Median AUC Research Articles

Exploring neural markers of dereification in meditation based on EEG and personalized models of electrophysiological brain states

With mounting evidence for the benefits of meditation, there has been a growing interest in measuring and quantifying meditative states. This study introduces the Inner Dereification Index (IDI), a class of personalized models designed to quantify the distance from non-meditative states such as mind wandering based on a single individual’s neural activity. In addition to demonstrating high classification accuracy (median AUC: 0.996) at distinguishing meditation from thinking states moment by moment, IDI can accurately stratify meditator cohorts by experience, and correctly identify the practices most effective at training the dereification aspect of meditation (decentering from immersion with thoughts and perceptions and recognizing them as mental constructs). These results suggest that IDI models may be a useful real-time proxy for dereification and meditation progress, requiring only 1 min of mind wandering data (and no meditation data) during model training. Thus, they show promise for applications such as real-time meditation feedback, progress tracking, personalization of practices, and potential therapeutic applications of neurofeedback-assisted generation of positive states of consciousness.

Understanding overfitting in random forest for probability estimation: a visualization and simulation study

BackgroundRandom forests have become popular for clinical risk prediction modeling. In a case study on predicting ovarian malignancy, we observed training AUCs close to 1. Although this suggests overfitting, performance was competitive on test data. We aimed to understand the behavior of random forests for probability estimation by (1) visualizing data space in three real-world case studies and (2) a simulation study.MethodsFor the case studies, multinomial risk estimates were visualized using heatmaps in a 2-dimensional subspace. The simulation study included 48 logistic data-generating mechanisms (DGM), varying the predictor distribution, the number of predictors, the correlation between predictors, the true AUC, and the strength of true predictors. For each DGM, 1000 training datasets of size 200 or 4000 with binary outcomes were simulated, and random forest models were trained with minimum node size 2 or 20 using the ranger R package, resulting in 192 scenarios in total. Model performance was evaluated on large test datasets (N = 100,000).ResultsThe visualizations suggested that the model learned “spikes of probability” around events in the training set. A cluster of events created a bigger peak or plateau (signal), isolated events local peaks (noise). In the simulation study, median training AUCs were between 0.97 and 1 unless there were 4 binary predictors or 16 binary predictors with a minimum node size of 20. The median discrimination loss, i.e., the difference between the median test AUC and the true AUC, was 0.025 (range 0.00 to 0.13). Median training AUCs had Spearman correlations of around 0.70 with discrimination loss. Median test AUCs were higher with higher events per variable, higher minimum node size, and binary predictors. Median training calibration slopes were always above 1 and were not correlated with median test slopes across scenarios (Spearman correlation − 0.11). Median test slopes were higher with higher true AUC, higher minimum node size, and higher sample size.ConclusionsRandom forests learn local probability peaks that often yield near perfect training AUCs without strongly affecting AUCs on test data. When the aim is probability estimation, the simulation results go against the common recommendation to use fully grown trees in random forest models.

Diagnostic performance of radiologists in distinguishing post-COVID-19 residual abnormalities from interstitial lung abnormalities

Abstract Objective Distinguishing post-COVID-19 residual abnormalities from interstitial lung abnormalities (ILA) on CT can be challenging if clinical information is limited. This study aimed to evaluate the diagnostic performance of radiologists in distinguishing post-COVID-19 residual abnormalities from ILA. Methods This multi-reader, multi-case study included 60 age- and sex-matched subjects with chest CT scans. There were 40 cases of ILA (20 fibrotic and 20 non-fibrotic) and 20 cases of post-COVID-19 residual abnormalities. Fifteen radiologists from multiple nations with varying levels of experience independently rated suspicion scores on a 5-point scale to distinguish post-COVID-19 residual abnormalities from fibrotic ILA or non-fibrotic ILA. Interobserver agreement was assessed using the weighted κ value, and the scores of individual readers were compared with the consensus of all readers. Receiver operating characteristic curve analysis was conducted to evaluate the diagnostic performance of suspicion scores for distinguishing post-COVID-19 residual abnormalities from ILA and for differentiating post-COVID-19 residual abnormalities from both fibrotic and non-fibrotic ILA. Results Radiologists’ diagnostic performance for distinguishing post-COVID-19 residual abnormalities from ILA was good (area under the receiver operating characteristic curve (AUC) range, 0.67–0.92; median AUC, 0.85) with moderate agreement (κ = 0.56). The diagnostic performance for distinguishing post-COVID-19 residual abnormalities from non-fibrotic ILA was lower than that from fibrotic ILA (median AUC = 0.89 vs. AUC = 0.80, p = 0.003). Conclusion Radiologists demonstrated good diagnostic performance and moderate agreement in distinguishing post-COVID-19 residual abnormalities from ILA, but careful attention is needed to avoid misdiagnosing them as non-fibrotic ILA. Key Points QuestionHow good are radiologists at differentiating interstitial lung abnormalities (ILA) from changes related to COVID-19 infection? FindingsRadiologists had a median AUC of 0.85 in distinguishing post-COVID-19 abnormalities from ILA with moderate agreement (κ = 0.56). Clinical relevanceRadiologists showed good diagnostic performance and moderate agreement in distinguishing post-COVID-19 residual abnormalities from ILA; nonetheless, caution is needed in distinguishing residual abnormalities from non-fibrotic ILA.

