Predictive Logistic Regression Model Research Articles

Optimizing thyroid AUS nodules malignancy prediction: a comprehensive study of logistic regression and machine learning models.

The accurate diagnosis of thyroid nodules with indeterminate cytology, particularly in the atypia of undetermined significance (AUS) category, remains challenging. This study aims to predict the risk of malignancy in AUS nodules by comparing two machine learning (ML) and three conventional logistic regression (LR) models. A retrospective study on 356 AUS nodules in 342 individuals from 6728 patients who underwent thyroid surgery in 2021. All the clinical, ultrasonographic, and molecular data were collected and randomly separated into training and validation cohorts at a ratio of 7: 3. ML (random forest and XGBoost) and LR (lasso regression, best subset selection, and backward stepwise regression) models were constructed and evaluated using area under the curve (AUC), calibration, and clinical utility metrics. Approximately 90% (321/356) of the AUS nodules were malignant, predominantly papillary thyroid carcinoma with 68.6% BRAF V600E mutations. The final LR prediction model based on backward stepwise regression exhibited superior discrimination with AUC values of 0.83 (95% CI: 0.73-0.92) and 0.80 (95% CI: 0.67-0.94) in training and validation, respectively. Well calibration, and clinical utility were also confirmed. The ML models showed moderate performance. A nomogram was developed on the final LR model. The LR model developed using the backward stepwise regression, outperformed ML models in predicting malignancy in AUS thyroid nodules. The corresponding nomogram based on this model provides a valuable and practical tool for personalized risk assessment, potentially reducing unnecessary surgeries and enhancing clinical decision-making.

Logistic Regression Prediction Model for Illegal Wildlife Trade

The aim of this study is to build a model to assess the success rate of the strategy to combat illegal wildlife trade, and to predict the probability of effective implementation of the strategy by quantifying the indicators of successful implementation of the project and building a probabilistic generative model. First, this paper defines successful implementation of the project as a 30% reduction in illegal wildlife trade, and considers the increase in public environmental awareness and the strength of law enforcement agencies as indicators of successful implementation. Based on these indicators, this paper establishes a probability-based model that combines a large amount of data and expert advice to improve the prediction of the probability of success using a Bayesian approach. Further, this paper hypothesises that the probability of strategy success is determined by enforcement efforts, the level of public environmental awareness, strategy interventions and external factors, and develops a logistic regression model for prediction. Multiple simplified datasets were used to demonstrate the effectiveness of model fitting, and the statsmodels library in Python was utilised to calculate the probability of success. The results of the sensitivity analysis show that increased public environmental awareness has a significant impact on the success of the strategy, the enforcement effort should be in the appropriate interval, and project interventions and external factors also have an impact on the probability of success. By analysing the sensitivity intervals and saturation points, decision-making guidance is proposed to rationally allocate resources to increase the probability of project success.

A novel scoring system proposal to guide surgical treatment indications for high grade gliomas in elderly patients: DAK-75.

High-grade gliomas are the most prevalent neurooncological desease in adults, their incidence increases with age, peaking in the seventh decade. This paper aims to address how to select patients for surgical resection by identifying pre-surgical predictors of 12-month mortality in newly diagnosed HGG patients aged ≥ 75years. A prognostic score will be proposed to guide surgical decisions based on expected survival. Retrospective observational single-center cohort study was carried out at the "Città della Salute e della Scienza-Molinette" University Hospital in Turin, Italy. All consecutive patients aged ≥ 75years newly diagnosed with HGG were included, regardless of whether they underwent surgical resection. Clinical, radiological, histological and molecular data were collected.Variables potentially available at the time of diagnosis were considered to develop a multivariable logistic regression predictive model, with 12-months overall survival as the dependent variable. 102 patients aged 75years or older received a new diagnosis of high-grade glioma, of whom 68 underwent surgical resection. Patients undergoing surgery were slightly younger (76.9 vs 79.0years, p = 0.007) and had better performance status (median KPS 80 vs 70). Most tumors undergoing surgery were localized in cortical or subcortical non-motor areas (p < 0.001) and less frequently deep-seated (p = 0.023) or multifocal (p < 0.001). A predictive model, the DAK-75 score, was developed: the AUROC of the final model was 0.822 (95% CI 0.741-0.902). The score includes clinical presentation, tumor location, and KPS, ranging from 0 to 20, categorizing risk scores into low-risk and high-risk groups (< or > 8). Higher scores corresponded to fewer surgical patients and higher one-year mortality rates (92.2% vs 47.1%, p < 0.001). DAK-75 score may represent a valuable tool in the decision-making process for neurosurgical intervention in elderly patients diagnosed with HGG. Further studies are needed to externally and prospectively validate the scoring system.

