ROC AUC Research Articles

Integrating Deep Learning with Electronic Health Records for Early Glaucoma Detection: A Multi-Dimensional Machine Learning Approach.

Recent advancements in deep learning have significantly impacted ophthalmology, especially in glaucoma, a leading cause of irreversible blindness worldwide. In this study, we developed a reliable predictive model for glaucoma detection using deep learning models based on clinical data, social and behavior risk factor, and demographic data from 1652 participants, split evenly between 826 control subjects and 826 glaucoma patients. We extracted structural data from control and glaucoma patients' electronic health records (EHR). Three distinct machine learning classifiers, the Random Forest and Gradient Boosting algorithms, as well as the Sequential model from the Keras library of TensorFlow, were employed to conduct predictive analyses across our dataset. Key performance metrics such as accuracy, F1 score, precision, recall, and the area under the receiver operating characteristics curve (AUC) were computed to both train and optimize these models. The Random Forest model achieved an accuracy of 67.5%, with a ROC AUC of 0.67, outperforming the Gradient Boosting and Sequential models, which registered accuracies of 66.3% and 64.5%, respectively. Our results highlighted key predictive factors such as intraocular pressure, family history, and body mass index, substantiating their roles in glaucoma risk assessment. This study demonstrates the potential of utilizing readily available clinical, lifestyle, and demographic data from EHRs for glaucoma detection through deep learning models. While our model, using EHR data alone, has a lower accuracy compared to those incorporating imaging data, it still offers a promising avenue for early glaucoma risk assessment in primary care settings. The observed disparities in model performance and feature significance show the importance of tailoring detection strategies to individual patient characteristics, potentially leading to more effective and personalized glaucoma screening and intervention.

Childhood severe asthma: relationship among asthma control scores, FeNO, spirometry and impulse oscillometry

IntroductionThe evaluation of the asthmatic patient is usually based on clinical and functional parameters that do not necessarily evidence the degree of airway inflammation. The aim of this study was to analyze whether clinical scores (CS) correlate with spirometry (S), impulse oscillometry (IO) and FeNO, in severe asthmatic children.Material and methodsA multicentric, prospective, cross-sectional study was conducted over a 12-month period. All SA patients (6–18 years old) followed-up in the Pulmonology Department were recruited. CS, FeNO measurements, IO and S were consecutively performed on the same day. Asthma control was ascertained using ACT and GINAq. A cut-off value of ≥ 25 parts per billion (ppb) was used to define airway inflammation.ResultsEighty-one patients were included. ACT: 75% (n 61) were controlled; GINAq: 44.5% (n 36) were controlled; 39.5% (n 32) were partly controlled, and 16% (n 13) were uncontrolled. FeNO had a median value of 24 ppb (IQR 14–41); FeNO ≥ 25 ppb was observed in 49% of patients (n 39). ROC AUC for FeNO vs. ACT was 0.71 (95%CI 0.57–0.86), PPV 0.47, NPV 0.87, SE 0.61, SP 0.80; FeNO vs. GINAq was ROC AUC 0.69 (95%CI 0.54–0.85), PPV 0.34, NPV 0.91, SE 0.62, SP 0.77; Youden cut-off FeNO > 39 ppb for both CS.ConclusionIn severe asthmatic children, current symptoms control as evidenced by ACT and GINA correlates with low FeNO values. Clinical scores showed good correlation with airway inflammation.

Post-processing of quantitative 4D-CT for initial evaluation of scapholunate Instability: Assessment of simplified approaches to data analysis

ObjectivesTo evaluate the diagnostic performance of simplified post-processing approaches for quantitative wrist 4D-CT in the assessment of scapholunate instability (SLI). MethodsA prospective monocentric case-control study included 60 patients with suspected post-traumatic scapholunate ligament (SLL) tears and persistent pain. Of these, 40 patients exhibited SLL tears, subdivided into two groups of 20 each: one group with completely torn ligaments and the other with partially torn ligaments. The remaining 20 patients, whose SLLs were intact, served as controls. 4D-CT and CT arthrography were performed, and post-processed by two readers using three approaches: the standard method with full data assessment and dedicated software, partial data assessment with post-processing software (bone locking), and partial data assessment without post-processing software (no bone locking). The scapholunate gap (SLG) parameter was measured in millimeters to evaluate scapholunate diastasis during radioulnar deviation (RUD). The scapholunate ligament status on CT arthrography was considered the gold standard. ResultsThe SLG-derived parameters (range, mean, and maximal values) were significantly increased in patients with both intact and torn scapholunate ligaments across all post-processing approaches (P values ranging from 0.001 to 0.004). SLG range was the best parameter for diagnosing SLL tears, with ROC AUC values ranging from 0.7 to 0.88 across the three post-processing methods. The interobserver reproducibility was better with the alternative approaches (ICC values 0.93–0.96) compared to the standard approach (ICC values 0.65–0.72). Additionally, post-processing time was shorter with the alternative approaches, especially when specific software was not used (reduced from 10 to three minutes). ConclusionSimpler approaches to wrist 4D-CT data analysis yielded acceptable diagnostic performances and improved interobserver reproducibility compared to the standard approach.

