Regularized Ordinal Regression and the ordinalNet R Package

Dysregulated STAT3 signaling and T cell immunometabolic dysfunction define a targetable, high mortality subphenotype of critically ill children

Sepsis is the leading cause of death of hospitalized children worldwide. Despite the established link between immune dysregulation and mortality in pediatric sepsis, it remains unclear which host immune factors contribute causally to adverse sepsis outcomes. Identifying modifiable pathobiology is an essential first step to successful translation of biologic insights into precision therapeutics. We designed a prospective, longitudinal cohort study of 88 critically ill pediatric patients with multiple organ dysfunction syndrome (MODS), including patients with and without sepsis, to define subphenotypes associated with targetable mechanisms of immune dysregulation. We first assessed plasma proteomic profiles and identified shared features of immune dysregulation in MODS patients with and without sepsis. We then employed consensus clustering to define three subphenotypes based on protein expression at disease onset and identified a strong association between subphenotype and clinical outcome. We next identified differences in immune cell frequency and activation state by MODS subphenotype and determined the association between hyperinflammatory pathway activation and cellular immunophenotype. Using single cell transcriptomics, we demonstrated STAT3 hyperactivation in lymphocytes from the sickest MODS subgroup and then identified an association between STAT3 hyperactivation and T cell immunometabolic dysregulation. Finally, we compared proteomics findings between patients with MODS and patients with inborn errors of immunity that amplify cytokine signaling pathways to further assess the impact of STAT3 hyperactivation in the most severe patients with MODS. Overall, these results identify a potentially pathologic and targetable role for STAT3 hyperactivation in a subset of pediatric patients with MODS who have high severity of illness and poor prognosis.

Visual detection of seizures in mice using supervised machine learning

Seizures are caused by abnormally synchronous brain activity that can result in changes in muscle tone, such as twitching, stiffness, limpness, or rhythmic jerking. These behavioral manifestations are clear on visual inspection and the most widely used seizure scoring systems in preclinical models, such as the Racine scale in rodents, use these behavioral patterns in semiquantitative seizure intensity scores. However, visual inspection is time-consuming, low-throughput, and partially subjective, and there is a need for rigorously quantitative approaches that are scalable. In this study, we used supervised machine learning approaches to develop automated classifiers to predict seizure severity directly from noninvasive video data. Using the PTZ-induced seizure model in mice, we trained video-only classifiers to predict ictal events, combined these events to predict an univariate seizure intensity for a recording session, as well as time-varying seizure intensity scores. Our results show, for the first time, that seizure events and overall intensity can be rigorously quantified directly from overhead video of mice in a standard open field using supervised approaches. These results enable high-throughput, noninvasive, and standardized seizure scoring for downstream applications such as neurogenetics and therapeutic discovery.

Scoring system for diagnosis and pretreatment risk assessment of neuroblastoma using urinary biomarker combinations

The urinary catecholamine metabolites, homovanillic acid (HVA) and vanillylmandelic acid (VMA), are used for the adjunctive diagnosis of neuroblastomas. We aimed to develop a scoring system for the diagnosis and pretreatment risk assessment of neuroblastoma, incorporating age and other urinary catecholamine metabolite combinations. Urine samples from 227 controls (227 samples) and 68 patients with neuroblastoma (228 samples) were evaluated. First, the catecholamine metabolites vanillactic acid (VLA) and 3-methoxytyramine sulfate (MTS) were identified as urinary marker candidates through comprehensive analysis using liquid chromatography-mass spectrometry. The concentrations of these marker candidates and conventional markers were then compared among controls, patients, and numerous risk groups to develop a scoring system. Participants were classified into four groups: control, low risk, intermediate risk, and high risk, and the proportional odds model was fitted using the L2-penalized maximum likelihood method, incorporating age on a monthly scale for adjustment. This scoring model using the novel urine catecholamine metabolite combinations, VLA and MTS, had greater area under the curve values than the model using HVA and VMA for diagnosis (0.978 vs. 0.964), pretreatment risk assessment (low and intermediate risk vs. high risk: 0.866 vs. 0.724; low risk vs. intermediate and high risk: 0.871 vs. 0.680), and prognostic factors (MYCN status: 0.741 vs. 0.369, histology: 0.932 vs. 0.747). The new system also had greater accuracy in detecting missing high-risk neuroblastomas, and in predicting the pretreatment risk at the time of screening. The new scoring system employing VLA and MTS has the potential to replace the conventional adjunctive diagnostic method using HVA and VMA.

