Cross-validated Logistic Regression Research Articles

Automated Video-Based Approach for the Diagnosis of Tourette Syndrome.

The occurrence of tics is the main basis for the diagnosis of Gilles de la Tourette syndrome (GTS). Video-based tic assessments are time consuming. The aim was to assess the potential of automated video-based tic detection for discriminating between videos of adults with GTS and healthy control (HC) participants. The quantity and temporal structure of automatically detected tics/extra movements in videos from adults with GTS (107 videos from 42 participants) and matched HCs were used to classify videos using cross-validated logistic regression. Videos were classified with high accuracy both from the quantity of tics (balanced accuracy of 87.9%) and the number of tic clusters (90.2%). Logistic regression prediction probability provides a graded measure of diagnostic confidence. Expert review of about 25% of lower-confidence predictions could ensure an overall classification accuracy above 95%. Automated video-based methods have a great potential to support quantitative assessment and clinical decision-making in tic disorders.

Development of an algorithm combining blood-based biomarkers, fecal immunochemical test, and age for population-based colorectal cancer screening

Implementation of screening modalities have reduced the burden of colorectal cancer (CRC), but high false positive rates pose a major problem for colonoscopy capacity. We aimed to create a tailored screening algorithm that expands the fecal immunochemical test (FIT) with a blood specimen and current age to improve selection of individuals for diagnostic colonoscopy. In this prospective multi-center study, eight blood-based biomarkers (CEA, Ferritin, hsCRP, HE4, Cyfra21-1, Hepsin, IL-8 and OPG) were investigated in 1,977 FIT positive individuals from the Danish national CRC screening program undergoing follow-up colonoscopy. Specimens were analyzed on ARCHITECT i2000®, ARCHITECT c8000® or Luminex xMAP® machines. FIT analyses and blood-based biomarker data were combined with clinical data (i.e., age and colonoscopy findings) in a cross-validated logistic regression model (algorithm) benchmarked against a model solely using the FIT result (FIT model) applying different cutoffs for FIT positivity. The cohort included individuals with CRC (n = 240), adenomas (n = 938) or no neoplastic lesions (n = 799). The cross-validated algorithm combining the eight biomarkers, quantitative FIT result and age performed superior to the FIT model in discriminating CRC versus non-CRC individuals (AUC 0.77 versus 0.67, p < 0.001). When discriminating individuals with either CRC or high- or medium-risk adenomas versus low-risk adenomas or clean colorectum, the AUCs were 0.68 versus 0.64 for the algorithm and FIT model, respectively. The algorithm presented here can improve patient allocation to colonoscopy, reducing colonoscopy burden without compromising cancer and adenomas detection rates or vice versa.

Algorithm Training and Testing for a Nonendoscopic Barrett’s Esophagus Detection Test in Prospective Multicenter Cohorts

Endoscopic Barrett's esophagus (BE) and esophageal adenocarcinoma (EAC) detection is invasive and expensive. Nonendoscopic BE/EAC detection tools are guideline-endorsed alternatives. We previously described a 5-methylated DNA marker (MDM) panel assayed on encapsulated sponge cell collection device (CCD) specimens. We aimed to train a new algorithm using a 3-MDM panel and test its performance in an independent cohort. Algorithm training and test samples were from 2 prospective multicenter cohorts. All BE cases had esophageal intestinal metaplasia (with or without dysplasia/EAC); control subjects had no endoscopic evidence of BE. The CCD procedure was followed by endoscopy. From CCD cell lysates, DNA was extracted, bisulfite treated, and MDMs were blindly assayed. The algorithm was set and locked using cross-validated logistic regression (training set) and its performance was assessed in an independent test set. Training (N= 352) and test (N= 125) set clinical characteristics were comparable. The final panel included 3 MDMs (NDRG4, VAV3, ZNF682). Overall sensitivity was 82% (95% CI, 68%-94%) at 90% (79%-98%) specificity and 88% (78%-94%) sensitivity at 84% (70%-93%) specificity in training and test sets, respectively. Sensitivity was 90% and 68% for all long- and short-segment BE, respectively. Sensitivity for BE with high-grade dysplasia and EAC was 100% in training and test sets. Overall sensitivity for nondysplastic BE was 82%. Areas under the receiver operating characteristic curves for BE detection were 0.92 and 0.94 in the training and test sets, respectively. A locked 3-MDM panel algorithm for BE/EAC detection using a nonendoscopic CCD demonstrated excellent sensitivity for high-risk BE cases in independent validation samples. (Clinical trials.gov: NCT02560623, NCT03060642.).

Identification of preclinical dementia according to ATN classification for stratified trial recruitment: A machine learning approach.

The Amyloid/Tau/Neurodegeneration (ATN) framework was proposed to identify the preclinical biological state of Alzheimer's disease (AD). We investigated whether ATN phenotype can be predicted using routinely collected research cohort data. 927 EPAD LCS cohort participants free of dementia or Mild Cognitive Impairment were separated into 5 ATN categories. We used machine learning (ML) methods to identify a set of significant features separating each neurodegeneration-related group from controls (A-T-(N)-). Random Forest and linear-kernel SVM with stratified 5-fold cross validations were used to optimize model whose performance was then tested in the ADNI database. Our optimal results outperformed ATN cross-validated logistic regression models by between 2.2% and 8.3%. The optimal feature sets were not consistent across the 4 models with the AD pathologic change vs controls set differing the most from the rest. Because of that we have identified a subset of 10 features that yield results very close or identical to the optimal. Our study demonstrates the gains offered by ML in generating ATN risk prediction over logistic regression models among pre-dementia individuals.

