Prediction of Poststroke Urinary Tract Infection Risk in Immobile Patients Using Machine Learning: Observational Cohort Study (Preprint)

BACKGROUND Poststroke immobility gets patients more vulnerable to stroke-relevant complications. Urinary tract infection (UTI) is one of major nosocomial infections significantly affecting the outcomes of immobile stroke patients. Previous studies have identified several risk factors, but it is still challenging to accurately estimate personal UTI risk due to unclear interaction of various factors and variability of individual characteristics. This calls for more precise and trust-worthy predictive models to assist with potential UTI identification. OBJECTIVE The aim of this study was to develop predictive models for UTI risk identification for immobile stroke patients. A prospective analysis was conducted to evaluate the effectiveness and clinical interpretability of the models. METHODS The data used in this study were collected from the Common Complications of Bedridden Patients and the Construction of Standardized Nursing Intervention Model (CCBPC). Derivation cohort included data of 3982 immobile stroke patients collected during CCBPC-I, from November 1, 2015 to June 30, 2016; external validation cohort included data of 3837 immobile stroke patients collected during CCBPC-II, from November 1, 2016 to July 30, 2017. 6 machine learning models and an ensemble learning model were derived based on 80% of derivation cohort and its effectiveness was evaluated with the remaining 20% of derivation cohort data. We further compared the effectiveness of predictive models in external validation cohort. The performance of logistic regression without regularization was used as a reference. We used Shapley additive explanation values to determine feature importance and examine the clinical significance of prediction models. Shapely values of the factors were calculated to represent the magnitude, prevalence, and direction of their effects, and were further visualized in a summary plot. RESULTS A total of 103(2.59%) patients were diagnosed with UTI in derivation cohort(N=3982); the internal validation cohort (N=797) shared the same incidence. The external validation cohort had a UTI incidence of 1.38% (N=53). Evaluation results showed that the ensemble learning model performed the best in area under the receiver operating characteristic (ROC) curve in internal validation, up to 82.2%; second best in external validation, 80.8%. In addition, the ensemble learning model performed the best sensitivity in both internal and external validation sets (80.9% and 81.1%, respectively). We also identified seven UTI risk factors (pneumonia, glucocorticoid use, female sex, mixed cerebrovascular disease, increased age, prolonged length of stay, and duration of catheterization) contributing most to the predictive model, thus demonstrating the clinical interpretability of model. CONCLUSIONS Our ensemble learning model demonstrated promising performance. Identifying UTI risk and detecting high risk factors among immobile stroke patients would allow more selective and effective use of preventive interventions, thus improving clinical outcomes. Future work should focus on developing a more concise scoring tool and prospectively examining the model in practical use.

Download Full-text

Development of a machine learning model for predicting pediatric mortality in the early stages of intensive care unit admission

Scientific Reports ◽

10.1038/s41598-020-80474-z ◽

2021 ◽

Vol 11 (1) ◽

Author(s):

Bongjin Lee ◽

Kyunghoon Kim ◽

Hyejin Hwang ◽

You Sun Kim ◽

Eun Hee Chung ◽

...

Keyword(s):

Machine Learning ◽

Intensive Care Unit ◽

Intensive Care ◽

Validation Cohort ◽

External Validation ◽

Learning Model ◽

Derivation Cohort ◽

Icu Admission ◽

Early Stages ◽

Machine Learning Model

AbstractThe aim of this study was to develop a predictive model of pediatric mortality in the early stages of intensive care unit (ICU) admission using machine learning. Patients less than 18 years old who were admitted to ICUs at four tertiary referral hospitals were enrolled. Three hospitals were designated as the derivation cohort for machine learning model development and internal validation, and the other hospital was designated as the validation cohort for external validation. We developed a random forest (RF) model that predicts pediatric mortality within 72 h of ICU admission, evaluated its performance, and compared it with the Pediatric Index of Mortality 3 (PIM 3). The area under the receiver operating characteristic curve (AUROC) of RF model was 0.942 (95% confidence interval [CI] = 0.912–0.972) in the derivation cohort and 0.906 (95% CI = 0.900–0.912) in the validation cohort. In contrast, the AUROC of PIM 3 was 0.892 (95% CI = 0.878–0.906) in the derivation cohort and 0.845 (95% CI = 0.817–0.873) in the validation cohort. The RF model in our study showed improved predictive performance in terms of both internal and external validation and was superior even when compared to PIM 3.

