Stepwise Variable Selection Method Research Articles

Etiology, Model for End-Stage Liver Disease and Hospitalization Requiring Intensive Care are Predictors of 30-day Readmission in Patients with Liver Disease

Purpose: Preventable hospital readmissions is an important issue in health care policy, and practice reform is vital to reducing cost, and improving quality of care. This study attempts to identify inpatient process issues as well as disease specific risk factors which may predispose a patient with liver disease to 30-readmission. Methods: Patients admitted to the liver service between August 2008 and September 2010 in patients with chronic liver disease were considered for the study. Patients were excluded if they had an admission within 30 days of the index admission and should have had at least one outpatient clinic visit during the time to follow-up. All outpatient and inpatient encounters were tracked until either 1) death, 2) loss to follow-up, 3) liver transplantation, or 4) end of study period (31st, May 2012). Retrospective chart review of demographics, disease severity, complications, medication history, admission testing and treatment, follow-up and compliance to follow-up, as well as readmission data were collected. Descriptive statistics were computed for all factors. A time-to-readmission analysis was performed to assess factors associated with read-mission within 30 days; An automated stepwise variable selection method performed on 1000 bootstrap samples was used to choose the final multivariable model; A P < 0.05 was considered statistically significant. Results: A total of 306 patients were included in the analysis. Fifty-two percent were male and mean age on admission was 54.4 ± 11.0 years; 27.8% had a readmission within 30 days of being discharged. After adjusting for all variables in the multivariate model, HCV (p=0.006), ICU admission (p=0.007) and higher MELD at discharge (p=0.007) were found to be significantly associated with higher likelihood of readmission. Liver transplant (p=0.004) was more common in subjects who were readmitted within 30 days although mortality was not significant (p=0.26). Four percent of subjects died within 3 months of first discharge. Conclusion: More than one-fourth of patients hospitalized for liver disease get readmitted within a month, Subjects with HCV were 2.5 times more likely to be readmitted within 30 days than those without HCV. Subjects who required ICU admission during their hospital stay had almost 3 times higher odds of readmission than those who were not in the ICU. For every 1 unit increase in MELD score at time of discharge, the likelihood of being readmitted within 30 days increases by 8%. These criteria can be used to identify patients at highest risk for readmission that may benefit from aggressive discharge and transition of care planning.

Comprehensive structure–activity relationship analysis of substituted 5-(biphenyl-4-ylmethyl) pyrazoles derivatives as AT1 selective angiotensin II receptor antagonists: 2D and kNNMFA QSAR approach

Two- and three-dimensional quantitative structure activity relationship studies of 5-(biphenyl-4-ylmethyl) pyrazoles derivatives were performed for their AT1 selective angiotensin II receptor antagonists activity using VlifeMDS 3.5™ software. The best 2D QSAR model for the prediction of AT1 selective angiotensin II receptor antagonists activity was obtained by applying stepwise variable selection with partial least square analysis giving r2 = 0.8962 and q2 = 0.7831, r2_se = 0.3986, q2_se = 0.2004, with external predictive ability of pred_r2 = 0.8293, pred_r2se = 0.5852. By using kNN-MFA with stepwise variable selection method approach, 3D QSAR models were generated to study the effect of steric, electrostatic, and hydrophobic descriptors on anti-hypertensive activity. The model with good external and internal predictivity for the training and test set has shown cross validation (q2) and external validation (pred_r2) values of 0.74 and 0.68, respectively. The SW-kNNMFA contour maps suggest some important structural features-like steric, electropositive, electronegative, hydrophobic substituent’s requirements of antihypertensive activity. Thus, this validated models brings important structural insight to aid in the design of novel antihypertensive agents.

A novel serum metabolomics-based diagnostic approach for colorectal cancer.

