Assessing the improved discriminatory power of a new biomarker in prognostic models

Although the area under the receiver operating characteristic (AUC) is the most popular measure of the performance of prediction models, it has limitations, especially when it is used to evaluate the added discrimination of a new biomarker in the model. Pencina et al. (2008) proposed two indices, the net reclassification improvement (NRI) and integrated discrimination improvement (IDI), to supplement the improvement in the AUC (IAUC). Their NRI and IDI are based on binary outcomes in case-control settings, which do not involve time-to-event outcome. However, many disease outcomes are time-dependent and the onset time can be censored. Measuring discrimination potential of a prognostic marker without considering time to event can lead to biased estimates. In this dissertation, we have extended the NRI and IDI to survival analysis settings and derived the corresponding sample estimators and asymptotic tests. Simulation studies were conducted to compare the performance of the time-dependent NRI and IDI with Pencina’s NRI and IDI. For illustration, we have applied the proposed method to a breast cancer study.^ Key words: Prognostic model, Discrimination, Time-dependent NRI and IDI ^

Gender associated risk factors of aortic aneurysms and dissections

Objectives. Predict who will develop a dissection. To create male and female prediction models using the risk factors: age, ethnicity, hypertension, high cholesterol, smoking, alcohol use, diabetes, heart attack, congestive heart failure, congenital and non-congenital heart disease, Marfan syndrome, and bicuspid aortic valve. ^ Methods. Using 572 patients diagnosed with aortic aneurysms, a model was developed for each of males and females using 80% of the data and then verified using the remaining 20% of the data. ^ Results. The male model predicted the probability of a male in having a dissection (p=0.076) and the female model predicted the probability of a female in having a dissection (p=0.054). The validation models did not support the choice of the developmental models. ^ Conclusions. The best models obtained suggested that those who are at a greater risk of having a dissection are males with non-congenital heart disease and who drink alcohol, and females with non-congenital heart disease and bicuspid aortic valve.^