Download Full-text

Machine learning identifies clinical parameters to predict mortality in patients undergoing transcatheter mitral valve repair

European Heart Journal ◽

10.1093/eurheartj/ehab724.2221 ◽

2021 ◽

Vol 42 (Supplement_1) ◽

Author(s):

E Zweck ◽

M Spieker ◽

P Horn ◽

C Iliadis ◽

C Metze ◽

...

Keyword(s):

Machine Learning ◽

Clinical Trials ◽

Mitral Valve ◽

Risk Stratification ◽

Mitral Valve Repair ◽

Validation Cohort ◽

External Validation ◽

Valve Repair ◽

Derivation Cohort ◽

Transcatheter Mitral Valve Repair

Abstract Background Transcatheter Mitral Valve Repair (TMVR) with MitraClip is an important treatment option for patients with severe mitral regurgitation. The lack of appropriate, validated and specific means to risk stratify TMVR patients complicates the evaluation of prognostic benefits of TMVR in clinical trials and practice. Purpose We aimed to develop an optimized risk stratification model for TMVR patients using machine learning (ML). Methods We included a total of 1009 TMVR patients from three large university hospitals, of which one (n=317) served as an external validation cohort. The primary endpoint was all-cause 1-year mortality, which was known in 95% of patients. Model performance was assessed using receiver operating characteristics. In the derivation cohort, different ML algorithms, including random forest, logistic regression, support vectors machines, k nearest neighbors, multilayer perceptron, and extreme gradient boosting (XGBoost) were tested using 5-fold cross-validation in the derivation cohort. The final model (Transcatheter MITral Valve Repair MortALIty PredicTion SYstem; MITRALITY) was tested in the validation cohort with respect to existing clinical scores. Results XGBoost was selected as the final algorithm for the MITRALITY Score, using only six baseline clinical features for prediction (in order of predictive importance): blood urea nitrogen, hemoglobin, N-terminal prohormone of brain natriuretic peptide (NT-proBNP), mean arterial pressure, body mass index, and creatinine. In the external validation cohort, the MITRALITY Score's area under the curve (AUC) was 0.783, outperforming existing scores which yielded AUCs of 0.721 and 0.657 at best. 1-year mortality in the MITRALITY Score quartiles across the total cohort was 0.8%, 1.3%, 10.5%, and 54.6%, respectively. Odds of mortality in MITRALITY Score quartile 4 as compared to quartile 1 were 143.02 [34.75; 588.57]. Survival analyses showed that the differences in outcomes between the MITRALITY Score quartiles remained even over a timeframe of 3 years post intervention (log rank: p<0.005). With each increase by 1% in the MITRALITY score, the respective proportional hazard ratio for 3-year survival was 1.06 [1.05, 1.07] (Cox regression, p<0.05). Conclusion The MITRALITY Score is a novel, internally and externally validated ML-based tool for risk stratification of patients prior to TMVR. These findings may potentially allow for more precise design of future clinical trials, may enable novel treatment strategies tailored to populations of specific risk and thereby serve future daily clinical practice. FUNDunding Acknowledgement Type of funding sources: None. Summary Figure

Download Full-text

Machine-learning model derived gene signature predictive of paclitaxel survival benefit in gastric cancer: results from the randomised phase III SAMIT trial

Gut ◽

10.1136/gutjnl-2021-324060 ◽

2021 ◽

pp. gutjnl-2021-324060

Author(s):

Raghav Sundar ◽

Nesaretnam Barr Kumarakulasinghe ◽

Yiong Huak Chan ◽

Kazuhiro Yoshida ◽

Takaki Yoshikawa ◽

...