BackgroundTo improve the quality of life of colorectal cancer patients, it is important to establish new screening methods for early diagnosis of colorectal cancer.Methodology/Principal FindingsWe performed serum metabolome analysis using gas-chromatography/mass-spectrometry (GC/MS). First, the accuracy of our GC/MS-based serum metabolomic analytical method was evaluated by calculating the RSD% values of serum levels of various metabolites. Second, the intra-day (morning, daytime, and night) and inter-day (among 3 days) variances of serum metabolite levels were examined. Then, serum metabolite levels were compared between colorectal cancer patients (N = 60; N = 12 for each stage from 0 to 4) and age- and sex-matched healthy volunteers (N = 60) as a training set. The metabolites whose levels displayed significant changes were subjected to multiple logistic regression analysis using the stepwise variable selection method, and a colorectal cancer prediction model was established. The prediction model was composed of 2-hydroxybutyrate, aspartic acid, kynurenine, and cystamine, and its AUC, sensitivity, specificity, and accuracy were 0.9097, 85.0%, 85.0%, and 85.0%, respectively, according to the training set data. In contrast, the sensitivity, specificity, and accuracy of CEA were 35.0%, 96.7%, and 65.8%, respectively, and those of CA19-9 were 16.7%, 100%, and 58.3%, respectively. The validity of the prediction model was confirmed using colorectal cancer patients (N = 59) and healthy volunteers (N = 63) as a validation set. At the validation set, the sensitivity, specificity, and accuracy of the prediction model were 83.1%, 81.0%, and 82.0%, respectively, and these values were almost the same as those obtained with the training set. In addition, the model displayed high sensitivity for detecting stage 0–2 colorectal cancer (82.8%).Conclusions/SignificanceOur prediction model established via GC/MS-based serum metabolomic analysis is valuable for early detection of colorectal cancer and has the potential to become a novel screening test for colorectal cancer.

Juvenile Offender Recidivism: An Examination of Risk Factors

One hundred and seventy three male juvenile offenders were followed two years postrelease from a residential treatment facility to assess recidivism and factors related to recidivism. The overall recidivism rate was 23.9%. Logistic regression with stepwise and backward variable selection methods was used to examine the relationship between recidivism and nine specific variables: offense type, age at initial involvement in juvenile justice, child welfare system involvement, termination of parental rights, parental criminal history, family support, program completion status, length of treatment stay, and discharge placement. Offender type was the only factor found to have a significant impact on recidivism with general and substance-involved offenders more likely to recidivate than sex offenders. Implications for future research are discussed.

Severe chronic heart failure in patients considered for heart transplantation in Poland

Based on the results of clinical trials, the prognosis for patients with severe heart failure (HF) has improved over the last 20 years. However, clinical trials do not reflect 'real life' due to patient selection. Thus, the aim of the POLKARD-HF registry was the analysis of survival of patients with refractory HF referred for orthotopic heart transplantation (OHT). Between 1 November 2003 and 31 October 2007, 983 patients with severe HF, referred for OHT in Poland, were included into the registry. All patients underwent routine clinical and hemodynamic evaluation, with NT-proBNP and hsCRP assessment. Death or an emergency OHT were assumed as the endpoints. The average observation period was 601 days. Kaplan-Meier curves with log-rank and univariate together with multifactor Cox regression model the stepwise variable selection method were used to determine the predictive value of analyzed variables. Among the 983 patients, the probability of surviving for one year was approximately 80%, for two years 70%, and for three years 67%. Etiology of the HF did not significantly influence the prognosis. The patients in NYHA class IV had a three-fold higher risk of death or emergency OHT. The univariate/multifactor Cox regression analysis revealed that NYHA IV class (HR 2.578, p < 0.0001), HFSS score (HR 2.572, p < 0.0001) and NT-proBNP plasma level (HR 1.600, p = 0.0200), proved to influence survival without death or emergency OHT. Despite optimal treatment, the prognosis for patients with refractory HF is still not good. NYHA class IV, NT-proBNP and HFSS score can help define the highest risk group. The results are consistent with the prognosis of patients enrolled into the randomized trials.

Culvert Asset Management Practices and Deterioration Modeling

Drainage infrastructure systems and culverts constitute an important portion of highway assets that require routine inspections, maintenance, and timely repair and renewal. Variations in structural characteristics (i.e., material type, shape, and dimension), environmental exposure, and wide geospatial distribution of these infrastructure assets accompanied with strict budget restrictions pose significant challenges for transportation agency officials. Deterioration models constitute one of the most essential components of any infrastructure asset management strategy; these models provide insight on the significant factors that affect infrastructure condition states and expected behavior of infrastructure assets under different conditions. The objectives of this study are to provide a review of previous studies on factors that affect culvert performance and durability and those on culvert asset management, to present current practices of transportation agencies with regard to management of culverts, and, last, to present development of a preliminary deterioration model that will allow decision makers to identify significant factors that affect deterioration of metal culverts and prioritize inspection procedures. The preliminary deterioration model presented in this study is developed by employing binary logistic regression with a forward stepwise variable selection method on data obtained from the Ohio Department of Transportation.