Importance of OCT-derived biomarkers for the recurrence of central serous chorioretinopathy using statistics and predictive modelling

Central serous chorioretinopathy (CSCR) is a retinal disease characterised by the accumulation of subretinal fluid, which often resolves spontaneously in acute cases. However, approximately one-third of patients experience recurrences that may cause severe and irreversible vision. This study aimed to identify parameters derived from optical coherence tomography (OCT) that are associated with CSCR recurrence. Our dataset included 5211 OCT scans from 344 eyes of 255 patients diagnosed with CSCR. 178 eyes were identified as recurrent, 109 as non-recurrent, and 57 were excluded. We extracted parameters using artificial intelligence algorithms based on U-Nets, convolutional kernels, and morphological operators. We applied inferential statistics to evaluate differences between the recurrent and non-recurrent groups, and we used a logistic regression predictive model, reporting the coefficients as a measure of biomarker importance. We identified nine predictive biomarkers for CSCR recurrence: age, intraretinal fluid, subretinal fluid, pigment epithelial detachments, choroidal vascularity index, integrity of photoreceptors and retinal pigment epithelium layer, choriocapillaris and choroidal stroma thickness, and thinning of the outer nuclear layer, and of the inner nuclear layer combined with the outer plexiform layer. These results could enable future developments in the automatic detection of CSCR recurrence, paving the way for translational medical applications.

Fetal Fraction of Cell-Free DNA in the Prediction of Adverse Pregnancy Outcomes: A Nationwide Retrospective Cohort Study.

To assess the added value of fetal fraction of cell-free DNA in the maternal circulation in the prediction of adverse pregnancy outcomes. Retrospective cohort study. Nationwide implementation study on non-invasive prenatal testing (NIPT; TRIDENT-2 study). Pregnant women in the Netherlands opting for NIPT between June 2018 and June 2019. Two logistic regression prediction models were constructed for each adverse pregnancy outcome. The first model (base model) included prognostic clinical parameters that were selected from existing first-trimester prediction models for adverse pregnancy outcomes. The second model (fetal fraction model) included fetal fraction as a predictor on top of the prognostic clinical parameters included in the base model. The added prognostic value of fetal fraction was assessed by comparing the base and fetal fraction models in terms of goodness of fit and predictive performance. Likelihood ratio test (LRT), area under the curve (AUC) and Integrated Discrimination Improvement (IDI) index. The study cohort consisted of 56 110 pregnancies. The incidence of adverse pregnancy outcomes was 5.7% for hypertensive disorders of pregnancy (HDP; n = 3207), 10.2% for birthweight < p10 (n = 5726), 3.2% for birthweight < p2.3 (n = 1796), 3.4% for spontaneous preterm birth (sPTB; n = 1891), 3.4% for diabetes (n = 1902) and 1.3% for congenital anomalies (n = 741). Adding fetal fraction to the base model improved model fit for HDP, birthweight < p10, birthweight < p2.3, all sPTB, and diabetes, but not for congenital anomalies (LRT p < 0.05). For HDP, the AUC improved from 0.67 to 0.68 by adding fetal fraction to the base model (p = 0.14) with an IDI of 0.0018 (p < 0.0001). For birthweight < p10, the AUC improved from 0.65 to 0.66 (p < 0.0001) with an IDI of 0.0023 (p < 0.0001). For birthweight < p2.3, the AUC improved from 0.67 to 0.69 (p < 0.0001) with an IDI of 0.0011 (p < 0.0001). For all sPTB, the AUC was similar for both models (AUC 0.63, p = 0.021) with an IDI of 0.00028 (p = 0.0023). For diabetes, the AUC was similar (AUC 0.72, p = 0.35) with an IDI of 0.00055 (p = 0.00015). Fetal fraction has statistically significant but limited prognostic value in the prediction of adverse pregnancy outcomes in addition to known prognostic clinical parameters.