Gross Hematuria Does not Affect the Selection of Nephrectomy Types for Clinical Stage 1 Clear Cell Renal Cell Carcinoma: A Multicenter, Retrospective Cohort Study

PURPOSE

This study aimed to discuss the correlation between gross hematuria and postoperative upstaging (from T1 to T3a) in patients with cT1 clear cell renal cell carcinoma (ccRCC) and to compare oncologic outcomes of partial nephrectomy (PN) and radical nephrectomy (RN) in patients with gross hematuria.

METHODS

A total of 2145 patients who met the criteria were enrolled in the study (including 363 patients with gross hematuria). The least absolute selection and shrinkage operator logistic regression was used to evaluate the risk factor of postoperative pathological upstaging. The propensity score matching (PSM) and stable inverse probability of treatment weighting (IPTW) analysis were used to balance the confounding factors. The Kaplan-Meier analysis and multivariate Cox proportional risk regression model were used to assess the prognosis.

RESULTS

Gross hematuria was a risk factor of postoperative pathological upstaging (odds ratio [OR] = 3.96; 95% confidence interval [CI] 2.44-6.42; P < 0.001). After PSM and stable IPTW adjustment, the characteristics were similar in corresponding patients in the PN and RN groups. In the PSM cohort, PN did not have a statistically significant impact on recurrence-free survival (hazard ratio [HR] = 1.48; 95% CI 0.25-8.88; P = 0.67), metastasis-free survival (HR = 1.24; 95% CI 0.33-4.66; P = 0.75), and overall survival (HR = 1.46; 95% CI 0.31-6.73; P = 0.63) compared with RN. The results were confirmed in sensitivity analyses.

CONCLUSIONS

Although gross hematuria was associated with postoperative pathological upstaging in patients with cT1 ccRCC, PN should still be the preferred treatment for such patients.

The influence of resolution on the predictive power of spatial heterogeneity measures as biomarkers of liver fibrosis

Spatial heterogeneity of cells in liver biopsies can be used as biomarker for disease severity of patients. This heterogeneity can be quantified by non-parametric statistics of point pattern data, which make use of an aggregation of the point locations. The method and scale of aggregation are usually chosen ad hoc, despite values of the aforementioned statistics being heavily dependent on them. Moreover, in the context of measuring heterogeneity, increasing spatial resolution will not endlessly provide more accuracy. The question then becomes how changes in resolution influence heterogeneity indicators, and subsequently how they influence their predictive abilities. In this paper, cell level data of liver biopsy tissue taken from chronic Hepatitis B patients is used to analyze this issue. Firstly, Morisita-Horn indices, Shannon indices and Getis-Ord statistics were evaluated as heterogeneity indicators of different types of cells, using multiple resolutions. Secondly, the effect of resolution on the predictive performance of the indices in an ordinal regression model was investigated, as well as their importance in the model. A simulation study was subsequently performed to validate the aforementioned methods. In general, for specific heterogeneity indicators, a downward trend in predictive performance could be observed. While for local measures of heterogeneity a smaller grid-size is outperforming, global measures have a better performance with medium-sized grids. In addition, the use of both local and global measures of heterogeneity is recommended to improve the predictive performance.

Sparse ordinal discriminant analysis

Ordinal class labels are frequently observed in classification studies across various fields. In medical science, patients' responses to a drug can be arranged in the natural order, reflecting their recovery postdrug administration. The severity of the disease is often recorded using an ordinal scale, such as cancer grades or tumor stages. We propose a method based on the linear discriminant analysis (LDA) that generates a sparse, low-dimensional discriminant subspace reflecting the class orders. Unlike existing approaches that focus on predictors marginally associated with ordinal labels, our proposed method selects variables that collectively contribute to the ordinal labels. We employ the optimal scoring approach for LDA as a regularization framework, applying an ordinality penalty to the optimal scores and a sparsity penalty to the coefficients for the predictors. We demonstrate the effectiveness of our approach using a glioma dataset, where we predict cancer grades based on gene expression. A simulation study with various settings validates the competitiveness of our classification performance and demonstrates the advantages of our approach in terms of the interpretability of the estimated classifier with respect to the ordinal class labels.