Transfer learning approach in pre-treatment CT images to predict therapeutic response in advanced malignant pleural mesothelioma.

Malignant pleural mesothelioma (MPM) is a poor-prognosis disease. Owing to the recent availability of new therapeutic options, there is a need to better assess prognosis. The initial clinical response could represent a useful parameter. We proposed a transfer learning approach to predict an initial treatment response starting from baseline CT scans of patients with advanced/unresectable MPM undergoing first-line systemic therapy. The therapeutic response has been assessed according to the mRECIST criteria by CT scan at baseline and after two to three treatment cycles. We used three slices of baseline CT scan as input to the pre-trained convolutional neural network as a radiomic feature extractor. We identified a feature subset through a double feature selection procedure to train a binary SVM classifier to discriminate responders (partial response) from non-responders (stable or disease progression). The performance of the prediction classifiers was evaluated with an 80:20 hold-out validation scheme. We have evaluated how the developed model was robust to variations in the slices selected by the radiologist. In our dataset, 25 patients showed an initial partial response, whereas 13 patients showed progressive or stable disease. On the independent test, the proposed model achieved a median AUC and accuracy of 86.67% and 87.50%, respectively. The proposed model has shown high performance even by varying the reference slices. Novel tools could help to improve the prognostic assessment of patients with MPM and to better identify subgroups of patients with different therapeutic responsiveness.

Combined single-cell RNA sequencing and mendelian randomization to identify biomarkers associated with necrotic apoptosis in intervertebral disc degeneration

BackgroundIntervertebral disc degeneration (IDD) is associated with back pain; back pain is a world-wide contributor to poor quality of life, while necroptosis has the characteristics of necroptosis and apoptosis, however, its role in IDD is still unclear. Therefore, the aim of this study was to identify biomarkers associated with necroptosis in IDD. PurposeTo explore biomarkers associated with necroptosis in IDD, reveal the pathogenesis of IDD, as well as provide new directions for the diagnosis and treatment of this disease. Study design/settingsRetrospective cohort study. Our study employs scRNA-seq coupled with MR analysis to investigate the causal relationship between necroptosis and IDD, laying a foundational groundwork for unveiling the intricate pathogenic mechanisms of this condition. MethodsData quality control and normalisation was executed in single-cell dataset, GSE205535. Then, different cell types were obtained by cell annotation through marker genes. Subsequently, chi-square test was employed to assess the distribution difference of different cell types between IDD and control to screen key cells. AUCell was applied to calculate necroptosis-related genes (NRGs) scores of all cell types, further key cells were divided into high and low NRGs groups according to the median AUC scores of different cell types. Afterwards, the differentially expressed genes (DEGs) within the two score groups were screened. Then, the genes that had causal relationship with IDD were selected as biomarkers by univariate and multivariate Mendelian randomization (MR) analysis. Finally, the expression of biomarkers in different cell types and pseudo-time analysis was analyzed separately. ResultsIn GSE205535, 16 different cell populations identified by UMAP cluster analysis were further annotated to 8 cell types using maker genes. Afterwards, 53 DEGs were screened between the high and low NRGs groups. In addition, 9 genes with causal relationship with IDD were obtained by univariate MR analysis, further multivariate MR analysis proved that NT5E and TMEM158 had a direct causal relationship with IDD, which were used as biomarkers in this study. This study not only found that the expression levels of NT5E and TMEM158 were higher in IDD group, but also found that fibrochondrocytes and inflammatory chondrocytes were the key cells of NT5E and TMEM158, respectively. In the end, the biomarkers had the same expression trend in the quasi-time series, and both of them from high to low and then increased. ConclusionNT5E and TMEM158, as biomarkers of necroptotic apoptotic IDD, were causally associated with IDD. clinical significanceThe understanding of chondrocytes as key cells provides new perspectives for deeper elucidation of the pathogenesis of IDD, improved diagnostic methods, and the development of more effective treatments. These findings are expected to provide a more accurate and personalised approach to clinical diagnosis and treatment, thereby improving the prognosis and quality of life of patients with IDD.

