Binary Predictors Research Articles

Spatial Analysis of Driving Factors on Land Cover Change’s Clusters in West Java Province

The study of land cover change as a phenomenon is necessary to do in order to understand global environmental change. With a cluster, the pattern and concentration of land cover change can be indicated, so that decisions can be made on target. The understanding of land cover change can be improved by identifying the factors affecting it. Land cover change depends on the physical and socio-economic characteristics of a region. Thus, the driving factor of land cover change in each region is different. This research is conducted to identify the driving factors on West Java Province’s land cover change based on it is land cover change clusters. The relation between each factor and land cover change can be evaluated using the binary logistic regression, which is a data analysis method used to find a relationship between a binary response variable (y) and predictor variable(s) (x). In this case, the land cover change, as a dichotomous phenomenon, acts as the response variable, while the driving factors act as the predictor variables. The result of this study indicates that in general, the distance to the nearest central business district, as a factor, is the dominant driving factor of land cover change on West Java Province.

Abstract PD10-08: Baseline patient reported outcomes predict aromatase inhibitor adherence failure among women in enrolled in a randomized trial of text-messaging

Abstract Background: Non-adherence to aromatase inhibitors (AIs) for breast cancer is common and increases risk of recurrence. Few prospective studies have systematically evaluated reasons for adherence failure. S1105 randomized women to receive text message reminders or usual care, and no effect of the intervention was observed on the rates of adherence failure. We analyzed baseline characteristics to evaluate factors associated with non-adherence prospectively over 3-years.Methods: Patients enrolled in SWOG S1105 were required to have been on an AI for ≥30 days at enrollment. Patients were assessed for non-adherence of AIs every 3 months for 36 months, with non-adherence defined as urine AI metabolite assay results satisfying any of the following: <10 [units], undetectable, specimen submitted outside of the ± 21 day follow-up appointment window, or no submitted specimen. At baseline patients were asked about insurance, pill number dispensed (30 vs 90) and medication cost, and they completed patient reported outcomes (PROs) focused on pain and endocrine symptoms (BPI (Brief Pain Inventory), FACT-ES (Endocrine Symptoms)), quality of life (FACT-G), as well as their beliefs about medications (TSQM (Treatment Satisfaction Questionnaire for Medicine) and BMQ (Brief Medication Questionnaire)). PRO scales were split at the median creating high (>median) vs low (<=median) binary predictors. We examined the association of baseline factors with both adherence failure at 36 months with a logistic regression and time-to-adherence failure (TTAF) with a cox proportional hazard regression controlling for stratification factors. Results: In total, 724 patients were registered from 40 institutions between May, 2012 and September, 2013. The median age was 60.9 years, and 64.5% were on AI <12 months prior to registration. Observed adherence at 36 months was 55.4% (41.2% in-window, 14.2% out of window). Neither age nor race nor ethnicity was associated with long-term adherence. Of the 29 baseline factors we analyzed, 12 were associated with non-adherence at 36 months, including symptoms (FACT-ES total, OR=1.64, 95% CI:1.2, 2.22 p=.002; BPI Interference OR=1.68, 95% CI:1.23,2.28 p=.001; FACT G, OR=1.70, 95% CI:1.25, 2.31 p<.001), satisfaction with medications (TSQM global, OR=1.55, 95% CI:1.1,2.17 p=.01) and beliefs about medication (BMQ, OR=1.51, 95% CI:1.11, 2.05 p=.009). Nine of 29 baseline factors were associated with TTAF, including lack of private insurance (HR=1.20, 95% CI:1.01,1.41 p=.04) and 30 day vs. 90 day pill supply (HR=1.2, 95% CI:1.01,1.42 p=.04). Factors associated with self-reported adherence will be presented. Conclusions: Baseline endocrine symptoms, pain, medication satisfaction, beliefs and concerns about side effects are strongly associated with long-term AI adherence. Targeted behavioral and symptom management interventions to improve adherence should be focused on women at the highest risk of adherence failure. Citation Format: Dawn L Hershman, Joeseph M Unger, Anna Moseley, Grace C Hillyer, Kathryn B. Arnold, Julie R Gralow, N. Lynn Henry, Scott D Ramsey, Alfred I Neugut. Baseline patient reported outcomes predict aromatase inhibitor adherence failure among women in enrolled in a randomized trial of text-messaging [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr PD10-08.

