Method For Outcome Prediction Research Articles

On Predicting Soccer Outcomes in the Greek League Using Machine Learning

The global expansion of the sports betting industry has brought the prediction of outcomes of sport events into the foreground of scientific research. In this work, soccer outcome prediction methods are evaluated, focusing on the Greek Super League. Data analysis, including data cleaning, Sequential Forward Selection (SFS), feature engineering methods and data augmentation is conducted. The most important features are used to train five machine learning models: k-Nearest Neighbor (k-NN), LogitBoost (LB), Support Vector Machine (SVM), Random Forest (RF) and CatBoost (CB). For comparative reasons, the best model is also tested on the English Premier League and the Dutch Eredivisie, exploiting data statistics from six seasons from 2014 to 2020. Convolutional neural networks (CNN) and transfer learning are also tested by encoding tabular data to images, using 10-fold cross-validation, after applying grid and randomized hyperparameter tuning: DenseNet201, InceptionV3, MobileNetV2 and ResNet101V2. This is the first time the Greek Super League is investigated in depth, providing important features and comparative performance between several machine and deep learning models, as well as between other leagues. Experimental results in all cases demonstrate that the most accurate prediction model is the CB, reporting 67.73% accuracy, while the Greek Super League is the most predictable league.

Will artificial intelligence help us in predicting outcomes in cardiac surgery?

Will artificial intelligence help us in predicting outcomes in cardiac surgery?

Association between SLICC/ACR damage index and outcomes for lupus patients after cardiac valve surgery.

Systemic lupus erythematosus (SLE) is associated with multi-organ damage including cardiac valve, which may need valvular operation. However, methods for outcome prediction and prosthetic valve selection are unclear in SLE patients undergoing cardiac valve surgery. Twenty-five SLE patients receiving valvular operation in a single institute between 2002 and 2020 were enrolled. Systemic Lupus International Collaborative Clinics/American College of Rheumatology Damage Index (SLICC/ACR damage index, SDI) was applied to evaluate the damage severity. Clinical outcomes were compared between patients with different SDI. The hospital survival rate was 88%, and long-term survival rate was 59.5% and 40.2% at 5 and 10 years. The median SDI was 4 (interquartile range 3–6) in our study, patients were then grouped into higher SDI (defined as SDI ≥ 5, n = 11) and lower SDI group (defined as SDI < 5, n = 14). The in-hospital survival rate (72.2% vs 100%, P = 0.074) and 5-year survival rate (18.2% vs 92.9%, P < 0.001) were lower in higher SDI group, compared to lower SDI group. SDI score was associated with long-term outcome for SLE patients receiving cardiac valve surgery. SDI ≥ 5 was associated with very poor long-term outcomes. This finding implicates that xenograft might be a reasonable choice for SLE patients with SDI ≥ 5.

Comparative evaluation of lethal outcome prediction methods in severely burned patients

Background: Regarding burn injury there are many approaches to assessing the possibility of death in severely burned patients. Despite the ease of use and the maximum prevalence of existing models, the assessment of the outcome in each of them is questionable, since the emphasis in different indices is on different indicators, avoiding the overall clinical picture of the disease.Objective: Comparative analysis of the effectiveness of methods for predicting a lethal outcome in patients with extensive skin burns.Material and мethods: Calculated characteristics of known in the literature and widely used Baux rules, Frank index, probit analysis and a new method of logistic regression were obtained and applied to evaluate the results of treatment of 282 adult patients with extensive skin burns, hospitalized in the Department of Anesthesiology and Intensive Care of the Thermal Injuries Unit, Saint-Petersburg I. I. Dzhanelidze Research Institute of Emergency Medicine in the period 2015–2021.Results: During the study a descriptive characteristic of methods for predicting a lethal outcome was obtained. Based on the data obtained, four-field contingency tables were compiled and a comparative analysis of the effectiveness of the models was carried out.Conclusion: Despite the high frequency of use of such forecasting methods as the Baux score, the Frank index and probit analysis in the combustiology practice, the results of the calculation indicate that these methods have low efficiency: the Baux score method accuracy of a lethal outcome prediction is 49.7%, the Frank index method – 41.5%, probit analysis method – 60%. The logistic regression model developed by us showed high efficiency compared to those presented earlier (the accuracy of predicting a lethal outcome was 93%), which gives grounds for recommending it for practical application.