Predicting Hearing Loss in Testicular Cancer Patients after Cisplatin-Based Chemotherapy.

Testicular cancer is predominantly curable, but the long-term side effects of chemotherapy have a severe impact on life quality. In this research study, we focus on hearing loss as a part of overall chemotherapy-induced ototoxicity. This is a unique approach where we combine clinical data from the acclaimed nationwide Danish Testicular Cancer (DaTeCa)-Late database. Clinical and genetic data on 433 patients were collected from hospital files in October 2014. Hearing loss was classified according to the FACT/GOG-Ntx-11 version 4 self-reported Ntx6. Machine learning models combining a genome-wide association study within a nested cross-validated logistic regression were applied to identify patients at high risk of hearing loss. The model comprising clinical and genetic data identified 67% of the patients with hearing loss; however, this was with a false discovery rate of 49%. For the non-affected patients, the model identified 66% of the patients with a false omission rate of 19%. An area under the receiver operating characteristic (ROC-AUC) curve of 0.73 (95% CI, 0.71-0.74) was obtained, and the model suggests genes SOD2 and MGST3 as important in improving prediction over the clinical-only model with a ROC-AUC of 0.66 (95% CI, 0.65-0.66). Such prediction models may be used to allow earlier detection and prevention of hearing loss. We suggest a possible biological mechanism for cisplatin-induced hearing loss development. On confirmation in larger studies, such models can help balance treatment in clinical practice.

Plasma biomarkers for prediction of Alzheimer's disease neuropathologic change.

While plasma biomarkers for Alzheimer's disease (AD) are increasingly being evaluated for clinical diagnosis and prognosis, few population-based autopsy studies have evaluated their utility in the context of predicting neuropathological changes. Our goal was to investigate the utility of clinically available plasma markers in predicting Braak staging, neuritic plaque score, Thal phase, and overall AD neuropathological change (ADNC).We utilized a population-based prospective study of 350 participants with autopsy and antemortem plasma biomarker testing using clinically available antibody assay (Quanterix) consisting of Aβ42/40 ratio, p-tau181, GFAP, and NfL. We utilized a variable selection procedure in cross-validated (CV) logistic regression models to identify the best set of plasma predictors along with demographic variables, and a subset of neuropsychological tests comprising the Mayo Clinic Preclinical Alzheimer Cognitive Composite (Mayo-PACC). ADNC was best predicted with plasma GFAP, NfL, p-tau181 biomarkers along with APOE ε4 carrier status and Mayo-PACC cognitive score (CV AUC = 0.798). Braak staging was best predicted using plasma GFAP, p-tau181, and cognitive scores (CV AUC = 0.774). Neuritic plaque score was best predicted using plasma Aβ42/40 ratio, p-tau181, GFAP, and NfL biomarkers (CV AUC = 0.770). Thal phase was best predicted using GFAP, NfL, p-tau181, APOE ε4 carrier status and Mayo-PACC cognitive score (CV AUC = 0.754). We found that GFAP and p-tau provided non-overlapping information on both neuritic plaque and Braak stage scores whereas Aβ42/40 and NfL were mainly useful for prediction of neuritic plaque scores. Separating participants by cognitive status improved predictive performance, particularly when plasma biomarkers were included. Plasma biomarkers can differentially inform about overall ADNC pathology, Braak staging, and neuritic plaque score when combined with demographics and cognitive variables and have significant utility for earlier detection of AD.

Correlation of gene signatures of Kaiso and LC3A/B protein expression with distinct tumor ecosystems and to predict response to breast cancer therapy.

566 Background: The multifunctional transcriptional repressor, Kaiso, is a member of the BTB/POZ (Broad complex, Tramtrak, Bric `aBrac/Pox zinc finger) family of zinc finger proteins. Kaiso protein levels in both the nucleus and the cytoplasm have each been shown to be highly sensitive and independent predictors of overall breast cancer survival. Through the use of gene signatures and/or gene modules developed through the analysis of RNA-seq data from breast cancer patients following biomarker stratification by either nuclear Kaiso, cytoplasmic Kaiso, or the Kaiso’s functional downstream target LC3A/B, we demonstrate the utility of Kaiso and LC3A/B derived signatures as predictors of treatment outcome. Methods: We used a machine Learning approach to assess a cohort of racially diverse 555 BC patients who underwent surgery for their primary BC in Greenville, NC and develop proteomics-based genomics (PbG) signatures. The cross-validated logistic regression modeling using publicly available gene expression data compiled from neoadjuvant clinical trials (N=996, JCO:30(16):1996-2004). Results: Kaiso and LC3A/B derived gene modules show accuracy in predicting pathological complete response (pCR, AUC = 0.728, 95% CI: 0.689-0.764). Moreover, predictive estimates of distant metastasis-free survival utilizing Kaiso and LC3A/B derived gene modules are also significant (overall DMFS, AUC= 0.737, 95% CI: 0.665-0.819) and (DMFS of non-pCR patients AUC= 0.71, 95% CI: 0.633-0.773). We then validate these findings in an independent, racially diverse breast cancer cohort receiving neoadjuvant treatment (N=443). Gene modules based on Kaiso and LC3A/B were also highly predictive in this independent cohort (total cohort pCR, AUC= 0.775, 95% CI: 0.687-0.838; pCR NHW (N=257), AUC= 0.762, 95% CI: 0.659-0.855; pCR Hispanic (N=125), AUC= 0.725, 95% CI: 0.493-0.873; and pCR NHB (N=42), AUC= 0.681, 95% CI:0.405-0.904). Conclusions: Combined analysis by multiplex immunofluorescence and deconvolution of gene expression data show that the protein expression of Kaiso and LC3A/B in tumors are associated with distinct tumor ecosystems where high levels cytoplasmic Kaiso are linked to an immune-suppressed tumor microenvironment.