Keyword(s):

Machine Learning ◽

Gastric Cancer ◽

Random Forest ◽

Survival Benefit ◽

Validation Cohort ◽

External Validation ◽

Gene Signature ◽

Learning Model ◽

Phase Iii ◽

Machine Learning Model

ObjectiveTo date, there are no predictive biomarkers to guide selection of patients with gastric cancer (GC) who benefit from paclitaxel. Stomach cancer Adjuvant Multi-Institutional group Trial (SAMIT) was a 2×2 factorial randomised phase III study in which patients with GC were randomised to Pac-S-1 (paclitaxel +S-1), Pac-UFT (paclitaxel +UFT), S-1 alone or UFT alone after curative surgery.DesignThe primary objective of this study was to identify a gene signature that predicts survival benefit from paclitaxel chemotherapy in GC patients. SAMIT GC samples were profiled using a customised 476 gene NanoString panel. A random forest machine-learning model was applied on the NanoString profiles to develop a gene signature. An independent cohort of metastatic patients with GC treated with paclitaxel and ramucirumab (Pac-Ram) served as an external validation cohort.ResultsFrom the SAMIT trial 499 samples were analysed in this study. From the Pac-S-1 training cohort, the random forest model generated a 19-gene signature assigning patients to two groups: Pac-Sensitive and Pac-Resistant. In the Pac-UFT validation cohort, Pac-Sensitive patients exhibited a significant improvement in disease free survival (DFS): 3-year DFS 66% vs 40% (HR 0.44, p=0.0029). There was no survival difference between Pac-Sensitive and Pac-Resistant in the UFT or S-1 alone arms, test of interaction p<0.001. In the external Pac-Ram validation cohort, the signature predicted benefit for Pac-Sensitive (median PFS 147 days vs 112 days, HR 0.48, p=0.022).ConclusionUsing machine-learning techniques on one of the largest GC trials (SAMIT), we identify a gene signature representing the first predictive biomarker for paclitaxel benefit.Trial registration numberUMIN Clinical Trials Registry: C000000082 (SAMIT); ClinicalTrials.gov identifier, 02628951 (South Korean trial)

Download Full-text

Development and Validation of a Simple-to-Use Nomogram to Predict Early Death in Metastatic Pancreatic Adenocarcinoma

Frontiers in Oncology ◽

10.3389/fonc.2021.729175 ◽

2021 ◽

Vol 11 ◽

Author(s):

Zhong Zhang ◽

Juan Pu ◽

Haijun Zhang

Keyword(s):

Risk Factors ◽

Pancreatic Adenocarcinoma ◽

Validation Cohort ◽

External Validation ◽

Significant Risk ◽

Early Death ◽

Seer Database ◽

Data Set ◽

Internal Validation ◽

Chinese Cohort

BackgroundPancreatic adenocarcinoma (PCa) is a highly aggressive malignancy with high risk of early death (survival time ≤3 months). The present study aimed to identify associated risk factors and develop a simple-to-use nomogram to predict early death in metastatic PCa patients.MethodsPatients diagnosed with metastatic PCa between 2010 and 2015 from the Surveillance, Epidemiology, and End Results (SEER) database were collected for model construction and internal validation. An independent data set was obtained from China for external validation. Independent risk variables contributed to early death were identified by logistic regression models, which were then used to construct a nomogram. Internal and external validation was performed to evaluate the nomogram using calibration curves and the receiver operating characteristic curves.ResultsA total of 19,464 patients in the SEER cohort and 67 patients in the Chinese cohort were included. Patients from the SEER database were randomly divided into the training cohort (n = 13,040) and internal validation cohort (n = 6,424). Patients in the Chinese cohort were selected for the external validation cohort. Overall, 10,484 patients experienced early death in the SEER cohort and 35 in the Chinese cohort. A reliable nomogram was constructed on the basis of 11 significant risk factors. Internal validation and external validation of the nomogram showed high accuracy in predicting early death. Decision curve analysis demonstrated that this predictive nomogram had excellent and potential clinical applicability.ConclusionThe nomogram provided a simple-to-use tool to distinguish early death in patients with metastatic PCa, assisting clinicians in implementing individualized treatment regimens.

