Binary Clinical Outcome Research Articles

A Reference Scaled Difference in Means Approach to Equivalence Test for Binary Clinical Outcomes

A proposed generic drug product needs to be compared with some reference product through equivalence test to support marketing approval. One important type of treatment outcome is the binary response indicating whether a favorable outcome is achieved. The binary response rates of the test and reference treatments are often compared via their difference with some margin. While the difference can usually be estimated in a clinical study, it is frequently difficult to determine a proper margin. Existing approaches suggest a fixed margin or a margin as a function of the reference response rate, such as step-wise constant margin, piece-wise smooth margin, etc. The issues with existing margin choices were discussed recently and a variable margin was proposed for non-inferiority test, which is a constant multiple of the standard deviation of reference response rate. In this paper, we extended the discussion to equivalence test, which can be formulated as two one-sided tests. Our discussion revealed some unique features of the equivalence test for binary responses with a variable margin. For instance, this approach may improve power control when the reference product has a high response rate. On the other hand, when both sample size and margin multiplier is small, the rejection rate of the equivalence test is nearly zero.

Associations between low body weight, weight loss, and medical instability in adults with eating disorders.

The physical complications of atypical anorexia nervosa remain understudied, with most studies completed in adolescents. This study seeks to examine the impact of various weight measures as predictors of medical instability in a large cohort of adult eating disorder patients. In this retrospective cohort study, the impact of admission body mass index (BMI), weight suppression, and recent weight loss (the rate of weight loss within the last 12 months) toward the development of medical complications of malnutrition were examined. Multivariable logistic regression assessed the association of binary clinical outcomes of interest with recent weight loss and weight suppression (adjusting for age, admission BMI, gender, and purging behaviors). Odds ratios (OR) and 99% confidence intervals were reported. Greater recent weight loss increased the odds of developing low prealbumin and reduced hand grip strength. A greater weight suppression was associated with increased likelihood of amenorrhea, reduced systolic blood pressure, nadir hemoglobin, and weekly weight gain upon nutritional rehabilitation. Lower admission BMI was predictive of all the medical outcomes examined, with the exception of bradycardia, and was generally the strongest predictor based on standardized coefficients. Recent weight loss and weight suppression are predictive of some of the physiologic changes of malnutrition, although low BMI is seemingly the greatest predictor for the development of these complications. These findings suggest that some patients with aggressive weight suppression and/or acute weight loss would benefit from medical stabilization, although this needs to be further defined. In adults, low BMI seems to be a better predictor of medical complications than weight suppression or aggressive recent weight loss. In adults, greater weight suppression is associated with increased likelihood of amenorrhea, reduced systolic blood pressure, nadir hemoglobin, and weight gain upon nutritional rehabilitation.

Biomarkers of Endometriosis

Background: Endometriosis is an under-recognised and often misdiagnosed chronic disease that affects one in ten women. It occurs when cells similar to those that line a uterus grow in other parts of the body, usually around the pelvis. The current gold standard for diagnosis of endometriosis is direct visualisation by histological analysis which can only be achieved by an invasive laparoscopy. It is estimated 176 million women and girls around the world are living with endometriosis. The condition costs $7.7 billion a year in direct medical expenses and lost productivity in Australia alone. Aim: To analyse a series of plasma samples from patients with clinically diagnosed endometriosis, others with gynaecological symptoms without endometriosis and otherwise healthy controls to identify biomarkers to differentiate these groups and potentially aid in endometriosis diagnosis. Method: A standard proteomic discovery workflow was used to search for unique biomarkers of endometriosis. Initially, pooled plasma samples from three distinct groups of patients were tested. The first group had symptoms of endometriosis with a laparoscopic/histology confirmed diagnosis. The second group also had symptoms, but the laparoscopy confirmed no endometriosis or an alternate diagnosis. The third group were healthy matched controls. The pooled plasma samples were quantitatively profiled against each other with a triplicate experiment. A validation of the biomarkers identified as significant in the discovery phase, was then undertaken in a larger clinical cohort (n=902) using a targeted multiplex mass spectrometry assay using multiple reaction monitoring (MRM). Statistical analyses were caried out using Stata and logistical regression was used to evaluate associations of biomarkers with binary clinical outcomes. Results: Out of the 943 proteins identified with [Formula: see text] 2 significant peptides in the initial discovery workflow, 47 were determined as being differential biomarkers across the three clinical groups. The derived targeted MRM assay was tested on the clinical validation cohort, and a number of candidate biomarkers were significantly associated, thereby indicating their potential for use in the diagnosis of endometriosis.