A novel cellular automata model integrated with deep learning for dynamic spatio-temporal land use change simulation

Land use change (LUC) exhibits obvious spatio-temporal dependency. Previous cellular automata (CA)-based methods usually treated the LUC dynamics as Markov processes and proposed a series of CA-Markov models, which however, were intrinsically unable to capture the long-term temporal dependency. Meanwhile, such models used only numerical proportion of neighboring land use (LU) types to represent neighborhood effects of LUC, which inevitably neglected the complicated spatial heterogeneity and thus caused inaccurate simulation results. To address these problems, this paper presents a novel CA model integrated with deep learning (DL) techniques to model spatio-temporal LUC dynamics. Our DL-CA model firstly uses a convolutional neural network to capture latent spatial features for complete representation of neighborhood effects. A recurrent neural network then extracts historical information of LUC from time-series land use maps. A random forest is appended as binary change predictor to avoid the imbalanced sample problem during model training.Land use data collected from 2000 to 2014 of the Dongguan City, China were used to verify our proposed DL-CA model. The input data from 2000 to 2009 were used for model training, the 2010 data for model validation, and the data collected from 2011 to 2014 were used for model evaluation. In addition, four traditional CA models of multilayer perceptron (MLP)-CA, support vector machine (SVM)-CA, logistic regression (LR)-CA and random forest (RF)-CA were also developed for accuracy comparisons. The simulation results demonstrate that the proposed DL-CA model accurately captures long-term spatio-temporal dependency for more accurate LUC prediction results. The DL-CA model raised prediction accuracy by 9.3%–11.67% in 2011–2014 in contrast to traditional CA models.

ROC and AUC with a Binary Predictor: a Potentially Misleading Metric.

In analysis of binary outcomes, the receiver operator characteristic (ROC) curve is heavily used to show the performance of a model or algorithm. The ROC curve is informative about the performance over a series of thresholds and can be summarized by the area under the curve (AUC), a single number. When a predictor is categorical, the ROC curve has one less than number of categories as potential thresholds; when the predictor is binary there is only one threshold. As the AUC may be used in decision-making processes on determining the best model, it important to discuss how it agrees with the intuition from the ROC curve. We discuss how the interpolation of the curve between thresholds with binary predictors can largely change the AUC. Overall, we show using a linear interpolation from the ROC curve with binary predictors corresponds to the estimated AUC, which is most commonly done in software, which we believe can lead to misleading results. We compare R, Python, Stata, and SAS software implementations. We recommend using reporting the interpolation used and discuss the merit of using the step function interpolator, also referred to as the "pessimistic" approach by Fawcett (2006).

A New Principal Stratum Estimand Investigating the Treatment Effect in Patients Who Would Comply, If Treated With a Specific Treatment

The draft ICH E9 (R1) addendum by the International Conference on Harmonisation working group opens for the use of a principal stratum (PS) in the analysis of study data for regulatory purpose, if a relevant estimand can be justified. Inspired by the so-called complier average causal effect and work within this framework, we propose a new estimator—Extrapolation based on propensity to comply—that estimates the treatment effect of an active treatment A relative to a comparator B (active or placebo), in the PS of patients who would comply, if they were treated with treatment A. Sensitivity of the approach to the number of covariates and their ability to predict PS membership is investigated based on data from a placebo-controlled study of brexpiprazole in schizophrenia. The performance of the estimator is compared with another estimator that is also based on principal stratification. A simulation study supports that the proposed estimator has a negligible bias even with a small sample size, except when the covariate predicting compliance is very weak. Not surprisingly, precision of the estimate increases substantially with stronger predictors of compliance. It is shown analytically that the estimator used for comparison is biased in general, and the bias is explicitly derived in a simple case with a binary predictor of compliance. While the proposed methodology is technically easy to implement, choosing predictors for modeling compliance is a key issue in practice.

841 The Impact of Inpatient Intravenous Opioids on Future Outpatient Opioid Exposure Among Hospitalized Patients With Inflammatory Bowel Disease

