L1 penalized continuation ratio models for ordinal response prediction using high-dimensional datasets

Mechanistic insights into the interactions between cancer drivers and the tumour immune microenvironment

BACKGROUND

The crosstalk between cancer and the tumour immune microenvironment (TIME) has attracted significant interest in the latest years because of its impact on cancer evolution and response to treatment. Despite this, cancer-specific tumour-TIME interactions and their mechanistic insights are still poorly understood.

METHODS

Here, we compute the significant interactions occurring between cancer-specific genetic drivers and five anti- and pro-tumour TIME features in 32 cancer types using Lasso regularised ordinal regression. Focusing on head and neck squamous cancer (HNSC), we rebuild the functional networks linking specific TIME driver alterations to the TIME state they associate with.

RESULTS

The 477 TIME drivers that we identify are multifunctional genes whose alterations are selected early in cancer evolution and recur across and within cancer types. Tumour suppressors and oncogenes have an opposite effect on the TIME and the overall anti-tumour TIME driver burden is predictive of response to immunotherapy. TIME driver alterations predict the immune profiles of HNSC molecular subtypes, and perturbations in keratinization, apoptosis and interferon signalling underpin specific driver-TIME interactions.

CONCLUSIONS

Overall, our study delivers a comprehensive resource of TIME drivers, gives mechanistic insights into their immune-regulatory role, and provides an additional framework for patient prioritisation to immunotherapy. The full list of TIME drivers and associated properties are available at http://www.network-cancer-genes.org .

Machine learning-based tumor-infiltrating immune cell-associated lncRNAs for predicting prognosis and immunotherapy response in patients with glioblastoma

Long noncoding ribonucleic acids (RNAs; lncRNAs) have been associated with cancer immunity regulation. However, the roles of immune cell-specific lncRNAs in glioblastoma (GBM) remain largely unknown. In this study, a novel computational framework was constructed to screen the tumor-infiltrating immune cell-associated lncRNAs (TIIClnc) for developing TIIClnc signature by integratively analyzing the transcriptome data of purified immune cells, GBM cell lines and bulk GBM tissues using six machine learning algorithms. As a result, TIIClnc signature could distinguish survival outcomes of GBM patients across four independent datasets, including the Xiangya in-house dataset, and more importantly, showed superior performance than 95 previously established signatures in gliomas. TIIClnc signature was revealed to be an indicator of the infiltration level of immune cells and predicted the response outcomes of immunotherapy. The positive correlation between TIIClnc signature and CD8, PD-1 and PD-L1 was verified in the Xiangya in-house dataset. As a newly demonstrated predictive biomarker, the TIIClnc signature enabled a more precise selection of the GBM population who would benefit from immunotherapy and should be validated and applied in the near future.

The number of response categories in ordered response models

The choice of the number m of response categories is a crucial issue in categorization of a continuous response. The paper exploits the Proportional Odds Models' property which allows to generate ordinal responses with a different number of categories from the same underlying variable. It investigates the asymptotic efficiency of the estimators of the regression coefficients and the accuracy of the derived inferential procedures when m varies. The analysis is based on models with closed-form information matrices so that the asymptotic efficiency can be analytically evaluated without need of simulations. The paper proves that a finer categorization augments the information content of the data and consequently shows that the asymptotic efficiency and the power of the tests on the regression coefficients increase with m. The impact of the loss of information produced by merging categories on the efficiency of the estimators is also considered, highlighting its risks especially when performed in its extreme form of dichotomization. Furthermore, the appropriate value of m for various sample sizes is explored, pointing out that a large number of categories can offset the limited amount of information of a small sample by a better quality of the data. Finally, two case studies on the quality of life of chemotherapy patients and on the perception of pain, based on discretized continuous scales, illustrate the main findings of the paper.

psupertime: supervised pseudotime analysis for time-series single-cell RNA-seq data

Motivation

Improvements in single-cell RNA-seq technologies mean that studies measuring multiple experimental conditions, such as time series, have become more common. At present, few computational methods exist to infer time series-specific transcriptome changes, and such studies have therefore typically used unsupervised pseudotime methods. While these methods identify cell subpopulations and the transitions between them, they are not appropriate for identifying the genes that vary coherently along the time series. In addition, the orderings they estimate are based only on the major sources of variation in the data, which may not correspond to the processes related to the time labels.