A proof-of-concept study on mortality prediction with machine learning algorithms using burn intensive care data.

IntroductionBurn injuries are a common traumatic injury. Large burns have high mortality requiring intensive care and accurate mortality predictions. To assess if machine learning (ML) could improve predictions, ML algorithms were tested and compared with the original and revised Baux score.MethodsAdmission data and mortality outcomes were collected from patients at Uppsala University Hospital Burn Centre from 2002 to 2019. Prognostic variables were selected, ML algorithms trained and predictions assessed by analysis of the area under the receiver operating characteristic curve (AUC). Comparison was made with Baux scores using DeLong testResultsA total of 17 prognostic variables were selected from 92 patients. AUCs in leave-one-out cross-validation for a decision tree model, an extreme boosting model, a random forest model, a support-vector machine (SVM) model and a generalised linear regression model (GLM) were 0.83 (95% confidence interval [CI] = 0.72–0.94), 0.92 (95% CI = 0.84–1), 0.92 (95% CI = 0.84–1), 0.92 (95% CI = 0.84–1) and 0.84 (95% CI = 0.74–0.94), respectively. AUCs for the Baux score and revised Baux score were 0.85 (95% CI = 0.75–0.95) and 0.84 (95% CI = 0.74–0.94). No significant differences were observed when comparing ML algorithms with Baux score and revised Baux score. Secondary variable selection was made to analyse model performance.ConclusionThis proof-of-concept study showed initial credibility in using ML algorithms to predict mortality in burn patients. The sample size was small and future studies are needed with larger sample sizes, further variable selections and prospective testing of the algorithms.Lay SummaryBurn injuries are one of the most common traumatic injuries especially in countries with limited prevention and healthcare resources. To treat a patient with large burns who has been admitted to an intensive care unit, it is often necessary to assess the risk of a fatal outcome. Physicians traditionally use simplified scores to calculate risks. One commonly used score, the Baux score, uses age of the patient and the size of the burn to predict the risk of death. Adding the factor of inhalation injury, the score is then called the revised Baux score. However, there are a number of additional causes that can influence the risk of fatal outcomes that Baux scores do not take into account. Machine learning is a method of data modelling where the system learns to predict outcomes based on previous cases and is a branch of artificial intelligence. In this study we evaluated several machine learning methods for outcome prediction in patients admitted for burn injury. We gathered data on 93 patients at admission to the intensive care unit and our experiments show that machine learning methods can reach an accuracy comparable with Baux scores in calculating the risk of fatal outcomes. This study represents a proof of principle and future studies on larger patient series are required to verify our results as well as to evaluate the methods on patients in real-life situations.

Superiority of Leukemic Stem Cell-Based Minimal Residual Disease Assay to Traditional Multiparameter Flow Cytometry-Based Method for Relapse Prediction in AML Patients: A Prospective Study with Randomized Training and Validation Sets