INTRODUCTION: Patients hospitalized for inflammatory bowel disease (IBD) often receive opioids (OPIs) to control pain. However, OPI use is associated with excess mortality and infections in IBD patients (AJG 2012;107(9):1409-22). Given the high potency of intravenous (IV) OPIs and their unclear impact on future OPI exposure, we sought to determine if IV OPI use is associated with subsequent outpatient OPI prescriptions (Rx) among hospitalized IBD patients. METHODS: We performed a retrospective cohort study of patients aged >18 years hospitalized at a large urban academic health system from 3/1/17 to 4/10/18. Patients were included if they were admitted to observation, general medical, general surgical, or gastrointestinal surgical services, had a discharge diagnosis of IBD, and had >1 outpatient office visit in the department of gastroenterology within 12 months of discharge. IBD history, patient-reported pain scores, and demographic, medication administration, and outpatient Rx data were abstracted from the electronic health record. We used multivariable mixed effect logistic regression modeling (MVMELR) to account for repeat encounters and to assess the association between IV OPI exposure and an outpatient OPI Rx provided within 12 months after discharge while adjusting for confounders. IV OPI exposure was represented both as a binary variable and a continuous variable in total IV morphine mg equivalents/length of stay in days (IVMMEs/d). Kaplan-Meier (KM) analysis was performed to compare time to outpatient OPI Rx by inpatient OPI exposure. RESULTS: We included 709 encounters among IBD patients between 3/1/17 and 4/10/18. Patient characteristics and unadjusted outcomes by OPI exposure are presented in Table 1. MVMELR demonstrated a significant association between IV OPI exposure (binary predictor: OR 2.9, 95% CI 1.4-5.9; continuous predictor in IVMMEs/d: OR 1.0, 95% CI 1.0-1.1) and an outpatient OPI Rx within 12 months of discharge. Receiving only non-IV OPIs was not associated with an opioid Rx (OR 0.8, 95% CI 0.3-2.4; Table 2). KM analysis demonstrated significant separation in rate curves for OPI Rx between OPI exposure groups (log rank test P < 0.01; Figure 1). CONCLUSION: Administration of IV OPIs to hospitalized IBD patients may increase their risk for further OPI exposure after discharge. IV OPIs should be used judiciously in this high-risk population with consideration of appropriate alternatives. Additional research is needed to identify long-term risks of IV OPI use in patients with IBD.

APRIDIT Unsupervised Classification with Asymmetric Valuation of Variable Discriminatory Worth

Sometimes one needs to classify individuals into groups, but there is no available grouping information due to social desirability bias in reporting behavior like unethical or dishonest intentions or unlawful actions. Assessing hard-to-detect behaviors is useful; however it is methodologically difficult because people are unlikely to self-disclose bad actions. This paper presents an unsupervised classification methodology utilizing ordinal categorical predictor variables. It allows for classification, individual respondent ranking, and grouping without access to a dependent group indicator variable. The methodology also measures predictor variable worth (for determining target behavior group membership) at a predictor variable category-by-category level, so different variable response categories can contain different amounts of information about classification. It is asymmetric in that a “0” on a binary predictor does not have a similar impact toward signaling “membership in the target group” as a “1” has for signaling “membership in the non-target group.” The methodology is illustrated by identifying Spanish consumers filing fraudulent insurance claims. A second illustration classifies Portuguese high school student’s propensity to alcohol abuse. Results show the methodology is useful when it is difficult to get dependent variable information, and is useful for deciding which predictor variables and categorical response options are most important.

Models with interactions overestimated heterogeneity of treatment effects and were prone to treatment mistargeting

ObjectivesWe aimed to compare the performance of different regression modeling approaches for the prediction of heterogeneous treatment effects. Study Design and SettingWe simulated trial samples (n = 3,600; 80% power for a treatment odds ratio of 0.8) from a superpopulation (N = 1,000,000) with 12 binary risk predictors, both without and with six true treatment interactions. We assessed predictions of treatment benefit for four regression models: a “risk model” (with a constant effect of treatment assignment) and three “effect models” (including interactions of risk predictors with treatment assignment). Three novel performance measures were evaluated: calibration for benefit (i.e., observed vs. predicted risk difference in treated vs. untreated), discrimination for benefit, and prediction error for benefit. ResultsThe risk modeling approach was well-calibrated for benefit, whereas effect models were consistently overfit, even with doubled sample sizes. Penalized regression reduced miscalibration of the effect models considerably. In terms of discrimination and prediction error, the risk modeling approach was superior in the absence of true treatment effect interactions, whereas penalized regression was optimal in the presence of true treatment interactions. ConclusionA risk modeling approach yields models consistently well calibrated for benefit. Effect modeling may improve discrimination for benefit in the presence of true interactions but is prone to overfitting. Hence, effect models—including only plausible interactions—should be fitted using penalized regression.