Characterization of the intestinal microbiota in MSM with HIV infection

BackgroundHIV-infected persons demonstrate notable disturbances in their intestinal microbiota; however, the impact of intestinal microbiota on HIV susceptibility in men who have sex with men (MSM), as well as the effects of HIV and antiretroviral therapy (ART) on their gut microbiota, remains under active study. Thus, our research focuses on clarifying the distinctions in intestinal microbiota composition among uninfected MSM and non-MSM healthy controls, investigating the alterations in early-stage intestinal microbial communities following HIV infection, and assessing how ART affects the intestinal microbiota.MethodsThis study enrolled four participant groups: uninfected MSM, Recent HIV-1 infection (RHI) MSM, MSM on ART, and non-MSM healthy controls, with 30 individuals in each group. We utilized 16S ribosomal DNA (16S rDNA) amplicon sequencing to analyze fecal microbiota and employed Luminex multiplex assays to measure plasma markers for microbial translocation (LBP, sCD14) and the inflammatory marker CRP.FindingsComparing uninfected MSM to non-MSM healthy controls, no substantial variances were observed in α and β diversity. Uninfected MSM had higher average relative abundances of Bacteroidetes, Prevotella, and Alloprevotella, while Bacteroides, Firmicutes, and Faecalibacterium had lower average relative abundances. MSM on ART had lower intestinal microbiota diversity than RHI MSM and uninfected MSM. In MSM on ART, Megasphaera and Fusobacterium increased, while Faecalibacterium and Roseburia decreased at genus level. Additionally, treatment with a non-nucleoside reverse transcriptase inhibitor (NNRTI) led to significant alterations in intestinal microbiota diversity and composition compared to RHI MSM. The random forest model showed that HIV infection biomarkers effectively distinguished between newly diagnosed HIV-infected MSM and HIV-negative MSM, with an ROC AUC of 76.24% (95% CI: 61.17-91.31%).ConclusionsMSM showed early intestinal microbiota imbalances after new HIV infection. MSM on ART experienced worsened dysbiosis, indicating a combined effect of HIV and ART. NNRTI-based treatment notably changed intestinal microbiota, suggesting a potential direct impact of NNRTI drugs on intestinal microbiota.

Predicting problem-solving success in an office simulation applying N-grams and a random forest to behavioral process data

Predicting students' problem-solving success in computer-based simulations at an early stage allows adaptive educational systems to provide learners with personalized support. In this paper, we predict students' problem-solving success by applying a machine-learning model, the random forest, to produce a binary classification (more vs. less successful students). During a business-related problem scenario that lasted 55 min, early behavioral data (during the first 5, 10, and 20 min) such as mouse clicks and keyboard strokes (approx. 29,800 early-window clickstreams and keystrokes during the first 20 min) of 234 trainees were recorded, mirroring the students' problem-solving behavior. We used the n-gram sequence mining technique, which was originally introduced within the emerging disciplines of natural language processing, text mining, and machine learning and has proven to be effective, particularly in the examination of online behavior. We trained the random forest model with training datasets that included all features (bigrams), as well as selected features (the most predictable bigrams explaining inter-group differences). Our results show that early predictions based on the first 10 and 20 min contained sufficient information to accurately predict problem-solving success, while predictions that are too early (based on the first 5 min) do not. As the size of the initial time window expanded, the classification performance improved. Moreover, the selection of the most predictable features improved the models' performance for all three time intervals. The model that was trained with only selected robust features that occurred in the first 20 min achieved the highest ROC AUC score of almost 0.70. This result falls within the range of accuracy scores observed in similar studies. From the instructor's perspective, predictions help in the early identification of weak students and can provide them with personalized learning prompts. For more successful students, tasks can be enriched adaptively.

A blind retrospective analysis of a novel predictive marker to ICB response in NSCLC, calculated directly from histopathological slides.