Factors Affecting Adherence to National Colorectal Cancer Screening: A 12-Year Longitudinal Study Using Multi-Institutional Pooled Data in Korea

BACKGROUND

Consistent uptake of colorectal cancer (CRC) screening is important to reduce the incidence and mortality from advanced-stage CRC and increase the survival rate of the patients. We conducted a longitudinal study to determine the factors affecting CRC screening compliance in Korean adults using individual-level linked data from the Korean National Health and Nutrition Examination Survey, Korean National Health Insurance Service, and Korean Health Insurance Review and Assessment Service.

METHODS

We selected 3,464 adults aged 50-79 years as the study population and followed them for 12 years (January 2007-December 2018). The outcome variable was the level of adherence to CRC screening, categorized as nonadherent, intermittently adherent, and consistently adherent. An ordinal logistic regression model was designed to determine the socioeconomic factors, family history of CRC, and medical conditions that could facilitate the consistent uptake of CRC screening.

RESULTS

The results showed a significant and positive association between consistent uptake of CRC screening and the 100-150% income category (odds ratio [OR], 1.710; 95% confidence interval [CI], 1.401-2.088); clerical, sales and service job category (OR, 1.962; 95% CI, 1.582-2.433); residency at medium-sized cities (OR, 1.295; 95% CI, 1.094-1.532); high-school graduation (OR, 1.440; 95% CI, 1.210-1.713); married status (OR, 2.281; 95% CI, 1.946-2.674); use of employment-based national health insurance (OR, 1.820; 95% CI, 1.261-2.626); use of private insurance (OR, 2.259; 95% CI, 1.970-2.589); no disability (OR, 1.428; 95% CI, 1.175-1.737); family history of CRC (OR, 2.027; 95% CI, 1.514-2.714); and history of colorectal neoplasm (OR, 1.216; 95% CI; 1.039-1.422).

CONCLUSION

The lack of regular participation in CRC screening programs in the Republic of Korea was found to be an issue that requires attention. Policies on CRC screening must place increased emphasis on strengthening educational and public relations initiatives.

Development and validation of a LASSO-based prediction model for immunosuppressive medication nonadherence in kidney transplant recipients

INTRODUCTION

To establish a prediction model to predict immunosuppressive medication (IM) nonadherence in kidney transplant recipients (KTRs) based on a combined theory framework.

METHODS

This polycentric, cross-sectional study included 1191 KTRs from October 2020 to February 2021 in China, with 1011 KTRs enrolled in the derivation set and 180 in the external validation set. Variables selected based on the combined theory of planned behavior (TPB)/health belief model (HBM) theory were analyzed by the least absolute shrinkage and selection operator (LASSO). Internal 10 cross-validation was conducted to determine the optimal lambda value. The receiver operating characteristic (ROC) curve, specificity, and sensitivity were used to evaluate the prediction model, and further assessment was run by external validation.

RESULTS

IM nonadherence rate was 38.48% in the derivation set and 37.22% in the validation set. The LASSO model was developed with eight predictors for IM nonadherence: age, preoperative drinking history, education, marital status, perceived barriers, social support, perceived behavioral control, and perceived susceptibility. The model demonstrated acceptable discrimination with the area under the ROC curve of 0.797 (95% CI: 0.745-0.850) in the internal validation set and 0.757 (95% CI: 0.684-0.829) in the external validation set. The specificity and sensitivity in the internal validation and external validation set were 0.741, 0.748, 0.673, and 0.716, respectively.

CONCLUSIONS

The LASSO model was developed to guide identifying high-risk nonadherent patients and timely and effective interventions to improve their prognosis and survival.

