[Overview of logistic regression model analysis and application]

Hybrid model of CT-fractional flow reserve, pericoronary fat attenuation index and radiomics for predicting the progression of WMH: a dual-center pilot study

Objective

To develop and validate a hybrid model incorporating CT-fractional flow reserve (CT-FFR), pericoronary fat attenuation index (pFAI), and radiomics signatures for predicting progression of white matter hyperintensity (WMH).

Methods

A total of 226 patients who received coronary computer tomography angiography (CCTA) and brain magnetic resonance imaging from two hospitals were divided into a training set (n = 116), an internal validation set (n = 30), and an external validation set (n = 80). Patients who experienced progression of WMH were identified from subsequent MRI results. We calculated CT-FFR and pFAI from CCTA images using semi-automated software, and segmented the pericoronary adipose tissue (PCAT) and myocardial ROI. A total of 1,073 features were extracted from each ROI, and were then refined by Elastic Net Regression. Firstly, different machine learning algorithms (Logistic Regression [LR], Support Vector Machine [SVM], Random Forest [RF], k-nearest neighbor [KNN] and eXtreme Gradient Gradient Boosting Machine [XGBoost]) were used to evaluate the effectiveness of radiomics signatures for predicting WMH progression. Then, the optimal machine learning algorithm was used to compare the predictive performance of individual and hybrid models based on independent risk factors of WMH progression. Receiver operating characteristic (ROC) curve analysis, calibration and decision curve analysis were used to evaluate predictive performance and clinical value of the different models.

Results

CT-FFR, pFAI, and radiomics signatures were independent predictors of WMH progression. Based on the machine learning algorithms, the PCAT signatures led to slightly better predictions than the myocardial signatures and showed the highest AUC value in the XGBoost algorithm for predicting WMH progression (AUC: 0.731 [95% CI: 0.603-0.838] vs.0.711 [95% CI: 0.584-0.822]). In addition, pFAI provided better predictions than CT-FFR (AUC: 0.762 [95% CI: 0.651-0.863] vs. 0.682 [95% CI: 0.547-0.799]). A hybrid model that combined CT-FFR, pFAI, and two radiomics signatures provided the best predictions of WMH progression [AUC: 0.893 (95%CI: 0.815-0.956)].

Conclusion

pFAI was more effective than CT-FFR, and PCAT signatures were more effective than myocardial signatures in predicting WMH progression. A hybrid model that combines pFAI, CT-FFR, and two radiomics signatures has potential use for identifying WMH progression.

Construction of an Early Alert System for Intradialytic Hypotension before Initiating Hemodialysis Based on Machine Learning

Introduction

Intradialytic hypotension (IDH) is prevalent and associated with high hospitalization and mortality rates. The purpose of this study was to explore the risk factors for IDH and use artificial intelligence to establish an early alert system before hemodialysis sessions to identify patients at high risk of IDH.

Materials and Methods

We obtained data on 314,534 hemodialysis sessions conducted at Sichuan Provincial People's Hospital from the renal disease treatment information system. IDH was defined as a systolic blood pressure drop ≥20 mm Hg, a mean arterial pressure drop ≥10 mm Hg during dialysis, or the occurrence of clinical hypotensive events requiring nursing intervention. After pre-processing, the data were randomly divided into training (80%) and testing (20%) sets. Four interpolation methods, three feature selection methods, and 18 machine learning algorithms were used to construct predictive models. The area under the receiver operating characteristic curve (AUC) was the main indicator for evaluating the performance of the models, while Shapley Additive ExPlanation was used to explain the contribution of each variable to the best predictive model.

Results

A total of 3,906 patients and 314,534 dialysis sessions were included, of which 142,237 cases showed IDH (incidence rate, 45.2%). Nineteen parameters were identified through artificial intelligence feature screening. They included age, pre-dialysis weight, dry weight, pre-dialysis blood pressure, heart rate, prescribed ultrafiltration, blood cell counts (neutrophil, lymphocyte, monocyte, eosinophil, lymphocyte, and platelet counts), hematocrit, serum calcium, creatinine, urea, glucose, and uric acid. Random forest, gradient boosting, and logistic regression were the three best models, and the AUCs were 0.812 (95% confidence interval [CI], 0.811-0.813), 0.748 (95% CI, 0.747-0.749), and 0.743 (95% CI, 0.742-0.744), respectively.

Conclusion

Our dialysis software-based artificial intelligence alert system can be used to predict IDH occurrence, enabling the initiation of relevant interventions.