2665 Background: Immune checkpoint blockers (ICB), and primarily PD-1/PD-L1 inhibitors, are in the forefront of contemporary clinical oncology and have become an integral part of treatment of many malignancies, including non-small cell lung cancer (NSCLC). Nevertheless, tumor response to ICB varies widely. Predictive markers commonly used to distinguish patients likely to respond to ICB, such as PD-L1 expression and tumor mutational burden (TMB) have limited predictive value, which calls for the development of practical and more accurate tests. We present results of a blind retrospective analysis of a novel predictive marker of ICB response in NSCLC, relying solely on histopathological slides. Methods: We obtained high resolution Hematoxylin and Eosin (H&E) slides from tumor-tissue samples of 50 cases of metastatic NSCLC patients treated with first-line PD-1 inhibitors. We retrospectively applied our ENLIGHT-DeepPT (ENLIGHT-DP for short) pipeline to generate, in a blinded manner, an individual response score to PD-1 inhibition for each slide. ENLIGHT-DP is composed of two main steps: (i) predict mRNA expression directly from an H&E slide using DeepPT, our digital-pathology based algorithm; and (ii) use these values as input to ENLIGHT, our transcriptome-based precision oncology platform, which generates a score that predicts response to targeted therapies and ICB (based on a 10-gene signature in this case). We then unblinded the clinical outcome (RECIST1.1), and evaluated ENLIGHT-DP’s performance vs. standard markers. Results: ENLIGHT-DP’s score is predictive of response in this cohort, which had an overall response rate of 68% (34 of 50), with ROC AUC = 0.69 (p = 0.01, one-sided permutation test). Using a predefined threshold for binary classification of response derived from independent data, all 15 patients that were predicted to respond by ENLIGHT-DP indeed responded (100% PPV, 44% sensitivity). In comparison, predicting response according to PD-L1 > 1% achieves 68% PPV and 62% sensitivity, while PD-L1 > 50% achieves 65% PPV and 38% sensitivity, i.e, both thresholds exhibit no predictive power (PPV <= baseline response rate). Patients with high TMB (>10) had 82% PPV and 26% sensitivity, showing lower predictive benefit than ENLIGHT-DP. ENLIGHT-DP was particularly good at stratifying patients with PD-L1 < 1% (18 patients, ROC AUC = 0.8, p = 0.03). Conclusions: ENLIGHT-DP demonstrates high predictive power for response to ICB in NSCLC relying solely on accessible H&E slides, outperforming the commonly used PD-L1 and TMB markers. ENLIGHT-DP is also able to identify responders within patients with PD-L1 < 1%, for whom ICB is usually considered ineffective. Importantly, our approach does not require training on prior treatment outcomes, and can therefore be generalized to drugs for which such data is unavailable or scarce.

Usefulness of extended inflammatory parameters related to neutrophil activation reported by Sysmex XN-1000 hematology analyzer for predicting complicated acute appendicitis. Comparison with canonical inflammatory laboratory tests

AimAccurate diagnosis of complicated appendicitis is of importance to ensure that patients receive early and effective treatment, minimizing the risk of postoperative complications to promote successful recovery. Biochemical markers are a promising tool to identify complicated appendicitis. We aimed to evaluate the potential role of novel parameters related with neutrophil activation, known as “Extended Inflammation Parameters” (EIP), included in blood cell count reported by Sysmex XN-Series analyzers, compared to other canonical biomarkers in identifying complicated appendicitis. MethodProspective observational study including patients with confirmed diagnosis of acute appendicitis. C-reactive protein (CRP), procalcitonin, cell blood count, including white blood cell (WBC), absolute neutrophil (ANC) and immature granulocyte (IG) count and EIP (neutrophil reactivity [NEUT-RI] and granularity intensity [NEUT-GI]) were analyzed before surgery. Their accuracy to diagnose complicated appendicitis was tested in an ROC curve analysis. ResultsOur population study included 119 patients, and appendicitis was complicated in 58 (48.7%). NLR, CRP and procalcitonin levels, ANC and IG count and NEUT-RI and NEUT-GI were higher in patients with complicated appendicitis. Regarding accuracy for complicated appendicitis, CRP was the biomarker with the highest performance (ROC AUC: 0.829), with an optimal cutoff of 73.1 mg/L (sensitivity: 63.8%, specificity: 88.5%). NEUT-RI and NEUT-GI achieved both significant but poor accuracy, with ROC AUC of 0.606 and 0.637, respectively. ConclusionsNovel laboratory tests reported by Sysmex XN-Series analyzers have poor accuracy for identifying complicated appendicitis. In this study, CRP was the biomarker with the highest performance and may be useful as predictor of the severity of acute appendicitis.

Comparison of the ability of short time low PEEP challenge and mini fluid challenge to predict fluid responsiveness in patients undergoing open pancreaticoduodenectomy: an observational cohort study.

The mini-fluid challenge (MFC), which assesses the change in stroke volume index (SVI) following the administration of 100 mL of crystalloids, and the short-time low positive end-expiratory pressure (PEEP) challenge (SLPC), which evaluates the temporary reduction in SVI due to a PEEP increment, are two functional hemodynamic tests used to predict fluid responsiveness in the operating room. However, SLPC has not been assessed in patients undergoing abdominal surgery, and there is no study comparing these two methods during laparotomy. Therefore, we aimed to compare the SLPC and MFC in patients undergoing open pancreaticoduodenectomy. All patients received a standard hemodynamic management. The study protocol evaluated the percentage change in SVI following the application of an additional 5 cmH2O PEEP (SVIΔ%-SLPC) and the infusion of 100 mL crystalloid (SVIΔ%-MFC). Challenges that resulted in an increase of more than 15% in SVI after the 500 ml of fluid loading were classified as positive challenges (PC). Areas under the receiver operating characteristics curves (ROC AUCs) were used for the comparison of the methods. Thirty-three patients completed the study with 94 challenges. Fifty-five (58.5%) of them were PCs. The ROC AUC of SVIΔ%-MFC was observed to be significantly higher than that of SVIΔ%-SLPC (0.97 vs. 0.64, p < 0.001). The best cut-off value for SVIΔ%-MFC was 5.6%. If we had stopped the bolus fluid administration when SVIΔ%-MFC ≤ 5% was observed (lower limit of the gray zone), we would have postponed the fluid loading in 35 (89.7%) of 39 negative challenges. The amount of fluid deferred would have corresponded to up to 40% of the total fluid given. SVIΔ%-MFC predicts fluid responsiveness with high diagnostic performance and is better than SVIΔ%-SLPC in patients undergoing open pancreatoduodenectomy. Additionally, the use of SVIΔ%-MFC has the potential to defer up to 40% of the total fluid given. gov: NCT05419570.