Role and molecular mechanisms of HuangQiSiJunZi decoction for treating triple-negative breast cancer as explored via network pharmacology and bioinformatics analyses

ObjectiveIn this study, we evaluated the molecular mechanisms of HuangQiSiJunZi Decoction (HQSJZD) for treating triple-negative breast cancer (TNBC) using network pharmacology and bioinformatics analyses.MethodsEffective chemical components together with action targets of HQSJZD were selected based on the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP). Meanwhile, differentially expressed genes (DEGs) were extracted from TNBC sample data in The Cancer Genome Atlas (TCGA) database. Additionally, we built a protein-protein interaction (PPI) network and acquired hub genes. Gene Expression Omnibus(GEO) datasets were utilized to verify the accuracy of hub gene expression. Additionally, enrichment analyses were conducted on key genes. Furthermore, TNBC severity-related high-risk factors were screened through univariate together with multivariate Cox regressions; next, the logistic regression prediction model was built. Moreover, differential levels of 22 immune cell types in TNBC tissues compared with normal tissues were analyzed. The hub gene levels within pan-cancer and the human body were subsequently visualized and analyzed. Finally, quantitative PCR (RT-qPCR) was used to validate the correlation of the hub genes in TNBC cells.ResultsThe study predicted 256 targets of active ingredients and 1791 DEGs in TNBC, and obtained 16 hub genes against TNBC. The prognostic signature based on FOS, MMP9, and PGR was independent in predicting survival. A total of seven types of immune cells, such as CD4 + memory T cells, showed a significant difference in infiltration (p < 0.05), and immune cells were related to the hub genes. The HPA database was adopted for hub gene analyses, and as determined, FOS was highly expressed in most human organs. The results of RT-qPCR validation for the FOS hub gene were consistent with those of bioinformatic analyses.ConclusionHQSJZD might regulate the interleukin-17 and aging pathways via FOS genes to increase immune cell infiltration in TNBC tissues, and thus, may treat TNBC and improve the prognosis. The FOS genes are likely to be a new marker for TNBC.

Modeling the importance of physician training in practice location for Ohio otolaryngologists.

Underserved communities in southeastern Ohio and Appalachia face significant healthcare accessibility challenges, with the Midwest offering a lower density of healthcare providers compared to coastal regions. Specifically, underserved communities in southeastern Ohio and Appalachia are disadvantaged in otolaryngology care. This analysis aims to identify factors that lead otolaryngologists to a respective practice location, and if any of that influence comes from where otolaryngologists completed their medical education. The proportion of otolaryngologists who performed medical school, residency, and/or fellowship in Ohio was analyzed utilizing a three-sample test for equality. Multivariate logistic regression and Pearson prediction models were produced to analyze the impact of performing medical training (medical school, residency, and fellowship) in Ohio. Going to medical school in Ohio significantly increases the odds of going to an otolaryngology residency in the state (p<0.001). Moreover, between medical school and residency, medical school was a significantly better predictor of otolaryngologists practicing in Appalachia (Δ Bayesian Information Criterion [BIC]>2) and southeast Ohio (ΔBIC>10). Medical school in state was also a better predictor of percent rural and median household income than residency (ΔBIC>10). The multivariate model of medical school and residency was significantly better than either predictor alone for the population (ΔBIC>2). All models predicting percent rural were significantly improved with the addition of a Doctor of Osteopathy (DO) degree (ΔBIC>10). Where physicians complete their medical training (medical school, residency, and fellowship) in state has a significant impact on predicting their future place of practice. This study found that the location of such training has a positive predictive nature as to whether that physician will practice in a rural and underserved area in the future. Notably, the addition of being licensed as a DO also increased the probability of that physician practicing in a rural area.

