Feature selection and prediction of treatment failure in tuberculosis.
ABSTRACT: BACKGROUND:Tuberculosis is a major cause of morbidity and mortality in the developing world. Drug resistance, which is predicted to rise in many countries worldwide, threatens tuberculosis treatment and control. OBJECTIVE:To identify features associated with treatment failure and to predict which patients are at highest risk of treatment failure. METHODS:On a multi-country dataset managed by the National Institute of Allergy and Infectious Diseases we applied various machine learning techniques to identify factors statistically associated with treatment failure and to predict treatment failure based on baseline demographic and clinical characteristics alone. RESULTS:The complete-case analysis database consisted of 587 patients (68% males) with a median (p25-p75) age of 40 (30-51) years. Treatment failure occurred in approximately one fourth of the patients. The features most associated with treatment failure were patterns of drug sensitivity, imaging findings, findings in the microscopy Ziehl-Nielsen stain, education status, and employment status. The most predictive model was forward stepwise selection (AUC: 0.74), although most models performed at or above AUC 0.7. A sensitivity analysis using the 643 original patients filling the missing values with multiple imputation showed similar predictive features and generally increased predictive performance. CONCLUSION:Machine learning can help to identify patients at higher risk of treatment failure. Closer monitoring of these patients may decrease treatment failure rates and prevent emergence of antibiotic resistance. The use of inexpensive basic demographic and clinical features makes this approach attractive in low and middle-income countries.
Project description:BACKGROUND:Meningiomas are stratified according to tumor grade and extent of resection, often in isolation of other clinical variables. Here, we use machine learning (ML) to integrate demographic, clinical, radiographic and pathologic data to develop predictive models for meningioma outcomes. METHODS AND FINDINGS:We developed a comprehensive database containing information from 235 patients who underwent surgery for 257 meningiomas at a single institution from 1990 to 2015. The median follow-up was 4.3 years, and resection specimens were re-evaluated according to current diagnostic criteria, revealing 128 WHO grade I, 104 grade II and 25 grade III meningiomas. A series of ML algorithms were trained and tuned by nested resampling to create models based on preoperative features, conventional postoperative features, or both. We compared different algorithms' accuracy as well as the unique insights they offered into the data. Machine learning models restricted to preoperative information, such as patient demographics and radiographic features, had similar accuracy for predicting local failure (AUC = 0.74) or overall survival (AUC = 0.68) as models based on meningioma grade and extent of resection (AUC = 0.73 and AUC = 0.72, respectively). Integrated models incorporating all available demographic, clinical, radiographic and pathologic data provided the most accurate estimates (AUC = 0.78 and AUC = 0.74, respectively). From these models, we developed decision trees and nomograms to estimate the risks of local failure or overall survival for meningioma patients. CONCLUSIONS:Clinical information has been historically underutilized in the prediction of meningioma outcomes. Predictive models trained on preoperative clinical data perform comparably to conventional models trained on meningioma grade and extent of resection. Combination of all available information can help stratify meningioma patients more accurately.
Project description:Tinnitus is a complex condition that is associated with major psychological and economic impairments - partly through various comorbidities such as depression. Understanding the interaction between tinnitus and depression may thus improve either symptom cluster's prevention, diagnosis and treatment. In this study, we developed and validated a machine learning model to predict depression severity after outpatient therapy (T1) based on variables obtained before therapy (T0). 1,490 patients with chronic tinnitus (comorbid major depressive disorder: 52.2%) who completed a 7-day multimodal treatment encompassing tinnitus-specific components, cognitive behavioural therapy, physiotherapy and informational counselling were included. 185 variables were extracted from self-report questionnaires and socio-demographic data acquired at T0. We used 11 classification methods to train models that reliably separate between subclinical and clinical depression at T1 as measured by the general depression questionnaire. To ensure highly predictive and robust classifiers, we tuned algorithm hyperparameters in a 10-fold cross-validation scheme. To reduce model complexity and improve interpretability, we wrapped model training around an incremental feature selection mechanism that retained features that contributed to model prediction. We identified a LASSO model that included all 185 features to yield highest predictive performance (AUC = 0.87?±?0.04). Through our feature selection wrapper, we identified a LASSO model with good trade-off between predictive performance and interpretability that used only 6 features (AUC = 0.85?±?0.05). Thus, predictive machine learning models can lead to a better understanding of depression in tinnitus patients, and contribute to the selection of suitable therapeutic strategies and concise and valid questionnaire design for patients with chronic tinnitus with or without comorbid major depressive disorder.