Background: Minimal/measurable residual disease (MRD) determined by multiparameter flow cytometry (MFC) is an important variable for relapse prediction and treatment approach selection in patients with acute myeloid leukemia (AML). We aimed to investigate whether leukemia-stem cell (LSC)-based assay is superior to traditional MFC methods, including LAIP and D-F-N assays, for MRD evaluation in predicting clinical outcomes.Methods: In this cohort study, a total of 360 AML patients who received allogeneic stem cell transplantation (allo-SCT) between July 2018 and November 2019 were prospectively enrolled. The patients were randomized (1:1) and classified into a training set (n=180) and a validation set (n=180). Posttransplantation MRD were according to LSC based assay, mainly including a cocktail of CD7, CD11b, CD22, CD56, Tim-3, and CLL-1 on CD34 +CD38 - cells, and traditional assay determined by MFC, respectively.Findings: In the training set, patients were classified as LSC positive group (group A) and LSC negative group (group B) according to a cutoff value of CD34 +CD38 -cocktail + LSCs as 0.004%. Subjects in group A had a higher cumulative incidence of relapse (CIR, 42.7% vs. 2.6%, P<0.001) and comparable non-relapse mortality (NRM, 0% vs. 8.1%, P=0.154) compared with cases in group B, leading to inferior leukemia-free survival (LFS, 57.3% vs. 89.3%, P<0.001) and overall survival (OS, 70.8% vs. 90%, P=0.009) of cases in group A to group B. Multivariate analysis showed that positive LSCs after transplantation could independently predict CIR (P<0.001), LFS (P<0.001), and OS (P=0.021). The predictive value of positive LSCs following allo-HSCT for CIR (P<0.001), LFS (P<0.001), and OS (P=0.004) was further confirmed in the validation set. In the total case cohort, multivariate analysis also showed that positive LSCs after transplantation could independently predict CIR (HR=13.999, P<0.001), LFS (HR=5.429, P<0.001), and OS (HR=3.761, P=0.021). The total patients were classified into positive MRD and negative MRD groups according to the traditional MFC method. The results showed that the CIR of patients in the traditional MRD positive group was significantly higher than the CIR of patients in the traditional MRD negative group (44.9% vs. 7.3%, P<0.001), leading to lower LFS (55.1% vs. 85.6%, P<0.001) and OS (55.6% vs. 87.5%, P<0.001). Compared with MRD detected by the traditional MFC method, using LSCs for MRD evaluation has high sensitivity (66.7% vs. 43%), a high C-index (0.76 vs. 0.69) and a high Youden index (0.58 vs. 0.37). The median time from CD34 +CD38cocktail + LSC positivity to relapse was longer than the median time from traditional MRD positivity to relapse by 141.5 days (range, 18-465 days) vs. 64.5 days (range, 13-144 days) (P=0.003). The median level between CD34 +CD38 -cocktail + LSCs and traditional MRD detected by MFC was 0.0072% (range, 0.0007%-3.742%) and 0.16% (range, 0.01%-3.75%) (P<0.001).Interpretation: Our data suggest the superiority of LSC-based MRD assays such as higher sensitivity, low false negativity, and longer time for MRD positivity to relapse to traditional MFC MRD methods for outcome prediction in AML patients received allograft. [Display omitted] DisclosuresNo relevant conflicts of interest to declare.

An interpretable outcome prediction model based on electronic health records and hierarchical attention

Outcome prediction aims to predict the future health condition of patients from Electronic Health Record (EHR) data. Because of the sequential characteristic of EHR data, recurrent neural network (RNN)-based outcome prediction methods have achieved state-of-the-art results. However, the major drawback of RNN-based outcome prediction methods is lack of interpretability, which would lead to trust issues. Aiming at this problem, this paper proposes interpretable outcome prediction model with hierarchical attention (IoHAN), an interpretable outcome prediction model by leveraging attention mechanism. The main novelty of IoHAN is that it can pinpoint the fine-grained influence on the final prediction result of each medical component by decomposing the attention weights hierarchically into hospital visits, medical variables, and interactions between medical variables. We evaluated IoHAN on MIMIC-III, a large real-world EHR data set. The experiment results demonstrate that IoHAN can achieve higher prediction accuracy than state-of-the-art outcome prediction models. In addition, the hierarchical decomposed attention weights can interpret the prediction results in a more natural and understandable way.

