Flexible cost-penalized Bayesian model selection: Developing inclusion paths with an application to diagnosis of heart disease.

We propose a Bayesian model selection approach that allows medical practitioners to select among predictor variables while taking their respective costs into account. Medical procedures almost always incur costs in time and/or money. These costs might exceed their usefulness for modeling the outcome of interest. We develop Bayesian model selection that uses flexible model priors to penalize costly predictors a priori and select a subset of predictors useful relative to their costs. Our approach (i) gives the practitioner control over the magnitude of cost penalization, (ii) enables the prior to scale well with sample size, and (iii) enables the creation of our proposed inclusion path visualization, which can be used to make decisions about individual candidate predictors using both probabilistic and visual tools. We demonstrate the effectiveness of our inclusion path approach and the importance of being able to adjust the magnitude of the prior's cost penalization through a dataset pertaining to heart disease diagnosis in patients at the Cleveland Clinic Foundation, where several candidate predictors with various costs were recorded for patients, and through simulated data.

Bayesian model selection for generalized linear mixed models.

We propose a Bayesian model selection approach for generalized linear mixed models (GLMMs). We consider covariance structures for the random effects that are widely used in areas such as longitudinal studies, genome-wide association studies, and spatial statistics. Since the random effects cannot be integrated out of GLMMs analytically, we approximate the integrated likelihood function using a pseudo-likelihood approach. Our Bayesian approach assumes a flat prior for the fixed effects and includes both approximate reference prior and half-Cauchy prior choices for the variances of random effects. Since the flat prior on the fixed effects is improper, we develop a fractional Bayes factor approach to obtain posterior probabilities of the several competing models. Simulation studies with Poisson GLMMs with spatial random effects and overdispersion random effects show that our approach performs favorably when compared to widely used competing Bayesian methods including deviance information criterion and Watanabe-Akaike information criterion. We illustrate the usefulness and flexibility of our approach with three case studies including a Poisson longitudinal model, a Poisson spatial model, and a logistic mixed model. Our proposed approach is implemented in the R package GLMMselect that is available on CRAN.

Longitudinal Validation of a Specific Measure of Fear Avoidance in Athletes: Predicting Time from Injury to Return to Sports Competition.

Purpose: We developed the Athlete Fear Avoidance Questionnaire (AFAQ) to measure fear avoidance in athletes. Previous fear avoidance scales were developed for the general population and have demonstrated significant predictive capabilities regarding rehabilitation. No research to date has examined the association between athlete fear avoidance as measured by the AFAQ and the rehabilitation time in athletes. Patients and Methods: Fifty-nine athletes who were injured during sport season participated in the study (40 males and 19 females). At injury onset, all participants completed self-report functional questionnaires. In addition, we measured multiple aspects of fear avoidance including athlete fear avoidance (AFAQ), kinesiophobia (TSK), and pain catastrophizing (PCS). Finally, we assessed pain severity and interference, as well as depression. Once the athletes were able to return to competition all participants answered the questionnaires again. Pearson correlations and a regression analysis were used to identify relationships between function, psychological variables, pain, and return to competition time. Results: The AFAQ yielded the strongest correlation with return to competition time (r=0.544, p<0.001). In addition, function at initial injury time and pain interference were also significantly correlated with return to competition time (r=0.442, p<0.001 and r=0.356, p=0.006 respectively). Athlete fear-avoidance combined with function at the time of injury explained 34% of the variance of return to competition time in the multivariate regression model (p<0.001). Conclusion: Athlete fear-avoidance as measured by the AFAQ is associated with rehabilitation time and returning to competition in injured athletes. Psychosocial factors including athlete fear avoidance may explain why some athletes take longer to rehabilitate than others and should be evaluated in athletes who are taking longer than anticipated to complete their rehabilitation. Reducing athlete fear avoidance may facilitate rehabilitation in future studies.