Project description:The aim of this study is to predict early distant failure in early stage non-small cell lung cancer (NSCLC) treated with stereotactic body radiation therapy (SBRT) using clinical parameters by machine learning algorithms.The dataset used in this work includes 81 early stage NSCLC patients with at least 6months of follow-up who underwent SBRT between 2006 and 2012 at a single institution. The clinical parameters (n=18) for each patient include demographic parameters, tumor characteristics, treatment fraction schemes, and pretreatment medications. Three predictive models were constructed based on different machine learning algorithms: (1) artificial neural network (ANN), (2) logistic regression (LR) and (3) support vector machine (SVM). Furthermore, to select an optimal clinical parameter set for the model construction, three strategies were adopted: (1) clonal selection algorithm (CSA) based selection strategy; (2) sequential forward selection (SFS) method; and (3) statistical analysis (SA) based strategy. 5-cross-validation is used to validate the performance of each predictive model. The accuracy was assessed by area under the receiver operating characteristic (ROC) curve (AUC), sensitivity and specificity of the system was also evaluated.The AUCs for ANN, LR and SVM were 0.75, 0.73, and 0.80, respectively. The sensitivity values for ANN, LR and SVM were 71.2%, 72.9% and 83.1%, while the specificity values for ANN, LR and SVM were 59.1%, 63.6% and 63.6%, respectively. Meanwhile, the CSA based strategy outperformed SFS and SA in terms of AUC, sensitivity and specificity.Based on clinical parameters, the SVM with the CSA optimal parameter set selection strategy achieves better performance than other strategies for predicting distant failure in lung SBRT patients.
Project description:BACKGROUND:Multiple organ failure (MOF) is a serious complication of moderately severe (MASP) and severe acute pancreatitis (SAP). This study aimed to develop and assess three machine-learning models to predict MOF. METHODS:Patients with MSAP and SAP who were admitted from July 2014 to June 2017 were included. Firstly, parameters with significant differences between patients with MOF and without MOF were screened out by univariate analysis. Then, support vector machine (SVM), logistic regression analysis (LRA) and artificial neural networks (ANN) models were constructed based on these factors, and five-fold cross-validation was used to train each model. RESULTS:A total of 263 patients were enrolled. Univariate analysis screened out sixteen parameters referring to blood volume, inflammatory, coagulation and renal function to construct machine-learning models. The predictive efficiency of the optimal combinations of features by SVM, LRA, and ANN was almost equal (AUC = 0.840, 0.832, and 0.834, respectively), as well as the Acute Physiology and Chronic Health Evaluation II score (AUC = 0.814, P > 0.05). The common important predictive factors were HCT, K-time, IL-6 and creatinine in three models. CONCLUSIONS:Three machine-learning models can be efficient prognostic tools for predicting MOF in MSAP and SAP. ANN is recommended, which only needs four common parameters.
Project description:PURPOSE:Reliable and accurate predictive models are necessary to drive the success of radiomics. Our aim was to identify the optimal radiomics-based machine learning method for isocitrate dehydrogenase (IDH) genotype prediction in diffuse gliomas. METHODS:Eight classical machine learning methods were evaluated in terms of their stability and performance for pre-operative IDH genotype prediction. A total of 126 patients were enrolled for analysis. Overall, 704 radiomic features extracted from the pre-operative MRI images were analyzed. The patients were randomly assigned to either the training set or the validation set at a ratio of 2:1. Feature selection and classification model training were done using the training set, whereas the predictive performance and stability of the model were independently assessed using the validation set. RESULTS:Random Forest (RF) showed high predictive performance (accuracy 0.885 ± 0.041, AUC 0.931 ± 0.036), whereas neural network (NN) (accuracy 0.829 ± 0.064, AUC 0.878 ± 0.052) and flexible discriminant analysis (FDA) (accuracy 0.851 ± 0.049, AUC 0.875 ± 0.057) displayed low predictive performance. With regard to stability, RF also showed high robustness against data perturbation (relative standard deviations, RSD 3.87%). CONCLUSIONS:RF is a promising machine learning method in predicting IDH genotype. Development of an accurate and reliable model can assist in the initial diagnostic evaluation and treatment planning for diffuse glioma patients.