DAGBagM: learning directed acyclic graphs of mixed variables with an application to identify protein biomarkers for treatment response in ovarian cancer

BackgroundApplying directed acyclic graph (DAG) models to proteogenomic data has been shown effective for detecting causal biomarkers of complex diseases. However, there remain unsolved challenges in DAG learning to jointly model binary clinical outcome variables and continuous biomarker measurements.ResultsIn this paper, we propose a new tool, DAGBagM, to learn DAGs with both continuous and binary nodes. By using appropriate models, DAGBagM allows for either continuous or binary nodes to be parent or child nodes. It employs a bootstrap aggregating strategy to reduce false positives in edge inference. At the same time, the aggregation procedure provides a flexible framework to robustly incorporate prior information on edges.ConclusionsThrough extensive simulation experiments, we demonstrate that DAGBagM has superior performance compared to alternative strategies for modeling mixed types of nodes. In addition, DAGBagM is computationally more efficient than two competing methods. When applying DAGBagM to proteogenomic datasets from ovarian cancer studies, we identify potential protein biomarkers for platinum refractory/resistant response in ovarian cancer. DAGBagM is made available as a github repository at https://github.com/jie108/dagbagM.

Sensitive and dynamic CTCs measurement for prediction and monitoring of PD-1 therapy in GC/EGJC patients: Results from a pilot phase 2 study.

e16094 Background: Programmed death-ligand 1 (PD-L1) expression is currently being explored as a predictive biomarker for anti-PD-1antibody treatment in patients with advanced gastric or esophagogastric junction cancer (GC/EGJC) patients. Circulating tumor cells (CTCs) can be collected non-invasively and sequentially, and may allow real-time monitoring of immune activation in tumor. The aim of this study was to investigate whether PD-L1 expression on CTCs can accurately represent tumor PD-L1 expression and the potential of CTCs as a companion diagnosis (CDx) biomarker. Methods: We evaluated the CTCs counts and PD-L1 status on CTCs during treatment in a phase 2 clinical trial, where patients with previously treated gastric cancer received JS001 in combination with Apatinib. The multi-center trial was registered with ClinicalTrials.gov, number NCT04190745. Blood samples were collected from patients. Sixteen patients with GC were analyzed to investigate the association between clinical outcome and changes of CTC counts pre and post JS001 plus Apatinib. The correlation between the binary clinical outcome (SD/PR/CR vs. PD) and the change in CTCs counts pre and post treatment was analyzed using a logistic regression adjusting for baseline CTCs counts. PD-L1 expression on CTCs were analyzed to determine whether they were in accordance with tissue biopsy. Results: CTCs were detected in all the patients (16/16，100%) ranging from 1 to 33 CTCs/ml. PD-L1+ CTCs were detected in 12 of 16 patients (75%) ranging from 1 to 5 PD-L1 + CTCs/ml. PD-L1 expression in tumor tissue was positive in 11 of 13 patients (84.6%, positive means CPS > = 1), the sensitivity of PD-L1+ CTCs is 90.9% (10/11) if using biopsy as gold standard. The only false negative patient though with CPS 40%, got no benefit from immunotherapy. The specificity of PD-L1+ CTCs detection is 100%, and the accuracy is 90.9%. Overall response rate of this regimen is 26.6% (4/15). Disease control rate is 53.3% (8/15). Progression free survival (PFS) was 2.43 months in all the evaluable patients. However, patients with PD-L1+ CTCs had a longer PFS compared to patients without PD-L1+ CTC (3.30months vs 1.37months, p = 0.001). Overall Survival is not reached. Change of CTCs counts after treatment was significantly correlated with clinical response (p = 0.024). The logistic model McFadden goodness of fit score was 0.435, which is a strong correlation value. Notably, one of the patients who had pseudo-progression shows remarked decreased CTCs count. Conclusions: Our results reveal the potential of CTCs as a noninvasive CDx biomarker in patients with advanced GC/EGJC.