MP17-08 CLINICAL VALIDATION OF A NOVEL SIX-GENE RISK-STRATIFICATION SIGNATURE TO SUPPORT TREATMENT DECISIONS FOR MEN DIAGNOSED WITH PROSTATE CANCER

You have accessJournal of UrologyCME1 Apr 2023MP17-08 CLINICAL VALIDATION OF A NOVEL SIX-GENE RISK-STRATIFICATION SIGNATURE TO SUPPORT TREATMENT DECISIONS FOR MEN DIAGNOSED WITH PROSTATE CANCER Agnieszka Krzyzanowska, Stephen Barron, Debra F. Higgins, Tony Loughman, Katherine M. Sheehan, Angel Chan-Ju Wang, Bozena Fender, Leah M. McGuire, David Galvin, Amanda O'Neil, Joanna Fay, Anthony O'Grady, Desmond O'Leary, R. William. G. Watson, Anders Bjartell, and William M. Gallagher Agnieszka KrzyzanowskaAgnieszka Krzyzanowska More articles by this author , Stephen BarronStephen Barron More articles by this author , Debra F. HigginsDebra F. Higgins More articles by this author , Tony LoughmanTony Loughman More articles by this author , Katherine M. SheehanKatherine M. Sheehan More articles by this author , Angel Chan-Ju WangAngel Chan-Ju Wang More articles by this author , Bozena FenderBozena Fender More articles by this author , Leah M. McGuireLeah M. McGuire More articles by this author , David GalvinDavid Galvin More articles by this author , Amanda O'NeilAmanda O'Neil More articles by this author , Joanna FayJoanna Fay More articles by this author , Anthony O'GradyAnthony O'Grady More articles by this author , Desmond O'LearyDesmond O'Leary More articles by this author , R. William. G. WatsonR. William. G. Watson More articles by this author , Anders BjartellAnders Bjartell More articles by this author , and William M. GallagherWilliam M. Gallagher More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003237.08AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Prostate cancer (PCa) is diagnosed in over 1.2 million men worldwide annually. Most men have indolent PCa and can safely defer treatment with active surveillance. However, many men are commonly treated with surgery or radiation therapy which are associated with reduced quality of life. Clinicians are increasingly using molecular information to further understand a patient’s risk of aggressive PCa, and while these signatures have shown promise, there remains a significant opportunity for improving prognostic capacity. METHODS: Expression of a novel six-gene signature was measured by RT-qPCR in prostate samples from an Irish cohort of men diagnosed with PCa and treated with radical prostatectomy (RP) (PCRC, n=426). Cross-validated logistic regression identified a six-gene molecular risk score (MRS) with strong prognostic performance to predict aggressive PCa. The MRS was combined with the Cancer of the Prostate Risk Assessment (CAPRA) score to create a molecular plus clinical risk score (MCRS). The MRS and MCRS were clinically validated in an independent Swedish cohort of men diagnosed with PCa and treated by RP (UPCA, n=203 RP and n=100 biopsy samples). RESULTS: We identified a six-gene signature, four prognostic and two reference genes, with improved prognostic value over clinical features. In the clinical biopsy validation study, the AUC of the signature for AP was 0.72 (MRS) and 0.82 (MCRS) versus 0.72 (EAU) and 0.82 (CAPRA), and for high primary Gleason was 0.75 (MRS) and 0.85 (MCRS) versus 0.77 (EAU) and 0.83 (CAPRA). The C index for BCR was 0.62 (MRS) and 0.74 (MCRS) versus 0.65 (EAU) and 0.73 (CAPRA). In bivariable analysis, the six-gene signature added statistically significant (p<0.001) prognostic value to EAU and CAPRA. In the RP clinical validation study, the C index for BCR was 0.74 (MRS) and 0.83 (MCRS) versus 0.69 (EAU) and 0.77 (CAPRA). In bivariable analysis, the six-gene signature added statistically significant (p<0.0001) prognostic value to EAU and CAPRA. CONCLUSIONS: The six-gene molecular prognostic signature has strong performance for AP, high primary Gleason and BCR in independent clinical validation studies when tested on prostate biopsy or RP tissue. MCRS provides improved prognostic stratification and can inform optimal patient management post-diagnosis and post-treatment. Source of Funding: Support is acknowledged from the Swedish Cancer Society, the Swedish Scientific Council and Government Funding of Clinical Research within the Swedish National Health Service (ALF), the Irish Cancer Society, and OncoAssure Ltd © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e215 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Agnieszka Krzyzanowska More articles by this author Stephen Barron More articles by this author Debra F. Higgins More articles by this author Tony Loughman More articles by this author Katherine M. Sheehan More articles by this author Angel Chan-Ju Wang More articles by this author Bozena Fender More articles by this author Leah M. McGuire More articles by this author David Galvin More articles by this author Amanda O'Neil More articles by this author Joanna Fay More articles by this author Anthony O'Grady More articles by this author Desmond O'Leary More articles by this author R. William. G. Watson More articles by this author Anders Bjartell More articles by this author William M. Gallagher More articles by this author Expand All Advertisement PDF downloadLoading ...