Improved Prediction of Knee Osteoarthritis by the Machine Learning Model XGBoost

Objectives

The accurate prediction of osteoarthritis (OA) severity in patients can be helpful to make the proper decision of intervention. This study aims to build up a powerful model to assess predictive risk factors and severity of knee osteoarthritis (KOA) in the clinical scenario.

Methods

A total of 4796 KOA cases and 1205 features were selected by feature selections from the public OA database, Osteoarthritis Initiative (OAI). Six machine learning-based models were constructed and compared for the accuracy of OA prediction. The gradient-boosting decision tree was used to identify important prediction features in the extreme gradient boosting (XGBoost) model. The performance of models was evaluated by F1-score.

Results

Twenty features were determined as predictors for KOA risk and severity, including the subject characteristics, knee symptoms/risk factors and physical exam. The XGBoost model demonstrated 100% prediction accuracy for 54.7% of examined samples, and the remaining 45.3% of samples showed Kellgren and Lawrence (KL) gradings very close to the actual levels. It showed the highest prediction accuracy with an F1-score of 0.553 among the tested six models.

Conclusions

We demonstrate that the XGBoost is the best model for the prediction of KOA severity in the six examined models. In addition, 20 risk features were determined as the essential predictors of KOA, including the physical exam, knee symptoms/risk factors and subject characteristics, which may be useful for the identification of high-risk KOA cases and for making appropriate treatment decisions as well.

An ultrasound-based radiomics model to distinguish between sclerosing adenosis and invasive ductal carcinoma

Objectives

We aimed to develop an ultrasound-based radiomics model to distinguish between sclerosing adenosis (SA) and invasive ductal carcinoma (IDC) to avoid misdiagnosis and unnecessary biopsies.

Methods

From January 2020 to March 2022, 345 cases of SA or IDC that were pathologically confirmed were included in the study. All participants underwent pre-surgical ultrasound (US), from which clinical information and ultrasound images were collected. The patients from the study population were randomly divided into a training cohort (n = 208) and a validation cohort (n = 137). The US images were imported into MaZda software (Version 4.2.6.0) to delineate the region of interest (ROI) and extract features. Intragroup correlation coefficient (ICC) was used to evaluate the consistency of the extracted features. The least absolute shrinkage and selection operator (LASSO) logistic regression and cross-validation were performed to obtain the radiomics score of the features. Based on univariate and multivariate logistic regression analyses, a model was developed. 56 cases from April 2022 to December 2022 were included for independent validation of the model. The diagnostic performance of the model and the radiomics scores were evaluated by performing the receiver operating characteristic (ROC) analysis. The calibration curve and decision curve analysis (DCA) were used for calibration and evaluation. Leave-One-Out Cross-Validation (LOOCV) was used for the stability of the model.

Results

Three predictors were selected to develop the model, including radiomics score, palpable mass and BI-RADS. In the training cohort, validation cohort and independent validation cohort, AUC of the model and radiomics score were 0.978 and 0.907, 0.946 and 0.886, 0.951 and 0.779, respectively. The model showed a statistically significant difference compared with the radiomics score (p<0.05). The Kappa value of the model was 0.79 based on LOOCV. The Brier score, calibration curve, and DCA showed the model had a good calibration and clinical usefulness.

Conclusions

The model based on radiomics, ultrasonic features, and clinical manifestations can be used to distinguish SA from IDC, which showed good stability and diagnostic performance. The model can be considered a potential candidate diagnostic tool for breast lesions and can contribute to effective clinical diagnosis.

Comprehensive analysis of prediction of the EGFR mutation and subtypes based on the spinal metastasis from primary lung adenocarcinoma

Purpose

To investigate the use of multiparameter MRI-based radiomics in the in-depth prediction of epidermal growth factor receptor (EGFR) mutation and subtypes based on the spinal metastasis in patients with primary lung adenocarcinoma.

Methods

A primary cohort was conducted with 257 patients who pathologically confirmed spinal bone metastasis from the first center between Feb. 2016 and Oct. 2020. An external cohort was developed with 42 patients from the second center between Apr. 2017 and Jun. 2021. All patients underwent sagittal T1-weighted imaging (T1W) and sagittal fat-suppressed T2-weight imaging (T2FS) MRI imaging. Radiomics features were extracted and selected to build radiomics signatures (RSs). Machine learning classify with 5-fold cross-validation were used to establish radiomics models for predicting the EGFR mutation and subtypes. Clinical characteristics were analyzed with Mann-Whitney U and Chi-Square tests to identify the most important factors. Nomogram models were developed integrating the RSs and important clinical factors.

