Backward Elimination Approach Research Articles

Gender-Specific Differences for Risk of Disability and Death in Atrial Fibrillation-Related Stroke

In the latest American Heart Association/American College of Cardiology/Heart Rhythm Society atrial fibrillation (AF) guidelines, CHA2DS2-VASc replaced the CHADS2 stroke risk assessment to determine prophylactic anticoagulation, reflecting female gender's association with stroke incidence in AF. However, little investigation has been pursued of potential risk factors associated with worsened stroke severity. In this study, we examined patients with AF with ischemic stroke patient characteristics associated with increased stroke severity. Using the Get With The Guidelines-Stroke database, we retrospectively identified 221 consecutive patients with AF diagnosed with acute ischemic stroke and performed in depth chart review, evaluating demographics, labs, and co-morbidities. We analyzed the modified Rankin Scale (mRS) at discharge as a surrogate for stroke severity, defining severe stroke as fatal (mRS of 6) or disabling (mRS 4 to 5), requiring max assistance with ambulation or activities of daily living. Female gender, advanced age, and decreased body surface area were associated with disabling or fatal stroke (68.3% of patients with mRS 4 to 6 vs 50% with mRS 0 to 3, 78.4 vs 71.1year, and 1.83 vs 1.92, respectively). Using a backward elimination approach revealed a logistic regression model with statistically significant odds ratios (ORs) for female gender (OR 1.99) and age (OR 1.04), and borderline significant for a history of coronary artery disease (OR 1.89). In conclusion, female gender is associated in the AF population with a twofold risk of severe disabling or fatal ischemic stroke, a finding that persists after controlling for potential confounders. This finding highlights the potential benefit from appropriate anticoagulation use for stroke prophylaxis in the AF population.

Overall Survival (OS) of Acute Myeloid Leukemia (AML) Treated at Academic Center (AC) Versus Non-Academic Center (NAC)

Introduction:Prior studies in cancer have suggested better OS of patients treated at AC as compared to NAC. This may be related to the availability of physicians with expertise in specific malignancies, better multidisciplinary care and access to more resources and clinical trials. Whether academic status of the facility affects OS of AML is unknown.Methods:We utilized the National Cancer Database Participant User File (NCDB PUF) to extract patient-level data of patients with AML reported between 1998 to 2011. Hospital facilities were classified as either AC (academic/research program) or NAC (community cancer program, comprehensive community cancer program, and other, per NCDB classification).We included only those patients, who had all of the first course treatment or a decision not to treat made at the reporting facility. Subjects with complete and known data for the variables sex, age, race, education, income, distance traveled for health care, hospital type, facility location, urban/rural location, insurance, Charlson co-morbidity score, chemotherapy use, time from diagnosis to treatment initiation, use of hematopoietic stem cell transplant, 30-day mortality, last contact, and vital status were included. These variables were analyzed in a univariate analysis. Kaplan Meier curves were drawn and compared using log rank test. Multivariate analysis was done using logistic regression for 30-day mortality and Cox regression with backward elimination approach for OS. Statistical analysis was done using PC SAS version 9.4.Results:A total of 7823 AML patients were studied, of which 4681 (60%) patients received treatment at AC. Patients treated at AC differed from NAC in the median age (62 vs. 67years; p <0.001), race (p <0.001), education (p=0.005), income (p <0.001), co-morbidity score (p=0.019), insurance (p<0.001), receipt of chemotherapy (p<0.001), transplant (p<0.001) and facility location (p<0.001). The median OS (12.6 vs. 7.0 months; p value <0.001) and 1-year OS (51% vs. 39%; p value <0.001) was better in AC as compared to NAC. In a multivariate analysis, the 30-day mortality was significantly worse in NAC as compared to AC (odds ratio, OR 1.52; 95% confidence interval, CI 1.33-1.74; p <0.001) (Table 1). Similarly, Cox regression showed that the OS was significantly worse in NAC as compared to AC (hazard ratio, HR 1.13; 95% CI 1.07-1.19; p <0.001) after adjusting for age, sex, Charlson co-morbidity score, receipt of chemotherapy, transplant, insurance and income status and facility location.Conclusion:Our population-based study suggests that the receipt of initial therapy at AC versus NAC is associated with lower 30-day mortality and higher 1-year OS. This may presumably be related to the provision of dedicated multidisciplinary leukemia teams, access to more resources and clinical trials in AC. The reasons behind such differences should be investigated in future studies, and necessary steps be taken to minimize this gap.Table 1Multivariate logistic regression of 30-day mortalityVariableOdds ratio95% confidence intervalP valueAcademic (ref) Non-Academic1 1.521.33-1..74<0.001Age - <60 years (ref) - > 60 years1 2.321.92-2.80<0.001Charlsonco-morbidity score -0 (ref) -1 - 2 or more1 1.45 2.141.23-1.69 1.74-2.63<0.001 <0.001Chemotherapy - Yes (ref) - No1 2.551.93-3.38<0.001Days until first treatment0.870.86-0.89<0.001Income status - $ 46,000 + (ref) - < $ 30,000 - 30,000-34,999 - 35,000-45,9991 1.31 1.31 1.211.06-1.62 1.09-1.58 1.03-1.430.011 0.004 0.023Insurance Status - Private insurance/managed care (ref) - Not insured - Medicaid - Medicare - Other government1 2.32 0.81 1.56 1.131.66-3.24 0.59-1.10 1.30-1.87 0.63-2.02<0.001 0.178 <0.001 0.686 [Display omitted] DisclosuresNo relevant conflicts of interest to declare.