Author Correction: Leaf nutrients, not specific leaf area, are consistent indicators of elevated nutrient inputs.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Discrimination learning in oxycodone-treated nonhuman primates.

BACKGROUND: Prescription opioid abuse continues to be a public health concern of epidemic proportions. Notwithstanding the extensive literature regarding opioid action, there has been little systematic research regarding the effects of opioid dependence and withdrawal on aspects of cognition-related behavior in laboratory animals. The present studies examined the effects of the prescription opioid oxycodone on learning processes in nonhuman primates. METHODS: The ability of subjects to repeatedly learn novel touchscreen-based visual discriminations was examined during three conditions of opioid exposure. Discrimination learning was examined, first, during oxycodone self-administration (3-hr sessions, 0.1 mg/kg/injection) and, next, during non-contingent chronic treatment with oxycodone (10 mg/kg/day). Finally, discrimination learning was re-examined during antagonist-precipitated opioid withdrawal (0.001-0.1 mg/kg naltrexone) and, subsequently, following abrupt discontinuation of oxycodone treatment. RESULTS: Although motoric behavior was disrupted by oxycodone, neither the development of discrimination learning nor steady-state performance were impaired following oxycodone self-administration or during non-contingent chronic oxycodone treatment. However, discrimination learning was substantially impaired during oxycodone withdrawal, whether elicited by naltrexone or by abrupt oxycodone discontinuation. Moreover, these learning impairments were concordant with autonomic signs of opioid withdrawal. CONCLUSIONS: Taken together, the present studies indicate that impairment of learning processes can accompany the unconditioned signs of opioid withdrawal.

Leaf nutrients, not specific leaf area, are consistent indicators of elevated nutrient inputs.

Leaf traits are frequently measured in ecology to provide a 'common currency' for predicting how anthropogenic pressures impact ecosystem function. Here, we test whether leaf traits consistently respond to experimental treatments across 27 globally distributed grassland sites across 4 continents. We find that specific leaf area (leaf area per unit mass)-a commonly measured morphological trait inferring shifts between plant growth strategies-did not respond to up to four years of soil nutrient additions. Leaf nitrogen, phosphorus and potassium concentrations increased in response to the addition of each respective soil nutrient. We found few significant changes in leaf traits when vertebrate herbivores were excluded in the short-term. Leaf nitrogen and potassium concentrations were positively correlated with species turnover, suggesting that interspecific trait variation was a significant predictor of leaf nitrogen and potassium, but not of leaf phosphorus concentration. Climatic conditions and pretreatment soil nutrient levels also accounted for significant amounts of variation in the leaf traits measured. Overall, we find that leaf morphological traits, such as specific leaf area, are not appropriate indicators of plant response to anthropogenic perturbations in grasslands.

Compliance with preoperative care measures reduces surgical site infection after colorectal operation.

BACKGROUND: Surgical site infections (SSIs) are a major cause of morbidity complicating colorectal operations. Several evidence-based preoperative strategies are associated with decreased SSI rates. We hypothesize that compliance with multiple strategies is associated with lower incidence of SSI after the elective colorectal operation. METHODS: Preoperative care measure compliance before colorectal operations were assessed. Measures included antiseptic wash the night before and day of operation, oral antibiotic, and mechanical bowel preparation, antibiotic prophylaxis, Chloraprep skin preparation, and hair clipping. Rates of SSI after colectomy and other pertinent outcomes were stratified by full and partial compliance with preoperative measures. Exclusion criteria included bowel perforation, ischemia, complete obstruction, intra-abdominal abscess, and no intraoperative skin closure. RESULTS: Eight hundred twenty-six subjects underwent colectomy between 2010 and 2016; 469 met inclusion criteria. Compliance with all measures occurred in 214 (46%) and was independently associated with lower postoperative infection rates (odds ratio [OR], 0.37; confidence interval [CI], 0.16-0.85; P = 0.02). SSI occurred in 51 (11%): was superficial in 35 (7%); deep in 5 (1%); and organ space in 11 (2%). SSI rates were reduced from 16% (partial or no compliance group) to 5% (full compliance group). No stand-alone intervention was independently associated with decreased SSI rate. Multivariate analysis found the following factors associated with a lower risk of SSI: full compliance with all five process measures, lower BMI, nonsmoker, and minimally invasive operation. DISCUSSION AND CONCLUSION: Compliance with preoperative care strategies reduces rates of SSI after colectomy with a cumulative effect more pronounced than any single intervention reinforcing the need for protocol-driven and evidence-based care for patients undergoing colorectal operations.