P37-F Predictors of cognitive impairment in patients with mine blast injury

Background Mine blast injury is accompanied by traumatic brain injury (TBI) in 60–100% of cases. TBI is associated with cognitive impairment (CI). However, predictors of CI in this patients (pts) are studied insufficiently. Material and methods 115 men aged 18 to 45 years with mild TBI caused by using of mine-blasting weapons. Cognitive functions were evaluated by using MoCA test, HADS scale. Cognitive dysfunction was defined as MoCa ⩽ 25 scores. Besides routine examination, neurophysiological methods such as quantitative electroencephalography (qEEG) and cognitive evoked potentials P300 (CEP P300), were performed. Results. In binary logistic regression predictors CI were complaints on headache and fatigability (OR 2.52, 95% CI 2.6–59.7, p = 0.02), increased tendon reflexes (OR, 2.6, 95% CI 1.24–2.32, p = 0.017), higher HADS anxiety level (OR-1.8, 95% CI 1.59–2.2, p = 0.04), changes of alpha rhythm frequency (OR-1.02, 95% CI 1.01–1.04, p = 0.004), decrease of alpha rhythm amplitudes (OR - 1.07, 95% CI 1.03–1.11, p Conclusions changes in qEEG and CEP P300 parameters are reliable independent predictors of CI in pts with TBI affected by mine blast trauma.

The relationship between statistical power and predictor distribution in multilevel logistic regression: a simulation-based approach

BackgroundDespite its popularity, issues concerning the estimation of power in multilevel logistic regression models are prevalent because of the complexity involved in its calculation (i.e., computer-simulation-based approaches). These issues are further compounded by the fact that the distribution of the predictors can play a role in the power to estimate these effects. To address both matters, we present a sample of cases documenting the influence that predictor distribution have on statistical power as well as a user-friendly, web-based application to conduct power analysis for multilevel logistic regression.MethodComputer simulations are implemented to estimate statistical power in multilevel logistic regression with varying numbers of clusters, varying cluster sample sizes, and non-normal and non-symmetrical distributions of the Level 1/2 predictors. Power curves were simulated to see in what ways non-normal/unbalanced distributions of a binary predictor and a continuous predictor affect the detection of population effect sizes for main effects, a cross-level interaction and the variance of the random effects.ResultsSkewed continuous predictors and unbalanced binary ones require larger sample sizes at both levels than balanced binary predictors and normally-distributed continuous ones. In the most extreme case of imbalance (10% incidence) and skewness of a chi-square distribution with 1 degree of freedom, even 110 Level 2 units and 100 Level 1 units were not sufficient for all predictors to reach power of 80%, mostly hovering at around 50% with the exception of the skewed, continuous Level 2 predictor.ConclusionsGiven the complex interactive influence among sample sizes, effect sizes and predictor distribution characteristics, it seems unwarranted to make generic rule-of-thumb sample size recommendations for multilevel logistic regression, aside from the fact that larger sample sizes are required when the distributions of the predictors are not symmetric or balanced. The more skewed or imbalanced the predictor is, the larger the sample size requirements. To assist researchers in planning research studies, a user-friendly web application that conducts power analysis via computer simulations in the R programming language is provided. With this web application, users can conduct simulations, tailored to their study design, to estimate statistical power for multilevel logistic regression models.

Association of OXTR rs53576 with the Developmental Trajectories of Callous-Unemotional Traits and Stressful Life Events in 3- to 9-Year-Old Community Children.

The objective was to obtain developmental trajectories combining callous-unemotional traits and the number of stressful life-events between ages 3 and 9years and to ascertain their association with the polymorphism rs53576 at the Oxytocin Receptor gene (OXTR). A total of 377 children were assessed yearly from ages 3 to 9years. Latent class growth analysis for parallel processes was used to identify distinct trajectories for callous-unemotional traits (assessed using the Inventory of Callous-Unemotional Traits, ICU) and number of stressful life-events, and then the influence of being an A allele carrier on class membership was included with OXTR genotypes as a binary time-invariant predictor, following a 3-step approach. A 3-class model showed the highest entropy (.859) and adequate posterior probabilities of class membership (≥.884). Class 1 (n = 226, 59.9%) included children with low and stable ICU scores and low and descending stressful life-events; class 2 (n = 127, 33.7%) included children with high and ascending ICU scores and low and slightly descending stressful life-events; and class 3 (n = 24, 6.4%) included children with persistently high profiles both for ICU scores and stressful life-events. Carrying an A allele (genotypes GA/AA) increased the odds of pertaining to class 3 (high and persistent ICU scores and stressful life-events) as opposed to class 2 (OR = 4.27, p = 0.034) or class 1 (OR = 3.81, p = 0.042). The results suggest the importance of considering callous-unemotional traits and stressful life-events in conjunction. In addition, the genetic variability of OXTR (rs53576) may help to understand individual differences in early development.