Overall Survival (OS) of Acute Lymphoblastic Leukemia (ALL) Is Better When Initial Therapy Is Administered in Academic Hospitals (AH) Versus Non-Academic Hospitals (NAH): A Retrospective Analysis of the National Cancer Data Base (NCDB)

IntroductionDuring initial diagnosis and therapy, patients with ALL can develop life-threatening complications such as sepsis, leukostasis, hemorrhage, and tumor lysis syndrome. Dedicated multidisciplinary leukemia teams may be needed to provide optimal management of such complications and selection of optimal therapeutic strategy. AH are more likely to have such expertise, adequate resources, standard operating policies and clinical trials, which may influence early mortality and OS. Such trends have been noted for other malignancies, such as lung cancer (Samson, Am Thorac Surg 2015 and Luchtenborg JCO 2013). In acute myeloid leukemia, OS may be better if patients are treated in a high-volume hospital (Giri Blood 2015). Whether OS of ALL differs based on the facility type remains unclear.MethodsUsing the NCDB Participant User File, we extracted patient-level data of patients with ALL reported between 1998 to 2012. Hospital facilities were classified as either AH (academic/research program) or NAH (community cancer program, comprehensive community cancer program, and other, as per NCDB classification). Patients, who received all of their first course treatment or a decision not to treat made at the reporting facility, were included. Subjects with complete and known data for the variables sex, age, race, education, income, distance traveled for health care, hospital type, facility location, urban/rural location, insurance, Charlson co-morbidity score, chemotherapy use, time from diagnosis to treatment initiation, use of hematopoietic stem cell transplant, 30-day mortality, last contact, and vital status were included. These variables were analyzed in a univariate analysis. Kaplan Meier curves were drawn and compared using log rank test. Multivariate analysis was performed using logistic regression for 30-day mortality and cox regression with backward elimination approach for OS. Statistical analysis was done using PC SAS version 9.4.ResultsOf 9863 patients with ALL, 5710 (57.9%) were treated in AH. Patients treated at AH versus NAH were more likely to be African-Americans, uninsured and Medicaid insured, travel long distance to receive health care and receive transplant as a part of their treatment. The median OS (23 vs. 17 months) and 1-year OS (67% vs. 59%) were better in AH as compared to NAH (Figure 1). In a multivariate analysis, the 30-day mortality was significantly worse in NAH as compared to AH (odds ratio, OR 1.206; 95% confidence interval, CI 1.011-1.44; p <0.0374) (Table 1). Similarly, Cox regression showed that the OS was significantly worse in NAH as compared to AH (hazard ratio, HR 1.14; 95% CI 1.08-1.19; p <0.001) after adjusting for other covariates.ConclusionOS of patients with ALL may be improved, if patients receive initial therapy in AH. Possible explanations may include increased provider experience, enhanced multidisciplinary care, and access to clinical trials, among others. Improved understanding of such factors may provide opportunity to improve OS of patients treated at NAH.Table 1Multivariate logistic regression of 30-day mortalityVariableOdds ratio95% confidence intervalP valueAcademic (ref) Non-Academic1 1.2061.01-1.440.0374Age - <60 years (ref)- > 60 years1 2.9072.28-3.71<0.001Charlson comorbidity score -0 (ref) -1- 2 or more1 1.47 2.131.18-1.82 1.57-2.880.0005 <0.0001Chemotherapy - Yes (ref) - No1 2.931.82-4.72<0.0001Days until first treatment0.9270.91-0.94<0.0001High school education - 29% or more (ref) - 20%-28.9% - 14%-19.9% - Less than 14%1 0.82 0.88 0.630.63-1.06 0.69-1.14 0.49-0.810.13 0.34 0.0004Insurance Status - Private insurance/managed care (ref)- Not insured - Medicaid - Medicare - Other government1 1.46 0.80 1.61 0.451.04-2.05 0.57-1.12 1.26-2.06 0.11-1.840.03 0.20 0.0002 0.26 [Display omitted] DisclosuresNo relevant conflicts of interest to declare.