Identification of patients at high risk for post-discharge venous thromboembolism after hepato-pancreato-biliary surgery: which patients benefit from extended thromboprophylaxis?

BACKGROUND: The objective of the current study was to define risk factors associated with the 30-day post-operative risk of VTE after HPB surgery and create a model to identify patients at highest risk of post-discharge VTE. METHODS: Patients who underwent hepatectomy or pancreatectomy in the ACS-NSQIP Participant Use Files 2011-2015 were identified. Logistic regression modeling was used; a model to predict post-discharge VTE was developed. Model discrimination was tested using area under the curve (AUC). RESULTS: Among 48,860 patients, the overall 30-day incidence of VTE after hepatectomy and pancreatectomy was 3.2% (n = 1580) with 1.1% (n = 543) of VTE events occurring after discharge. Patients who developed post-discharge VTE were more likely to be white, had a higher median BMI, have undergone pancreatic surgery, had longer median operative times, and to have had a transfusion. A weighted prediction model demonstrated good calibration and fair discrimination (AUC = 0.63). A score of ≥-4.50 had maximum sensitivity and specificity, resulting in 44% of patients being treating with prophylaxis for an overall VTE risk of 1.1%. CONCLUSIONS: Utilizing independent factors associated with post-discharge VTE, a prediction model was able to stratify patients according to risk of VTE and may help identify patients who are most likely to benefit from pharmacoprophylaxis.

Which Patients Require Extended Thromboprophylaxis After Colectomy? Modeling Risk and Assessing Indications for Post-discharge Pharmacoprophylaxis.

BACKGROUND: Given the conflicting nature of reported risk factors for post-discharge venous thromboembolism (VTE) and unclear guidelines for post-discharge pharmacoprophylaxis, we sought to determine risk factors for 30-day post-discharge VTE after colectomy to predict which patients will benefit from post-discharge pharmacoprophylaxis. METHODS: Patients who underwent colectomy in the American College of Surgeons National Surgical Quality Improvement Project Participant Use Files from 2011 to 2015 were identified. Logistic regression modeling was used. Receiver-operating characteristic curves were used and the best cut-points were determined using Youden's J index (sensitivity + specificity - 1). Hosmer-Lemeshow goodness-of-fit test was used to test model calibration. A random sample of 30% of the cohort was used as a validation set. RESULTS: Among 77,823 cases, the overall incidence of VTE after colectomy was 1.9%, with 0.7% of VTE events occurring in the post-discharge setting. Factors associated with post-discharge VTE risk including body mass index, preoperative albumin, operation time, hospital length of stay, race, smoking status, inflammatory bowel disease, return to the operating room and postoperative ileus were included in logistic regression equation model. The model demonstrated good calibration (goodness of fit P = 0.7137) and good discrimination (area under the curve (AUC) = 0.68; validation set, AUC = 0.70). A score of ≥-5.00 had the maxim sensitivity and specificity, resulting in 36.63% of patients being treated with prophylaxis for an overall VTE risk of 0.67%. CONCLUSION: Approximately one-third of post-colectomy VTE events occurred after discharge. Patients with predicted post-discharge VTE risk of ≥-5.00 should be recommended for extended post-discharge VTE prophylaxis.