Clinical large language model to predict loss to follow up for oncology patients discharged to skilled nursing facilities.

e13602 Background: The prognosis of cancer patients is heavily influenced by performance status which also informs an oncologist’s decision for initiating or continuing with cancer treatment. From SEER data, there is evidence that oncology patients who are discharged to a skilled nursing facility are less likely to receive chemotherapy and radiation. Patients with advanced cancer discharged from hospital also have a readmission rate as high as 34 percent. Those that are unlikely to follow up are a vulnerable population that should be identified to discuss how to best optimize post hospitalization care. Methods: University of Florida's Integrated Data Repository (IDR) was queried for established oncology patients at University of Florida Health from 2012 – 2023. The cohort was further narrowed down to patients who had exposures to antineoplastics within 6 months and loss to follow up was defined as not having a clinic visit within 6 months of hospital discharge. The study cohort included 194 patients and 97 patients were not lost to follow up and 97 patients were lost to follow up. Structured health data including ICD-9 and 10 codes, antineoplastic drug exposures, Area Deprivation Index (ADI), discharge vitals were extracted. Dataset was split into 70% training and 30% testing. Using logistic regression, random forest and XGBoost, loss to follow up was predicted with 5 fold cross validation and hyperparameter tuning with GridSearchCV for the best ROC AUC. Results: The median Charlson score for cohort was 11, median length of stay in the hospital was 7 days, median ED visits within 6 months was 8, median LACE score of 21. 86% of the cohort had Medicare, 6% with Medicaid, 6% Blue Cross, 2% Managed Care and 0.5% Other. There were 96 females and 98 males. 66% of the cohort was Caucasian and 31% Black, 3.9% Hispanic, 1% Asian, 0.5% multiracial. XGBoost and random forest performed better than logistic regression, both achieving a AUC ROC of 0.83 vs. 0.78. XGBoost had a higher F1 score of 0.75, sensitivity of 0.71, compared to 0.68 and 0.55 respectively for both Random Forest and logistic regression. The most important features for Random Forest were ED visits within 6 months, followed by diastolic blood pressure, heart rate, temperature then systolic blood pressure on day of discharge. The most important features for XGBoost were diagnoses of pulmonary congestion, pulmonary embolism, pleural effusion and type 2 diabetes with complications. Conclusions: Our XGBoost model achieved metrics which are in line with the higher performances seen with machine learning models predicting readmissions following hospital discharge. This serves as a good baseline model to compare against new state of the art models that incorporate unstructured data such as clinical notes and radiology image reports. These models are currently under development by the same authors using the clinical large language model GatorTron.

Prediction of therapeutic response and cancer outcomes in solid tumours via in silico clinical trials.

e13603 Background: Oncology therapeutic development has high failure rates with substantial costs. Advances in technology, including molecular characterisations of cancer and computational power, provide the opportunity to better model therapeutic response for existing and emerging therapies. Methods: Concr adapted Bayesian statistical principles used by astrophysicists to effectively integrate multi-dimensional data (genomic &amp; transcriptomic) for accurate modelling of cancer biology. Our proprietary computational framework was applied to predict clinical response and survival through construction of digital twins and in silico simulations of clinical trials. Predicted log odds ratios (LOR) were generated for drug response and compared to published data from 8 suitable historical clinical trials. Three distinct models underpin this framework: Drug Efficacy, Treatment Response &amp; Overall Survival (OS). Three publicly available datasets were used for training: Cancer Therapeutic Response Portal ( in vitro dose-response data), Cancer Cell Line Encyclopaedia and TCGA. Results: For all 8 clinical studies, the digital twin model accurately simulated both trial arms, compared drug efficacy across arms and predicted which treatment was most active. Unblinded evaluation: Our model accurately predicted gemcitabine would have greater clinical benefit than 5-FU (LOR -0.10, P &lt;0.0001) in metastatic pancreatic cancer. In advanced breast cancer, the model predicted docetaxel had a higher response rate than doxorubicin, and carboplatin response rates were similar to docetaxel in BRCA wild-type patients. In platinum-sensitive recurrent ovarian cancer, the model predicted cisplatin-based combination had a higher response rate than paclitaxel. Blinded evaluation: Using data for paclitaxel, the model correctly predicted that nab-paclitaxel+gemcitabine response rates were higher than gemcitabine in metastatic pancreatic cancer (predicted LOR -0.090, p = &lt;0.001). In all 3 early breast cancer studies, predictions mirrored clinical trial results, adjuvant ECF &gt; CMF, adjuvant TC &gt; AC and adjuvant tamoxifen was superior to capecitabine. The underlying pan-tumour Overall Survival model used a Random Survival Forest approach, trained on 23 cancers with predictive accuracy of 0.78 AUC ROC and C-index 0.71. To enable cohort enrichment for drug response, the model segmented patient cohorts into treatment responder and non-responder. Using this approach, the log odds increased for drug response rate for 16 of the 19 different cancers and 14 of the different 17 drugs. Conclusions: Our digital twin model represents transformative technology for clinical trial simulation with strong potential for accelerating and de-risking clinical drug development, and utility for synthetic control arms.