Abstract 10283: Motivational Interviewing Decreases Readmissions in Adults With Heart Failure

Introduction: In chronic heart failure (HF), 70% of patients are readmitted to the hospital within 12 months. At least 63% of these readmissions are for a condition other than HF. Hypothesis: We hypothesized that motivational interviewing (MI) would be effective in decreasing readmissions unrelated to HF in HF patients. Methods: Hospitalized adults with HF were enrolled if they were symptomatic (NYHA class II-IV), spoke and read English, lived in a setting where they could independently engage in self-care, and were cognitively intact. We enrolled 100 HF patients (mean age 60 ± 14.3 years; 57% Black, 33% female, 85% NYHA class III/IV symptoms, 64% were educated ≤ high school level) into a randomized controlled trial, using minimization to balance group assignment on NYHA class and gender. Participants in both groups received educational materials emphasizing self-care behavior change. The intervention group received MI emphasizing self-care (e.g. adherence, symptom monitoring) and tailored to individual goals. The intervention was provided by a nurse in a single home visit with 3-4 follow-up phone calls. Participants were followed for 3 months. For the primary outcome, readmission unrelated to HF, a multiple variable regression model was constructed using a backward variable elimination approach. Results: Over 3 months, 25% of the sample had a HF-readmission and 15% had a readmission unrelated to HF. There was an absolute difference of 21% (9 vs 30) in readmissions unrelated to HF between the MI intervention and control groups (p=0.006). The MI group had 94% lower odds of having a readmission unrelated to HF (p=0.001). Each additional unit of hemoglobin was associated with 48% lower odds of having a readmission unrelated to HF (p=0.006). For each additional year of age, the odds of having a readmission unrelated to HF were 7% lower. Participants with diabetes had 6.7 times the odds (p=0.012) of having a readmission unrelated to HF. Conclusion: An MI intervention emphasizing self-care and tailored to individual goals significantly reduced readmissions unrelated to HF in adults with HF, even after accounting for the influence of anemia, age, and diabetes.

Calcium, ARMS2 genotype, and Chlamydia pneumoniae infection in early age-related macular degeneration: a multivariate analysis from the Nagahama study.

Although various risk factors have been identified for the development of age-related macular degeneration (AMD), risk factors of early AMD have been relatively under studied. We aimed to investigate AMD risk factors by evaluating multiple factors in association with large drusen, an important component of AMD, simultaneously. In a community-based cross-sectional survey in Japan, 971 large drusen cases and 3,209 controls were compared for 65 variables, including systemic, environmental, and genetic factors. The association and the effect size of each factor were evaluated with logistic regression analysis using a backward-elimination approach. Multivariate analyses identified a significant association in serum calcium level (odds ratio [OR] = 0.932, P = 1.05 × 10−3), ARMS2 A69S (rs10490924) genotype (OR = 1.046, P < 0.001), Chlamydia pneumoniae IgG (OR = 1.020, P = 0.0440), and age (OR = 1.013, P < 0.001) for large drusen. Hypocalcemia was observed in 7.2% of large drusen cases and in 5.5% of controls (P = 0.0490). C. pneumoniae infections was more frequent in large drusen cases (56.4%) than in conrols (51.7%, P = 0.00956). These results suggest that calcium, ARMS2 genotype, C. pneumonia infection, and age are significant factors in the development of the early stages of AMD.

Etrolizumab Population Pharmacokinetics (Pop PK) and Covariate Analysis in Patients With Moderately to Severely Active Ulcerative Colitis (UC)