Estimation of the volume of distribution of some pharmacologically important compounds from their structural descriptors

Quantitative structure-activity relationship (QSAR) approaches were used to estimate the volume of distribution (Vd) using an artificial neural network (ANN). The data set consisted of the volume of distribution of 129 pharmacologically important compounds, i.e., benzodiazepines, barbiturates, NSAIDs, tricyclic anti-depressants and some antibiotics, such as betalactams, tetracyclines and quinolones. The descriptors, which were selected by stepwise variable selection methods, were: the Moriguchi octanol-water partition coefficient; the 3D-MoRSEsignal 30, weighted by atomic van der Waals volumes; the fragmentbased polar surface area; the d COMMA2 value, weighted by atomic masses; the Geary autocorrelation, weighted by the atomic Sanderson electronegativities; the 3D-MoRSE - signal 02, weighted by atomic masses, and the Geary autocorrelation - lag 5, weighted by the atomic van der Waals volumes. These descriptors were used as inputs for developing multiple linear regressions (MLR) and artificial neural network models as linear and non-linear feature mapping techniques, respectively. The standard errors in the estimation of Vd by the MLR model were: 0.104, 0.103 and 0.076 and for the ANN model: 0.029, 0.087 and 0.082 for the training, internal and external validation test, respectively. The robustness of these models were also evaluated by the leave-5-out cross validation procedure, that gives the statistics Q2 = 0.72 for the MLR model and Q2 = 0.82 for the ANN model. Moreover, the results of the Y-randomization test revealed that there were no chance correlations among the data matrix. In conclusion, the results of this study indicate the applicability of the estimation of the Vd value of drugs from their structural molecular descriptors. Furthermore, the statistics of the developed models indicate the superiority of the ANN over the MLR model.

Regression Analysis of Walking Parameters for the Age-Predictive Equation

We tried to establish an age-predictive equation from kinematic and kinetic parameters during walking using the 2007 Okada database of Japanese elderly. Some predictive equations were established for each gender (109 males and 139 females)--once for all variables, second for the stepwise variable selection method. The range of motion (ROM) of knee and hip and step length are the stable parameters in age-predictive models for both genders. Although the model achieved the Minimum Akaike's Information Criterion Estimate (MAICE), it did not reflect sufficiently the regimes of variable selection.

Application of an iterative Bayesian variable selection method in a genome-wide association study of rheumatoid arthritis

Genome-wide association studies usually involve several hundred thousand of single-nucleotide polymorphisms (SNPs). Conventional approaches face challenges when there are enormous number of SNPs but a relatively small number of samples and, in some cases, are not feasible. We introduce here an iterative Bayesian variable selection method that provides a unique tool for association studies with a large number of SNPs (p) but a relatively small sample size (n). We applied this method to the simulated case-control sample provided by the Genetic Analysis Workshop 15 and compared its performance with stepwise variable selection method. We demonstrated that the results of iterative Bayesian variable selection applied to when p t n are as comparable as those of stepwise variable selection implemented to when n t p. When n > p, the iterative Bayesian variable selection performs better than stepwise variable selection does.

Positive False Discovery Rate Estimate in Step-Wise Variable Selection

Selecting predictors to optimize the outcome prediction is an important statistical method. However, it usually ignores the false positives in the selected predictors. In this article, we advocate a conventional stepwise forward variable selection method based on the predicted residual sum of squares, and develop a positive false discovery rate (pFDR) estimate for the selected predictor subset, and a local pFDR estimate to prioritize the selected predictors. This pFDR estimate takes account of the existence of non null predictors, and is proved to be asymptotically conservative. In addition, we propose two views of a variable selection process: an overall and an individual test. An interesting feature of the overall test is that its power of selecting non null predictors increases with the proportion of non null predictors among all candidate predictors. Data analysis is illustrated with an example, in which genetic and clinical predictors were selected to predict the cholesterol level change after four months of tamoxifen treatment, and pFDR was estimated. Our method's performance is evaluated through statistical simulations.