Results

The RSs derived from T1W showed better performance for predicting the EGFR mutation and subtypes compared with those from T2FS in terms of AUC, accuracy and specificity. The nomogram models integrating RSs from combination of the two MRI sequences and important clinical factors achieved the best prediction capabilities in the training (AUCs, EGFR vs. Exon 19 vs. Exon 21, 0.829 vs. 0.885 vs.0.919), internal validation (AUCs, EGFR vs. Exon 19 vs. Exon 21, 0.760 vs. 0.777 vs.0.811), external validation (AUCs, EGFR vs. Exon 19 vs. Exon 21, 0.780 vs. 0.846 vs.0.818). DCA curves indicated potential clinical values of the radiomics models.

Conclusions

This study indicated potentials of multi-parametric MRI-based radiomics to assess the EGFR mutation and subtypes. The proposed clinical-radiomics nomogram models can be considered as non-invasive tools to assist clinicians in making individual treatment plans.

A Predictive Model for Abnormal Bone Density in Male Underground Coal Mine Workers

The dark and humid environment of underground coal mines had a detrimental effect on workers’ skeletal health. Optimal risk prediction models can protect the skeletal health of coal miners by identifying those at risk of abnormal bone density as early as possible. A total of 3695 male underground workers who attended occupational health physical examination in a coal mine in Hebei, China, from July to August 2018 were included in this study. The predictor variables were identified through single-factor analysis and literature review. Three prediction models, Logistic Regression, CNN and XG Boost, were developed to evaluate the prediction performance. The training set results showed that the sensitivity of Logistic Regression, XG Boost and CNN models was 74.687, 82.058, 70.620, the specificity was 80.986, 89.448, 91.866, the F1 scores was 0.618, 0.919, 0.740, the Brier scores was 0.153, 0.040, 0.156, and the Calibration-in-the-large was 0.104, 0.020, 0.076, respectively, XG Boost outperformed the other two models. Similar results were obtained for the test set and validation set. A two-by-two comparison of the area under the ROC curve (AUC) of the three models showed that the XG Boost model had the best prediction performance. The XG Boost model had a high application value and outperformed the CNN and Logistic regression models in prediction.

Prediction of Response to Radiotherapy by Characterizing the Transcriptomic Features in Clinical Tumor Samples across 15 Cancer Types

Purpose

Radiotherapy (RT) is one of the major cancer treatments. However, the responses to RT vary among individual patients, partly due to the differences of the status of gene expression and mutation in tumors of patients. Identification of patients who will benefit from RT will improve the efficacy of RT. However, only a few clinical biomarkers were currently used to predict RT response. Our aim is to obtain gene signatures that can be used to predict RT response by analyzing the transcriptome differences between RT responder and nonresponder groups.

Materials and Methods

We obtained transcriptome data of 1664 patients treated with RT from the TCGA database across 15 cancer types. First, the genes with a significant difference between RT responder (R group) and nonresponder groups (PD group) were identified, and the top 100 genes were used to build the gene signatures. Then, we developed the predictive model based on binary logistic regression to predict patient response to RT.

Results

We identified a series of differentially expressed genes between the two groups, which are involved in cell proliferation, migration, invasion, EMT, and DNA damage repair pathway. Among them, MDC1, UCP2, and RBM45 have been demonstrated to be involved in DNA damage repair and radiosensitivity. Our analysis revealed that the predictive model was highly specific for distinguishing the R and PD patients in different cancer types with an area under the curve (AUC) ranging from 0.772 to 0.972. It also provided a more accurate prediction than that from a single-gene signature for the overall survival (OS) of patients.

Conclusion

The predictive model has a potential clinical application as a biomarker to help physicians create optimal treatment plans. Furthermore, some of the genes identified here may be directly involved in radioresistance, providing clues for further studies on the mechanism of radioresistance.

Predictive Value of a Combined Model Based on Pre-Treatment and Mid-Treatment MRI-Radiomics for Disease Progression or Death in Locally Advanced Nasopharyngeal Carcinoma

Purpose

A combined model was established based on the MRI-radiomics of pre- and mid-treatment to assess the risk of disease progression or death in locally advanced nasopharyngeal carcinoma.