Project description:BACKGROUND:Urinary protein quantification is critical for assessing the severity of chronic kidney disease (CKD). However, the current procedure for determining the severity of CKD is completed through evaluating 24-h urinary protein, which is inconvenient during follow-up. OBJECTIVE:To quickly predict the severity of CKD using more easily available demographic and blood biochemical features during follow-up, we developed and compared several predictive models using statistical, machine learning and neural network approaches. METHODS:The clinical and blood biochemical results from 551 patients with proteinuria were collected. Thirteen blood-derived tests and 5 demographic features were used as non-urinary clinical variables to predict the 24-h urinary protein outcome response. Nine predictive models were established and compared, including logistic regression, Elastic Net, lasso regression, ridge regression, support vector machine, random forest, XGBoost, neural network and k-nearest neighbor. The AU-ROC, sensitivity (recall), specificity, accuracy, log-loss and precision of each of the models were evaluated. The effect sizes of each variable were analysed and ranked. RESULTS:The linear models including Elastic Net, lasso regression, ridge regression and logistic regression showed the highest overall predictive power, with an average AUC and a precision above 0.87 and 0.8, respectively. Logistic regression ranked first, reaching an AUC of 0.873, with a sensitivity and specificity of 0.83 and 0.82, respectively. The model with the highest sensitivity was Elastic Net (0.85), while XGBoost showed the highest specificity (0.83). In the effect size analyses, we identified that ALB, Scr, TG, LDL and EGFR had important impacts on the predictability of the models, while other predictors such as CRP, HDL and SNA were less important. CONCLUSIONS:Blood-derived tests could be applied as non-urinary predictors during outpatient follow-up. Features in routine blood tests, including ALB, Scr, TG, LDL and EGFR levels, showed predictive ability for CKD severity. The developed online tool can facilitate the prediction of proteinuria progress during follow-up in clinical practice.
Project description:We examined the hypothesis that plasma biomarkers and concomitant clinical findings after initial glucocorticoid therapy can accurately predict failure of graft-versus-host-disease (GVHD) treatment and mortality. We analyzed plasma samples and clinical data in 165 patients after 14 days of glucocorticoid therapy and used logistic regression and areas under receiver-operating characteristic curves (AUC) to evaluate associations with treatment failure and nonrelapse mortality (NRM). Initial treatment of GVHD was unsuccessful in 49 patients (30%). For predicting GVHD treatment failure, the best clinical combination (total serum bilirubin and skin GVHD stage: AUC, .70) was competitive with the best biomarker combination (T cell immunoglobulin and mucin domain 3 [TIM3] and [interleukin 1 receptor family encoded by the IL1RL1 gene, ST2]: AUC, .73). The combination of clinical features and biomarker results offered only a slight improvement (AUC, .75). For predicting NRM at 1 year, the best clinical predictor (total serum bilirubin: AUC, .81) was competitive with the best biomarker combination (TIM3 and soluble tumor necrosis factor receptor-1 [sTNFR1]: AUC, .85). The combination offered no improvement (AUC, .85). Infection was the proximate cause of death in virtually all patients. We conclude that after 14 days of glucocorticoid therapy, clinical findings (serum bilirubin, skin GVHD) and plasma biomarkers (TIM3, ST2, sTNFR1) can predict failure of GVHD treatment and NRM. These biomarkers reflect counter-regulatory mechanisms and provide insight into the pathophysiology of GVHD reactions after glucocorticoid treatment. The best predictive models, however, exhibit inadequate positive predictive values for identifying high-risk GVHD cohorts for investigational trials, as only a minority of patients with high-risk GVHD would be identified and most patients would be falsely predicted to have adverse outcomes.