Confident identification of subgroups from SNP testing in RCTs with binary outcomes

In modern drug development, genotype information becomes more frequently collected in randomized controlled trials (RCTs) for individualized risk prediction and personalized medicine development. Finding single nucleotide polymorphisms (SNPs) that are predictive of differential treatment efficacy, measured by a clinical outcome, is fundamentally different and more challenging than the traditional association test for a quantitative trait. With the objective to confidently identify and infer genetic subgroups with enhanced treatment efficacy from a large RCT for an eye disease, age-related macular degeneration (AMD), where the clinical endpoint is binary (progressed or not), we propose a novel SNP-testing procedure for binary clinical outcomes. Specifically, we formulate four contrasts to simultaneously assess all possible genetic effects on a logic-respecting efficacy measure, the relative risk (between treatment and control). Our method controls both within- and across-SNP multiplicity rigorously. We then use real genotype data to perform chromosome-wide simulations to evaluate our method performance and to provide practical recommendations. Finally, we apply the proposed method to perform a genome-wide SNP testing for the AMD trial and successfully identify multiple gene regions with genetic subgroups exhibiting enhanced efficacy in terms of decreasing the AMD progressionrate.

In small cell lung cancer patients treated with RRx-001, a downregulator of CD47, decreased expression of PD-L1 on circulating tumor cells significantly correlates with clinical benefit

BackgroundSmall cell lung cancer (SCLC) is the most aggressive lung tumor, characterized by a rapid doubling time and the development of widespread metastases, for which immune checkpoint inhibitors have been approved to overcome T cell anergy. In light of its dismal prognosis, and lack of curative options, new therapies for extensive-disease SCLC are desperately needed.MethodsRRx-001 is a small molecule Myc inhibitor and down-regulates CD47 expression on tumor cells. We evaluated the programmed death-ligand 1 (PD-L1) status of circulating tumor cells (CTCs) pre and post RRx-001 treatment in a phase 2 clinical trial, called QUADRUPLE THREAT, where patients with previously treated SCLC received RRx-001 in combination with a platinum doublet. The trial was registered with ClinicalTrials.gov, number NCT02489903. Fourteen patients with SCLC were analyzed to investigate the association between clinical outcome and PD-L1 expression on CTCs pre and post RRx-001. The correlation between the binary clinical outcome (clinical benefit vs. progressive disease) and the change of PD-L1 expression on CTCs after RRx-001 was analyzed using a logistic regression adjusting for baseline PD-L1 expression.ResultsThe logistic model McFadden goodness of fit score was 0.477. The logistic model analyzing the association between decreased PD-L1 expression on CTCs after RRx-001 and response to reintroduced platinum doublet had an approximate 92.8% accuracy in its prediction of clinical benefit. The estimated receiver operating characteristic (ROC) displayed a ROC area under the curve (AUC) of 0.93 (95% confidence interval, 0.78–0.99).ConclusionsThese results suggest that PD-L1 expression on CTCs decreased after RRx-001 was significantly correlated with response to reintroduced platinum-based doublet therapy. Monitoring PD-L1 expression on CTCs during RRx-001 treatment may serve as a biomarker to predict response to RRx-001-based cancer therapy.

Bayesian biomarker-driven outcome-adaptive randomization with an imperfect biomarker assay.