Download Full-text

Development and Validation of a Prognostic Model for Transplant-Associated Thrombotic Microangiopathy Following Allogeneic Hematopoietic Stem Cell Transplantation

Blood ◽

10.1182/blood-2020-138408 ◽

2020 ◽

Vol 136 (Supplement 1) ◽

pp. 16-17

Author(s):

Peng Zhao ◽

Ye-Jun Wu ◽

Qing-Yuan Qu ◽

Shan Chong ◽

Xiao-Wan Sun ◽

...

Keyword(s):

Overall Survival ◽

Prognostic Model ◽

Validation Cohort ◽

Risk Group ◽

External Validation ◽

Internal Validation ◽

Hematopoietic Stem ◽

Derivation Cohort ◽

Peking University ◽

Score Model

Introduction Transplant-associated thrombotic microangiopathy (TA-TMA) is a potentially life-threatening complication of allogeneic hematopoietic stem cell transplantation (allo-HSCT), which can result in multiorgan injury and increased risk for mortality. Renewed interest has emerged in the prognostication of TA-TMA with the development of novel diagnostic and management algorithms. Our previous study reported an adverse outcome in patients with TA-TMA and concomitant acute graft-versus-host disease (Eur J Haematol, 2018). However, information on markers for the early identification of severe cases remains limited. Therefore, this study is concentrated on the development and validation of a prognostic model for TA-TMA, which might facilitate risk stratification and contribute to individualized management. Methods Patients receiving allo-HSCT in Peking University People's Hospital with 1) a diagnosis of microangiopathic hemolytic anemia (MAHA) or 2) evidence of microangiopathy were retrospectively identified from 2010 to 2018. The diagnosis of TA-TMA was reviewed according to the Overall-TMA criteria (Transplantation, 2010). Patients without fulfillment of the diagnostic criteria or complicated with other causes of MAHA were excluded from analysis. Prognostic factors for TA-TMA were determined among patients receiving HSCT between 2010 and 2014 (derivation cohort). Candidate predictors (univariate P < 0.1) were included in the multivariate analysis using a backward stepwise logistic regression model. A risk score model was then established according to the regression coefficient of each independent prognostic factor. The performance of this predictive model was evaluated through internal validation (bootstrap method with 1000 repetitions) and external temporal validation performed on data from those who received HSCT between 2015 and 2018 (validation cohort). Results 5337 patients underwent allo-HSCT at Peking University Institute of Hematology from 2010 to 2018. A total of 1255 patients with a diagnosis of MAHA and/or evidence of microangiopathy were retrospectively identified, among whom 493 patients met the inclusion criteria for this analysis (269 in the derivation cohort and 224 in the validation cohort). The median age at the time of TA-TMA diagnosis was 28 (IQR: 17-41) years. The median duration from the time of transplantation to the diagnosis of TA-TMA was 63 (IQR: 38-121) days. The 6-month overall survival rate was 42.2% (208/493), and the 1-year overall survival rate was 45.0% (222/493). In the derivation cohort, patient age (≥35 years), anemia (hemoglobin <70 g/L), severe thrombocytopenia (platelet count <15,000/μL), elevated lactic dehydrogenase (serum LDH >800 U/L) and elevated total bilirubin (TBIL >1.5*ULN) were identified by multivariate analysis as independent prognostic factors for the 6-month outcome of TA-TMA. A risk score model was constructed according to the regression coefficients (Table 1), and patients were stratified into a low-risk group (0-1 points), an intermediate-risk group (2-4 points) and a high-risk group (5-6 points). The Kaplan-Meier estimations of overall survival separated well between these risk groups (Figure 1). The prognostic model showed significant discriminatory capacity, with a cross-validated c-index of 0.770 (95%CI, 0.714-0.826) in the internal validation and 0.768 (95%CI, 0.707-0.829) in the external validation cohort. The calibration plots also indicated a good correlation between model-predicted and observed probabilities. Conclusions A prognostic model for TA-TMA incorporating several baseline laboratory factors was developed and evaluated, which demonstrated significant predictive capacity through internal and external validation. This predictive model might facilitate prognostication of TA-TMA and contribute to early identification of patients at higher risk for adverse outcomes. Further study may focus on whether these high-risk patients could benefit from early application of specific management. Disclosures No relevant conflicts of interest to declare.