Molecular modeling and QSAR analysis of some 4,5-dichloroimidazolyl-1,4-DHP-based calcium channel blockers

The effects of the structural features of some 4,5-dichloroimidazolyl-1,4-dihydropyridine on their calcium channel antagonist activity have been studied using molecular modeling and quantitative structure activity-relationship analysis. Both symmetrical and asymmetrical dihydropyridine derivatives were used. AM1 semi-empirical quantum chemical calculation was used to find the optimum 3-D geometry of the molecules. Four different sets of descriptors, including chemical, topological, quantum chemical and substituent constant, were then calculated for each molecule. For each set of descriptors, the best multilinear QSAR equations were obtained by the stepwise variable selection method using leave-one-out cross-validation as selection criterion. Separate QSAR models were first obtained for symmetrical and asymmetrical derivatives, after which a general model was proposed for the entire set of molecules. This model has root mean square error of 0.45 and reproduces more than 82% of the variances in the calcium channel antagonist activity data. The sum of the negative charges, the energy of the highest occupied molecular orbital, molecular volume and the least negative charge were identified as the most significant descriptors.

A prospective study on C-reactive protein as a prognostic factor for survival time of terminally ill cancer patients

C-reactive protein (CRP) has been investigated as a predictor of life expectancy in terminal cancer in one previous study. However, the other variables such as patient's symptom or physical examination findings were not considered. The aim of this study is to prove serum CRP level as a predictor of survival time, considering patient's symptoms, physical examination findings, and various serological variables in terminally ill cancer patients with a prospective cohort design. Forty-four terminally ill cancer patients were divided into two groups by serum CRP levels and followed up until death. We adjusted the influence of some clinical and laboratory variables on survival by use of Cox's proportional hazard model. Using the stepwise variable selection method, we found the final model. For 19 patients, CRP levels at 2 weeks and a week before death were compared by Wilcoxon signed ranks test. All 44 study subjects died during the study period, and the median survival time was 17 days. Survival time of the elevated CRP group (> or =2.2 mg/dl) was found to be significantly shorter than the lower CRP group in univariate analysis (hazard rate = 3.221, P = 0.001). In multivariate analysis, elevated CRP level (> or =2.2 mg/dl) was selected as one of the unfavorable indicators regarding survival. Dyspnea and hyperbilirubinemia were also found significant predictors of shorter life expectancy in the final model. Serum CRP levels were significantly increased between the first and the second week before death. Our results showed that serum CRP level can be a useful indicator predictive of survival time of terminally ill cancer patients.

A simple digital image processing system to aid in melanoma diagnosis in an everyday melanocytic skin lesion unit. A preliminary report

For early melanoma diagnosis, experienced dermatologists have an accuracy of 64-80% using clinical diagnostic criteria, usually the ABCD rule, while automated melanoma diagnosis systems are still considered to be experimental and serve as adjuncts to the naked-eye expert prediction. In an attempt to aid in early melanoma diagnosis, we developed an image processing program with the aim to discriminate melanoma from melanocytic nevi, establishing a mathematical model to come up with a melanoma probability. Digital images of 132 melanocytic skin lesions (23 melanomas and 109 melanocytic nevi) were studied in features of geometry, color, and color texture. A total of 43 variables were studied for all lesions, e.g., geometry, color texture, sharpness of border, and color variables. Univariate logistic regression analysis followed by "-2 log likelihood" test and Spearman's rank correlation coefficient were used to eliminate inappropriate variables, as the presence of multi-collinearity among variables could cause severe problems in any stepwise variable selection method. Initially, "-2 log likelihood" and nonparametric Spearman's rho picked five variables to be included in a multivariate model of prediction. The five-variable model was then reduced to three variables and the performance of each model was tested. The "jackknife" method was performed in order to validate the model with the three variables and its accuracy was weighed vs. the five-variable model by receiver-operating characteristics (ROC) curve plotting. It was concluded that the reduced model did not compromise discriminatory power. Not all variables contributed much to the model, therefore they were progressively eliminated and the model was finally reduced to three covariates of significance. A predictive equation was calculated, incorporating parameters of geometry, color, and color texture as independent covariates for the prediction of melanoma. The proposed model provides melanoma probability with a 60.9% sensitivity and 95.4% specificity of prediction, an overall accuracy of 89.4% (probability level 0.5), and 8% false-negative results. Through a digital image processing system and the development of a mathematical model of prediction, discrimination between melanomas and melanocytic nevi seems feasible with a high rate of accuracy using multivariate logistic regression analysis. The proposed model is an alternative method to aid in early melanoma diagnosis. Expensive and sophisticated equipment is not required and it can be easily implemented in a reasonably priced portable programmable computer, in order to predict previously undiagnosed skin melanoma before histopathology results confirm diagnosis.