ORTH.Ord: An R package for analyzing correlated ordinal outcomes using alternating logistic regressions with orthogonalized residuals

BACKGROUND AND OBJECTIVES

Marginal models with generalized estimating equations (GEE) are usually recommended for analyzing correlated ordinal outcomes which are commonly seen in a longitudinal study or clustered randomized trial (CRT). Within-cluster association is often of interest in longitudinal studies or CRTs, and can be estimated with paired estimating equations. However, the estimators for within-cluster association parameters and variances may be subject to finite-sample biases when the number of clusters is small. The objective of this article is to introduce a newly developed R package ORTH.Ord for analyzing correlated ordinal outcomes using GEE models with finite-sample bias corrections.

METHODS

The R package ORTH.Ord implements a modified version of alternating logistic regressions with estimation based on orthogonalized residuals (ORTH), which use paired estimating equations to jointly estimate parameters in marginal mean and association models. The within-cluster association between ordinal responses is modeled by global pairwise odds ratios (POR). The R package also provides a finite-sample bias correction to POR parameter estimates based on matrix multiplicative adjusted orthogonalized residuals (MMORTH) for correcting estimating equations, and bias-corrected sandwich estimators with different options for covariance estimation.

RESULTS

A simulation study shows that MMORTH provides less biased global POR estimates and coverage of their 95% confidence intervals closer to the nominal level than uncorrected ORTH. An analysis of patient-reported outcomes from an orthognathic surgery clinical trial illustrates features of ORTH.Ord.

CONCLUSIONS

This article provides an overview of the ORTH method with bias-correction on both estimating equations and sandwich estimators for analyzing correlated ordinal data, describes the features of the ORTH.Ord R package, evaluates the performance of the package using a simulation study, and finally illustrates its application in an analysis of a clinical trial.

An Item-Response Mapping from General Health Questionnaire Responses to EQ-5D-3L Using a General Population Sample from England

BACKGROUND

The 12-item General Health Questionnaire (GHQ-12) is widely used to measure mental health and well-being; however, it is not possible to estimate values on the full health = 1, dead = 0 scale used to construct quality-adjusted life-years (QALYs) from GHQ-12 responses as it is not preference-based.

OBJECTIVE

The aim of this study was to create an item-response mapping between GHQ-12 and EQ-5D-3L health states, for which several value sets exist.

METHODS

Data from the 2012 Health Survey for England with complete GHQ-12 and EQ-5D-3L descriptive system responses were used for analysis. Data were split 70/30 into estimation/test samples. Four modelling approaches, with EQ-5D-3L levels on each dimension as dependent variables and GHQ-12 responses as independent variables were assessed: non-parametric, simple ordered logit (OL), extended OL, and least absolute shrinkage and selection operator (LASSO). Approaches were assessed using Akaike and Bayesian information criteria, predictive accuracy measured using root mean squared error (RMSE), and simplicity.

RESULTS

A total of 8114 responses became 6924 after discarding missing values, with 4847 used in estimation and 2077 used for testing. LASSO had a better model fit on the pain/discomfort dimension, but no model had markedly superior predictive accuracy. The non-parametric approach was chosen for the mapping algorithm based on simplicity. Predicted and observed EQ-5D-3L values for the test sample had a correlation of 0.488. Prediction accuracy was better for GHQ-12 scores below 20 than scores above 20.

CONCLUSION

The mapping allows EQ-5D-3L responses to be predicted using GHQ-12 responses, which may be useful in estimating utility values and QALYs. An R script and Microsoft Excel spreadsheet are provided to facilitate calculations.

Mapping of disease-specific Oxford Knee Score onto EQ-5D-5L utility index in knee osteoarthritis

BACKGROUND

EQ5D is a generic measure of health. It provides a single index value for health status that can be used in the clinical and economic evaluation of healthcare. Oxford Knee Score (OKS) is a joint-specific outcome measure tool designed to assess symptoms and function in osteoarthritis patients after joint replacement surgery. Though widely used, it has the disadvantage of lacking health index value. To fill the gap between functional and generic questionnaires with economic value, we linked generic EQ-5D-5L to the specific OKS to give a single index value for health status in KOA patients.

QUESTIONS/PURPOSES

Developing and evaluating an algorithm to estimate EuroQoL generic health utility scores (EQ-5D-5L) from the disease-specific OKS using data from patients with knee osteoarthritis (KO).