We investigate the impact of biomarker assay's accuracy on the operating characteristics of a Bayesian biomarker-driven outcome-adaptive randomization design. In a simulation study, we assume a trial with two treatments, two biomarker-based strata, and a binary clinical outcome (response). Pbt denotes the probability of response for treatment t (t = 0 or 1) in biomarker stratum (b = 0 or 1). Four different scenarios in terms of true underlying response probabilities are considered: a null (P00 = P01 = 0.25, P10 = P11= 0.25) and consistent (P00 = P10 = 0.25, P01 = 0.5) treatment effect scenario, as well as a quantitative (P00 = P01 = P10 = 0.25, P11 = 0.5) and a qualitative (P00 = P11 = 0.5, P01 = P10 = 0.25) stratum-treatment interaction. For each scenario, we compare the case of a perfect with the case of an imperfect biomarker assay with sensitivity and specificity of 0.8 and 0.7, respectively. In addition, biomarker-positive prevalence values P(B = 1) = 0.2 and 0.5 are investigated. Results show that the use of an imperfect assay affects the operational characteristics of the Bayesian biomarker-based outcome-adaptive randomization design. In particular, the misclassification causes a substantial reduction in power accompanied by a considerable increase in the type-I error probability. The magnitude of these effects depends on the sensitivity and specificity of the assay, as well as on the distribution of the biomarker in the patient population. With an imperfect biomarker assay, the decision to apply a biomarker-based outcome-adaptive randomization design may require careful reflection.

Predictors of the Clinical Course of Tourette Syndrome: A Longitudinal Study.

Tourette syndrome (TS) is a chronic childhood neurodevelopmental disorder characterized by motor and vocal tics and frequent comorbidities. The clinical presentation of Tourette syndrome is heterogeneous and the prognosis for each individual child is difficult to define. This large prospective longitudinal study explores predictors in childhood of the clinical course of tics and comorbidities in early adulthood. The cohort was recruited at the Danish National Tourette Clinic. Data were collected at baseline (N = 314; ages, 5-19 years) and follow-up 6 years later (n = 227; ages, 11-26 years) to examine changes in the expression of tics and comorbidities. Childhood clinical factors, represented by 4 binary clinical outcomes, were selected as possible predictors of the clinical course of tics and comorbidities in early adulthood; these were tic severity and diagnoses of obsessive compulsive disorder (OCD), attention-deficit hyperactivity disorder (ADHD), and emotional disorders. The strongest predictors of high tic scores, OCD, or ADHD diagnoses in early adulthood were the corresponding tic (odds ratio [OR]: 1.09), OCD (OR: 1.08), and ADHD (OR: 1.13) severity scores (per scale point) in childhood. Being female (OR: 3.94) and childhood ADHD severity (OR: 1.11) predicted future emotional disorders. Special education, genetic factors, and psychosocial factors were also predictive for the clinical course of Tourette syndrome. We identified strong clinical predictors of Tourette syndrome-associated outcomes in early adulthood that are directly applicable to clinical Tourette syndrome populations and may help to guide new patients, plan early interventions, and implement preventive measures.

Leveraging Multilayered “Omics” Data for Atopic Dermatitis: A Road Map to Precision Medicine

Atopic dermatitis (AD) is a complex multifactorial inflammatory skin disease that affects ~280 million people worldwide. About 85% of AD cases begin in childhood, a significant portion of which can persist into adulthood. Moreover, a typical progression of children with AD to food allergy, asthma or allergic rhinitis has been reported (“allergic march” or “atopic march”). AD comprises highly heterogeneous sub-phenotypes/endotypes resulting from complex interplay between intrinsic and extrinsic factors, such as environmental stimuli, and genetic factors regulating cutaneous functions (impaired barrier function, epidermal lipid, and protease abnormalities), immune functions and the microbiome. Though the roles of high-throughput “omics” integrations in defining endotypes are recognized, current analyses are primarily based on individual omics data and using binary clinical outcomes. Although individual omics analysis, such as genome-wide association studies (GWAS), can effectively map variants correlated with AD, the majority of the heritability and the functional relevance of discovered variants are not explained or known by the identified variants. The limited success of singular approaches underscores the need for holistic and integrated approaches to investigate complex phenotypes using trans-omics data integration strategies. Integrating omics layers (e.g., genome, epigenome, transcriptome, proteome, metabolome, lipidome, exposome, microbiome), which often have complementary and synergistic effects, might provide the opportunity to capture the flow of information underlying AD disease manifestation. Overlapping genes/candidates derived from multiple omics types include FLG, SPINK5, S100A8, and SERPINB3 in AD pathogenesis. Overlapping pathways include macrophage, endothelial cell and fibroblast activation pathways, in addition to well-known Th1/Th2 and NFkB activation pathways. Interestingly, there was more multi-omics overlap at the pathway level than gene level. Further analysis of multi-omics overlap at the tissue level showed that among 30 tissue types from the GTEx database, skin and esophagus were significantly enriched, indicating the biological interconnection between AD and food allergy. The present work explores multi-omics integration and provides new biological insights to better define the biological basis of AD etiology and confirm previously reported AD genes/pathways. In this context, we also discuss opportunities and challenges introduced by “big omics data” and their integration.