Introduction: Etrolizumab is a humanized IgG1 monoclonal antibody that specifically binds the β7 subunit of α4β7 and αEβ7 integrins. Clinical activity of etrolizumab has been observed in patients with moderate-to-severe UC in a phase 2 trial. The objective of this Pop PK analysis was to characterize covariate effects on the PK profile of etrolizumab. Methods: This analysis included data from 145 etrolizumab-treated patients from 3 studies: a phase 1 dose-escalation study (single doses at 0.3, 1, 3 or 10 mg/kg intravenously [IV] or 3 mg/kg subcutaneously [SC] or 3 monthly doses at 0.5, 1.5 or 3 mg/kg SC or 4 mg/kg IV), a phase 2 efficacy and safety study (2 SC dose cohorts of 100 mg at weeks 0, 4 and 8 or 420 mg at week 0 followed by 300 mg at weeks 2, 4 and 8) and a phase 2 open-label extension study (300 mg SC monthly). A stepwise forward addition and backward elimination approach was used to evaluate the impact of the following baseline covariates on etrolizumab PK: age, sex, body weight (BW), Mayo Clinic score, albumin, C-reactive protein, fecal calprotectin, prior use and response to anti-tumor necrosis factor α biologics and concomitant treatment with steroids or immunosuppressants. The Pop PK analysis was performed using NONMEM 7.1.2. Results: A 2-compartmental model with first-order absorption and elimination kinetics described etrolizumab PK well. BW and baseline serum albumin were identified as significant covariates for clearance (CL) (p<0.001); each explained 16% and 30% of the interindividual variability (IIV) of CL, respectively. Simulation showed similar exposure variability for a BW-based-dose or a fixed-dose. No other covariates were found to have significant impact on etrolizumab PK. The final model estimated the typical population CL value and central volume of distribution to be 0.245 L/day and 3.2 L, respectively. The elimination half-life and SC bioavailability were estimated to be 11 days and 53%, respectively. The IIV of evaluated PK parameters ranged from 19% to 67%. Pop PK simulation estimated that 100 mg nominal dose (105 mg actual dose) every 4 weeks, the proposed phase 3 dose regimen, can maintain a steady-state trough concentration of >1.7 μg/mL in 85% of patients (1.7 μg/mL is the observed minimum drug level associated with full β7 receptor occupancy in peripheral blood and colonic tissue). Conclusion: Etrolizumab exhibited linear PK at the target therapeutic dose. Based on the available data, BW and baseline serum albumin were identified as statistically significant covariates for CL. However, given the favorable safety and efficacy profiles for etrolizumab, the impact of these covariates on the therapeutic outcome is expected to be minimal and not clinically relevant. Disclosure - Xiaohui (Tracey) Wei, Diana Luca, Yehong Wang, Sharon O’Byrne, Rich Erickson, John D. Davis and Meina T. Tang are employees of Genentech, Inc.

A study of the combined effects of machining parameters on cutting force components during high speed robotic trimming of CFRPs

In this study, the combined effects of machining parameters on cutting force components during high speed robotic trimming of CFRPs were experimentally investigated. A statistical technique based on the evaluation of the rank and the condition number of the observation matrix followed by the backward elimination approach was used for selecting the most important variables of the models. The developed models were then used to predict the cutting force components for different values of cutting speed, feed and robot configuration. Outcomes from the models were analyzed and compared. A very good agreement with experimental data was achieved. Furthermore, results show that the amplitude for the normal force is higher than that for the feed force. The feed rate exhibits greater influence on cutting force components as compared to the cutting speed and robot configuration. The two-level interaction effect of the feed, cutting speed and robot configuration are also found significant.

Feature selection for high-dimensional class-imbalanced data sets using Support Vector Machines

Feature selection and classification of imbalanced data sets are two of the most interesting machine learning challenges, attracting a growing attention from both, industry and academia. Feature selection addresses the dimensionality reduction problem by determining a subset of available features to build a good model for classification or prediction, while the class-imbalance problem arises when the class distribution is too skewed. Both issues have been independently studied in the literature, and a plethora of methods to address high dimensionality as well as class-imbalance has been proposed. The aim of this work is to simultaneously explore both issues, proposing a family of methods that select those attributes that are relevant for the identification of the target class in binary classification. We propose a backward elimination approach based on successive holdout steps, whose contribution measure is based on a balanced loss function obtained on an independent subset. Our experiments are based on six highly imbalanced microarray data sets, comparing our methods with well-known feature selection techniques, and obtaining a better prediction with consistently fewer relevant features.

Chemometric classification of casework arson samples based on gasoline content

Detection and identification of ignitable liquids (ILs) in arson debris is a critical part of arson investigations. The challenge of this task is due to the complex and unpredictable chemical nature of arson debris, which also contains pyrolysis products from the fire. ILs, most commonly gasoline, are complex chemical mixtures containing hundreds of compounds that will be consumed or otherwise weathered by the fire to varying extents depending on factors such as temperature, air flow, the surface on which IL was placed, etc. While methods such as ASTM E-1618 are effective, data interpretation can be a costly bottleneck in the analytical process for some laboratories. In this study, we address this issue through the application of chemometric tools.Prior to the application of chemometric tools such as PLS-DA and SIMCA, issues of chromatographic alignment and variable selection need to be addressed. Here we use an alignment strategy based on a ladder consisting of perdeuterated n-alkanes. Variable selection and model optimization was automated using a hybrid backward elimination (BE) and forward selection (FS) approach guided by the cluster resolution (CR) metric.In this work, we demonstrate the automated construction, optimization, and application of chemometric tools to casework arson data. The resulting PLS-DA and SIMCA classification models, trained with 165 training set samples, have provided classification of 55 validation set samples based on gasoline content with 100% specificity and sensitivity.