Low discriminative power of WISC cognitive profile in developmental dyscalculia

BackgroundThe role of domain-general cognitive abilities in the etiology of Developmental Dyscalculia (DD) is a hotly debated issue. AimsIn the present study, we tested whether WISC-IV cognitive profiles can be useful to single out DD. Methods and proceduresUsing a stringent 2-SD cutoff in a standardized numeracy battery, we identified children with DD (N = 43) within a clinical sample referred for assessment of learning disability and compared them in terms of WISC cognitive indexes to the remaining children without DD (N = 100) employing cross-validated logistic regression. Outcomes and resultsBoth groups showed higher Verbal Comprehension and Perceptual Reasoning than Working Memory and Processing Speed, and DD scores were generally lower. Predictive accuracy of WISC indexes in identifying DD individuals was low (AUC = 0.67) and it dropped to chance level in discriminating DD from selected controls (N = 43) with average math performance but matched on global IQ. The inclusion of a visuospatial memory score as an additional predictor did not improve classification accuracy. Conclusions and implicationsThese results demonstrate that cognitive profiles do not reliably discriminate DD from non-DD children, thereby weakening the appeal of domain-general accounts.

Development and clinical validation of a novel biopsy-based six-gene signature to accurately risk-stratify men with early-stage prostate cancer to aid treatment decision making.

374 Background: Prostate cancer (PCa) is diagnosed in over 1.2 million men worldwide annually, making it the second-most common non-cutaneous cancer in men. Most men have indolent PCa and can safely defer treatment with active surveillance. However, many patients are commonly treated with surgery (radical prostatectomy, RP) or radiation therapy, which are associated with a significant reduction in quality of life. Molecular signatures have been developed to assist clinicians in stratifying patients based on their risk of aggressive PCa; while these have shown promise, there remains a significant opportunity for improving prognostic capacity. Methods: A panel of fourteen prognostic genes was identified from The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA PRAD) database. Gene expression was measured by reverse-transcription quantitative polymerase chain reaction in RP samples from an Irish cohort of men diagnosed with PCa and treated with RP (PCRC, n=426). Cross-validated logistic regression analysis identified a six-gene molecular risk score (MRS) with strong prognostic performance to predict aggressive PCa, i.e., adverse pathology (AP) at RP or biochemical recurrence (BCR). The MRS was combined with the Cancer of the Prostate Risk Assessment (CAPRA) score to create a molecular plus clinical risk score (MCRS). The MRS and MCRS were clinically validated in biopsy samples from an independent Swedish cohort of men diagnosed with PCa and treated by RP (UPCA, n=100). Results: We identified a six-gene signature, consisting of four prognostic and two reference genes, with improved prognostic value over clinical features. In the clinical biopsy validation study univariable analysis, the AUC of the signature for AP was 0.72 (MRS) and 0.82 (MCRS) versus 0.72 for EAU risk categories and 0.82 for CAPRA. The C-index for BCR was 0.62 (MRS) and 0.74 (MCRS) versus 0.65 for EAU risk categories and 0.73 for CAPRA. Furthermore, in bivariable analysis, the six-gene signature added statistically significant (p &lt; 0.001) prognostic value to EAU risk categories and CAPRA. Conclusions: The six-gene prognostic signature has strong performance for AP and BCR in an independent clinical biopsy validation study. MCRS can provide improved prognostic stratification and inform optimal patient management post-diagnosis.

Severe Developmental Dyscalculia Is Characterized by Core Deficits in Both Symbolic and Nonsymbolic Number Sense

A long-standing debate concerns whether developmental dyscalculia is characterized by core deficits in processing nonsymbolic or symbolic numerical information as well as the role of domain-general difficulties. Heterogeneity in recruitment and diagnostic criteria make it difficult to disentangle this issue. Here, we selected children (n = 58) with severely compromised mathematical skills (2 SD below average) but average domain-general skills from a large sample referred for clinical assessment of learning disabilities. From the same sample, we selected a control group of children (n = 42) matched for IQ, age, and visuospatial memory but with average mathematical skills. Children with dyscalculia showed deficits in both symbolic and nonsymbolic number sense assessed with simple computerized tasks. Performance in the digit-comparison task and the numerosity match-to-sample task reliably separated children with developmental dyscalculia from controls in cross-validated logistic regression (area under the curve = .84). These results support a number-sense-deficit theory and highlight basic numerical abilities that could be targeted for early identification of at-risk children as well as for intervention.