Neighborhood Death Rate: A Big-Data Based Health Informatics Pipeline Helps Identify Factors Impacting Clinical Outcomes and Cost Effectiveness in Acute Myeloid Leukemia Patients

BackgroundAcute Myeloid Leukemia (AML) affects 18,860 adults in the US and is expected to account for 10,460 deaths of all mortalities due to leukemia. Several models have been constructed comparing different treatment strategies with their economic burden. We sought to investigate the early deaths in AML using a unique model utilizing the predictability of sum total of patient’s neighborhood characteristics at the level of 5% and its applicability to assess the correlation between the overall outcomes and length of stay.MethodsThe patient data was queried from METEOR (Methodist Environment for Translational Enhancement and Outcomes Research), a clinical data warehouse and analytics environment that integrates existing business data warehouse with internal and external research databases and national registries to support clinical research and outcomes studies for improving patient care cost-effectively. METEOR data warehouse contains records dating back to January 1, 2006, with over 1 million unique patients and over 4 million unique patient encounters. In order to evaluate the correlation between integral/continuous variables (e.g. length of stay a.k.a. LoS) and binary clinical outcomes (e.g. survival/expired), we developed a statistical pipeline based on the concept of “Neighborhood Death Rate (NDR)”. For a specific patient X, the mean length of stay across all his/her encounters serves as the model input; and a series of neighborhood patient population are defined as all patients with mean length of stay within ±5%, ±15% or ±25% range of the number for X; the death rates within the neighborhood population then serve as a output. Thus for each individual patient, a series of NDRs are defined, and the correlation coefficients between mean LoS and 5%, 15% and 25% NDR are available. These correlations are then individually corrected for baseline demographic factors sex and race; number of encounters; and the first order statistics (mean, median, sum, minimum and maximum) for total cost and cost without bone marrow transplant corresponding to each encounter. Such statistics corrections help identify the driving force underlying the “LoS-NDR” correlations.ResultsOur database has records for 12943 encounters from 1141 patients, 838 of which are currently alive survival 73%) . The groups of survived and expired patients show significant difference on the mean LoS (p-value 2.04-E22 from Wilcoxon rank sum test). Figure 1A plots the distributions of survived (green) and expired (red) patient population separately. X-axis shows the log2 value of mean LoS; for each patient X, we calculate the number of patient with both 1) the same outcome as X and 2) a mean length of stay within ±15% range of X; and this number is then normalized using the total size of patients with the same outcome as X. [Display omitted] [Display omitted] The two outcome groups were then merged together to calculate the ±15% NDR for mean LoS (Figure 1B, X-axis show the log2 value of mean LoS, red nodes indicates expired patients). The Spearman’s Rho correlation between mean LoS and its ±15% NDR is 0.8390; meanwhile between mean LoS of 2 and 3 days, the NDR makes a significant jump from below 0.1 to above 0.3, which may represent a critical predictor for patient outcomes.During the statistical corrections, the Spearman correlations changed less than 0.01 when corrected by Sex, Race and #of encounters, indicating limited impact from these baseline factors. However, the correlation reduced to 0.6906 when corrected by mean total cost, and to 0.7301 when corrected by mean total cost without bone marrow transplant. Table 1 show the results for individual statistics corrections for Spearman correlations of “mean LoS-±15% NDR” and “median LoS-±15% NDR”. [Display omitted] ConclusionsPublished models for longevity and cost effectiveness often do not apply to the induction treatment as this a time-lapse hospital data for the duration is not usually available. Our unique analysis of Neighborhood death rates revealed that total cost is the main read-out that underlies the high correlation between length of stay and neighborhood death rates in patients with AML. We further hypothesis that medication cost (infection, critical care) and intensive care infrastructure are the main driving forces underlying the high correlation between LoS and NDR; and this model may help us obtain novel strategies to improve the cost effectiveness for AML patient care. DisclosuresNo relevant conflicts of interest to declare.

