Prognostic modelling with logistic regression analysis: a comparison of selection and estimation methods in small data sets

Developing Risk Prediction Models for Postoperative Pancreatic Fistula: a Systematic Review of Methodology and Reporting Quality

Postoperative pancreatic fistula is still a major complication after pancreatic surgery, despite improvements of surgical technique and perioperative management. We sought to systematically review and critically access the conduct and reporting of methods used to develop risk prediction models for predicting postoperative pancreatic fistula. We conducted a systematic search of PubMed and EMBASE databases to identify articles published before January 1, 2015, which described the development of models to predict the risk of postoperative pancreatic fistula. We extracted information of developing a prediction model including study design, sample size and number of events, definition of postoperative pancreatic fistula, risk predictor selection, missing data, model-building strategies, and model performance. Seven studies of developing seven risk prediction models were included. In three studies (42 %), the number of events per variable was less than 10. The number of candidate risk predictors ranged from 9 to 32. Five studies (71 %) reported using univariate screening, which was not recommended in building a multivariate model, to reduce the number of risk predictors. Six risk prediction models (86 %) were developed by categorizing all continuous risk predictors. The treatment and handling of missing data were not mentioned in all studies. We found use of inappropriate methods that could endanger the development of model, including univariate pre-screening of variables, categorization of continuous risk predictors, and model validation. The use of inappropriate methods affects the reliability and the accuracy of the probability estimates of predicting postoperative pancreatic fistula.

Review and evaluation of penalised regression methods for risk prediction in low-dimensional data with few events

Risk prediction models are used to predict a clinical outcome for patients using a set of predictors. We focus on predicting low-dimensional binary outcomes typically arising in epidemiology, health services and public health research where logistic regression is commonly used. When the number of events is small compared with the number of regression coefficients, model overfitting can be a serious problem. An overfitted model tends to demonstrate poor predictive accuracy when applied to new data. We review frequentist and Bayesian shrinkage methods that may alleviate overfitting by shrinking the regression coefficients towards zero (some methods can also provide more parsimonious models by omitting some predictors). We evaluated their predictive performance in comparison with maximum likelihood estimation using real and simulated data. The simulation study showed that maximum likelihood estimation tends to produce overfitted models with poor predictive performance in scenarios with few events, and penalised methods can offer improvement. Ridge regression performed well, except in scenarios with many noise predictors. Lasso performed better than ridge in scenarios with many noise predictors and worse in the presence of correlated predictors. Elastic net, a hybrid of the two, performed well in all scenarios. Adaptive lasso and smoothly clipped absolute deviation performed best in scenarios with many noise predictors; in other scenarios, their performance was inferior to that of ridge and lasso. Bayesian approaches performed well when the hyperparameters for the priors were chosen carefully. Their use may aid variable selection, and they can be easily extended to clustered-data settings and to incorporate external information.

Modern modeling techniques had limited external validity in predicting mortality from traumatic brain injury

BACKGROUND AND OBJECTIVE

Prediction of medical outcomes may potentially benefit from using modern statistical modeling techniques. We aimed to externally validate modeling strategies for prediction of 6-month mortality of patients suffering from traumatic brain injury (TBI) with predictor sets of increasing complexity.

METHODS

We analyzed individual patient data from 15 different studies including 11,026 TBI patients. We consecutively considered a core set of predictors (age, motor score, and pupillary reactivity), an extended set with computed tomography scan characteristics, and a further extension with two laboratory measurements (glucose and hemoglobin). With each of these sets, we predicted 6-month mortality using default settings with five statistical modeling techniques: logistic regression (LR), classification and regression trees, random forests (RFs), support vector machines (SVM) and neural nets. For external validation, a model developed on one of the 15 data sets was applied to each of the 14 remaining sets. This process was repeated 15 times for a total of 630 validations. The area under the receiver operating characteristic curve (AUC) was used to assess the discriminative ability of the models.