Etiology and characteristics of non-aneurysmal thunderclap headache presenting to an acute setting

ObjectivesTo describe the characteristics and causes of non-aneurysmal thunderclap headache (TCH) and compare serious from benign underlying causes. MethodsRetrospective cohort study of consecutive adult patients with TCH presenting to a tertiary care academic medical center between 2010 and 2020. Aneurysmal subarachnoid hemorrhage cases were excluded. Cases were categorized into serious or benign; serious TCH was defined as any condition in which delayed diagnosis and treatment could result in neurological disability or death. Risk factors for serious TCH were analyzed. We adhere to standardized guidelines for reporting observational studies. ResultsA total of 932 patients presented with TCH. After exclusion of 393 patients with aneurysmal-type subarachnoid hemorrhage, 539 were included in the analysis. One-half (n = 275, 51.0 %) had a serious cause. Median age was 51 years, 69.0 % were female. Most frequent diagnoses were intracranial hemorrhage (n = 102, 18.9 %), reversible cerebral vasoconstriction syndrome (n = 97, 18.0 %), and idiopathic TCH (n = 102, 38.6 %). A multivariable logistic regression model for prediction of serious TCH included age, hypertension, migraines, recurrent TCH, level of consciousness and other clinical exam findings, and achieved an AUROC of 0.732. This score had a sensitivity of 79.9 % (95 % CI 73.5–83.5 %) for the identification of serious TCH. A 0.5-point increase in the risk score was associated with a 73 % increase in the odds of serious TCH (odds ratio 1.73, 95 % CI 1.53–1.95, p < 0.001). ConclusionOur study describes the relative frequency of presentation and etiologies among patients with TCH This score can aide clinicians in recognising patients with potentially serious cause of TCH, for whom additional imaging and neurological consultation is necessary.

Development of predictive model for the neurological deterioration among mild traumatic brain injury patients using machine learning algorithms.

Mild traumatic brain injury (mTBI) comprises a majority of traumatic brain injury (TBI) cases. While some mTBI would suffer neurological deterioration (ND) and therefore have poorer prognosis. This study was designed to develop the predictive model for the ND among mTBI using machine learning algorithms. mTBI patients recorded in the Medical Information Mart for Intensive Care-III were selected for the study. The ND was defined as a drop of Glasgow Coma Scale ≥ 2 within the first 7day after admission. Eight machine learning algorithms were trained and validated with 5-fold cross validation including extreme gradient boosting, logistic regression, light gradient boosting machine, random forest, adaptive boosting, decision tree, complement naïve Bayes, and support vector machine. The value of eight machine learning algorithms was compared by the area under the receiver operating characteristic curve (AUC). 361 mTBI patients suffered the ND with the incidence of 30.7%. The ND group had higher 30-day mortality (p = 0.001). In the training cohort of mTBI patients, the random forest performed the best on predicting the ND with the AUC of 1.000. The XGBoost and AdaBoost had an AUC of 0.827 and 0.815, respectively. The logistic regression performed the best on predicting the ND in the validation cohort with the AUC of 0.741. The XGBoost, random forest and AdaBoost had an AUC of 0.729, 0.735, 0.736 in the validation cohort, respectively. After adjusting confounding effects, the multivariate logistic regression found only two independent risk factors for the ND including Sequential Organ Failure Assessment (SOFA) (p < 0.001) and hypertension (p = 0.001). The logistic regression predictive model composed of SOFA and hypertension had an AUC of 0.741. SOFA score and complicated hypertension are two independent risk factors for the neurological deterioration among mTBI patients. The logistic regression predictive model incorporating SOFA and hypertension is helpful to identify mTBI patients with the high risk of ND.

Comparison of machine-learning and logistic regression models for prediction of 30-day unplanned readmission in electronic health records: A development and validation study.