PATIENTS AND METHODS

This is a cross-sectional study of 571 patients with KO. We used four distinct mapping algorithms: Cumulative Probability for Ordinal Data, Penalized Ordinal Regression, CART (Classification and Regression Trees), and Ordinal random forest. We compared the resultant models' degrees of accuracy.

RESULTS

Mobility was best predicted by penalized regression with pre-processed predictors, usual activities by random forest, pain/discomfort by cumulative probability with pre-processed predictors, self-care by random forest with RFE (recursive feature elimination) predictors, and anxiety/depression by CART with RFE predictors. Model accuracy was lowest with anxiety/depression and highest with mobility and usual activities. Using available country value sets, the average MAE was 0.098 ± 0.022, ranging from 0.063 to 0.142; and the average MSE was 0.020 ± 0.008 ranging from 0.008 to 0.042.

CONCLUSIONS

The current study derived accurate mapping techniques from OKS to the domains of EQ-5D-5L, allowing for the computation of QALYs in economic evaluations. A machine learning-based strategy offers a viable mapping alternative that merits further exploration.

Single-cell sequencing and establishment of an 8-gene prognostic model for pancreatic cancer patients

Background

Single-cell sequencing (SCS) technologies enable analysis of gene structure and expression data at single-cell resolution. However, SCS analysis in pancreatic cancer remains largely unexplored.

Methods

We downloaded pancreatic cancer SCS data from different databases and applied appropriate dimensionality reduction algorithms. We identified 10 cell types and subsequently screened differentially expressed marker genes of these 10 cell types using FindAllMarkers analysis. Also, we evaluated the tumor immune microenvironment based on ESTIMATE and MCP-counter. Statistical enrichment was evaluated using Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway analysis. We used all candidate gene sets in KEGG database to perform gene set enrichment analysis. We used LASSO regression to reduce the number of genes in the pancreatic risk model by R package glmnet, followed by rtPCR to validate the expression of the signature genes in different pancreatic cancer cell lines.

Results

We identified 15 cell subpopulations by dimension reduction and data clustering. We divided the 15 subpopulations into 10 distinct cell types based on marker gene expression. Then, we performed functional enrichment analysis for the 352 marker genes in pancreatic cancer cells. Based on RNA expression data and prognostic information from TCGA and GEO datasets, we identified 42 prognosis-related genes, including 5 protective genes and 37 high-risk genes, which we used to identified two molecular subtypes. C1 subtype was associated with a better prognosis, whereas C2 subtype was associated with a worse prognosis. Moreover, chemokine and chemokine receptor genes were differentially expressed between C1 and C2 subtypes. Functional and pathway enrichment uncovered functional differences between C1 and C2 subtype. We identified eight genes that could serve as potential biomarkers for prognosis prediction in pancreatic cancer patients. These genes were used to establish an 8-gene pancreatic cancer prognostic model.

Conclusions

We established an 8-gene pancreatic cancer prognostic model. This model can meaningfully predict prognosis and treatment response in pancreatic cancer patients.

Machine learning predicts improvement of functional outcomes in traumatic brain injury patients after inpatient rehabilitation

Survivors of traumatic brain injury (TBI) have an unpredictable clinical course. This unpredictability makes clinical resource allocation for clinicians and anticipatory guidance for patients difficult. Historically, experienced clinicians and traditional statistical models have insufficiently considered all available clinical information to predict functional outcomes for a TBI patient. Here, we harness artificial intelligence and apply machine learning and statistical models to predict the Functional Independence Measure (FIM) scores after rehabilitation for traumatic brain injury (TBI) patients. Tree-based algorithmic analysis of 629 TBI patients admitted to a large acute rehabilitation facility showed statistically significant improvement in motor and cognitive FIM scores at discharge.