Provision of Out-of-hospital Analgesia to Older Fallers With Suspected Fractures: Above Par, but Opportunities for Improvement Exist

Paramedics frequently attend older patients who have fallen and sustained suspected fractures, a population of patients who may be at risk of inadequate analgesic care. This prospective study aimed to describe the rate and effectiveness of analgesia administered by paramedics to older patients with suspected fractures secondary to falls and to identify predictive factors associated with provision of analgesia. A cohort of older patients aged greater than 65 years with suspected fall-related fractures was extracted from a database of 1,610 cases collected during a prospective, nonconsecutive observational study of older people who had fallen and received an ambulance response from October 1, 2010, through June 30, 2011. Fall-specific data, collected on scene by paramedics using a specially designed data form, were linked to patient clinical records and dispatch information. Descriptive analyses were performed to describe rates and effectiveness of analgesic administration, and multivariate logistic regression was conducted to identify factors associated with provision of analgesia. Of 1,610 patients in the observational study database, there were 333 patients identified as having suspected fractures, thus forming the study population. The mean (±SD) age was 82 (±8) years, and 75% were female. Suspected fractures of the hip were most common (42%). An initial pain score was recorded in 67% of cases, and the median initial pain severity was 8 of 10 (interquartile range [IQR] = 5 to 9). Overall, 60% received analgesia, and 80% of those received parenteral opiates. Intravenous (IV) morphine was most common (63%), followed by methoxyflurane (39%) and intranasal fentanyl (17%). Administration of oral analgesics was uncommon. Analgesia was considered to be clinically effective (≥30% relative reduction in pain severity) in 62% of cases. Patients with suspected hip fractures had greater odds of receiving analgesia compared to those with suspected fractures at other anatomical sites (odds ratio [OR] = 2.7, 95% confidence interval [CI] = 1.17 to 6.32; p = 0.02). Compared to those with mild pain, the odds of receiving analgesia increased significantly for patients with moderate pain (OR = 6.5, 95% CI = 2.3 to 18.8; p < 0.0001) and severe pain (OR = 31.1, 95% CI = 9.9 to 97.6; p < 0.0001). In this population of older people who fell and sustained suspected fractures, two-thirds received paramedic-administered analgesia. The majority of patients received clinically effective analgesia, and the presence of a suspected hip fracture increased the likelihood of receiving pain relief.

Changes in plasma and urine globotriaosylceramide levels do not predict Fabry disease progression over 1 year of agalsidase alfa

Globotriaosylceramide concentrations were assessed as potential predictors of change from baseline after 12 months by estimated glomerular filtration rate and left-ventricular mass index using pooled data from three randomized, placebo-controlled agalsidase alfa trials and open-label extensions of patients with Fabry disease. Males (aged 18 years or older) with Fabry disease received agalsidase alfa (0.2 mg/kg every other week for 12 months). A backward-elimination approach evaluated potential predictors (baseline estimated glomerular filtration rate and left-ventricular mass index; age at first dose; baseline and change from baseline at 12 months of globotriaosylceramide (urine, plasma); urine protein excretion; and systolic and diastolic blood pressure). Subgroups included patients randomized to placebo or agalsidase alfa (double-blind phase), then to agalsidase alfa (open-label extensions; placebo→agalsidase alfa or agalsidase alfa→agalsidase alfa, respectively) and stage 2/3 chronic kidney disease patients. Baseline estimated glomerular filtration rate, age at first dose, baseline urine globotriaosylceramide excretion, and baseline and change from baseline urine protein excretion significantly predicted change from baseline estimated glomerular filtration rate in the analysis population (N = 73; all P<0.05), although not in all subgroups. Change from baseline urine and plasma globotriaosylceramide (baseline and change from baseline) concentrations did not predict change from baseline estimated glomerular filtration rate. No predictors of left-ventricular mass index were significant. Changes in globotriaosylceramide concentrations do not appear to be useful biomarkers for prediction of Fabry disease-related changes in estimated glomerular filtration rate or left-ventricular mass index.

Examining Multiple Parenting Behaviors on Young Children’s Dietary Fat Consumption

ObjectiveTo understand the association between parenting and children’s dietary fat consumption, this study tested a comprehensive model of parenting that included parent household rules, parent modeling of rules, parent mediated behaviors, and parent support. DesignCross-sectional. SettingBaseline data from the MOVE/me Muevo project, a recreation site-based obesity prevention and control intervention trial. ParticipantsFive hundred forty-one parents of children between the ages of 5 and 8 years and living in San Diego County. Main Outcome MeasureChildren’s fat consumption based on parent report using a short food frequency questionnaire. AnalysisA hierarchical linear regression was conducted. In exploratory analyses, a stepwise backward elimination approach was used. ResultsChildren’s fat consumption was positively associated with parent household rules (P < .01) and negatively associated with parent modeling of rules (P < .01). Conclusions and ImplicationsControlling parenting behaviors, such as rule setting, are associated with more frequent fat consumption, whereas role modeling healthful behaviors is associated with less frequent fat consumption. Changing parenting behaviors with regard to how they feed their children is a logical avenue for improving eating behaviors.