Download Full-text

Development and validation of an online model to predict critical COVID-19 with immune-inflammatory parameters

Journal of Intensive Care ◽

10.1186/s40560-021-00531-1 ◽

2021 ◽

Vol 9 (1) ◽

Author(s):

Yue Gao ◽

Lingxi Chen ◽

Jianhua Chi ◽

Shaoqing Zeng ◽

Xikang Feng ◽

...

Keyword(s):

Machine Learning ◽

Critical Illness ◽

Prediction Model ◽

Validation Cohort ◽

External Validation ◽

Validation Dataset ◽

Support Vector ◽

Internal Validation ◽

Inflammatory Parameters ◽

Severity Prediction

Abstract Background Immune and inflammatory dysfunction was reported to underpin critical COVID-19(coronavirus disease 2019). We aim to develop a machine learning model that enables accurate prediction of critical COVID-19 using immune-inflammatory features at admission. Methods We retrospectively collected 2076 consecutive COVID-19 patients with definite outcomes (discharge or death) between January 27, 2020 and March 30, 2020 from two hospitals in China. Critical illness was defined as admission to intensive care unit, receiving invasive ventilation, or death. Least Absolute Shrinkage and Selection Operator (LASSO) was applied for feature selection. Five machine learning algorithms, including Logistic Regression (LR), Support Vector Machine (SVM), Gradient Boosted Decision Tree (GBDT), K-Nearest Neighbor (KNN), and Neural Network (NN) were built in a training dataset, and assessed in an internal validation dataset and an external validation dataset. Results Six features (procalcitonin, [T + B + NK cell] count, interleukin 6, C reactive protein, interleukin 2 receptor, T-helper lymphocyte/T-suppressor lymphocyte) were finally used for model development. Five models displayed varying but all promising predictive performance. Notably, the ensemble model, SPMCIIP (severity prediction model for COVID-19 by immune-inflammatory parameters), derived from three contributive algorithms (SVM, GBDT, and NN) achieved the best performance with an area under the curve (AUC) of 0.991 (95% confidence interval [CI] 0.979–1.000) in internal validation cohort and 0.999 (95% CI 0.998–1.000) in external validation cohort to identify patients with critical COVID-19. SPMCIIP could accurately and expeditiously predict the occurrence of critical COVID-19 approximately 20 days in advance. Conclusions The developed online prediction model SPMCIIP is hopeful to facilitate intensive monitoring and early intervention of high risk of critical illness in COVID-19 patients. Trial registration This study was retrospectively registered in the Chinese Clinical Trial Registry (ChiCTR2000032161). Graphical abstracthelper lymphocytve vv

Download Full-text

Abstract WP62: Delayed Appearance of DWI Lesions in Clinically Suspected DWI-Negative Stroke: DWI-CONVERSION Score

Stroke ◽

10.1161/str.51.suppl_1.wp62 ◽

2020 ◽

Vol 51 (Suppl_1) ◽

Author(s):

Kitae Kim ◽

Beom Joon Kim ◽

Jaewon Huh ◽

Seong Kyu Yang ◽

Moon-ku Han ◽

...