Predictive value of baseline 68Ga-FAPI PET/CT in the first-line therapy of pancreatic ductal adenocarcinoma (PDAC).

e16300 Background: This study aimed to explore whether parameters from the baseline 68Ga-fibroblast activating protein (FAP) inhibitor (FAPI) PET/CT could predict outcomes in PDAC patients receiving first-line treatment based on a Phase II clinical trial (previously reported in 2023 ASCO, abstr 4142 and 2024 ASCO-GI, abstr 671). Methods: Patients underwent 68Ga-FAPI PET/CT scan before the first dose of first-line treatment of NASCA (nab-paclitaxel, S-1, surufatinib, and camrelizumab) or GnP (nab-paclitaxel and gemcitabine) regimen were enrolled. The parameters of 68Ga-FAPI-04 PET/CT, including SUVmax, SUVmean, target-to-background ratio (TBR) (background including liver, muscle, and blood pool), and FAP expression tumor volume (FTV) of each tumor lesion, were recorded. The highest SUVmax, SUVmean and TBR (H-SUVmax, H-SUVmean, H-TBR), and summed FTV in each region (primary tumor, liver, peritoneal, lymph nodes, lung and bone) and whole body (WB) were also evaluated. Results: From January 2022 to July 2023, a total of 28 out of 42 patients were enrolled. Nineteen patients had received the NASCA regimen, and 9 patients had received GnP regimen. The SUVmax of primary site and metastatic sites were 16.46 (7.56-28.03) and 7.83 (2.13-19.22), respectively. The SUVmean of primary site and metastatic sites were 9.24 (4.05-14.99) and 4.63 (1.44-11.25), respectively. Both the SUVmax and SUVmean of primary site were significantly higher than that of metastatic sites (P &lt; 0.001). 68Ga-FAPI PET/CT showed higher sensitivity in detecting metastases than CT/MRI, especially peritoneal metastases. Only 30% (3/10) of patients were diagnosed with peritoneal metastases by CT/MRI. The ROC curve and AUC for the H-SUVmax, H-SUVmean, and summed FTV in WB or each region were evaluated by separating patients into responders and non-responders. H-SUVmax-WB had the highest AUC for predicting treatment response with a cut-off value of 18.2. Patients with H-SUVmax-WB below 18.2 had significantly longer progeression-free survival (PFS) than those above [8.8 months vs 3.8 months, P = 0.026, HR = 0.30, 95%CI (0.10-0.87)]. However, H-SUVmax-WB was not a prognostic factor (P = 0.292). Among H-TBRblood, TBRmuscle and TBRliver, H-TBRblood-WB (25.9 as the cut-off) was the most pronounced predictive (P = 0.001) and prognostic factor (P = 0.039). The trend of differences in PFS continue exists both in NASCA (p = 0.047) and GnP group (p = 0.018). The multivariable analysis revealed that H-TBRblood-WB (P = 0.026) and liver meatastases (P = 0.026) were independent predictive factors. Conclusions: The highest TBRblood of tumor lesions on 68Ga-FAPI PET/CT was associated with poor response in first-line treatment of PDAC. The utility of 68Ga FAPI PET/CT before treatment could reflect the tumor burden and may predict the response of therapy. For those with high uptake of FAPI, therapy targeting FAP might have a promising future.

Prediction of Interactomic HUB Genes in Periodontitis With Acute Myocardial Infarction.

Acute myocardial infarction (AMI) risk correlates with C-reactive protein (CRP) levels, suggesting systemic inflammation is present well before AMI. Studying different types of periodontal disease (PD), extremely common in individuals at risk for AMI, has been one important research topic. According to recent research, AMI and PD interact via the systemic production of certain proinflammatory and anti-inflammatory cytokines, small signal molecules, and enzymes that control the onset and development of both disorders' chronic inflammatory reactions. This study uses machine learning to identify the interactome hub biomarker genes in acute myocardial infarction and periodontitis. GSE208194 and GSE222883 were chosen for our research after a thorough search using keywords related to the study's goal from the gene expression omnibus (GEO) datasets. DEGs were identified from the GEOR tool, and the hub gene was identified using Cytoscape-cytohubba. Using expression values, Random Forest, Adaptive Boosting, and Naive Bayes, widgets-generated transcriptomics data, were labelled, and divided into 80/20 training and testing data with cross-validation. ROC curve, confusion matrix, and AUC were determined. In addition, Functional Enrichment Analysis of Differentially Expressed Gene analysis was performed. Random Forest, AdaBoost, and Naive Bayes models with 99%, 100%, and 75% AUC, respectively. Compared to RF, AdaBoost, and NB classification models, AdaBoost had the highest AUC. Categorization algorithms may be better predictors than important biomarkers. Machine learning model predicts hub and non-hub genes from genomic datasets with periodontitis and acute myocardial infarction.