Superiority of Leukemic Stem Cell-Based Minimal Residual Disease Assay to Traditional Multiparameter Flow Cytometry-Based Method for Relapse Prediction in AML Patients: A Prospective Study with Randomized Training and Validation Sets

Background: Minimal/measurable residual disease (MRD) determined by multiparameter flow cytometry (MFC) is an important variable for relapse prediction and treatment approach selection in patients with acute myeloid leukemia (AML). We aimed to investigate whether leukemia-stem cell (LSC)-based assay is superior to traditional MFC methods, including LAIP and D-F-N assays, for MRD evaluation in predicting clinical outcomes. Methods: In this cohort study, a total of 360 AML patients who received allogeneic stem cell transplantation (allo-SCT) between July 2018 and November 2019 were prospectively enrolled. The patients were randomized (1:1) and classified into a training set (n=180) and a validation set (n=180). Posttransplantation MRD were according to LSC based assay, mainly including a cocktail of CD7, CD11b, CD22, CD56, Tim-3, and CLL-1 on CD34+CD38- cells, and traditional assay determined by MFC, respectively. Findings: In the training set, patients were classified as LSC positive group (group A) and LSC negative group (group B) according to a cutoff value of CD34+CD38- cocktail+ LSCs as 0.004%. Subjects in group A had a higher cumulative incidence of relapse (CIR, 42.7% vs. 2.6%, P <0.001) and comparable non-relapse mortality (NRM, 0% vs. 8.1%, P =0.154) compared with cases in group B, leading to inferior leukemia-free survival (LFS, 57.3% vs. 89.3%, P <0.001) and overall survival (OS, 70.8% vs. 90%, P=0.009) of cases in group A to group B. Multivariate analysis showed that positive LSCs after transplantation could independently predict CIR (P <0.001), LFS (P <0.001), and OS (P=0.021). The predictive value of positive LSCs following allo-HSCT for CIR (P <0.001), LFS (P <0.001), and OS (P=0.004) was further confirmed in the validation set. In the total case cohort, multivariate analysis also showed that positive LSCs after transplantation could independently predict CIR (HR=13.999,P <0.001), LFS (HR=5.429, P <0.001), and OS (HR=3.761, P=0.021). The total patients were classified into positive MRD and negative MRD groups according to the traditional MFC method. The results showed that the CIR of patients in the traditional MRD positive group was significantly higher than the CIR of patients in the traditional MRD negative group (44.9% vs. 7.3%, P <0.001), leading to lower LFS (55.1% vs. 85.6%,P <0.001) and OS (55.6% vs. 87.5%, P <0.001). Compared with MRD detected by the traditional MFC method, using LSCs for MRD evaluation has high sensitivity (66.7% vs. 43%), a high C-index (0.76 vs. 0.69) and a high Youden index (0.58 vs . 0.37). The median time from CD34+CD38 cocktail+ LSC positivity to relapse was longer than the median time from traditional MRD positivity to relapse by 141.5 days (range, 18-465 days) vs . 64.5 days (range, 13-144 days) ( P =0.003). The median level between CD34+CD38-cocktail+ LSCs and traditional MRD detected by MFC was 0.0072% (range, 0.0007%-3.742%) and 0.16% (range, 0.01%-3.75%) (P <0.001). Interpretation: Our data suggest the superiority of LSC-based MRD assays such as higher sensitivity, low false negativity, and longer time for MRD positivity to relapse to traditional MFC MRD methods for outcome prediction in AML patients received allograft. Trial Registration: This study was registered as #ChiCTR1800016458 at http://www.chictr.org.cn/showproj.aspx?proj=27944. Funding: This work was partly supported by grants from the National Key Research and Development Program of China (2017YFA0104500), Beijing Municipal Science and Technology Commission (Z181100009618032), National Natural Science Foundation of China (Grant No. 81621001, 81530046, 81770189, and 81670186), and Peking University Clinical Scientist Program (BMU2019LCKXJ003). Declaration of Interest: None to declare. Ethical Approval: The study was performed in accordance with the Declaration of Helsinki and was approved by the local institutional review board. All donors and recipients gave written informed consent before enrollment.