Longitudinal assessment of brain structure and behaviour in youth with rapid weight gain: Potential contributing causes and consequences.

Independent of weight status, rapid weight gain has been associated with underlying brain structure variation in regions associated with food intake and impulsivity among pre-adolescents. Yet, we lack clarity on how developmental maturation coincides with rapid weight gain and weight stability. We identified brain predictors of 2-year rapid weight gain and its longitudinal effects on brain structure and impulsivity in the Adolescent Brain Cognitive DevelopmentSM Study®. Youth were categorized as Healthy Weight/Weight Stable (WSHW , n=527) or Weight Gainers (WG, n=221, >38lbs); 63% of the WG group were healthy weight at 9-to-10-years-old. A fivefold cross-validated logistic elastic-net regression revealed that rapid weight gain was associated with structural variation amongst 39 brain features at 9-to-10-years-old in regions involved with executive functioning, appetitive control and reward sensitivity. Two years later, WG youth showed differences in change over time in several of these regions and performed worse on measures of impulsivity. These findings suggest that brain structure in pre-adolescence may predispose some to rapid weight gain and that weight gain itself may alter maturational brain change in regions important for food intake and impulsivity. Behavioural interventions that target inhibitory control may improve trajectories of brain maturation and facilitate healthier behaviours.

Pupil size as a robust marker of attentional bias toward nicotine-related stimuli in smokers

Spatial attention can be magnetically attracted by behaviorally salient stimuli. This phenomenon occasionally conflicts with behavioral goals, leading to maladaptive consequences, as in the case of addiction, in which attentional biases have been described and linked with clinically meaningful variables, such as craving level or dependence intensity. Here, we sought to probe the markers of attentional priority in smokers through eye-tracking measures, by leveraging the established link between eye movements and spatial attention. We were particularly interested in potential markers related to pupil size, because pupil diameter reflects a range of autonomic, affective, and cognitive/attentional reactions to behaviorally significant stimuli and is a robust marker of appetitive and aversive learning. We found that changes in pupil size to nicotine-related visual stimuli could reliably predict, in cross-validated logistic regression, the smoking status of young smokers (showing pupil constriction) better than more traditional proxy measures. The possibility that pupil constriction may reflect a bias toward central vision, for example, attentional capture, is discussed in terms of sensory tuning with respect to nicotine-related stimuli. Pupil size was more sensitive at lower nicotine dependence levels, and at increased abstinence time (though these two variables were collinear). We conclude that pupillometry can provide a robust marker for attentional priority computation and useful indications regarding motivational states and individual attitudes toward conditioned stimuli.

S410 Algorithm Training and Independent Test Set Performance for a Molecular Non-Endoscopic Test for Detection of Esophageal Adenocarcinoma and Barrett’s Esophagus in Multicenter Cohorts