Relapse and Survival in Bladder Cancer Patients Undergoing microRNA-129 and microRNA-145 Assays.

The lack of indicators to measure tumor's invasive biological behavior is an important issue. The aim of this study was to examine the effect of miRNAs 129 and 145 on tumor progression as well as patient survival. Seventy five breast cancer (BC) patients and 75 controls were included in this research. Two miRNA expressions were estimated using real-time PCR. Biomarkers for BC detection was tested using ROC curves and AUC. miR-129 and miR-145 expressions were significant. miR-129 and miR-145 classifiers (AUC = 0.943 and 0.748, respectively) help diagnose BC. Unlike miR-145, miR-129 did not affect the Kaplan-Meier survival curve analysis for progression-free survival at the end of the trial. The development of transitional cell carcinoma disease was found to have a strong correlation with miR-145 in both univariate and multivariate Cox regression analyses. Additionally, infiltrating + invasive urothelial carcinoma was also found to be correlated with miR-145. Conversely, elevated miR-129 expression in BC patients did not lead to an increase in cancer-specific recurrence or mortality, as observed in both univariate and multivariate Cox regression studies. The miRNA signature can help detect survival-associated miRNAs and develop BC miRNA therapeutics.

Systemic inflammation-modified PG-SGA index in the survival prediction of advanced cancers.

12056 Background: The tight association between inflammation and nutrition has been established. Whether the systemic inflammatory indicators could increase the predictive power of nutrition assessment tools in the survival of advanced cancers remains unclear. Methods: In this retrospective study, advanced cancer patients were enrolled from the West China Hospital between Nov 2019 and Sep 2023. Five systemic inflammatory indicators were selected including the systemic inflammatory index (SII), neutrophil to lymphocyte ratio (NLR), platelet to lymphocyte ratio (PLR), lymphocyte to monocyte ratio (LMR), and advanced lung cancer inflammation index (ALI). Nutritional status was evaluated by the PG-SGA. Kaplan-Meier curves, restricted cubic splines (RCS), and Cox regression analyses were used to estimate the effects of systemic inflammation indices as well as systemic inflammation-modified PG-SGA index on the survival of cancer patients. Stratified and sensitive analyses were conducted to determine the high-risk subgroups and check the robustness of the main findings. Results: In total, there were 1,065 advanced-stage cancer patients included in this study, with a mean age of 58.18 years old. Advanced lung cancer (265 cases, 24.9%) accounted for relatively higher proportions of the whole study population, followed by colorectal cancer (167 cases, 15.7%), esophagus cancer (143 cases, 13.4%), and gastric cancer (139 cases, 13.0%). With a median follow-up of 1,300 days, 501 deaths (47.04%) were observed. Multivariate Cox regression analysis showed that ALI maintained the best predictive power in predicting the all-cause mortality of advanced-stage cancer patients when compared with other systemic inflammation indicators (4th quartile: hazard ratio (HR)= 0.38, 95% confidence interval (CI): 0.29, 0.50, P&lt;0.001), with a C-index of 0.615. The novel inflammation-modified PG-SGA/ALI index showed better predictive value in identifying the high-risk subgroups of advanced-stage cancer patients, with an AUC of ROC of 0.68 in 365-day mortality, 0.65 in 730-day mortality, and 0.64 in 1300-day survival prediction, respectively. The C-index (0.626), calibration curves, and decision curve analysis validated the promising predictive accuracy of the PG-SGA/ALI index in predicting the all-cause mortality of advanced-stage cancer patients. No significant interaction was observed between subgroups. A series of sensitive analyses supported the consistent findings as we determined in the main regression. Conclusions: Our study highlights that the PG-SGA/ALI index has superior predictive value than original PG-SGA criteria in short- and long-term survival prediction of advanced cancers. The PG-SGA/ALI index would be a more accurate tool for nutritional evaluation and prognosis prediction among cancer patients during clinical practice.

Transcriptome-based response predictor to identify potential responders among patients with negative standard markers for response to immune checkpoint blockers.