Pharmacokinetics of extended-release clarithromycin in patients with Mycobacterium ulcerans infection

Clarithromycin extended-release (CLA-ER) was used as companion drug to rifampicin (RIF) for Mycobacterium ulcerans infection in the intervention arm of a WHO drug trial. RIF enhances CYP3A4 metabolism, thereby reducing CLA serum concentrations, and RIF concentrations might be increased by CLA co-administration. We studied the pharmacokinetics of CLA-ER at a daily dose of 15 mg/kg combined with RIF at a dose of 10 mg/kg in a subset of trial participants, and compared these to previously obtained pharmacokinetic data. Serial dried blood spot samples were obtained over a period of ten hours, and analyzed by LC–MS/MS in 30 study participants—20 in the RIF-CLA study arm, and 10 in the RIF-streptomycin study arm. Median CLA Cmax was 0.4 mg/L—and median AUC 3.9 mg*h/L, following 15 mg/kg CLA-ER. Compared to standard CLA dosed at 7.5 mg/kg previously, CLA-ER resulted in a non-significant 58% decrease in Cmax and a non-significant 30% increase in AUC. CLA co-administration did not alter RIF Cmax or AUC. Treatment was successful in all study participants. No effect of CLA co-administration on RIF pharmacokinetics was observed. Based on our serum concentration studies, the benefits CLA-ER over CLA immediate release are unclear.

Supervised machine learning to predict smoking lapses from Ecological Momentary Assessments and sensor data: Implications for just-in-time adaptive intervention development.

Specific moments of lapse among smokers attempting to quit often lead to full relapse, which highlights a need for interventions that target lapses before they might occur, such as just-in-time adaptive interventions (JITAIs). To inform the decision points and tailoring variables of a lapse prevention JITAI, we trained and tested supervised machine learning algorithms that use Ecological Momentary Assessments (EMAs) and wearable sensor data of potential lapse triggers and lapse incidence. We aimed to identify a best-performing and feasible algorithm to take forwards in a JITAI. For 10 days, adult smokers attempting to quit were asked to complete 16 hourly EMAs/day assessing cravings, mood, activity, social context, physical context, and lapse incidence, and to wear a Fitbit Charge 4 during waking hours to passively collect data on steps and heart rate. A series of group-level supervised machine learning algorithms (e.g., Random Forest, XGBoost) were trained and tested, without and with the sensor data. Their ability to predict lapses for out-of-sample (i) observations and (ii) individuals were evaluated. Next, a series of individual-level and hybrid (i.e., group- and individual-level) algorithms were trained and tested. Participants (N = 38) responded to 6,124 EMAs (with 6.9% of responses reporting a lapse). Without sensor data, the best-performing group-level algorithm had an area under the receiver operating characteristic curve (AUC) of 0.899 (95% CI = 0.871-0.928). Its ability to classify lapses for out-of-sample individuals ranged from poor to excellent (AUCper person = 0.524-0.994; median AUC = 0.639). 15/38 participants had adequate data for individual-level algorithms to be constructed, with a median AUC of 0.855 (range: 0.451-1.000). Hybrid algorithms could be constructed for 25/38 participants, with a median AUC of 0.692 (range: 0.523 to 0.998). With sensor data, the best-performing group-level algorithm had an AUC of 0.952 (95% CI = 0.933-0.970). Its ability to classify lapses for out-of-sample individuals ranged from poor to excellent (AUCper person = 0.494-0.979; median AUC = 0.745). 11/30 participants had adequate data for individual-level algorithms to be constructed, with a median AUC of 0.983 (range: 0.549-1.000). Hybrid algorithms could be constructed for 20/30 participants, with a median AUC of 0.772 (range: 0.444 to 0.968). In conclusion, high-performing group-level lapse prediction algorithms without and with sensor data had variable performance when applied to out-of-sample individuals. Individual-level and hybrid algorithms could be constructed for a limited number of individuals but had improved performance, particularly when incorporating sensor data for participants with sufficient wear time. Feasibility constraints and the need to balance multiple success criteria in the JITAI development and implementation process are discussed.

Fixed dose daptomycin: An opportunity for pharmacokinetic/pharmacodynamic optimization in Staphylococcus aureus infections.

Daptomycin is a high-use intravenous antimicrobial agent affording the convenience of once-daily dosing. Prior studies suggest an opportunity to use a more operationally convenient fixed rather than weight-based dosing but this approach has not been studied prospectively. This study quantified the probability of toxicity and efficacy end points by prospectively testing a fixed dose regimen of daptomycin (750 mg) in obese and non-obese adults. At least, three daptomycin concentrations were measured at steady-state for each patient. A population pharmacokinetic model was constructed to evaluate concentration-time profiles and investigate covariates of daptomycin clearance. Simulations were performed to evaluate the probability of achieving efficacy (24-h area under the curve (AUC0-24) ≥ 666 mg∙h/L) and toxicity (minimum concentration (C min) ≥24.3 mg/L) targets for fixed (500-1000 mg) and weight-based (6-12 mg/kg) daptomycin doses. Thirty-one patients (16 females, 15 males) with median (interquartile range (IQR)) age of 50 (30, 62) years and weight of 74 (54, 156) kg were included in the final analysis. Fixed dose daptomycin (750 mg) resulted in similar exposure across weights with a median (IQR) AUC0-24 of 819 (499, 1501) mg∙h/L and 749 (606, 1265) mg∙h/L in patients weighing ≤74 kg and >74 kg, respectively. Overall, male sex and increased kidney function necessitate higher fixed and weight-based doses to achieve efficacy. Creatine phosphokinase elevation was observed in two patients (6.5%) and predicted to be lower with fixed versus weight-based regimens. Fixed daptomycin dosing adjusted for sex and kidney function is expected to improve the efficacy-to-toxicity ratio, transitions of care, and costs compared to weight-based doses. However, no empiric dosing approach is predicted to achieve ≥90% efficacy while minimizing the risk of toxicity, so therapeutic drug monitoring should be considered on a patient-specific basis.