Evolutionary algorithms and decision trees for predicting poor outcome after endovascular treatment for acute ischemic stroke

Despite the large overall beneficial effects of endovascular treatment in patients with acute ischemic stroke, severe disability or death still occurs in almost one-third of patients. These patients, who might not benefit from treatment, have been previously identified with traditional logistic regression models, which may oversimplify relations between characteristics and outcome, or machine learning techniques, which may be difficult to interpret. We developed and evaluated a novel evolutionary algorithm for fuzzy decision trees to accurately identify patients with poor outcome after endovascular treatment, which was defined as having a modified Rankin Scale score (mRS) higher or equal to 5. The created decision trees have the benefit of being comprehensible, easily interpretable models, making its predictions easy to explain to patients and practitioners. Insights in the reason for the predicted outcome can encourage acceptance and adaptation in practice and help manage expectations after treatment. We compared our proposed method to CART, the benchmark decision tree algorithm, on classification accuracy and interpretability. The fuzzy decision tree significantly outperformed CART: using 5-fold cross-validation with on average 1090 patients in the training set and 273 patients in the test set, the fuzzy decision tree misclassified on average 77 (standard deviation of 7) patients compared to 83 (±7) using CART. The mean number of nodes (decision and leaf nodes) in the fuzzy decision tree was 11 (±2) compared to 26 (±1) for CART decision trees. With an average accuracy of 72% and much fewer nodes than CART, the developed evolutionary algorithm for fuzzy decision trees might be used to gain insights into the predictive value of patient characteristics and can contribute to the development of more accurate medical outcome prediction methods with improved clarity for practitioners and patients.

Failure of Lactate Clearance Predicts the Outcome of Critically Ill Septic Patients.

Purpose: Early lactate clearance is an important parameter for prognosis assessment and therapy control in sepsis. Patients with a lactate clearance >0% might differ from patients with an inferior clearance in terms of intensive care management and outcomes. This study analyzes a large collective with regards to baseline risk distribution and outcomes. Methods: In total, 3299 patients were included in this analysis, consisting of 1528 (46%) ≤0% and 1771 (54%) >0% patients. The primary endpoint was intensive care unit (ICU) mortality. Multilevel logistic regression analyses were used to compare both groups: A baseline model (model 1) with lactate clearance as a fixed effect and ICU as a random effect was installed. For model 2, patient characteristics (model 2) were included. For model 3, intensive care treatment (mechanical ventilation and vasopressors) was added to the model. Models 1 and 2 were used to evaluate the primary and secondary outcomes, respectively. Model 3 was only used to evaluate the primary outcomes. Adjusted odds ratios (aORs) with respective 95% confidence intervals (CI) were calculated. Results: The cohorts had no relevant differences regarding the gender, BMI, age, heart rate, body temperature, and baseline lactate. Neither the primary infection focuses nor the ethnic background differed between both groups. In both groups, the most common infection sites were of pulmonary origin, the urinary tract, and the gastrointestinal tract. Patients with lactate clearance >0% evidenced lower sepsis-related organ failure assessment (SOFA) scores (7 ± 6 versus 9 ± 6; p < 0.001) and creatinine (1.53 ± 1.49 versus 1.80 ± 1.67; p < 0.001). The ICU mortality differed significantly (14% versus 32%), and remained this way after multivariable adjustment for patient characteristics and intensive care treatment (aOR 0.43 95% CI 0.36–0.53; p < 0.001). In the additional sensitivity analysis, the lack of lactate clearance was associated with a worse prognosis in each subgroup. Conclusion: In this large collective of septic patients, the 6 h lactate clearance is an independent method for outcome prediction.

Prognostic Impact of Somatic Copy Number Alterations in Childhood B-Lineage Acute Lymphoblastic Leukemia.

The high prevalence of relapse in pediatric B-lineage acute lymphoblastic leukemia (B-ALL) despite the improvements achieved using current risk stratification schemes, demands more accurate methods for outcome prediction. Here, we provide a concise overview about the key advances that have expanded our knowledge regarding the somatic defects across B-ALL genomes, particularly focusing on copy number alterations (CNAs) and their prognostic impact. The identification of commonly altered genes in B-ALL has inspired the development of risk classifiers based on copy number states such as the IKZF1plus and the United Kingdom (UK) ALL-CNA classifiers to improve outcome prediction in B-ALL. CNA-risk classifiers have emerged as effective tools to predict disease relapse; though, their clinical applications are yet to be transferred to routine practice.