Epidemiologic research using probabilistic outcome definitions.

Epidemiologic studies using electronic healthcare data often define the presence or absence of binary clinical outcomes by using algorithms with imperfect specificity, sensitivity, and positive predictive value. This results in misclassification and bias in study results. We describe and evaluate a new method called probabilistic outcome definition (POD) that uses logistic regression to estimate the probability of a clinical outcome using multiple potential algorithms and then uses multiple imputation to make valid inferences about the risk ratio or other epidemiologic parameters of interest. We conducted a simulation to evaluate the performance of the POD method with two variables that can predict the true outcome and compared the POD method with the conventional method. The simulation results showed that when the true risk ratio is equal to 1.0 (null), the conventional method based on a binary outcome provides unbiased estimates. However, when the risk ratio is not equal to 1.0, the traditional method, either using one predictive variable or both predictive variables to define the outcome, is biased when the positive predictive value is <100%, and the bias is very severe when the sensitivity or positive predictive value is poor (less than 0.75 in our simulation). In contrast, the POD method provides unbiased estimates of the risk ratio both when this measure of effect is equal to 1.0 and not equal to 1.0. Even when the sensitivity and positive predictive value are low, the POD method continues to provide unbiased estimates of the risk ratio. The POD method provides an improved way to define outcomes in database research. This method has a major advantage over the conventional method in that it provided unbiased estimates of risk ratios and it is easy to use.

Peripheral Blood Absolute Lymphocyte/Monocyte Ratio Recovery During RCHOP Treatment Cycles Predicts Clinical Outcomes In Diffuse Large B-Cell Lymphoma

The peripheral blood absolute lymphocyte/monocyte ratio at diagnosis (ALC/AMC-DX) predicts survival in diffuse large B-cell lymphoma (DLBCL). However, a limitation of the ALC/AMC-DX is its inability to sequentially assess the host/tumor interaction during treatment. Therefore, we studied the ALC/AMC ratio, as a surrogate marker of host immunity (i.e., ALC) and tumor microenvironment (i.e., AMC), at each RCHOP treatment cycle as a predictor for clinical outcomes. From 2002 until 20011, 107 DLBCL patients originally diagnosed, treated with six cycles of RCHOP, and followed at Mayo Clinic qualified for the study. Using the receiver operating characteristic and internally validated using the k-fold cross validation; an ALC/AMC ratio ≥ 1.1 was identified as the cut-off value at each RCHOP treatment cycle to discriminate for the binary clinical outcome of death/survival. Univariate analysis revealed the ALC/AMC ratio ≥ 1.1 at each RCHOP treatment cycle was a predictor for overall survival (OS) and progression-free survival (PFS). To explore whether the patterns of ALC/AMC ratio recovery during the RCHOP treatment cycles had a prognostic value, we performed an unsupervised hierarchical clustering on the studied ALC/AMC ratio and identified four patient clusters, A-D. Eighty percent of patients in cluster C had an ALC/AMC ratio ≥ 1.1 during all the cycles, the other 20% had an ALC/AMC ratio ≥1.1 in one cycle; in cluster D patients, the ALC/AMC ratio was ≥ 1.1 between cycles 2 and 5 and none had an ALC/AMC ratio < 1.1; in cluster A patients, 73% of patients only had an ALC/AMC ratio ≥ 1.1 in one cycle and 27% in two cycles; and cluster B patients had 85% of patients with an ALC/AMC ratio < 1.1 during all the cycles. Inferior OS and PFS were progressively observed in clusters of patients with fewer cycles achieving an ALC/AMC ratio ≥ 1.1. Patients in clusters C had superior outcomes compared with patients in clusters B. Thus, we dichotomized patients into patients with an ALC/AMC ratio ≥ 1.1 during any cycle versus patients with an ALC/AMC ratio < 1.1 during all cycles. By multivariate analysis, the stratification of patients into ALC/AMC ≥ 1.1 during any cycle versus ALC/AMC < 1.1 during all cycles was an independent predictor for OS [Hazard ratio (HR) = 0.02, 95% CI (0.006-0.08), p< 0.0001], and PFS [HR = 0.06, 95% CI (0.02-0.15, p< 0.0001] when compared with the International Prognostic index The median OS for patients with an ALC/AMC ratio ≥ 1.1 during any cycle versus patients with an ALC/ACM ratio < 1.1 during all cycles was not reached versus 1.7 years with 5 years OS rates of 97% (95% CI, 90%-99%) versus 0%, p< 0.0001, respectively. The median PFS for patients with an ALC/AMC ratio ≥ 1.1 during any cycle versus patients with an ALC/AMC ratio < 1.1during all cycles was not reached versus 0.9 years with 5 years PFS rates of 83% (95% CI, 72%-90%) versus 0%, p< 0.0001, respectively. The ALC/AMC ratio during RCHOP treatment cycles is a predictor for survival and provides a platform to develop therapeutic modalities to manipulate the ALC/AMC ratio during chemotherapy to improve clinical outcomes in DLBCL. Disclosures:No relevant conflicts of interest to declare.