Network excitability of stimulation-induced spectral responses helps localize the seizure onset zone

ObjectiveWhile evoked potentials elicited by single pulse electrical stimulation (SPES) may assist seizure onset zone (SOZ) localization during intracranial EEG (iEEG) monitoring, induced high frequency activity has also shown promising utility. We aimed to predict SOZ sites using induced cortico-cortical spectral responses (CCSRs) as an index of excitability within epileptogenic networks. MethodsSPES was conducted in 27 epilepsy patients undergoing iEEG monitoring and CCSRs were quantified by significant early (10–200 ms) increases in power from 10 to 250 Hz. Using response power as CCSR network connection strengths, graph centrality measures (metrics quantifying each site’s influence within the network) were used to predict whether sites were within the SOZ. ResultsAcross patients with successful surgical outcomes, greater CCSR centrality predicted SOZ sites and SOZ sites targeted for surgical treatment with median AUCs of 0.85 and 0.91, respectively. We found that the alignment between predicted and targeted SOZ sites predicted surgical outcome with an AUC of 0.79. ConclusionsThese findings indicate that network analysis of CCSRs can be used to identify increased excitability of SOZ sites and discriminate important surgical targets within the SOZ. SignificanceCCSRs may supplement traditional passive iEEG monitoring in seizure localization, potentially reducing the need for recording numerous seizures.

International Multi-Specialty Expert Physician Preoperative Identification of Extranodal Extension n Oropharyngeal Cancer Patients using Computed Tomography: Prospective Blinded Human Inter-Observer Performance Evaluation.

Extranodal extension (pENE) is a critical prognostic factor in oropharyngeal cancer (OPC) that drives therapeutic disposition. Determination of pENE from radiological imaging has been associated with high inter-observer variability. However, the impact of clinician specialty on human observer performance of imaging-detected extranodal extension (iENE) remains poorly understood. To characterize the impact of clinician specialty on the accuracy of pre-operative iENE in human papillomavirus-positive (HPV+) OPC using computed tomography (CT) images. This prospective observational human performance study analyzed pre-therapy CT images from 24 HPV+ OPC patients, with duplication of 6 scans (n=30) of which 21 were pathologically confirmed pENE. Thirty-four expert observers, including 11 radiologists, 12 surgeons, and 11 radiation oncologists, independently assessed these scans for iENE and reported human-detected radiologic criteria and observer confidence. The primary outcomes included accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and Brier score for each physician, compared to ground-truth pENE. The significance of radiographic signs for prediction of pENE were determined through logistic regression analysis. Fleiss' kappa measured interobserver agreement, and Hanley-MacNeil AUC discrimination testing. Median accuracy across all specialties was 0.57 (95%CI 0.39 to 0.73), with no specialty showing discriminate performance greater than random estimation (median AUC 0.64, 95%CI 0.44 to 0.83). Significant differences between radiologists and surgeons in Brier scores (0.33 vs. 0.26, p < 0.01), radiation oncologists and surgeons in sensitivity (0.48 vs. 0.69, p > 0.1), and radiation oncologists and radiologists/surgeons in specificity (0.89 vs. 0.56, p > 0.1). Indistinct capsular contour and nodal necrosis were significant predictors of correct pENE status among all specialties. Interobserver agreement was weak for all the radiographic criteria, regardless of specialty (κ<0.6). Multiobserver testing shows physician discrimination of HPV+OPC pENE on pre-operative CT remains non-different than blind guessing, with high interrater variability and low diagnostic accuracy, regardless of clinician specialty. While minor differences in diagnostic performance among specialties are noted, they do not significantly affect the overall poor agreement and discrimination rates observed. The findings underscore the need for further research into automated detection systems or enhanced imaging techniques to improve the accuracy and reliability of iENE assessments in clinical practice.