Average pixel intensity method for prediction of outcome in secondary mitral regurgitation

BackgroundEchocardiographic grading of secondary mitral regurgitation (SMR) severity is challenging and involves multiple guideline-recommended parameters. We previously introduced the average pixel intensity (API) method for grading SMR. In this study,...

A Machine Learning Algorithm to Improve Risk Assessment for Patients with Sickle Cell Disease

Sickle cell disease (SCD) is a clinical syndrome that encompasses several different genotypes, the 3 most common being homozygosity for the bS allele (HbSS), compound heterozygosity of HbS and HbC (HbSC), and compound heterozygosity of HbS and HbSb thalassemia (HbSb+ or HbSb0 thalassemia). Generally, patients with HbSS and HbSb0 thalassemia genotypes have the most severe clinical manifestations, while patients with HbSC and HbSβ+-thalassemia are thought to be less severe. Within each of these genotypic groups, however, there are also substantial phenotypic differences. This heterogeneity makes it difficult to quantify the severity of the disease process and to guide therapeutics. As more intensive, high risk and costly treatments such as hematopoietic stem cell transplant and gene therapy are developing, the ability to assess patients at highest risk of early mortality becomes increasingly important. Integrating varied clinical, laboratory, and imaging markers for personalized risk prediction has been difficult, however, newer machine learning methods for outcome prediction take a more agnostic approach than traditional statistical methods and can detect complex, non-linear relationships in the data. In this study, we sought to apply machine learning methods to a well-characterized cohort of SCD patients followed at the National Institutes of Health in order to identify clinically meaningful subgroups of patients at highest risk of mortality. Between 2006 and 2017, 601 patients (age 35±13 years, 51% female) underwent echocardiogram, standard laboratory markers and hemoglobin electrophoresis resulting in 61 candidate variables. Among these patients, 488 had HbSS, 12 HbSb0 thalassemia, 80 HbSC, 20 HbSb+ thalassemia. All-cause mortality was ascertained by proxy interview, through medical records, and through the CDC National Death Index. Average follow-up time was 5 years and 130 patients were deceased. A random survival forest (RSF) algorithm followed by nested model selection and AIC Cox regression analysis identified 13 predictors of mortality (estimated right ventricular systolic pressure, peak tricuspid regurgitant (TR) velocity, mitral E velocity, septal and posterior wall thickness, IVC diameter, right atrial area, BUN, alkaline phosphatase, N-terminal-pro brain natriuretic peptide (BNP), creatinine, potassium and bicarbonate). This model performed better than individual clinical and laboratory variables with a C-statistic of 0.822 (genotype 0.524, eGFR 0.624, NT-proBNP 0.686, TR velocity 0.703). K-means clustering grouped all patients into 3 main clusters with significant survival differences. Survival at 8 years for the entire group was 70%; for individual clusters, survival was 43% for cluster 1, 72% for cluster 2, and 88% for cluster 3 (Figure 1A). Since TR velocity is recognized as one of the most specific independent predictors of mortality, we compared our results with this parameter. There was a better stratification of mortality risk using the 7 strongest parameters from RSF compared with TR velocity alone (Figure 1B), particularly for longer term outcomes. In this cohort of 601 patients with SCD, machine learning methods were used to show the heterogeneity of this disorder and the ability to detect phenotypic clusters with different mortality profiles. Although there are many individual predictors of mortality, few methods other than assessment by an expert clinician can integrate all known variables in deeply phenotyped patients. RSF and cluster analysis was used in this cohort to analyze a large amount of data in order to identify seven variables that could stratify patients into groups with significantly different outcomes. The specificity of this approach was high (c-statistic 0.822) and better than that of individual markers of end-organ involvement. Disclosures No relevant conflicts of interest to declare.