SNP Selection in Genome-Wide Association Studies via Penalized Support Vector Machine with MAX Test

One of main objectives of a genome-wide association study (GWAS) is to develop a prediction model for a binary clinical outcome using single-nucleotide polymorphisms (SNPs) which can be used for diagnostic and prognostic purposes and for better understanding of the relationship between the disease and SNPs. Penalized support vector machine (SVM) methods have been widely used toward this end. However, since investigators often ignore the genetic models of SNPs, a final model results in a loss of efficiency in prediction of the clinical outcome. In order to overcome this problem, we propose a two-stage method such that the the genetic models of each SNP are identified using the MAX test and then a prediction model is fitted using a penalized SVM method. We apply the proposed method to various penalized SVMs and compare the performance of SVMs using various penalty functions. The results from simulations and real GWAS data analysis show that the proposed method performs better than the prediction methods ignoring the genetic models in terms of prediction power and selectivity.

Bayesian semiparametric regression models for evaluating pathway effects on continuous and binary clinical outcomes.

Many statistical methods for microarray data analysis consider one gene at a time, and they may miss subtle changes at the single gene level. This limitation may be overcome by considering a set of genes simultaneously where the gene sets are derived from prior biological knowledge. Limited work has been carried out in the regression setting to study the effects of clinical covariates and expression levels of genes in a pathway either on a continuous or on a binary clinical outcome. Hence, we propose a Bayesian approach for identifying pathways related to both types of outcomes. We compare our Bayesian approaches with a likelihood-based approach that was developed by relating a least squares kernel machine for nonparametric pathway effect with a restricted maximum likelihood for variance components. Unlike the likelihood-based approach, the Bayesian approach allows us to directly estimate all parameters and pathway effects. It can incorporate prior knowledge into Bayesian hierarchical model formulation and makes inference by using the posterior samples without asymptotic theory. We consider several kernels (Gaussian, polynomial, and neural network kernels) to characterize gene expression effects in a pathway on clinical outcomes. Our simulation results suggest that the Bayesian approach has more accurate coverage probability than the likelihood-based approach, and this is especially so when the sample size is small compared with the number of genes being studied in a pathway. We demonstrate the usefulness of our approaches through its applications to a type II diabetes mellitus data set. Our approaches can also be applied to other settings where a large number of strongly correlated predictors are present.