Pharmacokinetics of dexmedetomidine in Chinese post-surgical intensive care unit patients

Dexmedetomidine is a rather new drug in China. We sought to describe the pharmacokinetics of dexmedetomidine in patients requiring post-operative sedation and ventilation in our surgical intensive care unit. Twenty-two patients received post-operative infusions of dexmedetomidine at 6 μg/kg/h for 10 min, followed by 0.4 μg/kg/h for 350 min. Venous blood samples were drawn and assayed for plasma concentration. The pharmacokinetics were analysed using a nonlinear mixed-effect model with an interindividual and intraindividual error model. An initial estimation was performed to determine which of the one-, two- or three-compartment models is best to describe the concentration-time data. The covariates age, gender, weight, height, lean body mass (LBM), body surface area (BSA) and body mass index (BMI) were tested for significant effects on parameters using a stepwise forward addition and backward elimination approach. Covariate effects were judged based on changes in the objective function value (OFV). The pharmacokinetics of dexmedetomidine were best described by a three-compartment model. The model was further improved when height was a covariate of systemic clearance (Cl1), with a decrease in OFV by -13.56 (P<0.01). From the heights of 155-178 cm, Cl1 increased by approximately 143%. The final pharmacokinetic parameter values were as follows: V1 =63.4 l, V2=41.3 l, V3 =284.3 l, Cl1=0.47×(height/160 cm)(6.42) l/min, Cl2=2.43 l/min and Cl3=0.086 l/min. This study identified (i) the effect of height on the pharmacokinetics of dexmedetomidine; (ii) that there is no influence of age, gender, weight, LBM, BSA and BMI on pharmacokinetic parameters; and (iii) it established a preliminary population pharmacokinetic model for Chinese patients.

A regression model to predict warfarin dose from clinical variables and polymorphisms in CYP2C9, CYP4F2, and VKORC1: Derivation in a sample with predominantly a history of venous thromboembolism

Background Pharmacogenomic warfarin dosing has been suggested to produce more accurate dosing and an improved patient safety profile; however, very few models have been derived in patients with venous thromboembolism. We sought to develop a new algorithm to predict maintenance dose in a cohort of patients, using clinical variables and genetic polymorphism in CYP2C9, VKORC1, and CYP4F2. Methods Patients on a stable maintenance dose of warfarin, with observed dose ranging from 0.6 to 12 mg were recruited from a specialized anticoagulation clinic (Ottawa Hospital Thrombosis Clinic) with genotyping and standardized patient interviews being conducted to collect clinical and genomic variables known to impact warfarin dose. Multivariate linear regression was used to develop the model using a stepwise backwards elimination approach. Results From 249 enrolled patients with a mean clinical maintenance dose of 5.58 mg/day, a model with an R 2 of 58% was developed as: Dose = 1.85-0.048(Age) + 0.041(BMI) + 0.05(Height in cm) - 0.73(Less Exercise) - 1.13(2C9*2 Hetero) - 2.09(2C9*2 Homo) - 1.51(2C9*3 Hetero) -1.43(VKORC1 GA) - 2.86(VKORC1 AA) - 1.33(4F2 CC) -1.24(4F2 CT) - 1.46(Angiotensin II Receptor Antagonist) - 0.84(β-Blockers). Analysis of residual plots revealed that prediction errors were a function of observed maintenance dose with the model tending to predict higher doses than observed in those with low dose requirements and lower doses than observed in those with higher dose requirement. Conclusion Our study confirms the importance of the CYP4F2 polymorphism. Our model may prove useful in clinical practice but further validation studies are required before implementation into clinical practice.

Population Pharmacokinetics of Siplizumab (MEDI-507), a Humanized Anti-CD2 Monoclonal Antibody, in Cancer Patients.