Technical considerations of multi-parametric tissue outcome prediction methods in acute ischemic stroke patients

Decisions regarding acute stroke treatment rely heavily on imaging, but interpretation can be difficult for physicians. Machine learning methods can assist clinicians by providing tissue outcome predictions for different treatment approaches based on acute multi-parametric imaging. To produce such clinically viable machine learning models, factors such as classifier choice, data normalization, and data balancing must be considered. This study gives comprehensive consideration to these factors by comparing the agreement of voxel-based tissue outcome predictions using acute imaging and clinical parameters with manual lesion segmentations derived from follow-up imaging. This study considers random decision forest, generalized linear model, and k-nearest-neighbor machine learning classifiers in conjunction with three data normalization approaches (non-normalized, relative to contralateral hemisphere, and relative to contralateral VOI), and two data balancing strategies (full dataset and stratified subsampling). These classifier settings were evaluated based on 90 MRI datasets from acute ischemic stroke patients. Distinction was made between patients recanalized using intraarterial and intravenous methods, as well as those without successful recanalization. For primary quantitative comparison, the Dice metric was computed for each voxel-based tissue outcome prediction and its corresponding follow-up lesion segmentation. It was found that the random forest classifier outperformed the generalized linear model and the k-nearest-neighbor classifier, that normalization did not improve the Dice score of the lesion outcome predictions, and that the models generated lesion outcome predictions with higher Dice scores when trained with balanced datasets. No significant difference was found between the treatment groups (intraarterial vs intravenous) regarding the Dice score of the tissue outcome predictions.

Divergent network properties that predict early surgical failure versus late recurrence in temporal lobe epilepsy.

The objectives of this study were to identify functional and structural network properties that are associated with early versus long-term seizure outcomes after mesial temporal lobe epilepsy (mTLE) surgery and to determine how these compare to current clinically used methods for seizure outcome prediction. In this case-control study, 26 presurgical mTLE patients and 44 healthy controls were enrolled to undergo 3-T MRI for functional and structural connectivity mapping across an 8-region network of mTLE seizure propagation, including the hippocampus (left and right), insula (left and right), thalamus (left and right), one midline precuneus, and one midline mid-cingulate. Seizure outcome was assessed annually for up to 3 years. Network properties and current outcome prediction methods related to early and long-term seizure outcome were investigated. A network model was previously identified across 8 patients with seizure-free mTLE. Results confirmed that whole-network propagation connectivity patterns inconsistent with the mTLE model predict early surgical failure. In those patients with networks consistent with the mTLE network, specific bilateral within-network hippocampal to precuneus impairment (rather than unilateral impairment ipsilateral to the seizure focus) was associated with mild seizure recurrence. No currently used clinical variables offered the same ability to predict long-term outcome. It is known that there are important clinical differences between early surgical failure that lead to frequent disabling seizures and late recurrence of less frequent mild seizures. This study demonstrated that divergent network connectivity variability, whole-network versus within-network properties, were uniquely associated with these disparate outcomes.

PCM-SABRE: a platform for benchmarking and comparing outcome prediction methods in precision cancer medicine

BackgroundNumerous publications attempt to predict cancer survival outcome from gene expression data using machine-learning methods. A direct comparison of these works is challenging for the following reasons: (1) inconsistent measures used to evaluate the performance of different models, and (2) incomplete specification of critical stages in the process of knowledge discovery. There is a need for a platform that would allow researchers to replicate previous works and to test the impact of changes in the knowledge discovery process on the accuracy of the induced models.ResultsWe developed the PCM-SABRE platform, which supports the entire knowledge discovery process for cancer outcome analysis. PCM-SABRE was developed using KNIME. By using PCM-SABRE to reproduce the results of previously published works on breast cancer survival, we define a baseline for evaluating future attempts to predict cancer outcome with machine learning. We used PCM-SABRE to replicate previous work that describe predictive models of breast cancer recurrence, and tested the performance of all possible combinations of feature selection methods and data mining algorithms that was used in either of the works. We reconstructed the work of Chou et al. observing similar trends – superior performance of Probabilistic Neural Network (PNN) and logistic regression (LR) algorithms and inconclusive impact of feature pre-selection with the decision tree algorithm on subsequent analysis.ConclusionsPCM-SABRE is a software tool that provides an intuitive environment for rapid development of predictive models in cancer precision medicine.