Keyword(s):

Validation Cohort ◽

External Validation ◽

Stroke Patients ◽

Derivation Cohort ◽

C Statistic ◽

Abcd2 Score ◽

Medical Histories ◽

Suspected Stroke ◽

Comparison Table

Introduction: Our aim was to determine the prevalence and factors associated with delayed appearance of DWI lesion among initially DWI-negative clinically suspected stroke patients in the follow-up DWIs during in-hospital care. Method: Among 5271 patients admitted to stroke unit as clinically suspected stroke/TIA within 7 days from symptom onset in our hospital via ER for 2010~2017, we selected subjects based on the following criteria 1) initial negative DWI (n=827), 2) follow-up DWI within 14 days (n=751). Then, we excluded 57 cases (hemorrhagic cases (n=4), cerebral angiography studies between MRIs (n=53)). Finally-included 694 cases were divided into two cohorts for temporal external validation (2010~2015 (n=488) as derivation; 2016~2017 (n=206) as validation). Results: Of 5271 cases, 827 cases (15.7%) showed initial negative DWI. In 694 finally-included cases, 22.5% (n=156) showed delayed appearance of DWI lesion. In derivation cohort, factors showing significant relationship with positive conversion comprised: medical histories such as atrial fibrillation (aOR 6.17, 3.23-12.01); symptoms including objective hemiparesis (aOR 4.39, 1.90-10.32) (Table 1). These factors were used to construct DWI-CONVERSION score (Table 2a). Its c-statistic was 0.813 in derivation cohort and 0.808 in validation cohort, which is significantly higher than that of ABCD2 score in validation cohort (c-statistic=0.678; P<0.01 for comparison; Table 2b). Conclusion: We identified prevalence and clinical factors significantly associated with delayed appearance of DWI lesions in clinically suspicious stroke patients. DWI-CONVERSION score is a simple tool to predict it.

Download Full-text

Machine learning based early warning system enables accurate mortality risk prediction for COVID-19

Nature Communications ◽

10.1038/s41467-020-18684-2 ◽

2020 ◽

Vol 11 (1) ◽

Cited By ~ 2

Author(s):

Yue Gao ◽

Guang-Yao Cai ◽

Wei Fang ◽

Hua-Yi Li ◽

Si-Yuan Wang ◽

...

Keyword(s):

Machine Learning ◽

Risk Prediction ◽

Mortality Risk ◽

Validation Cohort ◽

External Validation ◽

Warning System ◽

Support Vector ◽

Internal Validation ◽

Machine Learning Methods ◽

Boosted Decision Tree

Abstract Soaring cases of coronavirus disease (COVID-19) are pummeling the global health system. Overwhelmed health facilities have endeavored to mitigate the pandemic, but mortality of COVID-19 continues to increase. Here, we present a mortality risk prediction model for COVID-19 (MRPMC) that uses patients’ clinical data on admission to stratify patients by mortality risk, which enables prediction of physiological deterioration and death up to 20 days in advance. This ensemble model is built using four machine learning methods including Logistic Regression, Support Vector Machine, Gradient Boosted Decision Tree, and Neural Network. We validate MRPMC in an internal validation cohort and two external validation cohorts, where it achieves an AUC of 0.9621 (95% CI: 0.9464–0.9778), 0.9760 (0.9613–0.9906), and 0.9246 (0.8763–0.9729), respectively. This model enables expeditious and accurate mortality risk stratification of patients with COVID-19, and potentially facilitates more responsive health systems that are conducive to high risk COVID-19 patients.

Download Full-text

A Computed Tomography Radiomics-Based Prediction Model on Interstitial Lung Disease in Anti-MDA5-Positive Dermatomyositis

Frontiers in Medicine ◽

10.3389/fmed.2021.768052 ◽

2021 ◽

Vol 8 ◽

Author(s):

Wenwen Xu ◽

Wanlong Wu ◽

Yu Zheng ◽

Zhiwei Chen ◽

Xinwei Tao ◽

...

Keyword(s):