TIME SERIES ANALYSIS AS INPUT FOR PREDICTIVE MODELING: PREDICTING CARDIAC ARREST IN A PEDIATRIC INTENSIVE CARE UNIT

The first manuscript, entitled "Time-Series Analysis as Input for Clinical Predictive Modeling: Modeling Cardiac Arrest in a Pediatric ICU" lays out the theoretical background for the project. There are several core concepts presented in this paper. First, traditional multivariate models (where each variable is represented by only one value) provide single point-in-time snapshots of patient status: they are incapable of characterizing deterioration. Since deterioration is consistently identified as a precursor to cardiac arrests, we maintain that the traditional multivariate paradigm is insufficient for predicting arrests. We identify time series analysis as a method capable of characterizing deterioration in an objective, mathematical fashion, and describe how to build a general foundation for predictive modeling using time series analysis results as latent variables. Building a solid foundation for any given modeling task involves addressing a number of issues during the design phase. These include selecting the proper candidate features on which to base the model, and selecting the most appropriate tool to measure them. We also identified several unique design issues that are introduced when time series data elements are added to the set of candidate features. One such issue is in defining the duration and resolution of time series elements required to sufficiently characterize the time series phenomena being considered as candidate features for the predictive model. Once the duration and resolution are established, there must also be explicit mathematical or statistical operations that produce the time series analysis result to be used as a latent candidate feature. In synthesizing the comprehensive framework for building a predictive model based on time series data elements, we identified at least four classes of data that can be used in the model design. The first two classes are shared with traditional multivariate models: multivariate data and clinical latent features. Multivariate data is represented by the standard one value per variable paradigm and is widely employed in a host of clinical models and tools. These are often represented by a number present in a given cell of a table. Clinical latent features derived, rather than directly measured, data elements that more accurately represent a particular clinical phenomenon than any of the directly measured data elements in isolation. The second two classes are unique to the time series data elements. The first of these is the raw data elements. These are represented by multiple values per variable, and constitute the measured observations that are typically available to end users when they review time series data. These are often represented as dots on a graph. The final class of data results from performing time series analysis. This class of data represents the fundamental concept on which our hypothesis is based. The specific statistical or mathematical operations are up to the modeler to determine, but we generally recommend that a variety of analyses be performed in order to maximize the likelihood that a representation of the time series data elements is produced that is able to distinguish between two or more classes of outcomes. The second manuscript, entitled "Building Clinical Prediction Models Using Time Series Data: Modeling Cardiac Arrest in a Pediatric ICU" provides a detailed description, start to finish, of the methods required to prepare the data, build, and validate a predictive model that uses the time series data elements determined in the first paper. One of the fundamental tenets of the second paper is that manual implementations of time series based models are unfeasible due to the relatively large number of data elements and the complexity of preprocessing that must occur before data can be presented to the model. Each of the seventeen steps is analyzed from the perspective of how it may be automated, when necessary. We identify the general objectives and available strategies of each of the steps, and we present our rationale for choosing a specific strategy for each step in the case of predicting cardiac arrest in a pediatric intensive care unit. Another issue brought to light by the second paper is that the individual steps required to use time series data for predictive modeling are more numerous and more complex than those used for modeling with traditional multivariate data. Even after complexities attributable to the design phase (addressed in our first paper) have been accounted for, the management and manipulation of the time series elements (the preprocessing steps in particular) are issues that are not present in a traditional multivariate modeling paradigm. In our methods, we present the issues that arise from the time series data elements: defining a reference time; imputing and reducing time series data in order to conform to a predefined structure that was specified during the design phase; and normalizing variable families rather than individual variable instances. The final manuscript, entitled: "Using Time-Series Analysis to Predict Cardiac Arrest in a Pediatric Intensive Care Unit" presents the results that were obtained by applying the theoretical construct and its associated methods (detailed in the first two papers) to the case of cardiac arrest prediction in a pediatric intensive care unit. Our results showed that utilizing the trend analysis from the time series data elements reduced the number of classification errors by 73%. The area under the Receiver Operating Characteristic curve increased from a baseline of 87% to 98% by including the trend analysis. In addition to the performance measures, we were also able to demonstrate that adding raw time series data elements without their associated trend analyses improved classification accuracy as compared to the baseline multivariate model, but diminished classification accuracy as compared to when just the trend analysis features were added (ie, without adding the raw time series data elements). We believe this phenomenon was largely attributable to overfitting, which is known to increase as the ratio of candidate features to class examples rises. Furthermore, although we employed several feature reduction strategies to counteract the overfitting problem, they failed to improve the performance beyond that which was achieved by exclusion of the raw time series elements. Finally, our data demonstrated that pulse oximetry and systolic blood pressure readings tend to start diminishing about 10-20 minutes before an arrest, whereas heart rates tend to diminish rapidly less than 5 minutes before an arrest.

INTEGRATIVE BIOMARKER IDENTIFICATION AND CLASSIFICATION USING HIGH THROUGHPUT ASSAYS

It is well accepted that tumorigenesis is a multi-step procedure involving aberrant functioning of genes regulating cell proliferation, differentiation, apoptosis, genome stability, angiogenesis and motility. To obtain a full understanding of tumorigenesis, it is necessary to collect information on all aspects of cell activity. Recent advances in high throughput technologies allow biologists to generate massive amounts of data, more than might have been imagined decades ago. These advances have made it possible to launch comprehensive projects such as (TCGA) and (ICGC) which systematically characterize the molecular fingerprints of cancer cells using gene expression, methylation, copy number, microRNA and SNP microarrays as well as next generation sequencing assays interrogating somatic mutation, insertion, deletion, translocation and structural rearrangements. Given the massive amount of data, a major challenge is to integrate information from multiple sources and formulate testable hypotheses. This thesis focuses on developing methodologies for integrative analyses of genomic assays profiled on the same set of samples. We have developed several novel methods for integrative biomarker identification and cancer classification. We introduce a regression-based approach to identify biomarkers predictive to therapy response or survival by integrating multiple assays including gene expression, methylation and copy number data through penalized regression. To identify key cancer-specific genes accounting for multiple mechanisms of regulation, we have developed the integIRTy software that provides robust and reliable inferences about gene alteration by automatically adjusting for sample heterogeneity as well as technical artifacts using Item Response Theory. To cope with the increasing need for accurate cancer diagnosis and individualized therapy, we have developed a robust and powerful algorithm called SIBER to systematically identify bimodally expressed genes using next generation RNAseq data. We have shown that prediction models built from these bimodal genes have the same accuracy as models built from all genes. Further, prediction models with dichotomized gene expression measurements based on their bimodal shapes still perform well. The effectiveness of outcome prediction using discretized signals paves the road for more accurate and interpretable cancer classification by integrating signals from multiple sources.