The Tabu Search Procedure:An Alternative to the Variable Selection Methods

The effectiveness of the Tabu variable selection algorithm, to identify predictor variables related to a criterion variable, is compared with the stepwise variable selection method and the all possible regression approach. Considering results obtained from previous research, Tabu is more successful in identifying relevant variables than the stepwise method, and the adjusted R⊃2 and the Mallow Cp criteria from all possible regression models for certain conditions. Although not completely effective in identifying all variables related to the criterion variable, Tabu is less likely to select variables unrelated to the criterion variable than the alternative methods. We encourage researchers to consider theory, previous research, and professional judgment when selecting the final set of predictors. Limitations of the study as well as the need for further research are also discussed.

Mitotic rate and younger age are predictors of sentinel lymph node positivity: lessons learned from the generation of a probabilistic model.

Sentinel lymph node (SLN) biopsy allows surgeons to identify patients with subclinical nodal involvement who may benefit from lymphadenectomy and, possibly, adjuvant therapy. Several factors have been variably, and sometimes discordantly, reported to have predictive value for SLN metastasis to best select which patients require SLN biopsy. We reviewed 419 patients who underwent SLN biopsy for melanoma from a prospectively collected melanoma database. To derive a probabilistic model for the occurrence of a positive SLN, a multivariate logistic model was fit by using a stepwise variable selection method. The accuracy of each model was evaluated by using receiver operator characteristic curves. On univariate analysis, the number of mitoses per square millimeter, increasing Breslow depth, decreasing age, ulceration, and melanoma on the trunk showed a significant relationship to a positive SLN. Multivariate analysis revealed that once age, mitotic rate, and Breslow thickness were included, no other factor, including ulceration, was significantly associated with a positive SLN. The data suggest that younger patients with tumors <1 mm may still have a substantial risk for a positive SLN, especially if the mitotic rate is high. In addition to Breslow depth, mitoses per square millimeter and younger age were factors identified as independent predictors of a positive SLN. This model may identify patients with thin melanoma at sufficient risk for metastases to justify SLN biopsy.

Prediction of bioconcentration factor using genetic algorithm and artificial neural network

In this paper, genetic algorithm (GA) and stepwise multiple regression variable selection methods were used as a feature-selection tools and neural network was employed for feature mapping. To provide an extended test of these hybrid methods, a data set consists of the bioconcentration factors (BCF) for 53 molecules were selected. Suitable set of molecular descriptors were calculated and the important descriptors were selected by genetic algorithm and stepwise multiple regression methods. These variables serve as inputs to generated neural networks. After optimization and training of the networks, they were used for the calculation of bioconcentration factors for the prediction set. Comparison between results obtained showed the superiority of genetic algorithm over stepwise multiple regression method in feature-selection. For network that used the genetic algorithm for feature-selection methods the average relative error between predicted and experimental values of bioconcentration factor for training and prediction set are 0.07 and 0.13, respectively. Also the standard error of calibration and standard error of prediction are 0.26 and 0.40%, respectively, for this model. An analysis of the descriptor selected by genetic algorithm showed that they are essential to implies the steric and electrostatic attributes of studied molecules to obtain a satisfactory quantitative structure–activity relationship (QSAR) model.