Long cell-free DNA fragments predict early-progression in patients with advanced or metastatic cancer treated with immune-checkpoint inhibition.

e14550 Background: Immune-checkpoint inhibition (ICI) has significantly advanced cancer treatment, but many patients experience early progression (EP). Establishing reliable and early predictive biomarkers for guiding clinical practice is essential. Analysis of genome-wide cfDNA fragmentation profiles (fragmentome) is a promising non-invasive method for assessing treatment response independently of a specific molecular target, cancer type, and treatment. SChISM (Size CfDNA Immunotherapies Signature Monitoring) is a clinical study that monitors plasmatic cfDNA concentration and quantitative characteristics of the fragmentome in advanced pan-cancer patients, treated with ICI (n = 139). The aim is to predict EP, defined as progression at the first imaging evaluation, using pre-treatment data. Methods: Using BIABooster analysis technology from ADELIS, we assessed the predictive performances of plasmatic fragmentome-derived metrics: concentration, size distribution of first, second peaks and specific size ranges (total p=11 variables). Classical statistical analyses were conducted including hypothesis testing and modeling for association with EP (logistic regression) or progression-free-survival (PFS, Cox proportional hazard regression) in both the univariate and multivariate settings. Predictive analysis of EP was also performed. The dataset was split between a training (n=99) and test (n=40) set, using stratified sampling. Optimal thresholds were determined on the training dataset through receiver-operator characteristics (ROC) curve analysis, and confidence intervals determined using bootstrap resampling. Classification metrics were assessed in both the training and testing set. The entire process was bootstrapped 100 times to assess the robustness of the results. Results: Four out of the eleven features were associated with EP or PFS. Quantity of long fragments over 1650 bp showed the best discriminatory power (median AUC = 0.7, 95% CI: 0.59-0.80) of EP. Long fragments were significantly different between the EP groups (t-test p-val &lt; 0.001) and PFS groups (log-rank p-val &lt; 0.001). Patients with higher quantity of fragments over 1650 bp were likely not to progress before the first evaluation in both the univariate (odd ratio OR = 0.37 (0.22-0.64), p-val &lt; 0.001) and multivariate (OR = 0.45 (0.25-0.81), p-val = 0.008) settings. Accuracy was 0.71 (0.67-0.73), positive predictive value was 0.59 (0.53-0.63) and negative predictive value was 0.78 (0.75-0.79). Conclusions: These findings highlight the association of high-molecular-weight fragments with the early progressors ICI-treated. Thus, the improvement of the knowledge in the long fragments’ biology could potentially help in preventing or discontinuing unnecessary treatments for patients who are unlikely to respond to ICI.

Discriminating subtypes in advanced breast cancer with ctDNA methylation profiling.

1013 Background: Immunohistochemical subtyping of advanced breast cancer is critical to direct appropriate therapy. Yet, biopsy of recurrent cancer may not be technically possible, single-site tumour biopsies sample only a single metastasis and subtype may change through therapy. We developed a non-invasive blood test to subtype tumours based on circulating tumor DNA (ctDNA) methylation profiling. Methods: We interrogated TCGA-BRCA (level 3) methylation data from 783 tumours to identify regions with aberrant CpG methylation that differed between breast cancers and normal tissues, and between different breast cancer subtypes. A capture based enzymatic ctDNA based methylation profile was developed targeting 1257 differentially methylated regions (TWIST Methyl custom panel, TWIST Bioscience, panel size 0.4Mb). Plasma samples from patients with breast cancer, and healthy donors, were sequenced with the capture assay. A machine learning (ML) classifier was built to detect the presence of breast cancer, and a second classifier was built to discriminate ER positive v ER negative, and HER2 positive v HER2 negative from ctDNA methylation profiles. A 10-fold stratified nested cross-validation was used to assess classifier characteristics. Results: Plasma samples from 427 people (126 health donor, 27 early breast cancer patients and 291 metastatic breast cancer patients), along with 133 tumor tissue samples, were sequenced and a ML classifier built to detect the presence of breast cancer. In cross-validation, a random forest classifier yielded a median AUC of 0.996 (SD 0.020), with sensitivity of 94.9% at 99% specificity. A total of 231 ER positive and 63 ER negative plasma samples from patients were used for IHC ER status differentiation. In cross-validation a LGBM classifier yielded a median AUC of 0.928 (SD 0.051). Median sensitivity was 76.7% at specificity 86-99%. A total of 11% (25/231) ER positive and 8% (5/63) ER negative samples were misclassified. A total of 41 HER2 positive and 253 HER2 negative plasma samples were used for IHC HER2 status differentiation. In cross-validation a LGBM classifier yielded a median AUC of 0.934 (SD 0.075). Median sensitivity was 77.5% at specificity 85-92%, A total of 7% (3/41) HER-2 positive and 13% (33/253) HER2 negative samples were misclassified. In both IHC classifiers, subtype misclassifications were likely due to multiple concurrent BC tumours of differing IHC subtypes, biopsy proven subtype transitions and low purity. Effects of tumor purity on IHC-based classifiers will also be presented. Conclusions: Non-invasivemethylation based ctDNA subtyping can detect the presence of breast cancer, and discriminate IHC subtypes, with high accuracy. Further assessment is warranted to assess whether methylation-based subtyping has the potential to report IHC subtypes when repeat biopsy is not possible, detect subtype transitions and direct advanced breast cancer treatment.