RESULTS

For the most complex predictor set, the LR models performed best (median validated AUC value, 0.757), followed by RF and support vector machine models (median validated AUC value, 0.735 and 0.732, respectively). With each predictor set, the classification and regression trees models showed poor performance (median validated AUC value, <0.7). The variability in performance across the studies was smallest for the RF- and LR-based models (inter quartile range for validated AUC values from 0.07 to 0.10).

CONCLUSION

In the area of predicting mortality from TBI, nonlinear and nonadditive effects are not pronounced enough to make modern prediction methods beneficial.

High-Accuracy Detection of Early Parkinson's Disease through Multimodal Features and Machine Learning

Early (or preclinical) diagnosis of Parkinson's disease (PD) is crucial for its early management as by the time manifestation of clinical symptoms occur, more than 60% of the dopaminergic neurons have already been lost. It is now established that there exists a premotor stage, before the start of these classic motor symptoms, characterized by a constellation of clinical features, mostly non-motor in nature such as Rapid Eye Movement (REM) sleep Behaviour Disorder (RBD) and olfactory loss. In this paper, we use the non-motor features of RBD and olfactory loss, along with other significant biomarkers such as Cerebrospinal fluid (CSF) measurements and dopaminergic imaging markers from 183 healthy normal and 401 early PD subjects, as obtained from the Parkinson's Progression Markers Initiative (PPMI) database, to classify early PD subjects from normal using Naïve Bayes, Support Vector Machine (SVM), Boosted Trees and Random Forests classifiers. We observe that SVM classifier gave the best performance (96.40% accuracy, 97.03% sensitivity, 95.01% specificity, and 98.88% area under ROC). We infer from the study that a combination of non-motor, CSF and imaging markers may aid in the preclinical diagnosis of PD.

Diagnosis of TIA (DOT) score--design and validation of a new clinical diagnostic tool for transient ischaemic attack

BACKGROUND

The diagnosis of Transient Ischaemic Attack (TIA) can be difficult and 50-60% of patients seen in TIA clinics turn out to be mimics. Many of these mimics have high ABCD2 scores and fill urgent TIA clinic slots inappropriately. A TIA diagnostic tool may help non-specialists make the diagnosis with greater accuracy and improve TIA clinic triage. The only available diagnostic score (Dawson et al) is limited in scope and not widely used. The Diagnosis of TIA (DOT) Score is a new and internally validated web and mobile app based diagnostic tool which encompasses both brain and retinal TIA.

METHODS

The score was derived retrospectively from a single centre TIA clinic database using stepwise logistic regression by backwards elimination to find the best model. An optimum cutpoint was obtained for the score. The derivation and validation cohorts were separate samples drawn from the years 2010/12 and 2013 respectively. Receiver Operating Characteristic (ROC) curves and area under the curve (AUC) were calculated and the diagnostic accuracy of DOT was compared to the Dawson score. A web and smartphone calculator were designed subsequently.

RESULTS

The derivation cohort had 879 patients and the validation cohort 525. The final model had seventeen predictors and had an AUC of 0.91 (95% CI: 0.89-0.93). When tested on the validation cohort, the AUC for DOTS was 0.89 (0.86-0.92) while that of the Dawson score was 0.77 (0.73-0.81). The sensitivity and specificity of the DOT score were 89% (CI: 84%-93%) and 76% (70%-81%) respectively while those of the Dawson score were 83% (78%-88%) and 51% (45%-57%). Other diagnostic accuracy measures (DOT vs. Dawson) include positive predictive values (75% vs. 58%), negative predictive values (89% vs. 79%), positive likelihood ratios (3.67 vs. 1.70) and negative likelihood ratios (0.15 vs. 0.32).

CONCLUSION

The DOT score shows promise as a diagnostic tool for TIA and requires independent external validation before it can be widely used. It could potentially improve the triage of patients assessed for suspected TIA.