No Preventive Effect of Dietary Fiber Against Colon Cancer in the Japanese Population: A Cross-Sectional Analysis

The report of Fuchs et al. in 1999 on the protective effects of dietary fiber (DF) against colon carcinogenesis has led many researchers to question the benefits of DF. We analyzed the relationship between dietary intake and mortality from colon cancer in Japan cross-sectionally. Dietary data were taken from the National Nutrition Survey. The standardized mortality ratio (SMR) was calculated using data from "Vital Statistics" and "the Population Census in Japan." Multiple regression analysis (stepwise variable selection method) was performed with the SMR of colon cancer as the objective variable and intake of DF, nutrients, and food groups in 1966 as the explanatory variables. The beta regression coefficient was significantly positive for intakes of fat, protein, and vitamin C and significantly negative for intakes of calcium and vitamin A to the SMR of colon cancer. However, no significant correlation was observed for DF or for any of the various food groups analyzed. In conclusion, our data do not demonstrate any protective effect of DF on colon cancer in subjects with a low fat intake (Japanese subjects), which supports Fuchs' findings in subjects with high fat intake (U.S. subjects).

Risk Factors for Epistaxis during Nasotracheal Intubation

We performed a study to confirm which risk factors are significantly associated with epistaxis during nasotracheal intubation. One hundred patients who underwent nasotracheal intubation were included. Risk factors for epistaxis were analysed using the multiple logistic regression analysis with stepwise variable selection method. Epistaxis was most likely to occur if transit of the tube through the nasal passage was difficult (P=0.0001, odds ratio 625, 95% confidence interval 3.14-14.26). On the other hand, age and gender, obesity, smoking, tube size, repeated attempts of intubation, and intubation performed with the aid of Magillforceps were not significantly related with risk of epistaxis. The presence of nasal anatomical abnormalities also did not correlate significantly with epistaxis. Strategies to ensure smooth transit of the tube through the nasal passageways are essential to reduce the incidence of epistaxis.

분광특성 분석에 의한 논 잡초 검출의 기초연구

This is a fundamental study to develop a sensor to detect weeds in paddy field using machine vision adopted spectralphotometric technique in order to use the sensor to spread herbicide selectively. A set of spectral reflectance data was collected from dry and wet soil and leaves of rice and 6 kinds of weed to select desirable wavelengths to classify soil, rice and weeds. Stepwise variable selection method of discriminant analysis was applied to the data set and wavelengths of 680 and 802 m were selected to distinguish plants (including rice and weeds) from dry and wet soil, respectively. And wavelengths of 580 and 680 nm were selected to classify rice and weeds by the same method. Validity of the wavelengths to distinguish the plants from soil was tested by cross-validation test with built discriminant function to prove that all of soil and plants were classified correctly without any failure. Validity of the wavelengths for classification of rice and weeds was tested by the same method and the test resulted that 98% of rice and 83% of weeds were classified correctly. Feasibility of CCD color camera to detect weeds in paddy field was tested with the spectral reflectance data by the same statistical method as above. Central wavelengths of RGB frame of color camera were tried as tile effective wavelengths to distingush plants from soil and weeds from plants. The trial resulted that 100% and 94% of plants in dry soil and wet soil, respectively, were classified correctly by the central wavelength or R frame only, and 95% of rice and 85% of weeds were classified correctly by the central wavelengths of RGB frames. As a result, it was concluded that CCD color camera has good potential to be used to detect weeds in paddy field.

Variable selection in classification of environmental soil samples for partial least square and neural network models

Two variable selection methods were evaluated by comparing their predictions with respect to differentiating among environmental soil samples. The focus of this work is to determine which input variables are most relevant for prediction of soil sources using discriminant partial least square (D-PLS) and back-propagation artificial neural network (BP-ANN) models. The methods investigated were stepwise variable selection method and genetic algorithms (GAs). Microbial community DNA was extracted from 48 environmental soil samples derived from different field crops and soil sources. After amplification of bacterial ribosomal RNA genes by polymerase chain reaction (PCR), the products were separated by gel electrophoresis. Characteristic complex band patterns were obtained, indicating high bacterial diversity. Two hundred and twenty-three DNA band patterns produced in the gels of the soil samples were used in the analysis, after removal of included DNA standard markers. Based on the brightness of the bands, densitometric curves of the selected DNA band pattern were extracted from the gel images. The curves were smoothed using Savitsky–Golay method and scaled to the DNA standard markers. The prediction results based on the two variable selection methods for PLS and ANN models are presented and compared. Both methods gave good results before any variable selection methods, with the ANN being better than D-PLS. The prediction performance of both methods specially the D-PLS were improved by applying the stepwise variable selection and the GA variable selection method. The study also shows that GA variable selection had a significant improvement of the predictive ability than the stepwise variable selection method.