Results

We introduce psupertime, a supervised pseudotime approach based on a regression model, which explicitly uses time-series labels as input. It identifies genes that vary coherently along a time series, in addition to pseudotime values for individual cells, and a classifier that can be used to estimate labels for new data with unknown or differing labels. We show that psupertime outperforms benchmark classifiers in terms of identifying time-varying genes and provides better individual cell orderings than popular unsupervised pseudotime techniques. psupertime is applicable to any single-cell RNA-seq dataset with sequential labels (e.g. principally time series but also drug dosage and disease progression), derived from either experimental design and provides a fast, interpretable tool for targeted identification of genes varying along with specific biological processes.

Availability and implementation

R package available at github.com/wmacnair/psupertime and code for results reproduction at github.com/wmacnair/psupplementary.

Supplementary information

Supplementary data are available at Bioinformatics online.

ordinalbayes: Fitting Ordinal Bayesian Regression Models to High-Dimensional Data Using R

The stage of cancer is a discrete ordinal response that indicates the aggressiveness of disease and is often used by physicians to determine the type and intensity of treatment to be administered. For example, the FIGO stage in cervical cancer is based on the size and depth of the tumor as well as the level of spread. It may be of clinical relevance to identify molecular features from high-throughput genomic assays that are associated with the stage of cervical cancer to elucidate pathways related to tumor aggressiveness, identify improved molecular features that may be useful for staging, and identify therapeutic targets. High-throughput RNA-Seq data and corresponding clinical data (including stage) for cervical cancer patients have been made available through The Cancer Genome Atlas Project (TCGA). We recently described penalized Bayesian ordinal response models that can be used for variable selection for over-parameterized datasets, such as the TCGA-CESC dataset. Herein, we describe our ordinalbayes R package, available from the Comprehensive R Archive Network (CRAN), which enhances the runjags R package by enabling users to easily fit cumulative logit models when the outcome is ordinal and the number of predictors exceeds the sample size, P > N, such as for TCGA and other high-throughput genomic data. We demonstrate the use of this package by applying it to the TCGA cervical cancer dataset. Our ordinalbayes package can be used to fit models to high-dimensional datasets, and it effectively performs variable selection.

Respiratory Effects of Exposure to Aerosol From the Candidate Modified-Risk Tobacco Product THS 2.2 in an 18-Month Systems Toxicology Study With A/J Mice

Smoking cessation is the most effective measure for reducing the risk of smoking-related diseases. However, switching to less harmful products (modified-risk tobacco products [MRTP]) can be an alternative to help reduce the risk for adult smokers who would otherwise continue to smoke. In an 18-month chronic carcinogenicity/toxicity study in A/J mice (OECD Test Guideline 453), we assessed the aerosol of Tobacco Heating System 2.2 (THS 2.2), a candidate MRTP based on the heat-not-burn principle, compared with 3R4F cigarette smoke (CS). To capture toxicity- and disease-relevant mechanisms, we complemented standard toxicology endpoints with in-depth systems toxicology analyses. In this part of our publication series, we report on integrative assessment of the apical and molecular exposure effects on the respiratory tract (nose, larynx, and lungs). Across the respiratory tract, we found changes in inflammatory response following 3R4F CS exposure (eg, antimicrobial peptide response in the nose), with both shared and distinct oxidative and xenobiotic responses. Compared with 3R4F CS, THS 2.2 aerosol exerted far fewer effects on respiratory tract histology, including adaptive tissue changes in nasal and laryngeal epithelium and inflammation and emphysematous changes in the lungs. Integrative analysis of molecular changes confirmed the substantially lower impact of THS 2.2 aerosol than 3R4F CS on toxicologically and disease-relevant molecular processes such as inflammation, oxidative stress responses, and xenobiotic metabolism. In summary, this work exemplifies how apical and molecular endpoints can be combined effectively for toxicology assessment and further supports findings on the reduced respiratory health risks of THS 2.2 aerosol.