FERAL: network-based classifier with application to breast cancer outcome prediction.

Motivation: Breast cancer outcome prediction based on gene expression profiles is an important strategy for personalize patient care. To improve performance and consistency of discovered markers of the initial molecular classifiers, network-based outcome prediction methods (NOPs) have been proposed. In spite of the initial claims, recent studies revealed that neither performance nor consistency can be improved using these methods. NOPs typically rely on the construction of meta-genes by averaging the expression of several genes connected in a network that encodes protein interactions or pathway information. In this article, we expose several fundamental issues in NOPs that impede on the prediction power, consistency of discovered markers and obscures biological interpretation.Results: To overcome these issues, we propose FERAL, a network-based classifier that hinges upon the Sparse Group Lasso which performs simultaneous selection of marker genes and training of the prediction model. An important feature of FERAL, and a significant departure from existing NOPs, is that it uses multiple operators to summarize genes into meta-genes. This gives the classifier the opportunity to select the most relevant meta-gene for each gene set. Extensive evaluation revealed that the discovered markers are markedly more stable across independent datasets. Moreover, interpretation of the marker genes detected by FERAL reveals valuable mechanistic insight into the etiology of breast cancer.Availability and implementation: All code is available for download at: http://homepage.tudelft.nl/53a60/resources/FERAL/FERAL.zip.Contact: j.deridder@tudelft.nlSupplementary information: Supplementary data are available at Bioinformatics online.

Hypoxia and Outcome Prediction in Early-Stage Coma (Project HOPE): an observational prospective cohort study.

BackgroundThe number of resuscitated cardiac arrest patients suffering from anoxic-ischemic encephalopathy is considerable. However, outcome prediction parameters such as somatosensory evoked potentials need revision because they are based on data predating the implementation of mild therapeutical hypothermia and because data from our own laboratory suggest that they may fail to predict prognosis accurately. The present research project “Hypoxia and Outcome Prediction in Early-Stage Coma” is an ongoing observational prospective cohort study that aims to improve outcome prediction in anoxic coma by limiting the effects of falsely pessimistic predictions at the intensive care unit.MethodsOur outcome analysis is based on functional and behavioural definitions. This implies the analysis of the positive predictive value of prognostic markers yielding either positive or negative results. We also analyse the effect of covariates adjusted for age and sex such as sociodemographic variables, prognostic variables and treatment factors on functional and behavioural outcomes, with mixed effects regression models (i.e. fixed and random effects). We expect to enrol 172 patients based on the result of previous research. The null hypothesis is that there is a probability of <10 % that a positive outcome will be observed despite the presence of any of the predictors of a poor/negative outcome. We test the null hypothesis against a one-sided alternative using a Simon’s two-stage design to determine whether it is warranted to recruit the full number of patients suggested by a power analysis. The second stage has a design with a Type I error rate of 0.05 and 80 % power if the true response rate is 25 %.DiscussionWe aim to make a significant contribution to the revision and improvement of current outcome prediction methods in anoxic-ischemic encephalopathy patients. As a result, neurocritical care specialists worldwide will have considerably more accurate methods for prognosticating the outcome of anoxic-ischemic encephalopathy following cardiac arrest. This will facilitate the provision of treatment tailored to individual patients and the attainment of an optimal quality of life. It will also inform the decision to withdraw treatment with a level of accuracy never seen before in the field.Trial registrationClinicalTrials.gov NCT02231060 (registered 29 August 2014)

Rejoinder: Combining biomarkers to optimize patient treatment recommendations

Rejoinder: Combining biomarkers to optimize patient treatment recommendations

A New Method for Outcome Prediction of Acute Liver Failure - Pilot Study with the LiMAx Test

A New Method for Outcome Prediction of Acute Liver Failure - Pilot Study with the LiMAx Test