Anthropometric indices predicting incident hypertension in an Iranian population: The Isfahan cohort study.

Objective:The aim of the present study was to assess different obesity indices, as well as their best cut-off point, to predict the occurrence of hypertension (HTN) in an Iranian population.Methods:In a population-based study, subjects aged 35 years and older were followed for 7 years. Blood pressure was measured at baseline and after the follow-up. Anthropometry indices included body mass index (BMI), body adiposity index (BAI), the waist-to-height ratio (WHtR), the waist-to-hip ratio (WHpR), and waist and hip circumferences (WC and HC). Logistic regression was employed to calculate the odds ratio (OR) and 95% confidence intervals (CI) per standard deviation (SD) increment. The operating characteristic analysis was used to derive the best cut-off value for each index.Results:Among original 6504 participants, 2450 subjects who had no cardiovascular diseases (CVD) and HTN at baseline were revisited, and 542 (22.1%) new cases of HTN were detected. There were minimal differences between most indices in the adjusted models; however, the best HTN predictors were BMI (OR per SD 1.32; 95% CI 1.12–1.56) and almost equally WC (1.35; 1.13–1.60) in men and WC (1.20; 1.04–1.39) in women. As a binary predictor, BMI with a cut-off point of 24.9 kg/m2 in men (1.91; 1.40–2.62) and WC with a cut-off point of 98 cm in women (1.57; 1.17–2.10) were the best in adjusted models. WC, WHpR, and WHtR were significantly associated with an increased risk of HTN only in participants whose weight was normal (BMI, 18.5–24.9 kg/m2).Conclusion:Therefore, BMI in men and WC in women were the best predictors of HTN, both as continuous and binary factors at their appropriate cut-off points.

Dynamic BH3 Profiling Predicts for Clinical Response to Lenalidomide Plus Chemotherapy in Relapsed Acute Myeloid Leukemia

Background: Early prediction of response to novel chemotherapy combinations for patients diagnosed with relapsed or refractory acute myeloid leukemia (R/R AML) may have clinical utility. We conducted a phase 1 clinical trial (NCT01904643) of lenalidomide (LEN) given prior to MEC (Mitoxantrone + Etoposide + Cytarabine) salvage chemotherapy (LEN+MEC) for patients with R/R AML. Although the clinical trial was terminated due to administrative reasons, we explored the utility of FACS-based BH3 and dynamic BH3 profiling (DBP) assays to predict for response. Standard BH3 profiling is a functional assay that uses synthetic peptides derived from the BH3 domains of pro-apoptotic BCL-2 family members to measure mitochondrial “priming” (a cell's readiness for apoptosis) (Ni Chonghaile, Science, 2011). DBP measures whether short term incubation with drug enhances priming (Montero J, Cell, 2015).Methods: Primary objective was to assess the safety/toxicity profile (DLTs and MTD) of LEN+MEC. LEN was administered 5-7 days prior to MEC and dose was escalated using an mTPI design with a target toxicity level of 30%. Dose levels (DL) included LEN given at doses of 15mgx5d (DL1), 15mgx7d (DL2), 25 mgx5d (DL3), 25mgx7d (DL4), 50mgx5d (DL5), and 50 mgx7d (DL6). Dose-escalation decisions were made based on DLTs that occurred during induction chemotherapy. DLT was defined as persistent cytopenias in absence of disease, gr 3 or higher non-hematologic toxicity not due to underlying disease or intolerable gr 2 toxicities attributable to LEN. Blood/bone marrow samples were collected at pre-treatment, post LEN only, and post LEN+MEC therapy. Clinical response was defined per ELN guidelines. For correlative studies, standard BH3 and DBP assays were used to measure overall mitochondrial priming and individual anti-apoptotic protein dependencies. DBP was performed after overnight (~16h) treatment with LEN at various concentrations on pretreatment leukemic blasts. Delta priming, utilized as a readout, represents the change in mitochondrial outer membrane depolarization after LEN treatment as measured by cytochrome C release following exposure to BH3 peptides. Data analysis was performed blinded to clinical response. Receiver operating characteristic (ROC) curve analysis was used to determine the optimal cut-off level for respective clinical end points.Results: 17 total patients with R/R AML received LEN+MEC. Median age was 55 years (range: 22-72) with 12% Fav, 47% Int and 41% Adv risk by ELN risk stratification. 60% were heavily pre-treated. DLTs were observed in DL2 (n=1, gr 4 sepsis due to bacteremia), DL3 (n=1, gr 5 septic shock), and DL5 (n=1, gr3 bacteremia). MTD was not yet reached due to early study termination (although patients were treated at all 6 dose levels). No GVHD was observed among post-transplant patients. Composite remission rate among the 17 patients (CR + PR) was 35% (5 CR and 1 PR). Of the 5 patients who achieved CR, 3 of them eventually progressed. Median PFS and OS for the entire cohort was 2.30 (90% CI 1.84-3.47) and 4.60 (90% CI 2.84-22.10) months, respectively.To assess whether mitochondrial priming might correlate with clinical response, we first performed standard BH3 profiling with a series of peptides on 16 pretreatment bone marrow samples. Mitochondrial cytochrome c release in response to BIM-BH3 correlated weakly with clinical response. DBP on pretreatment myeloblasts in response to BIM (activator) or PUMA (sensitizer) peptides were able to stratify responders based on an increase in mitochondrial priming caused by ex vivo LEN treatment (P<0.01 and P<0.0005, respectively). Using ROC curve analysis, it was determined that BIM and PUMA peptides served as binary predictors of achievement of CR, with AUC 0.95 and 1.0, respectively.Conclusions: LEN+MEC was tolerable and had modest efficacy in this small study. Correlative studies showed that leukemic blasts briefly treated ex vivo with LEN which then underwent DBP were predictive of in vivo response among a heterogeneous patient population to LEN+MEC treatment. This study is an example of how a powerful biomarker can potentially identify those who are more likely to respond to novel therapies to avoid undesirable toxicity in those who would otherwise have less chance to benefit. Further validation is required to confirm the utility of DBP as a predictive biomarker to targeted agents given with chemotherapy. [Display omitted] DisclosuresGarcia:Celgene: Consultancy. Letai:AbbVie: Consultancy, Other: Lab research report; Vivid Biosciences: Equity Ownership; Flash Therapeutics: Equity Ownership; AstraZeneca: Consultancy, Other: Lab research report; Novartis: Consultancy, Other: Lab research report. Medeiros:Genentech: Employment; Celgene: Consultancy, Research Funding.