Smoothing in Ordinal Regression: An Application to Sensory Data

The so-called proportional odds assumption is popular in cumulative, ordinal regression. In practice, however, such an assumption is sometimes too restrictive. For instance, when modeling the perception of boar taint on an individual level, it turns out that, at least for some subjects, the effects of predictors (androstenone and skatole) vary between response categories. For more flexible modeling, we consider the use of a ‘smooth-effects-on-response penalty’ (SERP) as a connecting link between proportional and fully non-proportional odds models, assuming that parameters of the latter vary smoothly over response categories. The usefulness of SERP is further demonstrated through a simulation study. Besides flexible and accurate modeling, SERP also enables fitting of parameters in cases where the pure, unpenalized non-proportional odds model fails to converge.

Machine learning-assisted decision-support models to better predict patients with calculous pyonephrosis

Background

To develop a machine learning (ML)-assisted model capable of accurately identifying patients with calculous pyonephrosis before making treatment decisions by integrating multiple clinical characteristics.

Methods

We retrospectively collected data from patients with obstructed hydronephrosis who underwent retrograde ureteral stent insertion, percutaneous nephrostomy (PCN), or percutaneous nephrolithotomy (PCNL). The study cohort was divided into training and testing datasets in a 70:30 ratio for further analysis. We developed 5 ML-assisted models from 22 clinical features using logistic regression (LR), LR optimized by least absolute shrinkage and selection operator (Lasso) regularization (Lasso-LR), support vector machine (SVM), extreme gradient boosting (XGBoost), and random forest (RF). The area under the curve (AUC) was applied to determine the model with the highest discrimination. Decision curve analysis (DCA) was used to investigate the clinical net benefit associated with using the predictive models.

Results

A total of 322 patients were included, with 225 patients in the training dataset, and 97 patients in the testing dataset. The XGBoost model showed good discrimination with the AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of 0.981, 0.991, 0.962, 1.000, 1.000, and 0.989, respectively, followed by SVM [AUC =0.985, 95% confidence interval (CI): 0.970–1.000], Lasso-LR (AUC =0.977, 95% CI: 0.958–0.996), LR (AUC =0.936, 95% CI: 0.905–0.968), and RF (AUC =0.920, 95% CI: 0.870–0.970). Validation of the model showed that SVM yielded the highest AUC (0.977, 95% CI: 0.952–1.000), followed by Lasso-LR (AUC =0.959, 95% CI: 0.921–0.997), XGBoost (AUC =0.958, 95% CI: 0.902–1.000), LR (AUC =0.932, 95% CI: 0.878–0.987), and RF (AUC =0.868, 95% CI: 0.779–0.958) in the testing dataset.

Conclusions

Our ML-based models had good discrimination in predicting patients with obstructed hydronephrosis at high risk of harboring pyonephrosis, and the use of these models may be greatly beneficial to urologists in treatment planning, patient selection, and decision-making.

Bayesian penalized cumulative logit model for high-dimensional data with an ordinal response

Many previous studies have identified associations between gene expression, measured using high-throughput genomic platforms, and quantitative or dichotomous traits. However, we note that health outcome and disease status measurements frequently appear on an ordinal scale, that is, the outcome is categorical but has inherent ordering. Identification of important genes may be useful for developing novel diagnostic and prognostic tools to predict or classify stage of disease. Gene expression data are usually high-dimensional, meaning that the number of genes is much larger than the sample size or number of patients. Herein we describe some existing frequentist methods for modeling an ordinal response in a high-dimensional predictor space. Following Tibshirani (1996), who described the LASSO estimate as the Bayesian posterior mode when the regression coefficients have independent Laplace priors, we propose a new approach for high-dimensional data with an ordinal response that is rooted in the Bayesian paradigm. We show that our proposed Bayesian approach outperforms existing frequentist methods through simulation studies. We then compare the performance of frequentist and Bayesian approaches using a study evaluating progression to hepatocellular carcinoma in hepatitis C infected patients.