Using phylogenetics to infer HIV-1 transmission direction between known transmission pairs

Identifying the transmission direction between individuals provides unparalleled power to understand infectious disease epidemiology. With epidemiological and clinical information typically unavailable to infer transmission direction, phylogenetic analysis of pathogen sequence data offers an alternative approach. While the success of this phylogenetic analysis varies, the reasons remain unknown. We analyze sequence data from over 100 transmission pairs for which both the transmission direction of HIV is known and detailed additional information is available. We find that easily quantifiable phylogenetic and sampling characteristics discriminate whether a phylogenetically inferred transmission direction is correct. Our analysis highlights that while phylogenetic approaches to infer transmission direction are unsuitable for individual-level analysis, such as forensic investigations, confidence in source attribution can be incorporated in population-level analyses.

Inferring the transmission direction between linked individuals living with HIV provides unparalleled power to understand the epidemiology that determines transmission. Phylogenetic ancestral-state reconstruction approaches infer the transmission direction by identifying the individual in whom the most recent common ancestor of the virus populations originated. While these methods vary in accuracy, it is unclear why. To evaluate the performance of phylogenetic ancestral-state reconstruction to determine the transmission direction of HIV-1 infection, we inferred the transmission direction for 112 transmission pairs where transmission direction and detailed additional information were available. We then fit a statistical model to evaluate the extent to which epidemiological, sampling, genetic, and phylogenetic factors influenced the outcome of the inference. Finally, we repeated the analysis under real-life conditions with only routinely available data. We found that whether ancestral-state reconstruction correctly infers the transmission direction depends principally on the phylogeny's topology. For example, under real-life conditions, the probability of identifying the correct transmission direction increases from 32%—when a monophyletic–monophyletic or paraphyletic–polyphyletic tree topology is observed and when the tip closest to the root does not agree with the state at the root—to 93% when a paraphyletic–monophyletic topology is observed and when the tip closest to the root agrees with the root state. Our results suggest that documenting larger differences in relative intrahost diversity increases our confidence in the transmission direction inference of linked pairs for population-level studies of HIV. These findings provide a practical starting point to determine our confidence in transmission direction inference from ancestral-state reconstruction.

Validation of a Mobile, Sensor-based Neurobehavioral Assessment With Digital Signal Processing and Machine-learning Analytics

BACKGROUND

The Miro Health Mobile Assessment Platform consists of self-administered neurobehavioral and cognitive assessments that measure behaviors typically measured by specialized clinicians.

OBJECTIVE

To evaluate the Miro Health Mobile Assessment Platform's concurrent validity, test-retest reliability, and mild cognitive impairment (MCI) classification performance.

METHOD

Sixty study participants were evaluated with Miro Health version V.2. Healthy controls (HC), amnestic MCI (aMCI), and nonamnestic MCI (naMCI) ages 64-85 were evaluated with version V.3. Additional participants were recruited at Johns Hopkins Hospital to represent clinic patients, with wider ranges of age and diagnosis. In all, 90 HC, 21 aMCI, 17 naMCI, and 15 other cases were evaluated with V.3. Concurrent validity of the Miro Health variables and legacy neuropsychological test scores was assessed with Spearman correlations. Reliability was quantified with the scores' intraclass correlations. A machine-learning algorithm combined Miro Health variable scores into a Risk score to differentiate HC from MCI or MCI subtypes.

RESULTS

In HC, correlations of Miro Health variables with legacy test scores ranged 0.27-0.68. Test-retest reliabilities ranged 0.25-0.79, with minimal learning effects. The Risk score differentiated individuals with aMCI from HC with an area under the receiver operator curve (AUROC) of 0.97; naMCI from HC with an AUROC of 0.80; combined MCI from HC with an AUROC of 0.89; and aMCI from naMCI with an AUROC of 0.83.

CONCLUSION

The Miro Health Mobile Assessment Platform provides valid and reliable assessment of neurobehavioral and cognitive status, effectively distinguishes between HC and MCI, and differentiates aMCI from naMCI.

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.