1165. Comparing Patient Risk Factors, Sequence Type, and Resistance Loci Identification Approaches for Predicting Antibiotic Resistance in Escherichia coli Bloodstream Infections

BackgroundTo improve the adequacy of empiric antibiotic therapy, an important predictor of clinical outcome, rapid diagnostic tests of antibiotic resistance are increasingly being developed that identify the presence or absence of antibiotic resistance genes/Loci. Few approaches have utilized other sources of predictive information, which could be identified in shorter time periods, including patient epidemiologic risk factors for antibiotic resistance and markers of lineage (e.g., sequence type).MethodsUsing a dataset of 414 Escherichia coli isolated from separate episodes of bacteremia at a single academic institution in Toronto, Canada between 2010 and 2015, we compared the potential predictive ability of three approaches (epidemiologic, sequence type, and gene identification) for classifying antibiotic resistance to three commonly used classes of broad-spectrum antibiotic therapy (third-generation cephalosporins, fluoroquinolones, and aminoglycosides). We used logistic regression models with binary predictor variables to generate model receiver operating characteristic curves. Predictive discrimination was measured using apparent and corrected (bootstrapped) area under the curves (AUCs).ResultsUsing two simple epidemiologic risk factors (prior antibiotic exposure and recent prior Gram-negative susceptibility), modest predictive discrimination was achieved (AUCs 0.65–0.74). Sequence type demonstrated strong discrimination (AUCs 0.84–0.94) across all three antibiotic classes. Epidemiologic risk factors significantly improved sequence-type prediction for cephalosporins and aminoglycosides (P < 0.05). Gene identification approaches provided the highest degree of discrimination (AUCs 0.73–0.99), with no statistically significant benefit of adding epidemiologic predictors.ConclusionRapid identification of sequence type, or other lineage-based classification, could produce excellent discrimination of antibiotic resistance, and may be improved by incorporating readily available epidemiologic predictors. Disclosures All authors: No reported disclosures.