2666 Background: Immune checkpoint blockers (ICB) are revolutionizing cancer treatment, and are being approved for an increasingly wide range of cancer types. The most common biomarkers currently in use to select patients for ICB are PD-L1 expression on tumor cells, as measured by IHC, and tumor mutational burden (TMB) and microsatellite instability (MSI), both measured by NGS. However, some patients that are negative for these markers still respond to ICB. This calls for complementary biomarkers to better identify responders to ICB, especially in patients that are negative to current biomarkers. Here, we focus on predicting response to anti-PD1 in patients with negative PD-L1, TMB or MSI. Methods: We employ ENLIGHT, our transcriptome-based precision oncology platform, which identifies and utilizes clinically relevant genetic interactions to predict a patient’s response to a wide range of targeted drugs, including ICB. ENLIGHT generates an individual ICB response score calculated from a 10-gene expression signature - the ENLIGHT Matching Score (EMS). Patients with EMS above a predetermined threshold are considered matched by ENLIGHT to anti-PD-1 treatments. We have previously shown, based on more than 1000 cases analyzed retrospectively, that this signature can identify responders to anti-PD-1 with high accuracy. Here, we use ENLIGHT to perform a retrospective analysis of 125 cases from three different datasets, who had low PD-L1 presentation (&lt;1%), low TMB (&lt; 10) or microsatellite stable tumors (MSS), and were treated with anti-PD-1, to specifically assess ENLIGHT’s performance in this biomarker-negative sub-population. Results: Patients who responded to anti-PD-1 treatments had significantly higher EMS in all three datasets. Correspondingly, ENLIGHT is highly predictive of response to anti-PD-1 in patients with negative ICB markers in the three datasets (ROC AUCs of 0.80, 0.84, 0.77, respectively, for low PDL1, low TMB and MSS). It is important to note that ENLIGHT was not trained on any of these datasets, and is applied as-is using previously published parameters. Overall, we find that patients who are ENLIGHT-Matched to anti-PD1 in this biomarker-negative cohort are almost 3 times more likely to respond than patients who are not (31% vs. 11%, p=5e-13). Conclusions: ENLIGHT is a powerful tool for predicting response to anti-PD-1 treatment in patients with negative standard biomarkers for ICB, a currently unmet need with considerable clinical importance.

Identifying player skill of dota 2 using machine learning pipeline

The esports industry is one of the prominent business sectors in the digital era, particularly, Multiplayer Online Battle Arena (MOBA) games which gain much attention from gamers and streaming audiences. Among such games, Defense of the Ancient 2 or Dota 2 is the record holder for the highest prize esports tournament. Therefore, various companies and investors start their esports teams to compete in the Dota 2 tournaments, the Internationals. To success in the competition, player recruitment is a crucial process as it usually takes considerable effort to find a skillful player. Watching the game replay to evaluate the player’s skill is one of the approaches. However, it can be too exhaustive, also some player’s ranking, which represent the player’s skill, are often not available. In this paper, we propose an effective machine learning pipeline to evaluate the player’s skill. Our designed pipeline includes data collection, feature engineering, and machine learning modeling. We show the data collection process using open-source API. An effective method for feature engineering is proposed. Features, e.g., end-game, or tactical decision related statistics, are incorporated along with the resource in the game distribution, harassment tactic, or spatiotemporal features, in order to provide effective models. Subsequently, we apply major machine learning models based on a single game data, i.e., logistic regression and random forest, to the processed data. The most effective model can achieve up to 0.7091 precision, 0.5850 recall, 0.6411 F1-score, and 0.8123 ROC AUC score.

PD-1+ T lymphocyte proportions and hospitalized exacerbation of COPD: a prospective cohort study

ObjectiveTo evaluate the predictive value of PD-1 expression in T lymphocytes for rehospitalization due to acute exacerbations of COPD (AECOPD) in discharged patients.Methods115 participants hospitalized with COPD (average age 71.8 ± 6.0 years) were recruited at Fujian Provincial Hospital. PD1+T lymphocytes proportions (PD1+T%), baseline demographics and clinical data were recorded at hospital discharge. AECOPD re-admission were collected at 1-year follow-up. Kaplan-Meier analysis compared the time to AECOPD readmissions among groups stratified by PD1+T%. Multivariable Cox proportional hazards regression and stratified analysis determined the correlation between PD1+T%, potential confounders, and AECOPD re-admission. ROC and DCA evaluated PD1+T% in enhancing the clinical predictive values of Cox models, BODE and CODEX.Results68 participants (59.1%) were AECOPD readmitted, those with AECOPD readmission exhibited significantly elevated baseline PD-1+CD4+T/CD4+T% and PD-1+CD8 + T/CD8 + T% compared to non-readmitted counterparts. PD1+ T lymphocyte levels statistically correlated with BODE and CODEX indices. Kaplan-Meier analysis demonstrated that those in Higher PD1+ T lymphocyte proportions had reduced time to AECOPD readmission (logRank p < 0.05). Cox analysis identified high PD1+CD4+T and PD1+CD8+T ratios as risk factors of AECOPD readmission, with hazard ratios of 1.384(95%CI [1.043–1.725]) and 1.401(95%CI [1.013–1.789]), respectively. Notably, in patients aged < 70 years and with fewer than twice AECOPD episodes in the previous year, high PD1+T lymphocyte counts significantly increased risk for AECOPD readmission(p < 0.05). The AECOPD readmission predictive model, incorporating PD1+T% exhibited superior discrimination to the Cox model, BODE index and CODEX index, AUC of ROC were 0.763(95%CI [0.633–0.893]) and 0.734(95%CI [0.570–0.899]) (DeLong’s test p < 0.05).The DCA illustrates that integrating PD1+T% into models significantly enhances the utility in aiding clinical decision-making.ConclusionEvaluation of PD1+ lymphocyte proportions offer a novel perspective for identifying high-risk COPD patients, potentially providing insights for COPD management.Trial registrationChinese Clinical Trial Registry (ChiCTR, URL: www.chictr.org.cn/), Registration number: ChiCTR2200055611 Date of Registration: 2022-01-14.