Computed Tomography ◽

Interstitial Lung Disease ◽

Prediction Model ◽

Lung Disease ◽

Cox Regression ◽

Validation Cohort ◽

External Validation ◽

High Resolution Computed Tomography ◽

Internal Validation ◽

Derivation Cohort

Objectives: Anti-melanoma differentiation-associated gene 5-positive dermatomyositis-associated interstitial lung disease (MDA5+ DM-ILD) is a life-threatening disease. The current study aimed to quantitatively assess the pulmonary high-resolution computed tomography (HRCT) images of MDA5+ DM-ILD by applying the radiomics approach and establish a multidimensional risk prediction model for the 6-month mortality.Methods: This retrospective study was conducted in 228 patients from two centers, namely, a derivation cohort and a longitudinal internal validation cohort in Renji Hospital, as well as an external validation cohort in Guangzhou. The derivation cohort was randomly divided into training and testing sets. The primary outcome was 6-month all-cause mortality since the time of admission. Baseline pulmonary HRCT images were quantitatively analyzed by radiomics approach, and a radiomic score (Rad-score) was generated. Clinical predictors selected by univariable Cox regression were further incorporated with the Rad-score, to enhance the prediction performance of the final model (Rad-score plus model). In parallel, an idiopathic pulmonary fibrosis (IPF)-based visual CT score and ILD-GAP score were calculated as comparators.Results: The Rad-score was significantly associated with the 6-month mortality, outperformed the traditional visual score and ILD-GAP score. The Rad-score plus model was successfully developed to predict the 6-month mortality, with C-index values of 0.88 [95% confidence interval (CI), 0.79–0.96] in the training set (n = 121), 0.88 (95%CI, 0.71–1.0) in the testing set (n = 31), 0.83 (95%CI, 0.68–0.98) in the internal validation cohort (n = 44), and 0.84 (95%CI, 0.64–1.0) in the external validation cohort (n = 32).Conclusions: The radiomic feature was an independent and reliable prognostic predictor for MDA5+ DM-ILD.

Download Full-text

A Novel Prognostic Scoring System of Intrahepatic Cholangiocarcinoma With Machine Learning Basing on Real-World Data

Frontiers in Oncology ◽

10.3389/fonc.2020.576901 ◽

2021 ◽

Vol 10 ◽

Author(s):

Zhizhen Li ◽

Lei Yuan ◽

Chen Zhang ◽

Jiaxing Sun ◽

Zeyuan Wang ◽

...

Keyword(s):

Machine Learning ◽

Scoring System ◽

Cox Regression ◽

Validation Cohort ◽

External Validation ◽

Machine Learning Techniques ◽

Internal Validation ◽

Learning Techniques ◽

Staging Systems ◽

8Th Edition

Background and ObjectivesCurrently, the prognostic performance of the staging systems proposed by the 8th edition of the American Joint Committee on Cancer (AJCC 8th) and the Liver Cancer Study Group of Japan (LCSGJ) in resectable intrahepatic cholangiocarcinoma (ICC) remains controversial. The aim of this study was to use machine learning techniques to modify existing ICC staging strategies based on clinical data and to demonstrate the accuracy and discrimination capacity in prognostic prediction.Patients and MethodsThis is a retrospective study based on 1,390 patients who underwent surgical resection for ICC at Eastern Hepatobiliary Surgery Hospital from 2007 to 2015. External validation was performed for patients from 2015 to 2017. The ensemble of three machine learning algorithms was used to select the most important prognostic factors and stepwise Cox regression was employed to derive a modified scoring system. The discriminative ability and predictive accuracy were assessed using the Concordance Index (C-index) and Brier Score (BS). The results were externally validated through a cohort of 42 patients operated on from the same institution.ResultsSix independent prognosis factors were selected and incorporated in the modified scoring system, including carcinoembryonic antigen, carbohydrate antigen 19-9, alpha-fetoprotein, prealbumin, T and N of ICC staging category in 8th edition of AJCC. The proposed scoring system showed a more favorable discriminatory ability and model performance than the AJCC 8th and LCSGJ staging systems, with a higher C-index of 0.693 (95% CI, 0.663–0.723) in the internal validation cohort and 0.671 (95% CI, 0.602–0.740) in the external validation cohort, which was then confirmed with lower BS (0.103 in internal validation cohort and 0.169 in external validation cohort). Meanwhile, machine learning techniques for variable selection together with stepwise Cox regression for survival analysis shows a better prognostic accuracy than using stepwise Cox regression method only.ConclusionsThis study put forward a modified ICC scoring system based on prognosis factors selection incorporated with machine learning, for individualized prognosis evaluation in patients with ICC.

Download Full-text