#1756 Development and internal validation of machine learning algorithms for mortality prediction model of people with DM and CKD

Abstract Background and Aims Diabetes Mellitus (DM) and Chronic Kidney Disease (CKD) are leading causes of morbidity and mortality, presenting challenges in patient management. The aim of this study is to develop a machine learning-based predictive model for mortality in DM and CKD patients, improving early intervention and treatment personalization. Method 3,637 participants with DM and CKD from the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2018 were included. All-cause mortality was ascertained by linkage to National Death Index records through 31 December 2019 in NHANES. The dataset included clinical profiles, demographic details, and laboratory results of patients with DM and CKD. We performed 100 random groupings, dividing the data into a 75% training set and a 25% validation set each time. We then visualized the results of 100 AUC measurements using a boxplot. The objective was to forecast patient survival over 1, 3, 5, and 10-year horizons using a variety of machine learning algorithms. The algorithms tested included Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), Gaussian Naive Bayes (GNB), Multinomial Naive Bayes (MNB), Bayesian Network Classifier (BNC), AdaBoost (Ada), Gradient Boosting (Gradient), and Extreme Gradient Boosting (XG). The performance of each algorithm was evaluated based on the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC), with the results visualized in boxplot format. Results Upon result shown as the boxplots, it becomes evident that the Random Forest (RF) algorithm consistently exhibits one of the highest AUCs on average across all prediction intervals, suggesting a strong and stable predictive capacity. In the 1-year survival prediction, the RF algorithm demonstrates a highest AUC, implying consistent performance. The 3 and 5-year predictions reveal a slight dip in AUC for most models, including RF, but it still remains one of the top-performing algorithms with fewer outliers than XG and Gradient Boosting. The mean AUC of RF is particularly noteworthy at the 10-year mark, where most algorithms struggle with prediction accuracy, as indicated by lower median AUCs and greater variability. Notably, the RF algorithm identified six major factors for 1-year survival prediction: systolic blood pressure, uric acid, blood urea nitrogen, lactate dehydrogenase, lymphocyte count and total cholesterol, as well as three major factors for 10-year survival prediction: age, lymphocyte and gamma-glutamyl transferase. Conclusion Overall, the study indicates that while machine learning can be a powerful tool for survival prediction in patients with DM and CKD, the choice of algorithm is crucial, with RF standing out for its consistent and reliable performance.

Person-specific and pooled prediction models for binge eating, alcohol use and binge drinking in bulimia nervosa and alcohol use disorder.

Machine learning could predict binge behavior and help develop treatments for bulimia nervosa (BN) and alcohol use disorder (AUD). Therefore, this study evaluates person-specific and pooled prediction models for binge eating (BE), alcohol use, and binge drinking (BD) in daily life, and identifies the most important predictors. A total of 120 patients (BN: 50; AUD: 51; BN/AUD: 19) participated in an experience sampling study, where over a period of 12 months they reported on their eating and drinking behaviors as well as on several other emotional, behavioral, and contextual factors in daily life. The study had a burst-measurement design, where assessments occurred eight times a day on Thursdays, Fridays, and Saturdays in seven bursts of three weeks. Afterwards, person-specific and pooled models were fit with elastic net regularized regression and evaluated with cross-validation. From these models, the variables with the 10% highest estimates were identified. The person-specific models had a median AUC of 0.61, 0.80, and 0.85 for BE, alcohol use, and BD respectively, while the pooled models had a median AUC of 0.70, 0.90, and 0.93. The most important predictors across the behaviors were craving and time of day. However, predictors concerning social context and affect differed among BE, alcohol use, and BD. Pooled models outperformed person-specific models and the models for alcohol use and BD outperformed those for BE. Future studies should explore how the performance of these models can be improved and how they can be used to deliver interventions in daily life.

Hyperspectral imaging with machine learning for in vivo skin carcinoma margin assessment: a preliminary study

Surgical excision is the most effective treatment of skin carcinomas (basal cell carcinoma or squamous cell carcinoma). Preoperative assessment of tumoral margins plays a decisive role for a successful result. The aim of this work was to evaluate the possibility that hyperspectral imaging could become a valuable tool in solving this problem. Hyperspectral images of 11 histologically diagnosed carcinomas (six basal cell carcinomas and five squamous cell carcinomas) were acquired prior clinical evaluation and surgical excision. The hyperspectral data were then analyzed using a newly developed method for delineating skin cancer tumor margins. This proposed method is based on a segmentation process of the hyperspectral images into regions with similar spectral and spatial features, followed by a machine learning-based data classification process resulting in the generation of classification maps illustrating tumor margins. The Spectral Angle Mapper classifier was used in the data classification process using approximately 37% of the segments as the training sample, the rest being used for testing. The receiver operating characteristic was used as the method for evaluating the performance of the proposed method and the area under the curve as a metric. The results revealed that the performance of the method was very good, with median AUC values of 0.8014 for SCCs, 0.8924 for BCCs, and 0.8930 for normal skin. With AUC values above 0.89 for all types of tissue, the method was considered to have performed very well. In conclusion, hyperspectral imaging can become an objective aid in the preoperative evaluation of carcinoma margins.