It is expected but unknown whether machine-learning models can outperform regression models, such as a logistic regression (LR) model, especially when the number and types of predictor variables increase in electronic health records (EHRs). We aimed to compare the predictive performance of gradient-boosted decision tree (GBDT), random forest (RF), deep neural network (DNN), and LR with the least absolute shrinkage and selection operator (LR-LASSO) for unplanned readmission. We used EHRs of patients discharged alive from 38 hospitals in 2015-2017 for derivation and in 2018 for validation, including basic characteristics, diagnosis, surgery, procedure, and drug codes, and blood-test results. The outcome was 30-day unplanned readmission. We created six patterns of data tables having different numbers of binary variables (that ≥5% or ≥1% of patients or ≥10 patients had) with and without blood-test results. For each pattern of data tables, we used the derivation data to establish the machine-learning and LR models, and used the validation data to evaluate the performance of each model. The incidence of outcome was 6.8% (23,108/339,513 discharges) and 6.4% (7,507/118,074 discharges) in the derivation and validation datasets, respectively. For the first data table with the smallest number of variables (102 variables that ≥5% of patients had, without blood-test results), the c-statistic was highest for GBDT (0.740), followed by RF (0.734), LR-LASSO (0.720), and DNN (0.664). For the last data table with the largest number of variables (1543 variables that ≥10 patients had, including blood-test results), the c-statistic was highest for GBDT (0.764), followed by LR-LASSO (0.755), RF (0.751), and DNN (0.720), suggesting that the difference between GBDT and LR-LASSO was small and their 95% confidence intervals overlapped. In conclusion, GBDT generally outperformed LR-LASSO to predict unplanned readmission, but the difference of c-statistic became smaller as the number of variables was increased and blood-test results were used.

Identification of factors directly linked to incident chronic obstructive pulmonary disease: A causal graph modeling study.

Beyond exposure to cigarette smoking and aging, the factors that influence lung function decline to incident chronic obstructive pulmonary disease (COPD) remain unclear. Advancements have been made in categorizing COPD into emphysema and airway predominant disease subtypes; however, predicting which healthy individuals will progress to COPD is difficult because they can exhibit profoundly different disease trajectories despite similar initial risk factors. This study aimed to identify clinical, genetic, and radiological features that are directly linked-and subsequently predict-abnormal lung function. We employed graph modeling on 2,643 COPDGene participants (aged 45 to 80 years, 51.25% female, 35.1% African Americans; enrollment 11/2007-4/2011) with smoking history but normal spirometry at study enrollment to identify variables that are directly linked to future lung function abnormalities. We developed logistic regression and random forest predictive models for distinguishing individuals who maintain lung function from those who decline. Of the 131 variables analyzed, 6 were identified as informative to future lung function abnormalities, namely forced expiratory flow in the middle range (FEF25-75%), average lung wall thickness in a 10 mm radius (Pi10), severe emphysema, age, sex, and height. We investigated whether these features predict individuals leaving GOLD 0 status (normal spirometry according to Global Initiative for Obstructive Lung Disease (GOLD) criteria). Linear models, trained with these features, were quite predictive (area under receiver operator characteristic curve or AUROC = 0.75). Random forest predictors performed similarly to logistic regression (AUROC = 0.7), indicating that no significant nonlinear effects were present. The results were externally validated on 150 participants from Specialized Center for Clinically Oriented Research (SCCOR) cohort (aged 45 to 80 years, 52.7% female, 4.7% African Americans; enrollment: 7/2007-12/2012) (AUROC = 0.89). The main limitation of longitudinal studies with 5- and 10-year follow-up is the introduction of mortality bias that disproportionately affects the more severe cases. However, our study focused on spirometrically normal individuals, who have a lower mortality rate. Another limitation is the use of strict criteria to define spirometrically normal individuals, which was unavoidable when studying factors associated with changes in normalized forced expiratory volume in 1 s (FEV1%predicted) or the ratio of FEV1/FVC (forced vital capacity). This study took an agnostic approach to identify which baseline measurements differentiate and predict the early stages of lung function decline in individuals with previous smoking history. Our analysis suggests that emphysema affects obstruction onset, while airway predominant pathology may play a more important role in future FEV1 (%predicted) decline without obstruction, and FEF25-75% may affect both.