Data quality in trauma transfusion studies and the impact of missing data on predicting massive transfusion

Maximizing data quality may be especially difficult in trauma-related clinical research. Strategies are needed to improve data quality and assess the impact of data quality on clinical predictive models. This study had two objectives. The first was to compare missing data between two multi-center trauma transfusion studies: a retrospective study (RS) using medical chart data with minimal data quality review and the PRospective Observational Multi-center Major Trauma Transfusion (PROMMTT) study with standardized quality assurance. The second objective was to assess the impact of missing data on clinical prediction algorithms by evaluating blood transfusion prediction models using PROMMTT data. RS (2005-06) and PROMMTT (2009-10) investigated trauma patients receiving ≥ 1 unit of red blood cells (RBC) from ten Level I trauma centers. Missing data were compared for 33 variables collected in both studies using mixed effects logistic regression (including random intercepts for study site). Massive transfusion (MT) patients received ≥ 10 RBC units within 24h of admission. Correct classification percentages for three MT prediction models were evaluated using complete case analysis and multiple imputation based on the multivariate normal distribution. A sensitivity analysis for missing data was conducted to estimate the upper and lower bounds of correct classification using assumptions about missing data under best and worst case scenarios. Most variables (17/33=52%) had <1% missing data in RS and PROMMTT. Of the remaining variables, 50% demonstrated less missingness in PROMMTT, 25% had less missingness in RS, and 25% were similar between studies. Missing percentages for MT prediction variables in PROMMTT ranged from 2.2% (heart rate) to 45% (respiratory rate). For variables missing >1%, study site was associated with missingness (all p≤0.021). Survival time predicted missingness for 50% of RS and 60% of PROMMTT variables. MT models complete case proportions ranged from 41% to 88%. Complete case analysis and multiple imputation demonstrated similar correct classification results. Sensitivity analysis upper-lower bound ranges for the three MT models were 59-63%, 36-46%, and 46-58%. Prospective collection of ten-fold more variables with data quality assurance reduced overall missing data. Study site and patient survival were associated with missingness, suggesting that data were not missing completely at random, and complete case analysis may lead to biased results. Evaluating clinical prediction model accuracy may be misleading in the presence of missing data, especially with many predictor variables. The proposed sensitivity analysis estimating correct classification under upper (best case scenario)/lower (worst case scenario) bounds may be more informative than multiple imputation, which provided results similar to complete case analysis.^

Selection bias and the cross-validation of regression models for prediction

Strategies are compared for the development of a linear regression model with stochastic (multivariate normal) regressor variables and the subsequent assessment of its predictive ability. Bias and mean squared error of four estimators of predictive performance are evaluated in simulated samples of 32 population correlation matrices. Models including all of the available predictors are compared with those obtained using selected subsets. The subset selection procedures investigated include two stopping rules, C$\sb{\rm p}$ and S$\sb{\rm p}$, each combined with an 'all possible subsets' or 'forward selection' of variables. The estimators of performance utilized include parametric (MSEP$\sb{\rm m}$) and non-parametric (PRESS) assessments in the entire sample, and two data splitting estimates restricted to a random or balanced (Snee's DUPLEX) 'validation' half sample. The simulations were performed as a designed experiment, with population correlation matrices representing a broad range of data structures.^ The techniques examined for subset selection do not generally result in improved predictions relative to the full model. Approaches using 'forward selection' result in slightly smaller prediction errors and less biased estimators of predictive accuracy than 'all possible subsets' approaches but no differences are detected between the performances of C$\sb{\rm p}$ and S$\sb{\rm p}$. In every case, prediction errors of models obtained by subset selection in either of the half splits exceed those obtained using all predictors and the entire sample.^ Only the random split estimator is conditionally (on $\\beta$) unbiased, however MSEP$\sb{\rm m}$ is unbiased on average and PRESS is nearly so in unselected (fixed form) models. When subset selection techniques are used, MSEP$\sb{\rm m}$ and PRESS always underestimate prediction errors, by as much as 27 percent (on average) in small samples. Despite their bias, the mean squared errors (MSE) of these estimators are at least 30 percent less than that of the unbiased random split estimator. The DUPLEX split estimator suffers from large MSE as well as bias, and seems of little value within the context of stochastic regressor variables.^ To maximize predictive accuracy while retaining a reliable estimate of that accuracy, it is recommended that the entire sample be used for model development, and a leave-one-out statistic (e.g. PRESS) be used for assessment. ^