Applications and Performance of Machine Learning Algorithms in Emergency Medical Services: A Scoping Review.

The aim of this study was to summarize the literature on the applications of machine learning (ML) and their performance in Emergency Medical Services (EMS). Four relevant electronic databases were searched (from inception through January 2024) for all original studies that employed EMS-guided ML algorithms to enhance the clinical and operational performance of EMS. Two reviewers screened the retrieved studies and extracted relevant data from the included studies. The characteristics of included studies, employed ML algorithms, and their performance were quantitively described across primary domains and subdomains. This review included a total of 164 studies published from 2005 through 2024. Of those, 125 were clinical domain focused and 39 were operational. The characteristics of ML algorithms such as sample size, number and type of input features, and performance varied between and within domains and subdomains of applications. Clinical applications of ML algorithms involved triage or diagnosis classification (n = 62), treatment prediction (n = 12), or clinical outcome prediction (n = 50), mainly for out-of-hospital cardiac arrest/OHCA (n = 62), cardiovascular diseases/CVDs (n = 19), and trauma (n = 24). The performance of these ML algorithms varied, with a median area under the receiver operating characteristic curve (AUC) of 85.6%, accuracy of 88.1%, sensitivity of 86.05%, and specificity of 86.5%. Within the operational studies, the operational task of most ML algorithms was ambulance allocation (n = 21), followed by ambulance detection (n = 5), ambulance deployment (n = 5), route optimization (n = 5), and quality assurance (n = 3). The performance of all operational ML algorithms varied and had a median AUC of 96.1%, accuracy of 90.0%, sensitivity of 94.4%, and specificity of 87.7%. Generally, neural network and ensemble algorithms, to some degree, out-performed other ML algorithms. Triaging and managing different prehospital medical conditions and augmenting ambulance performance can be improved by ML algorithms. Future reports should focus on a specific clinical condition or operational task to improve the precision of the performance metrics of ML models.

Pharmacogenomic analysis in adrenocortical carcinoma reveals genetic features associated with mitotane sensitivity and potential therapeutics.

Adrenocortical carcinoma (ACC) is an aggressive endocrine malignancy with limited therapeutic options. Treating advanced ACC with mitotane, the cornerstone therapy, remains challenging, thus underscoring the significance to predict mitotane response prior to treatment and seek other effective therapeutic strategies. We aimed to determine the efficacy of mitotane via an in vitro assay using patient-derived ACC cells (PDCs), identify molecular biomarkers associated with mitotane response and preliminarily explore potential agents for ACC. In vitro mitotane sensitivity testing was performed in 17 PDCs and high-throughput screening against 40 compounds was conducted in 8 PDCs. Genetic features were evaluated in 9 samples using exomic and transcriptomic sequencing. PDCs exhibited variable sensitivity to mitotane treatment. The median cell viability inhibition rate was 48.4% (IQR: 39.3-59.3%) and -1.2% (IQR: -26.4-22.1%) in responders (n=8) and non-responders (n=9), respectively. Median IC50 and AUC were remarkably lower in responders (IC50: 53.4 µM vs 74.7 µM, P<0.0001; AUC: 158.0 vs 213.5, P<0.0001). Genomic analysis revealed CTNNB1 somatic alterations were only found in responders (3/5) while ZNRF3 alterations only in non-responders (3/4). Transcriptomic profiling found pathways associated with lipid metabolism were upregulated in responder tumors whilst CYP27A1 and ABCA1 expression were positively correlated to in vitro mitotane sensitivity. Furthermore, pharmacologic analysis identified that compounds including disulfiram, niclosamide and bortezomib exhibited efficacy against PDCs. ACC PDCs could be useful for testing drug response, drug repurposing and guiding personalized therapies. Our results suggested response to mitotane might be associated with the dependency on lipid metabolism. CYP27A1 and ABCA1 expression could be predictive markers for mitotane response, and disulfiram, niclosamide and bortezomib could be potential therapeutics, both warranting further investigation.

Identifying bias in models that detect vocal fold paralysis from audio recordings using explainable machine learning and clinician ratings.