Applying urinary ultrasound to predict the risk of spontaneous ureteral stone passage: a retrospective cohort study

BackgroundTo assess the value of urological ultrasound in predicting the risk of spontaneous passage of ureteral stones.MethodsClinical and ultrasound data were collected consecutively from patients receiving conservative treatment for ureteral stones, and the outcome of spontaneous passage was followed up for 1 month. Ultrasound variables independently associated with the risk of spontaneous stone passage were screened. A logistic regression prediction model was constructed based on the independent risk factors, and the discriminative efficacy and clinical utility of the prediction model in inferring the risk of spontaneous passing were assessed by the receiver operating characteristic (ROC) curve, calibration curve and clinical decision curve.ResultsA total of 163 patients undergoing conservative treatment for ureteral stones were included in the study, with a mean age of 45.95 ± 13.01 years. Among them, 47 cases (28.83%) experienced failure of spontaneous stone passage. Multivariable analysis revealed that stone length (OR: 2.622, P = 0.027), distal stone location (OR: 0.219, P = 0.003), and ureteral jetting frequency (OR: 6.541, P < 0.001) were independent risk factors for spontaneous stone passage. A prediction model incorporating stone length, stone location, and affected ureteral jetting frequency was developed to assess the risk of spontaneous stone passage. The area under the ROC curve was 0.814 (95% CI: 0.747–0.882), indicating good discriminatory power. The prediction model also demonstrated favorable net clinical benefit.ConclusionA prediction model based on ultrasound-derived stone length, location, and ureteral jetting frequency can accurately evaluate the risk of spontaneous stone passage in patients with ureteral stones, providing a basis for optimizing the clinical decision-making on ureteral stones, and has reliable clinical application value.

Do Positive Psychosocial Factors Contribute to the Prediction of Coronary Artery Disease? A UK Biobank-Based Machine Learning Approach.

Most prediction models for coronary artery disease (CAD) compile biomedical and behavioural risk factors, using linear multivariate models. This study explored the potential of integrating positive psychosocial factors (PPFs), including happiness, satisfaction with life, and social support, into conventional and machine learning-based CAD prediction models. We included UK Biobank participants without CAD at baseline. First, we estimated associations of individual PPFs with subsequent acute myocardial infarction (AMI) and chronic ischaemic heart disease (CIHD) using logistic regression. Then, we compared the performances of logistic regression and eXtreme Gradient Boosting (XGBoost) prediction models when adding PPFs as predictors to the Framingham Risk Score (FRS). Based on a sample size between 160,226 and 441,419 of UK Biobank participants, happiness, satisfaction with health and life, and participation in social activities were linked to lower AMI and CIHD risk (all p-for-trend ≤ 0.04), while social support was not. In a validation sample, adding PPFs to the FRS using logistic regression and XGBoost prediction models improved neither AMI (AUC change: 0.02% and 0.90%, respectively) nor CIHD (AUC change: -1.10% and -0.88%, respectively) prediction. PPFs were individually linked to CAD risk, in line with previous studies, and as reflected by the new European Society of Cardiology guidelines on cardiovascular disease prevention. However, including available PPFs in CAD-prediction models did not improve prediction compared to the FRS alone. Future studies should explore whether PPFs may act as CAD-risk modifiers, especially if the individual's risk is close to a decision threshold.

MASH Resolution Index: development and validation of a non-invasive score to detect histological resolution of MASH

BackgroundDynamic changes in non-invasive tests, such as changes in alanine aminotransferase (ALT) and MRI proton-density-fat-fraction (MRI-PDFF), may help to detect metabolic dysfunction-associated steatohepatitis (MASH) resolution, but a combination of non-invasive...

Comparative study of the risk prediction model of early postoperative frailty in elderly enterostomy patients based on machine learning methods.

Based on machine learning method, four types of early postoperative frailty risk prediction model of enterostomy patients were constructed to compare the performance of each model and provide the basis for preventing early postoperative frailty of elderly patients with enterostomy. The prospective convenience sampling method was conducted and 362 early postoperative enterostomy patients were selected in three hospitals from July 2020 to November 2023 in Shanghai, four different prediction models of Support Vector Machine (SVM), Bayes, XG Boost, and Logistic regression were used and compared the test effects of the four models (MCC, F1, AUC, and Brier index) to judge the classification performance of the four models in the data of this study. A total of 21 variables were included in this study, and the predictors mainly covered demographic information, stoma-related information, quality of life, anxiety and depression, and frailty. The validated models on the test set are XGBoost, Logistic regression, SVM prediction model, and Bayes on the MCC and F1 scores; on the AUC, XGBoost, Logistic regression, Bayes, and SVM prediction model; on the Brier scores, Bayes, Logistic regression, and XGBoost. XGBoost based on machine learning method is better than SVM prediction model, Logistic regression model and Bayes in sensitivity and accuracy. Quality of life in the early postoperative period can help guide clinical patients to identify patients at high risk of frailty and reduce the incidence of early postoperative frailty in elderly patients with enterostomy.