A Machine Learning Approach to Assess Differential Item Functioning in Psychometric Questionnaires Using the Elastic Net Regularized Ordinal Logistic Regression in Small Sample Size Groups

Assessing differential item functioning (DIF) using the ordinal logistic regression (OLR) model highly depends on the asymptotic sampling distribution of the maximum likelihood (ML) estimators. The ML estimation method, which is often used to estimate the parameters of the OLR model for DIF detection, may be substantially biased with small samples. This study is aimed at proposing a new application of the elastic net regularized OLR model, as a special type of machine learning method, for assessing DIF between two groups with small samples. Accordingly, a simulation study was conducted to compare the powers and type I error rates of the regularized and nonregularized OLR models in detecting DIF under various conditions including moderate and severe magnitudes of DIF (DIF = 0.4 and 0.8), sample size (N), sample size ratio (R), scale length (I), and weighting parameter (w). The simulation results revealed that for I = 5 and regardless of R, the elastic net regularized OLR model with w = 0.1, as compared with the nonregularized OLR model, increased the power of detecting moderate uniform DIF (DIF = 0.4) approximately 35% and 21% for N = 100 and 150, respectively. Moreover, for I = 10 and severe uniform DIF (DIF = 0.8), the average power of the elastic net regularized OLR model with 0.03 ≤ w ≤ 0.06, as compared with the nonregularized OLR model, increased approximately 29.3% and 11.2% for N = 100 and 150, respectively. In these cases, the type I error rates of the regularized and nonregularized OLR models were below or close to the nominal level of 0.05. In general, this simulation study showed that the elastic net regularized OLR model outperformed the nonregularized OLR model especially in extremely small sample size groups. Furthermore, the present research provided a guideline and some recommendations for researchers who conduct DIF studies with small sample sizes.

Bayesian variable selection for high-dimensional data with an ordinal response: identifying genes associated with prognostic risk group in acute myeloid leukemia

BACKGROUND

Acute myeloid leukemia (AML) is a heterogeneous cancer of the blood, though specific recurring cytogenetic abnormalities in AML are strongly associated with attaining complete response after induction chemotherapy, remission duration, and survival. Therefore recurring cytogenetic abnormalities have been used to segregate patients into favorable, intermediate, and adverse prognostic risk groups. However, it is unclear how expression of genes is associated with these prognostic risk groups. We postulate that expression of genes monotonically associated with these prognostic risk groups may yield important insights into leukemogenesis. Therefore, in this paper we propose penalized Bayesian ordinal response models to predict prognostic risk group using gene expression data. We consider a double exponential prior, a spike-and-slab normal prior, a spike-and-slab double exponential prior, and a regression-based approach with variable inclusion indicators for modeling our high-dimensional ordinal response, prognostic risk group, and identify genes through hypothesis tests using Bayes factor.

RESULTS

Gene expression was ascertained using Affymetrix HG-U133Plus2.0 GeneChips for 97 favorable, 259 intermediate, and 97 adverse risk AML patients. When applying our penalized Bayesian ordinal response models, genes identified for model inclusion were consistent among the four different models. Additionally, the genes included in the models were biologically plausible, as most have been previously associated with either AML or other types of cancer.

CONCLUSION

These findings demonstrate that our proposed penalized Bayesian ordinal response models are useful for performing variable selection for high-dimensional genomic data and have the potential to identify genes relevantly associated with an ordinal phenotype.

multiMarker: software for modelling and prediction of continuous food intake using multiple biomarkers measurements

Background

Metabolomic biomarkers offer potential for objective and reliable food intake assessment, and there is growing interest in using biomarkers in place of or with traditional self-reported approaches. Ongoing research suggests that multiple biomarkers are associated with single foods, offering great sensitivity and specificity. However, currently there is a dearth of methods to model the relationship between multiple biomarkers and single food intake measurements.

Results

Here, we introduce multiMarker, a web-based application based on the homonymous R package, that enables one to infer the relationship between food intake and two or more metabolomic biomarkers. Furthermore, multiMarker allows prediction of food intake from biomarker data alone. multiMarker differs from previous approaches by providing distributions of predicted intakes, directly accounting for uncertainty in food intake quantification. Usage of both the R package and the web application is demonstrated using real data concerning three biomarkers for orange intake. Further, example data is pre-loaded in the web application to enable users to examine multiMarker’s functionality.

Conclusion

The proposed software advance the field of Food Intake Biomarkers providing researchers with a novel tool to perform continuous food intake quantification, and to assess its associated uncertainty, from multiple biomarkers. To facilitate widespread use of the framework, multiMarker has been implemented as an R package and a Shiny web application.