Introduction: Sedated endoscopy for Barrett’s esophagus (BE) and esophageal adenocarcinoma (EAC) detection is invasive and expensive. Non-endoscopic BE/EAC detection tools have been guideline-endorsed to facilitate higher patient participation at lower cost. We previously described a promising panel of 5 methylated DNA markers (MDMs) assayed on esophageal specimens obtained by a sponge-on-a-string (SOS) cell collection device in phase II studies. We aimed to train an algorithm (establishing marker cut offs, to adjudicate samples as positive/negative) using a final MDM panel followed by testing in an independent sample set. Methods: Algorithm training samples (N=352) were prospectively collected from patients seen at 6 US medical centers. Test samples (N=125) were obtained from an independent, NIH-funded study conducted at 3 US medical centers. Cases had columnar metaplasia with intestinal metaplasia; controls had no endoscopic evidence of BE. Histology was reviewed by expert GI pathologists. The SOS device (25 mm, 10 ppi) was swallowed and withdrawn after 6-8 minutes followed by criterion standard endoscopy within 24 hours. DNA was extracted and bisulfite treated. Five MDMs were blindly assayed using the long probe quantitative amplified signal method. The algorithm was set using cross-validated logistic regression. The locked algorithm was applied to assay results from the test set. Results: Baseline characteristics of patients in training and test sets were comparable (Table). The final panel included 3 MDMs (NDRG4, VAV3, and ZNF682) and a reference marker (B3GALT6). Overall sensitivity for BE/EAC detection in the training set was 81% (95% CI 76-88%) at 90% (85-94%) specificity. Overall BE/EAC sensitivity in the test set was 88% (78-94%) at 84% (70-93%) specificity. Sensitivity for HGD and EAC was 100% in the training and test sets. Sensitivity for short segment NDBE in the test set was 63% (38-84%). Areas under the receiver operating characteristic (AUROC) curve for BE/EAC detection were 0.92 (95% CI 0.89-0.95) and 0.94 (0.90-0.98) in the training and test sets, respectively (Figure). The algorithm was not influenced by age, sex, or smoking history. 97% of participants in the training set and 85% in the test set successfully swallowed the SOS device, which was well tolerated and safe. Conclusion: A 3-MDM panel for BE/EAC detection demonstrated excellent sensitivity for high risk BE cases in multi-center case control training and test sets.Figure 1.: Area under the receiver operating characteristics curves for a 3-marker methylated DNA panel assayed from cytology specimen extracted DNA in training and test sets. Table 1. - Baseline characteristics of patients and performance characteristics of SOS test (overall sensitivity and specificity, and stratified by BE dysplasia grade) in training and test sets Variable Training Set (N=198 controls,154 cases) Test Set(N= 44 controls,81 cases) P value (comparing training and test sets) Control Case Control Case Mean (SD) age 55 (13) 65 (10) 52 (15) 65 (11) 0.312 Male Sex (%) 102 (52) 119 (77) 17 (39) 64 (79) 0.992 Mean (SD) BMI 29 (7) 30 (6) 30 (7) 30 (6) 0.283 Ever Smokers % 77 (39) 87 (56) 18 (41) 47 (58) 0.733 Mean *SD) BE length, cm - 4 (3) - 5 (3) 0.070 Long segment BE, N, (%) - 97 (63) - 56 (69) 0.426 Short segment BE, N (%) - 57 (37) - 25 (31) BE dysplasia grade, N (%) EAC - 12 (8) - 2 (2) HGD - 18 (12) - 11 (14) LGD - 7 (4) - 10 (12) IND - 18 (12) - 14 (17) NDBE (long segment) - 57 (37) - 25 (31) NDBE (short segment) - 42 (27) - 19 (24) Training Set% Positive (95% CI) Test Set% Positive (95% CI) Overall 81% (76-88%) 88 (78-94) Dysplasia grade EAC 100 (100-100) 100 (16-100) HGD 100 (100-100) 100 (72-100) LGD 71 (0-100) 90 (55-100) IND 74 (0-100) 93 (66-100%) NDBE (long segment) 91 (73-100) 96 (80-100%) NDBE (short segment) 61 (25-100) 63 (38-84%) Control (No BE) 10 (2-21) 16 (7-30)

Development and validation of prediction models for gestational diabetes treatment modality using supervised machine learning: a population-based cohort study

BackgroundGestational diabetes (GDM) is prevalent and benefits from timely and effective treatment, given the short window to impact glycemic control. Clinicians face major barriers to choosing effectively among treatment modalities [medical nutrition therapy (MNT) with or without pharmacologic treatment (antidiabetic oral agents and/or insulin)]. We investigated whether clinical data at varied stages of pregnancy can predict GDM treatment modality.MethodsAmong a population-based cohort of 30,474 pregnancies with GDM delivered at Kaiser Permanente Northern California in 2007–2017, we selected those in 2007–2016 as the discovery set and 2017 as the temporal/future validation set. Potential predictors were extracted from electronic health records at different timepoints (levels 1–4): (1) 1-year preconception to the last menstrual period, (2) the last menstrual period to GDM diagnosis, (3) at GDM diagnosis, and (4) 1 week after GDM diagnosis. We compared transparent and ensemble machine learning prediction methods, including least absolute shrinkage and selection operator (LASSO) regression and super learner, containing classification and regression tree, LASSO regression, random forest, and extreme gradient boosting algorithms, to predict risks for pharmacologic treatment beyond MNT.ResultsThe super learner using levels 1–4 predictors had higher predictability [tenfold cross-validated C-statistic in discovery/validation set: 0.934 (95% CI: 0.931–0.936)/0.815 (0.800–0.829)], compared to levels 1, 1–2, and 1–3 (discovery/validation set C-statistic: 0.683–0.869/0.634–0.754). A simpler, more interpretable model, including timing of GDM diagnosis, diagnostic fasting glucose value, and the status and frequency of glycemic control at fasting during one-week post diagnosis, was developed using tenfold cross-validated logistic regression based on super learner-selected predictors. This model compared to the super learner had only a modest reduction in predictability [discovery/validation set C-statistic: 0.825 (0.820–0.830)/0.798 (95% CI: 0.783–0.813)].ConclusionsClinical data demonstrated reasonably high predictability for GDM treatment modality at the time of GDM diagnosis and high predictability at 1-week post GDM diagnosis. These population-based, clinically oriented models may support algorithm-based risk-stratification for treatment modality, inform timely treatment, and catalyze more effective management of GDM.