Developing an instrument for an early prediction model of long-term functional outcomes in people with acquired injuries of the central nervous system: protocol and methodological aspects

Severe acquired brain injury (ABI) is a major cause of long-term disability and is the main determinant of health and societal costs. Early identification of favourable long-term recovery would allow personalized rehabilitative programs and better health care resources allocation. In light of the higher survival rate from intensive care units (ICU) in recent years, there is a growing need for early prognostication markers of functional recovery; to date, these data have been mainly collected at rehabilitation unit admission and not during the acute phase. We present the protocol and methodology to develop prediction models in people with severe acquired brain injury (GCS at admission to ICU < 8) for the functional and cognitive outcome at 12 months from the event. Predictors will be collected during the acute stage. Participants will be recruited within the first 72 h from the event in the ICUs of two teaching hospitals (Padova and Treviso). Participants will be followed up at discharge from ICU, admission and discharge from Neurorehabilitation and after 12 months from the event. Clinical and functional scales, electroencephalography, evoked potentials, magnetic resonance imaging and serological markers will be entered into a digital registry. Survival will be estimated using the Cox proportional hazard model. A multivariate prediction model will be developed for each of the functional and cognitive outcomes at 12 months from the event.

Predicting Ewe Body Condition Score Using Lifetime Liveweight and Liveweight Change, and Previous Body Condition Score Record

The body condition score (BCS) in sheep (Ovis aries) is a widely used subjective measure of body condition. Body condition score and liveweight have been reported to be statistically and often linearly related in ewes. Therefore, it was hypothesized that current BCS could be accurately and indirectly predicted using a ewe's lifetime liveweight, liveweight change, and previous BCS record. Ewes born between 2011 and 2012 (n = 11,798) were followed from 8 months to approximately 67 months of age in New Zealand. Individual ewe data was collected on liveweight and body condition scores at each stage of the annual cycle (pre-breeding, pregnancy diagnosis, pre-lambing, and weaning). Linear regression models were fitted to predict BCS at a given ewe age and stage of the annual cycle using a ewe's lifetime liveweight records (liveweight alone models). Further, linear models were then fitted using previous BCS and changes in liveweight, in addition to the lifetime liveweight records (combined models). Using the combined models improved (p < 0.01) the R² value by 39.8% (from 0.32 to 0.45) and lowered the average prediction error by 10% to 12% (from 0.29 to 0.26 body condition scores). However, a significant portion of the variability in BCS remained unaccounted for (39% to 89%) even in the combined models. The procedures found in this study, therefore, may overestimate or underestimate measures by 0.23 to 0.32 BCS, which could substantially change the status of the ewe, leading to incorrect management decisions. However, the findings do still suggest that there is potential for predicting ewe BCS from liveweight using linear regression if the key variables affecting the relationship between BCS and liveweight are accounted for.

High-dimensional variable selection for ordinal outcomes with error control

Many high-throughput genomic applications involve a large set of potential covariates and a response which is frequently measured on an ordinal scale, and it is crucial to identify which variables are truly associated with the response. Effectively controlling the false discovery rate (FDR) without sacrificing power has been a major challenge in variable selection research. This study reviews two existing variable selection frameworks, model-X knockoffs and a modified version of reference distribution variable selection (RDVS), both of which utilize artificial variables as benchmarks for decision making. Model-X knockoffs constructs a 'knockoff' variable for each covariate to mimic the covariance structure, while RDVS generates only one null variable and forms a reference distribution by performing multiple runs of model fitting. Herein, we describe how different importance measures for ordinal responses can be constructed that fit into these two selection frameworks, using either penalized regression or machine learning techniques. We compared these measures in terms of the FDR and power using simulated data. Moreover, we applied these two frameworks to high-throughput methylation data for identifying features associated with the progression from normal liver tissue to hepatocellular carcinoma to further compare and contrast their performances.