Using the area under an estimated ROC curve to test the adequacy of binary predictors

ABSTRACTWe consider using the area under an empirical receiver operating characteristic curve to test the hypothesis that a predictive index combined with a range of cutoffs performs no better than pure chance in forecasting a binary outcome. This corresponds to the null hypothesis that the area in question, denoted as AUC, is 1/2. We show that if the predictive index comes from a first-stage regression model estimated over the same data set, then testing the null based on the standard asymptotic normality results leads to severe size distortion in general settings. We then analytically derive the proper asymptotic null distribution of the empirical AUC in a special case; namely, when the first-stage regressors are Bernoulli random variables. This distribution can be utilised to construct a fully in-sample test of with correct size and more power than out-of-sample tests based on sample splitting, though practical application becomes cumbersome with more than two regressors.

Heckman imputation models for binary or continuous MNAR outcomes and MAR predictors

BackgroundMultiple imputation by chained equations (MICE) requires specifying a suitable conditional imputation model for each incomplete variable and then iteratively imputes the missing values. In the presence of missing not at random (MNAR) outcomes, valid statistical inference often requires joint models for missing observations and their indicators of missingness. In this study, we derived an imputation model for missing binary data with MNAR mechanism from Heckman’s model using a one-step maximum likelihood estimator. We applied this approach to improve a previously developed approach for MNAR continuous outcomes using Heckman’s model and a two-step estimator. These models allow us to use a MICE process and can thus also handle missing at random (MAR) predictors in the same MICE process.MethodsWe simulated 1000 datasets of 500 cases. We generated the following missing data mechanisms on 30% of the outcomes: MAR mechanism, weak MNAR mechanism, and strong MNAR mechanism. We then resimulated the first three cases and added an additional 30% of MAR data on a predictor, resulting in 50% of complete cases. We evaluated and compared the performance of the developed approach to that of a complete case approach and classical Heckman’s model estimates.ResultsWith MNAR outcomes, only methods using Heckman’s model were unbiased, and with a MAR predictor, the developed imputation approach outperformed all the other approaches.ConclusionsIn the presence of MAR predictors, we proposed a simple approach to address MNAR binary or continuous outcomes under a Heckman assumption in a MICE procedure.

Therapeutic Efficacy of Cranioplasty After Decompressive Craniectomy for Traumatic Brain Injury: A Retrospective Study

Autologous bone removed at the time of decompressive craniectomy (DC) is always the first choice for cranioplasty. The aim of this study was to evaluate the therapeutic efficacy of cranioplasty after DC by measuring the changes in the Functional Independence Measure (FIM) score, as well as to draw a comparison with the pre-cranioplasty FIM score and to evaluate the differences in the outcomes of cases managed by 2 methods: autologous bone (group I) or titanium mesh (group II). We included 47 patients (36 male and 11 female patients) who underwent unilateral cranioplasty after DC for traumatic brain injury at our institute from 2008 to 2017 in this analytical single-institution retrospective study. The primary binary predictor variable was cranioplasty reconstructive material (autologous bone or mesh). The primary outcome variable of interest was increased, decreased, or unchanged FIM score. The secondary outcome variables included evaluation of immediate complications. The Mann-Whitney U test was used to evaluate differences between scores. Group I (n=26) underwent cranioplasty using autologous bone flap, whereas group II (n=21) underwent cranioplasty using dynamic titanium mesh. Increases in FIM scores on the motor function subscale for group I (P=.01278) and group II (P=.00112) were statistically significant. Increases in FIM scores on the cognition subscale for group I (P=.17384) and group II (P=.9492) were statistically insignificant. Evaluation of the primary outcome variable (ie, increased, decreased, or unchanged FIM scores) and secondary outcome variables (ie, immediate complications) showed a statistically insignificant difference between the 2 groups with respect to improvement (P=.51). This study showed that cranioplasty, irrespective of the reconstructive material, after DC in patients with traumatic brain injury results in a significant functional improvement apart from form and esthetics.

Functional Sufficient Dimension Reduction for Functional Data Classification

We consider two novel functional classification methods for binary response and functional predictor. We extend the most popular functional sufficient dimension reduction methods such as functional sliced inverse regression (FSIR) and functional sliced average variance estimation (FSAVE) by introducing a regularized estimation procedure and incorporating the localized information of the functional predictor in the analysis. Compared to the existing FSIR and FSAVE, the proposed methods are appealing because they are capable of estimating more than one effective dimension reduction direction, whereas FSIR detects only one such direction and FSAVE produces inefficient estimation in the case of binary response. Moreover, our methods make use of the localized information of the functional predictor, thereby more efficiently capturing the nonlinear relation between the binary response and the functional predictor. Furthermore, the proposed methods can be extended to incorporate the ancillary unlabeled data in semi-supervised learning. The empirical performance and the applications of the proposed methods are demonstrated by simulation studies and real applications.