Glycoproteomics as a powerful liquid biopsy-based screening tool for non-small cell lung cancer.

e21148 Background: Protein glycosylation is the most abundant and complex form of post-translational protein modification. Glycosylation profoundly affects protein structure, conformation, and function. The elucidation of the potential role of differential protein glycosylation as biomarkers has so far been limited by the technical complexity of generating and interpreting this information. We have recently established a novel, powerful platform that combines liquid chromatography/mass spectrometry with a proprietary artificial-intelligence-based data processing engine that allows, for the first time, highly scalable interrogation of the glycoproteome. Methods: Using this platform, we interrogated 694 glycopeptide (GP) and non-glycosylated peptide transitions derived from 74 serum proteins in pre-treatment peripheral blood samples from a cohort of 316 individuals with non-small-cell lung cancer (NSCLC) (128 females, 187 males, 1 with unknown sex, median age 66 years, age range 31-89 years, stage 0-4 N’s: 1 / 99 / 80 / 84 / 49, 3 missing) and a comparison cohort of 194 healthy control samples (102 females, 92 males, median age 52 years, age range 30-63 years). Age- and sex-adjusted differential expression analysis for 596 normalized biomarkers were performed to evaluate statistically significant differential abundances using an FDR-adjusted q-value of 0.05 as a cutoff. Repeated five-fold cross-validated LASSO-regularized logistic regression was performed to create a multivariable classifier that predicts whether a serum sample belongs to the healthy or NSCLC cohort. Results: We identified 432 biomarkers with significant abundance differences at FDR ≤ 0.05 between samples with NSCLC and healthy controls. Using 70% of the complete cohort (balanced by case/control membership, NSCLC stage, sex, and age quartile) as a training set, we selected a total of 375 glycopeptide and non-glycosylated peptide biomarker features that remained differentially expressed at FDR-adjusted q-value ≤ 0.05 as input into a LASSO-regularized multivariable classifier. This resulting in a 19-biomarker model exhibiting an accuracy of 94.8% (96.9% sensitivity, 91.2% specificity) and AUC of 0.989. This classifier was validated in an independent test set comprising the remaining 30% of subjects, yielding an accuracy of 94.5% (95.5% sensitivity, 93.0% specificity) and AUC of 0.975. Sensitivity in the test set was 100% / 96% / 99% / 96% / 94% / 10%, in stages 0-4 and missing, respectively. Conclusions: Our results indicate that glycoproteomic biomarkers can be leveraged as a strong liquid biopsy-based screening tool for patients at high risk of NSCLC, as an alternative to imaging modalities.

Abstract P5-13-15: High dimensional flow cytometric analysis or the peripheral immune profile and response to HER2-targeted antibody therapy

Abstract Background: DAPHNe was a prospective trial designed to assess adherence to de-escalated antibody doublet therapy in the adjuvant setting among HER2+ breast cancer patients experiencing a pathologic complete response (pCR) following neoadjuvant taxol (T), trastuzumab (H) and pertuzumab (P). Peripheral blood mononuclear cells (PBMC), were collected from all patients at baseline and after THP completion. The goal of this study was to determine if a patient’s peripheral blood immune profile at baseline, or the longitudinal change with treatment, could predict response to THP. Methods: Blood samples were subjected to high dimensional (28-30 parameter) flow cytometry with comprehensive T- and NK-cell panels. A fully automated computational analysis strategy was undertaken consisting of unsupervised clustering of the high dimensional data into groups of cells with similar immunophenotypic signatures. Clustering was performed using 2 algorithms: Fingerprint-based clustering (Fluster) and High Throughput Mapper (HiTMapper). Clusters were tested using the Wilcoxon rank-sum test for correlation with the clinical response. Responders were those with pCR (=residual cancer burden [RCB] 0) or RCB 1; non-responders were those with RCB 2/3 disease. P values were adjusted with the Benjamini-Hochberg method to control for FDR. In addition to P values, effect size was evaluated using the nonparametric Cliff’s Delta measure. An effect was determined to be large if the magnitude of Delta was &amp;gt;0.4 which corresponded to one cohort coming ahead 70% of the time. In addition, groups of clusters were evaluated using multivariate statistical modeling or dimensionality reduction to determine if there was an association with pCR. Results: Matched baseline and pre-op PBMC were available to perform the NK panel in 66 patients and the T cell panel in 40. In both groups 70% were responders and 30% were non-responders. No cluster produced by Fluster or HiTMapper differed significantly between responders and non-responders however, in the T cell panel, several clusters had a large effect size (table) suggesting the clusters are good at differentiating some, responders from non-responders. Both algorithms agreed that the median responder has more CD4 naïve and CD8 naïve cells than the median non-responder. While no individual cluster differed significantly between responders and non-responders, cross-validated logistic regression analyses showed that 2 clusters, activated CD4 central memory clusters, and activated CD4 naïve clusters, predicted responder status with AUC of 0.70 and 0.68 respectively. Numerous clusters showed robust and significant longitudinal changes between baseline and pre-op samples. Stratifying longitudinal changes by response status revealed no significant differences between responders and non-responders, however evaluation of effect size suggested a naïve CD4 cluster that increased in non-responders and decreased in responders. The latter could be explained as naïve T cells acquiring a memory phenotype in response to treatment in responders. Conclusion: High dimensional flow cytometry suggested a potential role for monitoring several T cell subsets to predict response in HER2+ patients receiving THP. Additional analyses to include cyTOF evaluation of PBMCs are ongoing to further characterize the peripheral immune profile of these HER2+ patients. T cell clusters with high effect sizeMajor PhenotypeOther markersMethodp-valueeffect sizeCD4 CM.actCD38, CD226HiTMapper0.351-0.469CD4 Naive.act.2CD38, CD226HiTMAPPER0.4-0.413CD4 Naive.act.4CD226HiTMapper0.351-0.490CD4 Naive.act.5CD38HiTMapper0.351-0.524CD8 Naive 1CD226HiTMapper0.396-0.427CD3 Neg-Fluster0.3190.476CD3 Neg3CD45RA, CD185, CD197Fluster0.3190.476CD3 Neg4CD45RA, Eomes,tBETFluster0.4240.413CD4 Naive-Fluster0.319-0.517CD8 Naive1-Fluster0.319-0.469Unassigned 20 (CD4)CD45RA, CD27-, CD28-Fluster0.364-0.441 Citation Format: Esther R Ogayo, Adrienne Waks, Wade Rogers, Matei Ionita, Kenechukwu Adigwe, Jillian Alberti, Sapana Kadel, Jonni Moore, Tari King, Ian Krop, Sara Tolaney, Eric Winer, Jennifer Guerriero, Elizabeth Mittendorf. High dimensional flow cytometric analysis or the peripheral immune profile and response to HER2-targeted antibody therapy [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P5-13-15.