Materials and Methods

A total of 243 patients were analyzed. We extracted 10,400 radiomics features from the primary nasopharyngeal tumors and largest metastatic lymph nodes on the axial contrast-enhanced T1 weighted and T2 weighted in pre- and mid-treatment MRI, respectively. We used the SMOTE algorithm, center and scale and box-cox, Pearson correlation coefficient, and LASSO regression to construct the pre- and mid-treatment MRI-radiomics prediction model, respectively, and the risk scores named P score and M score were calculated. Finally, univariate and multivariate analyses were used for P score, M score, and clinical data to build the combined model and grouped the patients into two risk levels, namely, high and low.

Result

A combined model of pre- and mid-treatment MRI-radiomics successfully categorized patients into high- and low-risk groups. The log-rank test showed that the high- and low-risk groups had good prognostic performance in PFS (P<0.0001, HR: 19.71, 95% CI: 12.77–30.41), which was better than TNM stage (P=0.004, HR:1.913, 95% CI:1.250–2.926), and also had an excellent predictive effect in LRFS, DMFS, and OS.

Conclusion

Risk grouping of LA-NPC using a combined model of pre- and mid-treatment MRI-radiomics can better predict disease progression or death.

Genome-wide scanning for CHD1L gene in papillary thyroid carcinoma complicated with type 2 diabetes mellitus

PURPOSE

Papillary thyroid carcinoma (PTC) represents the most common subtype of thyroid cancer (TC). This study was set out to explore the potential effect of CHD1L on PTC and type 2 diabetes mellitus (T2DM).

METHODS

We searched for T2DM susceptibility genes through the GWAS database and obtained T2DM-related differentially expressed gene from the GEO database. The expression and clinical data of TC and normal samples were collated from the TCGA database. Receiver operating characteristic (ROC) curve analysis was subsequently applied to assess the sensitivity and specificity of the CHD1L for the diagnosis of PTC. The MCP-counter package in R language was then utilized to generate immune cell score to evaluate the relationship between CHD1L expression and immune cells. Then, we performed functional enrichment analysis of co-expressed genes and DEGs to determine significantly enriched GO terms and KEGG to predict the potential functions of CHD1L in PTC samples and T2DM adipose tissue.

RESULTS

From two genes (ABCB9, CHD1L) were identified to be DEGs (p < 1 * 10^-5) that exerted effects on survival (HR > 1, p < 0.05) in PTC and served as T2DM susceptibility genes. The gene expression matrix-based scoring of immunocytes suggested that PTC samples with high and low CHD1L expression presented with significant differences in the tumor microenvironment (TME). The enrichment analysis of CHD1L co-expressed genes and DEGs suggested that CHD1L was involved in multiple pathways to regulate the development of PTC. Among them, Kaposi sarcoma-associated herpesvirus infection, salmonella infection and TNF signaling pathways were highlighted as the three most relevant pathways. GSEA analysis, employed to analyze the genome dataset of PTC samples and T2DM adipose tissue presenting with high and low expression groups of CHD1L, suggests that these differential genes are related to chemokine signaling pathway, leukocyte transendothelial migration and TCELL receptor signaling pathway.

CONCLUSION

CHD1L may potentially serve as an early diagnostic biomarker for PTC, and a target of immunotherapy for PTC and T2DM.

Risk factors for de novo and therapy-related myelodysplastic syndromes (MDS)

PURPOSE

Myelodysplastic syndromes (MDS) are classified as de novo and therapy-related (tMDS). We evaluated associations between MDS risk factors separately for de novo and tMDS.

METHODS

The study population included 346 de novo MDS cases, 37 tMDS cases and 682 population controls frequency matched by age and sex. Polytomous logistic regression was performed to calculate odds ratios (OR) and 95% confidence intervals (CI).

RESULTS

After adjustment, former smoking status (OR = 1.45, 95% CI: 1.10-1.93), personal history of autoimmune disease (OR = 1.34, 95% CI: 0.99-1.82) and exposure to benzene (OR = 1.48, 95% CI: 1.00-2.19) were associated with de novo MDS. Risk estimates for the associations between smoking, autoimmune disease, and benzene exposure were similar in magnitude but non-significant in tMDS cases. Among individuals with a previous diagnosis of cancer, de novo MDS cases and controls were more likely to have had a previous solid tumor, while tMDS cases more commonly had a previous hematologic malignancy.

CONCLUSIONS

We observed similar associations between smoking, history of autoimmune disease and benzene exposure in de novo and tMDS although estimates for tMDS were imprecise due to small sample sizes. Future analyses with larger sample sizes will be required to confirm whether environmental factors influence risk of tMDS.