A four-group urine risk classifier for predicting outcomes in patients with prostate cancer

OBJECTIVES

To develop a risk classifier using urine-derived extracellular vesicle (EV)-RNA capable of providing diagnostic information on disease status prior to biopsy, and prognostic information for men on active surveillance (AS).

PATIENTS AND METHODS

Post-digital rectal examination urine-derived EV-RNA expression profiles (n = 535, multiple centres) were interrogated with a curated NanoString panel. A LASSO-based continuation ratio model was built to generate four prostate urine risk (PUR) signatures for predicting the probability of normal tissue (PUR-1), D'Amico low-risk (PUR-2), intermediate-risk (PUR-3), and high-risk (PUR-4) prostate cancer. This model was applied to a test cohort (n = 177) for diagnostic evaluation, and to an AS sub-cohort (n = 87) for prognostic evaluation.

RESULTS

Each PUR signature was significantly associated with its corresponding clinical category (P < 0.001). PUR-4 status predicted the presence of clinically significant intermediate- or high-risk disease (area under the curve = 0.77, 95% confidence interval [CI] 0.70-0.84). Application of PUR provided a net benefit over current clinical practice. In an AS sub-cohort (n = 87), groups defined by PUR status and proportion of PUR-4 had a significant association with time to progression (interquartile range hazard ratio [HR] 2.86, 95% CI 1.83-4.47; P < 0.001). PUR-4, when used continuously, dichotomized patient groups with differential progression rates of 10% and 60% 5 years after urine collection (HR 8.23, 95% CI 3.26-20.81; P < 0.001).

CONCLUSION

Urine-derived EV-RNA can provide diagnostic information on aggressive prostate cancer prior to biopsy, and prognostic information for men on AS. PUR represents a new and versatile biomarker that could result in substantial alterations to current treatment of patients with prostate cancer.

Bayesian personalized treatment selection strategies that integrate predictive with prognostic determinants

The evolution of "informatics" technologies has the potential to generate massive databases, but the extent to which personalized medicine may be effectuated depends on the extent to which these rich databases may be utilized to advance understanding of the disease molecular profiles and ultimately integrated for treatment selection, necessitating robust methodology for dimension reduction. Yet, statistical methods proposed to address challenges arising with the high-dimensionality of omics-type data predominately rely on linear models and emphasize associations deriving from prognostic biomarkers. Existing methods are often limited for discovering predictive biomarkers that interact with treatment and fail to elucidate the predictive power of their resultant selection rules. In this article, we present a Bayesian predictive method for personalized treatment selection that is devised to integrate both the treatment predictive and disease prognostic characteristics of a particular patient's disease. The method appropriately characterizes the structural constraints inherent to prognostic and predictive biomarkers, and hence properly utilizes these complementary sources of information for treatment selection. The methodology is illustrated through a case study of lower grade glioma. Theoretical considerations are explored to demonstrate the manner in which treatment selection is impacted by prognostic features. Additionally, simulations based on an actual leukemia study are provided to ascertain the method's performance with respect to selection rules derived from competing methods.

Quantifying risk factors in medical reports with a context-aware linear model

OBJECTIVE

We seek to quantify the mortality risk associated with mentions of medical concepts in textual electronic health records (EHRs). Recognizing mentions of named entities of relevant types (eg, conditions, symptoms, laboratory tests or behaviors) in text is a well-researched task. However, determining the level of risk associated with them is partly dependent on the textual context in which they appear, which may describe severity, temporal aspects, quantity, etc.

METHODS

To take into account that a given word appearing in the context of different risk factors (medical concepts) can make different contributions toward risk level, we propose a multitask approach, called context-aware linear modeling, which can be applied using appropriately regularized linear regression. To improve the performance for risk factors unseen in training data (eg, rare diseases), we take into account their distributional similarity to other concepts.