Using a machine learning approach to investigate factors associated with treatment-resistant depression among adults with chronic non-cancer pain conditions and major depressive disorder

Objective Presence of chronic non-cancer pain conditions (CNPC) among adults with major depressive disorder (MDD) may reduce benefits of antidepressant therapy, thereby increasing the possibility of treatment resistance. This study sought to investigate factors associated with treatment-resistant depression (TRD) among adults with MDD and CNPC using machine learning approaches. Methods This retrospective cohort study was conducted using a US claims database which included adults with newly diagnosed MDD and CNPC (January 2007–June 2017). TRD was identified using a clinical staging algorithm for claims data. Random forest (RF), a machine learning method, and logistic regression was used to identify factors associated with TRD. Initial model development included 42 known and/or probable factors that may be associated with TRD. The final refined model included 20 factors. Results Included in the sample were 23,645 patients (73% female mean age: 55 years; 78% with ≥2 CNPC, and 91% with joint pain/arthritis). Overall, 11.4% adults (N = 2684) met selected criteria for TRD. The five leading factors associated with TRD were the following: mental health specialist visits, polypharmacy (≥5 medications), psychotherapy use, anxiety, and age. Cross-validated logistic regression model indicated that those with TRD were younger, more likely to have anxiety, mental health specialist visits, polypharmacy, and psychotherapy use with adjusted odds ratios (AORs) ranging from 1.93 to 1.27 (all ps < .001). Conclusion Machine learning identified several factors that warrant further investigation and may serve as potential targets for clinical intervention to improve treatment outcomes in patients with TRD and CNPC.

Validating viral hepatitis B and C diagnosis codes: a retrospective analysis using Ontario's health administrative data.

We aimed to determine the criterion validity of using diagnosis codes for hepatitis B virus (HBV) and hepatitis C virus (HCV) to identify infections. Using linked laboratory and administrative data in Ontario, Canada, from January 2004 to December 2014, we validated HBV/HCV diagnosis codes against laboratory-confirmed infections. Performance measures (sensitivity, specificity, and positive predictive value) were estimated via cross-validated logistic regression and we explored variations by varying time windows from 1 to 5years before (i.e., prognostic prediction) and after (i.e., diagnostic prediction) the date of laboratory confirmation. Subgroup analyses were performed among immigrants, males, baby boomers, and females to examine the robustness of these measures. A total of 1,599,023 individuals were tested for HBV and 840,924 for HCV, with a resulting 41,714 (2.7%) and 58,563 (7.0%) infections identified, respectively. HBV/HCV diagnosis codes ± 3years of laboratory confirmation showed high specificity (99.9% HBV; 99.8% HCV), moderate positive predictive value (70.3% HBV; 85.8% HCV), and low sensitivity (12.8% HBV; 30.8% HCV). Varying the time window resulted in limited changes to performance measures. Diagnostic models consistently outperformed prognostic models. No major differences were observed among subgroups. HBV/HCV codes should not be the only source used for monitoring the population burden of these infections, due to low sensitivity and moderate positive predictive values. These results underscore the importance of ongoing laboratory and reportable disease surveillance systems for monitoring viral hepatitis in Ontario.

Cognitive profiles of patients with mild cognitive impairment due to Alzheimer's versus Parkinson's disease defined using a base rate approach: Implications for neuropsychological assessments.

IntroductionLarge studies on cognitive profiles of patients with mild cognitive impairment (MCI) due to Alzheimer's disease (AD‐MCI) compared to Parkinson's disease (PD‐MCI) are rare.MethodsData from two multicenter cohort studies in AD and PD were merged using a unified base rate approach for the MCI diagnosis. Cognitive profiles were compared using scores derived from the Consortium to Establish a Registry for Alzheimer's Disease battery.ResultsPatients with AD‐MCI showed lower standardized scores on all memory test scores and a language test. Patients with PD‐MCI showed lower standardized scores in a set‐shifting measure as an executive task. A cross‐validated logistic regression with test scores as predictors was able to classify 72% of patients correctly to AD‐MCI versus PD‐MCI.DiscussionThe applied test battery successfully discriminated between AD‐MCI and PD‐MCI. Neuropsychological test batteries in clinical practice should always include a broad spectrum of cognitive domains to capture any cognitive changes.