A prediction model for secondary invasive fungal infection among severe SARS-CoV-2 positive patients in ICU.

The global COVID-19 pandemic has resulted in over seven million deaths, and IFI can further complicate the clinical course of COVID-19. Coinfection of COVID-19 and IFI (secondary IFI) pose significant threats not only to healthcare systems but also to patient lives. After the control measures for COVID-19 were lifted in China, we observed a substantial number of ICU patients developing COVID-19-associated IFI. This creates an urgent need for predictive assessment of COVID-19 patients in the ICU environment for early detection of suspected fungal infection cases. This study is a single-center, retrospective research endeavor. We conducted a case-control study on severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) positive patients. The cases consisted of patients who developed any secondary IFI during their ICU stay at Jilin University China-Japan Union Hospital in Changchun, Jilin Province, China, from December 1st, 2022, to August 31st, 2023. The control group consisted of SARS-CoV-2 positive patients without secondary IFI. Descriptive and comparative analyses were performed, and a logistic regression prediction model for secondary IFI in COVID-19 patients was established. Additionally, we observed an increased incidence of COVID-19-associated pulmonary aspergillosis (CAPA) during this pandemic. Therefore, we conducted a univariate subgroup analysis on top of IFI, using non-CAPA patients as the control subgroup. From multivariate analysis, the prediction model identified 6 factors that are significantly associated with IFI, including the use of broad-spectrum antibiotics for more than 2 weeks (aOR=4.14, 95% CI 2.03-8.67), fever (aOR=2.3, 95%CI 1.16-4.55), elevated log IL-6 levels (aOR=1.22, 95% CI 1.04-1.43) and prone position ventilation (aOR=2.38, 95%CI 1.15-4.97) as independent risk factors for COVID-19 secondary IFI. High BMI (BMI ≥ 28 kg/m2) (aOR=0.85, 95% CI 0.75-0.94) and the use of COVID-19 immunoglobulin (aOR=0.45, 95% CI 0.2-0.97) were identified as independent protective factors against COVID-19 secondary IFI. The Receiver Operating Curve (ROC) area under the curve (AUC) of this model was 0.81, indicating good classification. We recommend paying special attention for the occurrence of secondary IFI in COVID-19 patients with low BMI (BMI < 28 kg/m2), elevated log IL-6 levels and fever. Additionally, during the treatment of COVID-19 patients, we emphasize the importance of minimizing the duration of broad-spectrum antibiotic use and highlight the potential of immunoglobulin application in reducing the incidence of IFI.

Functional Multivariable Logistic Regression With an Application to HIV Viral Suppression Prediction.

Motivated by improving the prediction of the human immunodeficiency virus (HIV) suppression status using electronic health records (EHR) data, we propose a functional multivariable logistic regression model, which accounts for the longitudinal binary process and continuous process simultaneously. Specifically, the longitudinal measurements for either binary or continuous variables are modeled by functional principal components analysis, and their corresponding functional principal component scores are used to build a logistic regression model for prediction. The longitudinal binary data are linked to underlying Gaussian processes. The estimation is done using penalized spline for the longitudinal continuous and binary data. Group-lasso is used to select longitudinal processes, and the multivariate functional principal components analysis is proposed to revise functional principal component scores with the correlation. The method is evaluated via comprehensive simulation studies and then applied to predict viral suppression using EHR data for people living with HIV in SouthCarolina.

Development and validation of a prediction model for mortality in critically ill COVID-19 patients.