RESULTS

The evaluation is based on a corpus of 531 reports from EHRs with 99 376 risk factors rated manually by experts. While context-aware linear modeling significantly outperforms single-task models, taking into account concept similarity further improves performance, reaching the level of human annotators' agreements.

CONCLUSION

Our results show that automatic quantification of risk factors in EHRs can achieve performance comparable to human assessment, and taking into account the multitask structure of the problem and the ability to handle rare concepts is crucial for its accuracy.

Penalized negative binomial models for modeling an overdispersed count outcome with a high-dimensional predictor space: Application predicting micronuclei frequency

Chromosomal aberrations, such as micronuclei (MN), have served as biomarkers of genotoxic exposure and cancer risk. Guidelines for the process of scoring MN have been presented by the HUman MicroNucleus (HUMN) project. However, these guidelines were developed for assay performance but do not address how to statistically analyze the data generated by the assay. This has led to the application of various statistical methods that may render different interpretations and conclusions. By combining MN with data from other high-throughput genomic technologies such as gene expression microarray data, we may elucidate molecular features involved in micronucleation. Traditional methods that can model discrete (synonymously, count) data, such as MN frequency, require that the number of explanatory variables (P) is less than the number of samples (N). Due to this limitation, penalized models have been developed to enable model fitting for such over-parameterized datasets. Because penalized models in the discrete response setting are lacking, particularly when the count outcome is over-dispersed, herein we present our penalized negative binomial regression model that can be fit when P > N. Using simulation studies we demonstrate the performance of our method in comparison to commonly used penalized Poisson models when the outcome is over-dispersed and applied it to MN frequency and gene expression data collected as part of the Norwegian Mother and Child Cohort Study. Our countgmifs R package is available for download from the Comprehensive R Archive Network and can be applied to datasets having a discrete outcome that is either Poisson or negative binomial distributed and a high-dimensional covariate space.

Alternative evolutionary histories in the sequence space of an ancient protein

To understand why molecular evolution turned out as it did, we must characterize not only the path that evolution followed across the space of possible molecular sequences but also the many alternative trajectories that could have been taken but were not. A large-scale comparison of real and possible histories would establish whether the outcome of evolution represents an optimal state driven by natural selection or the contingent product of historical chance events; it would also reveal how the underlying distribution of functions across sequence space shaped historical evolution. Here we combine ancestral protein reconstruction with deep mutational scanning to characterize alternative histories in the sequence space around an ancient transcription factor, which evolved a novel biological function through well-characterized mechanisms. We find hundreds of alternative protein sequences that use diverse biochemical mechanisms to perform the derived function at least as well as the historical outcome. These alternatives all require prior permissive substitutions that do not enhance the derived function, but not all require the same permissive changes that occurred during history. We find that if evolution had begun from a different starting point within the network of sequences encoding the ancestral function, outcomes with different genetic and biochemical forms would probably have resulted; this contingency arises from the distribution of functional variants in sequence space and epistasis between residues. Our results illuminate the topology of the vast space of possibilities from which history sampled one path, highlighting how the outcome of evolution depends on a serial chain of compounding chance events.

Subject-specific modelling of paired comparison data: A lasso-type penalty approach

In traditional paired comparison models heterogeneity in the population is simply ignored and it is assumed that all persons or subjects have the same preference structure. In the models considered here the preference of an object over another object is explicitly modelled as depending on subject-specific covariates, therefore allowing for heterogeneity in the population. Since by construction the models contain a large number of parameters we propose to use penalized estimation procedures to obtain estimates of the parameters. The used regularized estimation approach penalizes the differences between the parameters corresponding to single covariates. It enforces variable selection and allows to find clusters of objects with respect to covariates. We consider simple binary but also ordinal paired comparisons models. The method is applied to data from a pre-election study from Germany.