#909 Role of cystatin C in the development and progression of chronic heart failure in patients with diabetes

Abstract Background and Aims It has been known that cystatin C is a marker for early prediction of renal function. In recent years, studies have shown that cystatin C plays an important role in predicting new or worsening cardiovascular diseases. The aim of our study was to investigate the role of cystatin C in the development and progression of chronic heart failure (CHF) in patients with diabetes. Method A total of 499 patients with DT1 and DT2, aged 34 to 84 years (60 [60; 72]), were examined. All patients underwent standard clinical and laboratory examination, with an assessment of the levels of the natriuretic peptides (BNP and NTproBNP) in baseline plasma samples. Renal function was assessed based on the levels of serum creatinine and cystatin C. Patients were divided into quartiles according to cystatin C levels: Q1 &amp;lt;0.78 mg/l, Q2 0.78-0.92 mg/l, Q3 0.93-1.38 mg/l, Q4 &amp;gt;1.38 mg/l. Results Significant differences in the levels of BNP and NTproBNP were observed in patients with diabetes with cystine C levels ≥0.93 mg/l (32.20 [19.85; 69.35] pg/ml and 104.25 [57.70; 210.08] ng/l, respectively). There were significant correlations between cystatin C and the levels of BNP (r = 0.56, p &amp;lt; 0.05) and NTproBNP (r = 0.58, p &amp;lt; 0.05). Logistic regression analysis showed the prognostic value of cystatin C, and not creatinine, as a predictor of heart failure in patients with diabetes and GFR&amp;gt;60 ml/min/1.73 m2 (1 mg/l increase of cystatin C was associated with 7.085 times increase in CHF odds). ROC analysis revealed that the level of cystatin C ≥0.96 mg/l with a sensitivity of 84% and specificity of 75% predicted increased BNP and NTproBNP in patients with diabetes (ROC AUC = 0.844). Conclusion Cystatin C level could be used as a predictive indicator of CHF in patients with diabetes and GFR&amp;gt;60 ml/min/1.73 m2.

#959 Developing and validating a risk score model for prediction of AKI in non-ICU hospitalized patients

Abstract Background and aim : Acute kidney injury (AKI) is an increasingly common and potentially catastrophic complication in hospitalized patients. Several risk score models have been designed for intensive care unit (ICU) patients. Aim in this work was to establish a new risk prediction score for AKI patients in general wards. Methods All hospitalized ward patients from the first of July 2019 to 31st of December 2019 who developed AKI 24 hours after admission were included. To create a new prediction score model of AKI in hospitalized patients we used the data collected from 107 patients who developed AKI. We used our prospective cohort to develop and validate this prediction score model of AKI in patients admitted to general wards. Results Only 107 patients of 10,243 (1%) developed AKI 24 hours after admission to the general wards. Mortality rate was 26.2%. A score model of 15 points, based on a combination of clinical and laboratory data, was developed for prediction of AKI. We demonstrated that a cutoff value ≥ 4 out of 15 was a predictor of Acute Kidney Injury in non-ICU hospitalized patients. The area under the receiver operating characteristic (AUC ROC) value of the score model was 0.950, 95% CI (confidence interval) 0.922–0.978 and the P-value &amp;lt;0.001 with sensitivity of 94.39 and specificity of 81.43. Applying the score model in another group of patients (validation group n = 122) showed that the AUC ROC value in the validation group was 0.826. Conclusion We developed and validated a new risk score model with a cutoff value ≥ 4 out of 15 for prediction of Acute Kidney Injury in non-ICU hospitalized patients with a good sensitivity and specificity. It will help the early prediction of Acute Kidney Injury in non-ICU patients.

#971 Effects of arsenic trioxide on MMP9/IL-17A expression in lupus nephritis—results from bioinformatics analysis and in vitro studies

Abstract Background and Aims Previous research has indicated that ATO treatment at low doses may decrease the likelihood of flares in active systemic lupus erythematosus (SLE) patients, but the pharmacological mechanisms of such effect have not been studied. Method Machine learning and network pharmacology analysis were employed to explore the potential targets from differentially expressed genes (DEGs) of human SLE and LN peripheral blood mononuclear cells (PBMCs) microarray datasets and pharmacological databases. GO and KEGG analyses were performed, and the protein-protein interaction (PPI) network was constructed. The association of these characteristic genes and immune cells were further examined and validated. Results Three machine learning models (RF, SVM, and XGB) were utilized to identify five immunoregulatory genes in SLE among twelve intersection DEGs. Among these genes, MMP9 exhibited the highest ROC AUC (0.92) in predicting disease development. KEGG analysis revealed significant associations with the IL-17 signaling pathway (p = 1.67E-18). Furthermore, ssGSEA analysis indicated positive correlations between MMP9 and macrophages (correlation coefficient 0.88) as well as neutrophils (correlation coefficient 0.66). In vitro treatment of PBMCs from lupus nephritis (LN) patients with ATO resulted in downregulation of MMP9 and IL-17A, validating our bioinformatics findings. Conclusion ATO has the potential to mitigate the MMP/IL-17A pathways in PBMCs in LN patients, and hence maybe a promising and innovative therapeutic alternative.