Early prediction of prognosis may help early treatment measures to reduce mortality in critically ill coronavirus disease (COVID-19) patients. The study aimed to develop a mortality prediction model for critically ill COVID-19 patients. This retrospective study analyzed the clinical data of critically ill COVID-19 patients in an intensive care unit between April and June 2022. Propensity matching scores were used to reduce the effect of confounding factors. A predictive model was built using logistic regression analysis and visualized using a nomogram. Calibration and receiver operating characteristic (ROC) curves were used to estimate the accuracy and predictive value of the model. Decision curve analysis (DCA) was used to examine the value of the model for clinical interventions. In total, 137 critically ill COVID-19 patients were enrolled; 84 survived, and 53 died. Univariate and multivariate logistic regression analyses revealed that aspartate aminotransferase (AST), creatinine, and myoglobin levels were independent prognostic factors. We constructed logistic regression prediction models using the seven least absolute shrinkage and selection operator regression-selected variables (hematocrit, red blood cell distribution width-standard deviation, procalcitonin, AST, creatinine, potassium, and myoglobin; Model 1) and three independent factor variables (Model 2). The calibration curves suggested that the actual predictions of the two models were similar to the ideal predictions. The ROC curve indicated that both models had good predictive power, and Model 1 had better predictive power than Model 2. The DCA results suggested that the model intervention was beneficial to patients and patients benefited more from Model 1 than from Model 2. The predictive model constructed using characteristic variables screened using LASSO regression can accurately predict the prognosis of critically ill COVID-19 patients. This model can assist clinicians in implementing early interventions. External validation by prospective large-sample studies is required.

Understanding random resampling techniques for class imbalance correction and their consequences on calibration and discrimination of clinical risk prediction models

ObjectiveClass imbalance is sometimes considered a problem when developing clinical prediction models and assessing their performance. To address it, correction strategies involving manipulations of the training dataset, such as random undersampling or oversampling, are frequently used. The aim of this article is to illustrate the consequences of these class imbalance correction strategies on clinical prediction models’ internal validity in terms of calibration and discrimination performances. MethodsWe used both heuristic intuition and formal mathematical reasoning to characterize the relations between conditional probabilities of interest and probabilities targeted when using random undersampling or oversampling. We propose a plug-in estimator that represents a natural correction for predictions obtained from models that have been trained on artificially balanced datasets (“naïve” models). We conducted a Monte Carlo simulation with two different data generation processes and present a real-world example using data from the International Stroke Trial database to empirically demonstrate the consequences of applying random resampling techniques for class imbalance correction on calibration and discrimination (in terms of Area Under the ROC, AUC) for logistic regression and tree-based prediction models. ResultsAcross our simulations and in the real-world example, calibration of the naïve models was very poor. The models using the plug-in estimator generally outperformed the models relying on class imbalance correction in terms of calibration while achieving the same discrimination performance. ConclusionRandom resampling techniques for class imbalance correction do not generally improve discrimination performance (i.e., AUC), and their use is hard to justify when aiming at providing calibrated predictions. Improper use of such class imbalance correction techniques can lead to suboptimal data usage and less valid risk prediction models.

Predicting Kawasaki disease shock syndrome in children.

Kawasaki disease shock syndrome (KDSS) is a critical manifestation of Kawasaki disease (KD). In recent years, a logistic regression prediction model has been widely used to predict the occurrence probability of various diseases. This study aimed to investigate the clinical characteristics of children with KD and develop and validate an individualized logistic regression model for predicting KDSS among children with KD. The clinical data of children diagnosed with KDSS and hospitalized between January 2021 and December 2023 were retrospectively analyzed. The best predictors were selected by logistic regression and lasso regression analyses. A logistic regression model was built of the training set (n = 162) to predict the occurrence of KDSS. The model prediction was further performed by logistic regression. A receiver operating characteristic curve was used to evaluate the performance of the logistic regression model. We built a nomogram model by visualizing the calibration curve using a 1000 bootstrap resampling program. The model was validated using an independent validation set (n = 68). In the univariate analysis, among the 24 variables that differed significantly between the KDSS and KD groups, further logistic and Lasso regression analyses found that five variables were independently related to KDSS: rash, brain natriuretic peptide, serum Na, serum P, and aspartate aminotransferase. A logistic regression model was established of the training set (area under the receiver operating characteristic curve, 0.979; sensitivity=96.2%; specificity=97.2%). The calibration curve showed good consistency between the predicted values of the logistic regression model and the actual observed values in the training and validation sets. Here we established a feasible and highly accurate logistic regression model to predict the occurrence of KDSS, which will enable its early identification.