Cost-sensitive Performance Metric for Comparing Multiple Ordinal Classifiers

The surge of interest in personalized and precision medicine during recent years has increased the application of ordinal classification problems in biomedical science. Currently, accuracy, Kendall's τ_b , and average mean absolute error are three commonly used metrics for evaluating the effectiveness of an ordinal classifier. Although there are benefits to each, no single metric considers the benefits of predictive accuracy with the tradeoffs of misclassification cost. In addition, decision analysis that considers pairwise analysis of the metrics is not trivial due to inconsistent findings. A new cost-sensitive metric is proposed to find the optimal tradeoff between the two most critical performance measures of a classification task - accuracy and cost. The proposed method accounts for an inherent ordinal data structure, total misclassification cost of a classifier, and imbalanced class distribution. The strengths of the new methodology are demonstrated through analyses of three real cancer datasets and four simulation studies. The new cost-sensitive metric proved better performance in its ability to identify the best ordinal classifier for a given analysis. The performance metric devised in this study provides a comprehensive tool for comparative analysis of multiple (and competing) ordinal classifiers. Consideration of the tradeoff between accuracy and misclassification cost in decisions regarding ordinal classification problems is imperative in real-world application. The work presented here is a precursor to the possibility of incorporating the proposed metric into a prediction modeling algorithm for ordinal data as a means of integrating misclassification cost in final model selection.

Bayesian predictive modeling for genomic based personalized treatment selection

Efforts to personalize medicine in oncology have been limited by reductive characterizations of the intrinsically complex underlying biological phenomena. Future advances in personalized medicine will rely on molecular signatures that derive from synthesis of multifarious interdependent molecular quantities requiring robust quantitative methods. However, highly parameterized statistical models when applied in these settings often require a prohibitively large database and are sensitive to proper characterizations of the treatment-by-covariate interactions, which in practice are difficult to specify and may be limited by generalized linear models. In this article, we present a Bayesian predictive framework that enables the integration of a high-dimensional set of genomic features with clinical responses and treatment histories of historical patients, providing a probabilistic basis for using the clinical and molecular information to personalize therapy for future patients. Our work represents one of the first attempts to define personalized treatment assignment rules based on large-scale genomic data. We use actual gene expression data acquired from The Cancer Genome Atlas in the settings of leukemia and glioma to explore the statistical properties of our proposed Bayesian approach for personalizing treatment selection. The method is shown to yield considerable improvements in predictive accuracy when compared to penalized regression approaches.

Modeling discrete survival time using genomic feature data

Researchers have recently shown that penalized models perform well when applied to high-throughput genomic data. Previous researchers introduced the generalized monotone incremental forward stagewise (GMIFS) method for fitting overparameterized logistic regression models. The GMIFS method was subsequently extended by others for fitting several different logit link ordinal response models to high-throughput genomic data. In this study, we further extended the GMIFS method for ordinal response modeling using a complementary log-log link, which allows one to model discrete survival data. We applied our extension to a publicly available microarray gene expression dataset (GSE53733) with a discrete survival outcome. The dataset included 70 primary glioblastoma samples from patients of the German Glioma Network with long-, intermediate-, and short-term overall survival. We tested the performance of our method by examining the prediction accuracy of the fitted model. The method has been implemented as an addition to the ordinalgmifs package in the R programming environment.

ordinalgmifs: An R Package for Ordinal Regression in High-dimensional Data Settings

High-throughput genomic assays are performed using tissue samples with the goal of classifying the samples as normal < pre-malignant < malignant or by stage of cancer using a small set of molecular features. In such cases, molecular features monotonically associated with the ordinal response may be important to disease development; that is, an increase in the phenotypic level (stage of cancer) may be mechanistically linked through a monotonic association with gene expression or methylation levels. Though traditional ordinal response modeling methods exist, they assume independence among the predictor variables and require the number of samples (n) to exceed the number of covariates (P) included in the model. In this paper, we describe our ordinalgmifs R package, available from the Comprehensive R Archive Network, which can fit a variety of ordinal response models when the number of predictors (P) exceeds the sample size (n). R code illustrating usage is also provided.