Abstract 2670Poster Board II-646 Introduction:Siplizumab, a humanized IgG1k class monoclonal antibody that targets CD2 expressing T-and NK-cells, was evaluated in phase I dose-escalation trials in patients with CD2-positive lymphoproliferative disorder. The objective of this study was to develop a population pharmacokinetic (PK) model for siplizumab and identify covariates that could explain the variability in siplizumab pharmacokinetic parameters. Methods:A Phase 1, open label, dose-escalation study was conducted in 29 patients (14 males/15 females, age range 34–79 years) who received 0.2–4.8 mg/kg of siplizumab as 1-3 consecutive daily doses every 14 days for a total of 1-8 cycles. Siplizumab serum concentration data was analyzed using a nonlinear mixed effects modeling approach with software (NONMEM). Based on exploratory analysis, 1-and 2-compartment non-linear models were evaluated. Demographic covariates including body weight, age, sex and race (Caucasian/Black/Asian) were screened using Generalized Additive Model (GAM) analysis. Covariates selected during the GAM analysis were further tested for significance in NONMEM using the forward inclusion and backward elimination approach. Results:Siplizumab concentrations were obtained from all 29 patients in the study yielding a total of 619 serum concentration observations. Pronounced non-linearity in siplizumab serum concentrations was observed after the initial and later dosing cycles, with serum concentrations declining faster at lower dose levels. The data was best described by a two-compartment pharmacokinetic model with zero-order input with parallel linear and non-linear elimination pathways. Goodness of fit plots and model diagnostics indicated good agreement between observed and model predicted serum concentration values. The population estimates for linear clearance was 0.168 L/day with inter-subject variability (ISV) of 50%, and inter-compartmental clearance was 2.83 L/day. Nonlinear elimination parameters, Vmax and Km were 10.32 mcg/day (56% ISV) and 51.8 mcg/L, respectively. Sex of the patients was identified as a significant covariate impacting volumes of distribution. Male patients had higher central and peripheral volumes of distribution of 2.8 L and 3.0 L, respectively, compared to1.38 L and 2.4 L in females [32% vs 50% ISV]. Conclusion:The serum concentration-time profile of siplizumab was adequately described by a two-compartment non-linear PK model. Population parameters were precisely estimated and correspond well to reported PK behaviour of monocolonal antibodies with significant target mediated elimination. The lower volume distribution in females is most likely due to lower body weight compared to males in this study. The population PK model combined with pharmacodynamic data could serve as a tool to guide selection of optimal dose regimens for siplizumab. Disclosures:No relevant conflicts of interest to declare.

Using backward elimination with a new model order reduction algorithm to select best double mixture model for document clustering

Probabilistic latent semantic analysis (PLSA) is a double structure mixture model which has got a wide application in text and web mining. This method is capable of establishing hidden semantic relations among the observed features, using a number of latent variables. In this approach, the selection of the correct number of latent variables is critical. In the most of the previous researches, the number of latent topics was selected based on the number of invoked classes. This paper presents a method, based on backward elimination approach, which is capable of unsupervised order selection in PLSA. This method starts with a model having a number of components more than the needed value, and then prunes the mixtures to reach their optimum size. During the elimination process, proper selection of some latent variables which must be deleted is the most essential problem, and its relation to the final performance of the pruned model is straightforward. To treat this problem, we introduce a new combined pruning method which selects the best options for removal, while keeping a low computational cost, at all. We conducted some experiments on two datasets from Reuters-21578 corpus. The obtained results show that this algorithm leads to an optimized number of latent variables and in turn achieves better clustering performance compared to the conventional model selection methods. It also shows superiority over the case in which a PLSA model with a fixed number of latent variables, equal to the real number of clusters, is exploited.

Factors Associated With a Successful External Cephalic Version in the Esarly ECV Trial

Objectives:The objective of this research was to determine factors that were associated with a successful external cephalic version (ECV) procedure. Methods:We undertook a secondary analysis of data from a randomized controlled trial, The Early External Cephalic Version (Pilot) Trial. In this secondary analysis, we included data for the subset of 178 women who had an ECV as part of the pilot trial (123 nulliparous women with any breech presentation and 55 multiparous women with a frank breech presentation only). Using this dataset, we began with two separate univariate analyses, one of characteristics that could be determined before undertaking a procedure, and the other of factors associated with the ECV procedure itself. Variables that had a P value of ≤0.1 in the univariate analyses were included in two separate logistic regression models, one for preprocedural and one for procedural factors, using a backward elimination approach. Results:Multiparity and a non-engaged presenting part were significant preprocedural predictors of ECV success. Procedural factors predictive of ECV success included lower reported maternal pain scores during the procedure, a single attempt at ECV, and a more mobile fetus. Conclusion:Non-engagement of the presenting part was the only modifiable factor predicting ECV success that was identified in this analysis, and it supports the hypothesis that beginning the ECV procedure earlier in pregnancy, prior to engagement, may have merit. The Early ECV 2 Trial is in progress and will further test this hypothesis.

Genetic Determinants of Susceptibility to Cytarabine Arabinoside.