Project description:This study was designed to evaluate the predictive performance of 18F-fluorodeoxyglucose positron emission tomography (PET)-based radiomic features for local control of esophageal cancer treated with concurrent chemoradiotherapy (CRT). For each of the 30 patients enrolled, 440 radiomic features were extracted from both pre-CRT and mid-CRT PET images. The top 25 features with the highest areas under the receiver operating characteristic curve for identifying local control status were selected as discriminative features. Four machine-learning methods, random forest (RF), support vector machine, logistic regression, and extreme learning machine, were used to build predictive models with clinical features, radiomic features or a combination of both. An RF model incorporating both clinical and radiomic features achieved the best predictive performance, with an accuracy of 93.3%, a specificity of 95.7%, and a sensitivity of 85.7%. Based on risk scores of local failure predicted by this model, the 2-year local control rate and PFS rate were 100.0% (95% CI 100.0-100.0%) and 52.2% (31.8-72.6%) in the low-risk group and 14.3% (0.0-40.2%) and 0.0% (0.0-40.2%) in the high-risk group, respectively. This model may have the potential to stratify patients with different risks of local failure after CRT for esophageal cancer, which may facilitate the delivery of personalized treatment.
Project description:Based on a hollow-fiber system model of tuberculosis, we hypothesize that microbiologic failure and acquired drug resistance are primarily driven by low drug concentrations that result from pharmacokinetic variability.Clinical and pharmacokinetic data were prospectively collected from 142 tuberculosis patients in Western Cape, South Africa. Compartmental pharmacokinetic parameters of isoniazid, rifampin, and pyrazinamide were identified for each patient. Patients were then followed for up to 2 years. Classification and regression tree analysis was used to identify and rank clinical predictors of poor long-term outcome such as microbiologic failure or death, or relapse.Drug concentrations and pharmacokinetics varied widely between patients. Poor long-term outcomes were encountered in 35 (25%) patients. The 3 top predictors of poor long-term outcome, by rank of importance, were a pyrazinamide 24-hour area under the concentration-time curve (AUC) ? 363 mg·h/L, rifampin AUC ? 13 mg·h/L, and isoniazid AUC ? 52 mg·h/L. Poor outcomes were encountered in 32/78 patients with the AUC of at least 1 drug below the identified threshold vs 3/64 without (odds ratio = 14.14; 95% confidence interval, 4.08-49.08). Low rifampin and isoniazid peak and AUC concentrations preceded all cases of acquired drug resistance.Low drug AUCs are predictive of clinical outcomes in tuberculosis patients.
Project description:BACKGROUND:Current approaches to predict cardiovascular risk fail to identify many people who would benefit from preventive treatment, while others receive unnecessary intervention. Machine-learning offers opportunity to improve accuracy by exploiting complex interactions between risk factors. We assessed whether machine-learning can improve cardiovascular risk prediction. METHODS:Prospective cohort study using routine clinical data of 378,256 patients from UK family practices, free from cardiovascular disease at outset. Four machine-learning algorithms (random forest, logistic regression, gradient boosting machines, neural networks) were compared to an established algorithm (American College of Cardiology guidelines) to predict first cardiovascular event over 10-years. Predictive accuracy was assessed by area under the 'receiver operating curve' (AUC); and sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) to predict 7.5% cardiovascular risk (threshold for initiating statins). FINDINGS:24,970 incident cardiovascular events (6.6%) occurred. Compared to the established risk prediction algorithm (AUC 0.728, 95% CI 0.723-0.735), machine-learning algorithms improved prediction: random forest +1.7% (AUC 0.745, 95% CI 0.739-0.750), logistic regression +3.2% (AUC 0.760, 95% CI 0.755-0.766), gradient boosting +3.3% (AUC 0.761, 95% CI 0.755-0.766), neural networks +3.6% (AUC 0.764, 95% CI 0.759-0.769). The highest achieving (neural networks) algorithm predicted 4,998/7,404 cases (sensitivity 67.5%, PPV 18.4%) and 53,458/75,585 non-cases (specificity 70.7%, NPV 95.7%), correctly predicting 355 (+7.6%) more patients who developed cardiovascular disease compared to the established algorithm. CONCLUSIONS:Machine-learning significantly improves accuracy of cardiovascular risk prediction, increasing the number of patients identified who could benefit from preventive treatment, while avoiding unnecessary treatment of others.