Sparse kernel machine regression for ordinal outcomes

Ordinal outcomes arise frequently in clinical studies when each subject is assigned to a category and the categories have a natural order. Classification rules for ordinal outcomes may be developed with commonly used regression models such as the full continuation ratio (CR) model (fCR), which allows the covariate effects to differ across all continuation ratios, and the CR model with a proportional odds structure (pCR), which assumes the covariate effects to be constant across all continuation ratios. For settings where the covariate effects differ between some continuation ratios but not all, fitting either fCR or pCR may lead to suboptimal prediction performance. In addition, these standard models do not allow for nonlinear covariate effects. In this article, we propose a sparse CR kernel machine (KM) regression method for ordinal outcomes where we use the KM framework to incorporate nonlinearity and impose sparsity on the overall differences between the covariate effects of continuation ratios to control for overfitting. In addition, we provide data driven rule to select an optimal kernel to maximize the prediction accuracy. Simulation results show that our proposed procedures perform well under both linear and nonlinear settings, especially when the true underlying model is in-between fCR and pCR models. We apply our procedures to develop a prediction model for levels of anti-CCP among rheumatoid arthritis patients and demonstrate the advantage of our method over other commonly used methods.

Risk stratification using data from electronic medical records better predicts suicide risks than clinician assessments

BACKGROUND

To date, our ability to accurately identify patients at high risk from suicidal behaviour, and thus to target interventions, has been fairly limited. This study examined a large pool of factors that are potentially associated with suicide risk from the comprehensive electronic medical record (EMR) and to derive a predictive model for 1-6 month risk.

METHODS

7,399 patients undergoing suicide risk assessment were followed up for 180 days. The dataset was divided into a derivation and validation cohorts of 4,911 and 2,488 respectively. Clinicians used an 18-point checklist of known risk factors to divide patients into low, medium, or high risk. Their predictive ability was compared with a risk stratification model derived from the EMR data. The model was based on the continuation-ratio ordinal regression method coupled with lasso (which stands for least absolute shrinkage and selection operator).

RESULTS

In the year prior to suicide assessment, 66.8% of patients attended the emergency department (ED) and 41.8% had at least one hospital admission. Administrative and demographic data, along with information on prior self-harm episodes, as well as mental and physical health diagnoses were predictive of high-risk suicidal behaviour. Clinicians using the 18-point checklist were relatively poor in predicting patients at high-risk in 3 months (AUC 0.58, 95% CIs: 0.50 - 0.66). The model derived EMR was superior (AUC 0.79, 95% CIs: 0.72 - 0.84). At specificity of 0.72 (95% CIs: 0.70-0.73) the EMR model had sensitivity of 0.70 (95% CIs: 0.56-0.83).

CONCLUSION

Predictive models applied to data from the EMR could improve risk stratification of patients presenting with potential suicidal behaviour. The predictive factors include known risks for suicide, but also other information relating to general health and health service utilisation.

The classification of cancer stage microarray data

Correctly diagnosing the cancer stage is most important for selecting an appropriate cancer treatment option for a patient. Recent advances in microarray technology allow the cancer stage to be predicted using gene expression patterns. The cancer stage is in ordinal scale. In this paper, we employ strict ordinal regressions including cumulative logit model in traditional statistics with data dimensionality reduction, and distribution free approaches of large margin rank boundaries implemented by the support vector machine, as well as an ensemble ranking scheme to model the cancer stage using gene expression microarray data. Predictive genes included in models are selected by univariate feature ranking, and recursive feature elimination. We perform cross-validation experiments to assess and compare classification accuracies of ordinal and non-ordinal algorithms on five cancer stage microarray datasets. We conclude that a strict ordinal classifier trained by a validated approach can predict the cancer stage more accurately than traditional non-ordinal classifiers without considering the order of cancer stages.