Cytarabine arabinoside (ara-C) is an essential component of treatment for acute myeloid leukemia (AML) and resistance to this agent is a main reason for treatment failure. To identify genetic factors that are important in susceptibility to the cytotoxic effects of ara-C, we utilized an unbiased whole-genome approach using lymphoblastoid cell lines that are part of the International HapMap Project. We exposed 85 cell lines derived from Utah residents with ancestry from northern and western Europe (CEU) and 89 cell lines derived from individuals of Yoruba in Ibadan, Nigeria (YRI) to a range of concentrations of ara-C (1 – 80 μM) for 72 hours. We used a short-term proliferation assay, alamarBlue®, to measure cell growth inhibition at each concentration of ara-C. In addition, we used the Affymetrix Human Exon 1.0 ST array to determine gene expression of each cell line. We utilized publicly available HapMap (release #22) single nucleotide polymorphisms (SNPs) to perform genome wide association after filtering to remove Mendelian errors and SNPs with mean allele frequencies <5%, leaving approximately 2 million SNPs. We then performed a 3-step analysis to identify genetic signatures associated with gene expression important in susceptibility to ara-C cytotoxicity. The first step involved a Quantitative Transmission Disequilibrium Test (QTDT) association analysis between SNP genotype and percent survival at each concentration of ara-C. Using the SNPs identified in this initial step, we performed a second QTDT analysis between genotype and gene expression of approximately 17,000 transcripts from the exon array to identify SNPs that were associated with the expression of any genes. Finally, we performed a linear regression analysis between the expression of genes identified in this second step and percent survival to identify a subset of these genes whose expression is correlated with susceptibility to ara-C. Using a multivariable backward elimination approach we also determined how much variation in ara-C percent survival is explained by the SNPs identified in step 2. The mean (range) percent survival in the CEU compared to the YRI population was 53.3 (20.8 – 82.8) vs. 47.3 (25.2 – 73.7) at 5μM (p = .0018), 46.8 (18.7–73.2) vs. 39.9 (22.2 – 72.4) at 10μM (p = 0.0001), 40.3 (20.3 – 68.2)vs. 32.8 (17.5 – 53.9) at 40μM (p < 0.0001), and 37.3 (19.1 – 61.3) vs. 30.4 (15.8 – 50.0) (p < 0.0001) at 80μM ara-C. Among the 85 CEU cell lines we identified 177 SNPs for which genotype significantly associated with percent survival at 1μM, using an FDR cutoff of 0.05. Eleven of these 177 SNPs were significantly associated with the expression of a total of 24 genes and all were located trans to the associated gene. The expression of 23 of these genes significantly correlated with percent survival. Of the 11 SNPs that were associated with percent survival and gene expression, 4 SNPs explained 61.2% of the variation in susceptibility to ara-C cytotoxicity among the CEU cell lines. By using an cell-based model and a whole-genome approach that integrates genetic variation, expression variation and sensitivity to drug, we were able to identify a genetic signature for susceptibility to the cytotoxic effects of ara-C. We can now concentrate on the candidate genes identified from this analysis for further functional validation studies.

Variable selection in kernel Fisher discriminant analysis by means of recursive feature elimination

Variable selection serves a dual purpose in statistical classification problems: it enables one to identify the input variables which separate the groups well, and a classification rule based on these variables frequently has a lower error rate than the rule based on all the input variables. Kernel Fisher discriminant analysis (KFDA) is a recently proposed powerful classification procedure, frequently applied in cases characterised by large numbers of input variables. The important problem of eliminating redundant input variables before implementing KFDA is addressed in this paper. A backward elimination approach is recommended, and two criteria which can be used for recursive elimination of input variables are proposed and investigated. Their performance is evaluated on several data sets and in a simulation study.

Some Factors Affecting Reproductive Success in Crossbred Dairy Cows on Smallholder Farms in Coastal North-East Tanzania

A 2-year longitudinal survey was carried out to investigate factors affecting reproduction in crossbred cows on smallholder farms in and around an urban centre. Sixty farms were visited at approximately 2-week intervals and details of reproductive traits and body condition score (BCS) were collected. Fifteen farms were within the town (U), 23 farms were approximately 5 km from town (SU). and 22 farms approximately 10 km from town (PU). Sources of variation in reproductive traits were investigated using a general linear model (GLM) by a stepwise forward selection and backward elimination approach to judge important independent variables. Factors considered for the first step of formulation of the model included location (PU, SU and U), type of insemination, calving season, BCS at calving, at 3 months postpartum and at 6 months postpartum, calving year, herd size category, source of labour (hired and family labour), calf rearing method (bucket and partial suckling) and parity number of the cow. The effects of the independent variables identified were then investigated using a non-parametric survival technique. The number of days to first oestrus was increased on the U site (p = 0.045) and when family labour was used (p = 0.02). The non-parametric test confirmed the effect of site (p = 0.059), but effect of labour was not significant. The number of days from calving to conception was reduced by hiring labour (p = 0.003) and using natural service (p = 0.028). The non-parametric test confirmed the effects of type of insemination (p = 0.0001) while also identifying extended calving intervals on U and SU sites (p = 0.014). Labour source was again non-significant. Calving interval was prolonged on U and SU sites (p = 0.021), by the use of AI (p = 0.031) and by the use of family labour (p = 0.001). The nonparametric test confirmed the effect of site (p = 0.008) and insemination type (p > 0.0001) but not of labour source. It was concluded that under favourable conditions (PU site, hired labour and natural service) calving intervals of around 440 days could be achieved.