI, 80%-96%) compared with the gold-standard tumor tissue-based detection of TERT mutations. Upon longitudinal monitoring in 5 patients, we report that peripheral TERT-mutant allele frequency reflects the clinical course of the disease, with levels decreasing after surgical intervention and therapy and increasing with tumor progression. CONCLUSION: Our results demonstrate the feasibility of detecting circulating cfDNA TERT promoter mutations in patients with glioma with clinically relevant sensitivity and specificity.

Brainmarker-I Differentially Predicts Remission to Various Attention-Deficit/Hyperactivity Disorder Treatments: A Discovery, Transfer, and Blinded Validation Study

BackgroundAttention-deficit/hyperactivity disorder is characterized by neurobiological heterogeneity, possibly explaining why not all patients benefit from a given treatment. As a means to select the right treatment (stratification), biomarkers may aid in personalizing treatment prescription, thereby increasing remission rates. MethodsThe biomarker in this study was developed in a heterogeneous clinical sample (N = 4249) and first applied to two large transfer datasets, a priori stratifying young males (<18 years) with a higher individual alpha peak frequency (iAPF) to methylphenidate (N = 336) and those with a lower iAPF to multimodal neurofeedback complemented with sleep coaching (N = 136). Blinded, out-of-sample validations were conducted in two independent samples. In addition, the association between iAPF and response to guanfacine and atomoxetine was explored. ResultsRetrospective stratification in the transfer datasets resulted in a predicted gain in normalized remission of 17% to 30%. Blinded out-of-sample validations for methylphenidate (n = 41) and multimodal neurofeedback (n = 71) corroborated these findings, yielding a predicted gain in stratified normalized remission of 36% and 29%, respectively. ConclusionsThis study introduces a clinically interpretable and actionable biomarker based on the iAPF assessed during resting-state electroencephalography. Our findings suggest that acknowledging neurobiological heterogeneity can inform stratification of patients to their individual best treatment and enhance remission rates.