Genomic Biomarkers for a Binary Clinical Outcome in Early Drug Development Microarray Experiments

In this article, we discuss methods to select three different types of genes (treatment related, response related, or both) and investigate whether they can serve as biomarkers for a binary outcome variable. We consider an extension of the joint model introduced by Lin et al. (2010) and Tilahun et al. (2010) for a continuous response. As the model has certain drawbacks in a binary setting, we also present a way to use classical selection methods to identify subgroups of genes, which are treatment and/or response related. We evaluate their potential to serve as biomarkers by applying DLDA to predict the response level.

Multilocus association testing with penalized regression

In multilocus association analysis, since some markers may not be associated with a trait, it seems attractive to use penalized regression with the capability of automatic variable selection. On the other hand, in spite of a rapidly growing body of literature on penalized regression, most focus on variable selection and outcome prediction, for which penalized methods are generally more effective than their nonpenalized counterparts. However, for statistical inference, i.e. hypothesis testing and interval estimation, it is less clear how penalized methods would perform, or even how to best apply them, largely due to lack of studies on this topic. In our motivating data for a cohort of kidney transplant recipients, it is of primary interest to assess whether a group of genetic variants are associated with a binary clinical outcome, acute rejection at 6 months. In this article, we study some technical issues and alternative implementations of hypothesis testing in Lasso penalized logistic regression, and compare their performance with each other and with several existing global tests, some of which are specifically designed as variance component tests for high-dimensional data. The most interesting, and perhaps surprising, conclusion of this study is that, for low to moderately high-dimensional data, statistical tests based on Lasso penalized regression are not necessarily more powerful than some existing global tests. In addition, in penalized regression, rather than building a test based on a single selected "best" model, combining multiple tests, each of which is built on a candidate model, might be more promising.

Semiparametric estimation for joint modeling of colorectal cancer risk and functional biomarkers measured with errors.

This research is motivated by a pilot colorectal adenoma study, where the outcome of interest is the presence of colorectal adenoma representing risk for colorectal cancer, and the predictors of interest are protein biomarkers that are repeatedly measured with errors along the length of a microscopic structure in the human colon, the colon crypt. Biomarkers of this type are referred to as functional biomarkers. The investigators are interested in identifying features of functional biomarkers that are associated with risk for colorectal cancer. In this paper, we investigate a joint modeling approach, where the binary clinical outcome is modeled using a logistic regression model with the unobserved true functional biomarkers as the predictors. Most existing methods are developed either for linear models or for functional biomarkers measured without errors and cannot be directly applied to our data. The applicable methods include a two-step method and a maximum likelihood method, which have some limitations. We propose a robust semiparametric method to overcome the limitations of the existing methods. We study the properties of the proposed method, and show in simulations that it compares favorably with other methods and also offers significant savings in CPU time. We analyze the pilot colorectal adenoma data and show that expression levels of AFC, a tumor suppressor gene, in the transitional area from the proliferation zone to the differentiation zone of colon crypts are likely to be associated with risk for colorectal cancer. Given the relatively small sample size in the pilot study, our results need to be validated in the future full-scale studies.

Analysis of biomarker data: logs, odds ratios, and receiver operating characteristic curves.

We discuss two data analysis issues for studies that use binary clinical outcomes (whether or not an event occurred): the choice of an appropriate scale and transformation when biomarkers are evaluated as explanatory factors in logistic regression and assessing the ability of biomarkers to improve prediction accuracy for event risk. Biomarkers with skewed distributions should be transformed before they are included as continuous covariates in logistic regression models. The utility of new biomarkers may be assessed by measuring the improvement in predicting event risk after adding the biomarkers to an existing model. The area under the receiver operating characteristic (ROC) curve (C-statistic) is often cited; it was developed for a different purpose, however, and may not address the clinically relevant questions. Measures of risk reclassification and risk prediction accuracy may be more appropriate. The appropriate analysis of biomarkers depends on the research question. Odds ratios obtained from logistic regression describe associations of biomarkers with clinical events; failure to accurately transform the markers, however, may result in misleading estimates. Although the C-statistic is often used to assess the ability of new biomarkers to improve the prediction of event risk, other measures may be more suitable.