SURVIVAL PREDICTION FOR BRAIN TUMOR PATIENTS USING GENE EXPRESSION DATA

Brain tumor is one of the most aggressive types of cancer in humans, with an estimated median survival time of 12 months and only 4% of the patients surviving more than 5 years after disease diagnosis. Until recently, brain tumor prognosis has been based only on clinical information such as tumor grade and patient age, but there are reports indicating that molecular profiling of gliomas can reveal subgroups of patients with distinct survival rates. We hypothesize that coupling molecular profiling of brain tumors with clinical information might improve predictions of patient survival time and, consequently, better guide future treatment decisions. In order to evaluate this hypothesis, the general goal of this research is to build models for survival prediction of glioma patients using DNA molecular profiles (U133 Affymetrix gene expression microarrays) along with clinical information. First, a predictive Random Forest model is built for binary outcomes (i.e. short vs. long-term survival) and a small subset of genes whose expression values can be used to predict survival time is selected. Following, a new statistical methodology is developed for predicting time-to-death outcomes using Bayesian ensemble trees. Due to a large heterogeneity observed within prognostic classes obtained by the Random Forest model, prediction can be improved by relating time-to-death with gene expression profile directly. We propose a Bayesian ensemble model for survival prediction which is appropriate for high-dimensional data such as gene expression data. Our approach is based on the ensemble "sum-of-trees" model which is flexible to incorporate additive and interaction effects between genes. We specify a fully Bayesian hierarchical approach and illustrate our methodology for the CPH, Weibull, and AFT survival models. We overcome the lack of conjugacy using a latent variable formulation to model the covariate effects which decreases computation time for model fitting. Also, our proposed models provides a model-free way to select important predictive prognostic markers based on controlling false discovery rates. We compare the performance of our methods with baseline reference survival methods and apply our methodology to an unpublished data set of brain tumor survival times and gene expression data, selecting genes potentially related to the development of the disease under study. A closing discussion compares results obtained by Random Forest and Bayesian ensemble methods under the biological/clinical perspectives and highlights the statistical advantages and disadvantages of the new methodology in the context of DNA microarray data analysis.

Genetic and hormonal factors modulate spreading depression and transient hemiparesis in mouse models of familial hemiplegic migraine type 1.

Familial hemiplegic migraine type 1 (FHM1) is an autosomal dominant subtype of migraine with aura that is associated with hemiparesis. As with other types of migraine, it affects women more frequently than men. FHM1 is caused by mutations in the CACNA1A gene, which encodes the alpha1A subunit of Cav2.1 channels; the R192Q mutation in CACNA1A causes a mild form of FHM1, whereas the S218L mutation causes a severe, often lethal phenotype. Spreading depression (SD), a slowly propagating neuronal and glial cell depolarization that leads to depression of neuronal activity, is the most likely cause of migraine aura. Here, we have shown that transgenic mice expressing R192Q or S218L FHM1 mutations have increased SD frequency and propagation speed; enhanced corticostriatal propagation; and, similar to the human FHM1 phenotype, more severe and prolonged post-SD neurological deficits. The susceptibility to SD and neurological deficits is affected by allele dosage and is higher in S218L than R192Q mutants. Further, female S218L and R192Q mutant mice were more susceptible to SD and neurological deficits than males. This sex difference was abrogated by ovariectomy and senescence and was partially restored by estrogen replacement, implicating ovarian hormones in the observed sex differences in humans with FHM1. These findings demonstrate that genetic and hormonal factors modulate susceptibility to SD and neurological deficits in FHM1 mutant mice, providing a potential mechanism for the phenotypic diversity of human migraine and aura.