Detecting voice disorders from voice recordings could allow for frequent, remote, and low-cost screening before costly clinical visits and a more invasive laryngoscopy examination. Our goals were to detect unilateral vocal fold paralysis (UVFP) from voice recordings using machine learning, to identify which acoustic variables were important for prediction to increase trust, and to determine model performance relative to clinician performance. Patients with confirmed UVFP through endoscopic examination (N = 77) and controls with normal voices matched for age and sex (N = 77) were included. Voice samples were elicited by reading the Rainbow Passage and sustaining phonation of the vowel "a". Four machine learning models of differing complexity were used. SHapley Additive exPlanations (SHAP) was used to identify important features. The highest median bootstrapped ROC AUC score was 0.87 and beat clinician's performance (range: 0.74-0.81) based on the recordings. Recording durations were different between UVFP recordings and controls due to how that data was originally processed when storing, which we can show can classify both groups. And counterintuitively, many UVFP recordings had higher intensity than controls, when UVFP patients tend to have weaker voices, revealing a dataset-specific bias which we mitigate in an additional analysis. We demonstrate that recording biases in audio duration and intensity created dataset-specific differences between patients and controls, which models used to improve classification. Furthermore, clinician's ratings provide further evidence that patients were over-projecting their voices and being recorded at a higher amplitude signal than controls. Interestingly, after matching audio duration and removing variables associated with intensity in order to mitigate the biases, the models were able to achieve a similar high performance. We provide a set of recommendations to avoid bias when building and evaluating machine learning models for screening in laryngology.

COVID-19 antibody titers after tixagevimab–cilgavimab injection in patients with hematologic diseases; a single-center, prospective study

Knowledge of the SARS-CoV-2 antibody titers induced by tixagevimab–cilgavimab in patients with hematologic diseases remains insufficient. Here, we performed a single-center, prospective study to reveal the changes in antibody titer after administration of tixagevimab–cilgavimab in 78 patients with hematologic diseases. The median peak titer was 155.4 U/mL, and the median AUC was 46556 days·U/mL. First, we compared several characteristics between patients with low titers (peak titer ≤ 155.4 U/mL) and high titers (peak titer > 155.4 U/mL). We extracted 6 factors (patient age, sex, ECOG-PS, serum albumin level, and cross-sectional area and computed tomographic number of the psoas major muscle) as candidates influencing the antibody titers. Multiple regression analysis revealed that antibody titer was closely associated with these 6 factors (contribution rate = 0.76, p = 0.02). Our data support the inability of tixagevimab–cilgavimab to induce sufficient antibody titers against SARS-CoV-2, especially in older, frailer, female patients.

Vancomycin AUC0-24 estimation using first-order pharmacokinetic methods in pediatric patients.

The optimal dosing and monitoring of vancomycin in pediatrics is still unknown but has evolved to emphasize area under the curve over 24 h (AUC0-24) over minimum concentration (Cmin) monitoring. Real-world data supporting the feasibility of two-concentration kinetics with first-order equations for the estimation of vancomycin AUC0-24 in pediatric patients are lacking. To describe the interplay of vancomycin dose, AUC0-24, and Cmin using first-order equations within four pediatric age groups. This is a single-center, retrospective cohort study analyzing pediatric patients (<18 years) receiving intravenous vancomycin between 2020 and 2022. Included patients received at least 24 h of intravenous vancomycin with two concentrations obtained within 96 h of therapy initiation. Patients with baseline renal dysfunction were excluded. Patients were divided into four age categories: neonates (≤28 days), infants (29 days to <1 year), children (1-12 years), and adolescents (13-17 years). First-order equations were utilized to estimate pharmacokinetic parameters and AUC0-24. Overall, 219 patients (median age of 6 years [IQR 1-12]) met inclusion criteria. The median vancomycin daily dose was 30 mg/kg in neonates, 70 mg/kg in infants and children, and 52 mg/kg in adolescents. Median Cmin and AUC0-24 values among all age groups were 8.68 mg/L and 505 mg * h/L, respectively. For AUC0-24 values outside of the therapeutic range (400-600 mg * h/L), more values were SUPRAtherapeutic (>600 mg * h/L) than SUBtherapeutic (<400 mg * h/L). The overall trend within our data showed suboptimal correlation between Cmin and AUC0-24. However, 71% of patients with Cmin values of 5-10 mg/L had an AUC0-24 within the therapeutic range of 400-600 mg * h/L, whereas 23 patients (92%) with a SUPRAtherapeutic AUC0-24 had a Cmin value ≥15 mg/L. Approximately 10% of patients experienced acute kidney injury. Our data describe the relationship between vancomycin dose, Cmin, and AUC0-24 in pediatric patients. We demonstrated the feasibility of using first-order equations to estimate AUC0-24, using two concentrations obtained at steady state to monitor efficacy and safety in pediatric patients receiving intravenous vancomycin. Our data showed suboptimal correlation between AUC0-24 and Cmin, which indicates that Cmin should not be used as a surrogate marker for a therapeutic AUC0-24 in pediatric patients. In alignment with the 2020 vancomycin consensus guidelines, we suggest utilizing AUC0-24 for efficacy and safety monitoring.