Abstract 285: Uncovering the disorder of breast invasive carcinoma homeostasis proteins

Abstract Breast cancer is one of the most common cancers among women worldwide. Recent studies suggest that cancer cell homeostatic stable phenotypes may be implemented, controlled and maintained by Master Regulators (MR) proteins. In essence, all MR proteins are Transcription Factors, which frequently are intrinsically (IDP) or partially disordered proteins, depending on the extent of disordered regions (IDRs). IDPs/IDRs are known to be promiscuous binders that commonly function as central hubs in protein networks. Furthermore, they contain Molecular Recognition Features (MoRFs) and/or Short Linear Motifs (SLiMs), regions that occupy a unique structural and functional niche in which function is a direct consequence of intrinsic disorder. Gathering information on MR protein interactome has significant implications for understanding the molecular basis of the mechanisms responsible for the stability of tumor cell states. In this work, we focused on a subset of 128 MR proteins predicted from RNA-Seq profiles to be responsible for the Breast Invasive Carcinoma (BRCA) tumor homeostatic control across five distinct cancer subtypes and analyzed their disorder prevalence, functional importance, and potential roles in protein association networks. The disorder profile annotation was performed using the MobiDB-lite disorder meta-predictor. The proteins were classified according to the percentage of intrinsically disordered residues (PIDR) as highly ordered (PIDR &amp;lt; 10%), moderately disordered (10% ≤ PIDR &amp;lt; 30%), and highly disordered (PIDR ≥ 30%). We also evaluated target proteins using the CH-CDF plot, which is a combination of two binary disorder predictor methods: the charge-hydropathy plot (CH), that considers charge and hydrophobicity of amino acids, and the cumulative distribution function (CDF) plot, which is based on the distribution of per-residue disorder predictor scores. Disorder functional annotation was performed by mapping MoRFs and SLiMs using experimentally curated databases. The PrePPI database was used to predict interactivity of the BRCA MR proteins by constructing five cancer-specific protein-protein interaction (PPI) networks. Statistical significance of connectivity for the subtype interactomes was measured by the network-based distance method. Annotation of abundant/ housekeeping genes and network topology (hubs and bottlenecks) was introduced to correlate disorder, gene essentiality, and promiscuous binding activity. The result of this work is a comprehensive, functionally annotated inventory of the intrinsically disordered status of BRCA MR proteins. Our work revealed that the master regulator candidates are enriched in moderate (44%) and high (28%) intrinsic disorder and provide the groundwork for experimental studies that may uncover novel pathways associated with the aberrant protein activity that determine stable tumor states and predict IDPs as novel drug targets. Citation Format: Julia V. Castro, Evan O. Paull, Jose I. Garzon, Diana Murray, Andrea Califano, Barry Honig. Uncovering the disorder of breast invasive carcinoma homeostasis proteins [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 285.

Affective problems and decline in cognitive state in older adults: a systematic review and meta-analysis.

Evidence suggests that affective problems, such as depression and anxiety, increase risk for late-life dementia. However, the extent to which affective problems influence cognitive decline, even many years prior to clinical diagnosis of dementia, is not clear. The present study systematically reviews and synthesises the evidence for the association between affective problems and decline in cognitive state (i.e., decline in non-specific cognitive function) in older adults. An electronic search of PubMed, PsycInfo, Cochrane, and ScienceDirect was conducted to identify studies of the association between depression and anxiety separately and decline in cognitive state. Key inclusion criteria were prospective, longitudinal designs with a minimum follow-up period of 1 year. Data extraction and methodological quality assessment using the STROBE checklist were conducted independently by two raters. A total of 34 studies (n = 71 244) met eligibility criteria, with 32 studies measuring depression (n = 68 793), and five measuring anxiety (n = 4698). A multi-level meta-analysis revealed that depression assessed as a binary predictor (OR 1.36, 95% CI 1.05-1.76, p = 0.02) or a continuous predictor (B = -0.008, 95% CI -0.015 to -0.002, p = 0.012; OR 0.992, 95% CI 0.985-0.998) was significantly associated with decline in cognitive state. The number of anxiety studies was insufficient for meta-analysis, and they are described in a narrative review. Results of the present study improve current understanding of the temporal nature of the association between affective problems and decline in cognitive state. They also suggest that cognitive function may need to be monitored closely in individuals with affective disorders, as these individuals may be at particular risk of greater cognitive decline.