Prediction of sepsis in the burn intensive care unit patient

Sepsis is a significant cause for multiple organ failure and death in the burn patient, yet identification in this population is confounded by chronic hypermetabolism and impaired immune function. The purpose of this study was twofold: 1) determine the ability of the systemic inflammatory response syndrome (SIRS) and American Burn Association (ABA) criteria to predict sepsis in the burn patient; and 2) develop a model representing the best combination of clinical predictors associated with sepsis in the same population. A retrospective, case-controlled, within-patient comparison of burn patients admitted to a single intensive care unit (ICU) was conducted for the period January 2005 to September 2010. Blood culture results were paired with clinical condition: "positive-sick"; "negative-sick", and "screening-not sick". Data were collected for the 72 hours prior to each blood culture. The most significant predictors were evaluated using logistic regression, Generalized Estimating Equations (GEE) and ROC area under the curve (AUC) analyses to assess model predictive ability. Bootstrapping methods were employed to evaluate potential model over-fitting. Fifty-nine subjects were included, representing 177 culture periods. SIRS criteria were not found to be associated with culture type, with an average of 98% of subjects meeting criteria in the 3 days prior. ABA sepsis criteria were significantly different among culture type only on the day prior (p = 0.004). The variables identified for the model included: heart rate>130 beats/min, mean blood pressure<60 mmHg, base deficit<-6 mEq/L, temperature>36°C, use of vasoactive medications, and glucose>150 mg/d1. The model was significant in predicting "positive culture-sick" and sepsis state, with AUC of 0.775 (p < 0.001) and 0.714 (p < .001), respectively; comparatively, the ABA criteria AUC was 0.619 (p = 0.028) and 0.597 (p = .035), respectively. SIRS criteria are not appropriate for identifying sepsis in the burn population. The ABA criteria perform better, but only for the day prior to positive blood culture results. The time period useful to diagnose sepsis using clinical criteria may be limited to 24 hours. A combination of predictors is superior to individual variable trends, yet algorithms or computer support will be necessary for the clinician to find such models useful. ^

An exploratory analysis of variance and volatility in epileptic electroencephalograms

The electroencephalogram (EEG) is a physiological time series that measures electrical activity at different locations in the brain, and plays an important role in epilepsy research. Exploring the variance and/or volatility may yield insights for seizure prediction, seizure detection and seizure propagation/dynamics.^ Maximal Overlap Discrete Wavelet Transforms (MODWTs) and ARMA-GARCH models were used to determine variance and volatility characteristics of 66 channels for different states of an epileptic EEG – sleep, awake, sleep-to-awake and seizure. The wavelet variances, changes in wavelet variances and volatility half-lives for the four states were compared for possible differences between seizure and non-seizure channels.^ The half-lives of two of the three seizure channels were found to be shorter than all of the non-seizure channels, based on 95% CIs for the pre-seizure and awake signals. No discernible patterns were found the wavelet variances of the change points for the different signals. ^

Radiomics of NSCLC: Quantitative CT Image Feature Characterization and Tumor Shrinkage Prediction

Radiomics is the high-throughput extraction and analysis of quantitative image features. For non-small cell lung cancer (NSCLC) patients, radiomics can be applied to standard of care computed tomography (CT) images to improve tumor diagnosis, staging, and response assessment. The first objective of this work was to show that CT image features extracted from pre-treatment NSCLC tumors could be used to predict tumor shrinkage in response to therapy. This is important since tumor shrinkage is an important cancer treatment endpoint that is correlated with probability of disease progression and overall survival. Accurate prediction of tumor shrinkage could also lead to individually customized treatment plans. To accomplish this objective, 64 stage NSCLC patients with similar treatments were all imaged using the same CT scanner and protocol. Quantitative image features were extracted and principal component regression with simulated annealing subset selection was used to predict shrinkage. Cross validation and permutation tests were used to validate the results. The optimal model gave a strong correlation between the observed and predicted shrinkages with . The second objective of this work was to identify sets of NSCLC CT image features that are reproducible, non-redundant, and informative across multiple machines. Feature sets with these qualities are needed for NSCLC radiomics models to be robust to machine variation and spurious correlation. To accomplish this objective, test-retest CT image pairs were obtained from 56 NSCLC patients imaged on three CT machines from two institutions. For each machine, quantitative image features with concordance correlation coefficient values greater than 0.90 were considered reproducible. Multi-machine reproducible feature sets were created by taking the intersection of individual machine reproducible feature sets. Redundant features were removed through hierarchical clustering. The findings showed that image feature reproducibility and redundancy depended on both the CT machine and the CT image type (average cine 4D-CT imaging vs. end-exhale cine 4D-CT imaging vs. helical inspiratory breath-hold 3D CT). For each image type, a set of cross-machine reproducible, non-redundant, and informative image features was identified. Compared to end-exhale 4D-CT and breath-hold 3D-CT, average 4D-CT derived image features showed superior multi-machine reproducibility and are the best candidates for clinical correlation.