SYMPTOM AND FUNCTIONAL TRAITS OF BRIEF MAJOR DEPRESSIVE EPISODES AND DISCRIMINATION OF BEREAVEMENT

BACKGROUND

Despite the removal of the bereavement exclusion from DSM-5, clinicians may feel uncertain on how to proceed when caring for a patient who presents with depressive symptoms following the death of someone close. The ability to better distinguish, on a symptom and functional level, between patients who experience depression in the context of bereavement and those with nonbereavement-related depression, could help guide clinical decision making.

METHOD

Individual and clustered depressive symptom and impairment measures were used for modeling bereavement status within a nationally representative longitudinal cohort. Deviance, linear shrinkage factor, and bias-corrected c-statistic were used for identifying a well-calibrated and discriminating final model.

RESULTS

Of the 450 (1.2%) respondents with a single brief major depressive episode, 162 (38.4%) reported the episode as bereavement-related. The bereaved were less likely to endorse worthlessness (P < .001), social conflict (P < .001), distress (P < .001), thoughts of suicide (P = .001), wanting to die (P = .01), self-medicating (P = .01), and being withdrawn (P = .04). In a multivariate model, the bereaved were more likely to have thoughts of their own death (P = .003), guilt coupled with weight or appetite loss (P = .013), and were less likely to report social conflict (P < .001), worthlessness coupled with difficulty making decisions (P < .001), thoughts of suicide (P = .006), distress coupled with weight or appetite gain (P = .022), and self-medicating (P = .045).

CONCLUSIONS

Traits and trait combinations differentiate individuals who experience brief depressive episodes following the death of a loved one from other brief episodes. These differences can help guide clinical care of patients who present with depressive symptoms shortly after a loved one's death.

BMI, HOMA-IR, and Fasting Blood Glucose Are Significant Predictors of Peripheral Nerve Dysfunction in Adult Overweight and Obese Nondiabetic Nepalese Individuals: A Study from Central Nepal

Objective. Nondiabetic obese individuals have subclinical involvement of peripheral nerves. We report the factors predicting peripheral nerve function in overweight and obese nondiabetic Nepalese individuals. Methodology. In this cross-sectional study, we included 50 adult overweight and obese nondiabetic volunteers without features of peripheral neuropathy and 50 healthy volunteers to determine the normative nerve conduction data. In cases of abnormal function, the study population was classified on the basis of the number of nerves involved, namely, “<2” or “≥2.” Multivariable logistic regression analysis was carried out to predict outcomes. Results. Fasting blood glucose (FBG) was the significant predictor of motor nerve dysfunction (P = 0.039, 95% confidence interval (CI) = 1.003–1.127). Homeostatic model assessment of insulin resistance (HOMA-IR) was the significant predictor (P = 0.019, 96% CI = 1.420–49.322) of sensory nerve dysfunction. Body mass index (BMI) was the significant predictor (P = 0.034, 95% CI = 1.018–1.577) in case of ≥2 mixed nerves' involvement. Conclusion. FBG, HOMA-IR, and BMI were significant predictors of peripheral nerve dysfunction in overweight and obese Nepalese individuals.

First results about recovery of walking function in patients with intensive care unit-acquired muscle weakness from the General Weakness Syndrome Therapy (GymNAST) cohort study

OBJECTIVES

To describe the time course of recovery of walking function and other activities of daily living in patients with intensive care unit (ICU)-acquired muscle weakness.

DESIGN

This is a cohort study.

PARTICIPANTS

We included critically ill patients with ICU-acquired muscle weakness.

SETTING

Post-acute ICU and rehabilitation units in Germany.

MEASURES

We measured walking function, muscle strength, activities in daily living, motor and cognitive function.

RESULTS

We recruited 150 patients (30% female) who fulfilled our inclusion and exclusion criteria. The primary outcome recovery of walking function was achieved after a median of 28.5 days (IQR=45) after rehabilitation onset and after a median of 81.5 days (IQR=64) after onset of illness. Our final multivariate model for recovery of walking function included two clinical variables from baseline: the Functional Status Score ICU (adjusted HR=1.07 (95% CI 1.03 to 1.12) and the ability to reach forward in cm (adjusted HR=1.02 (95% CI 1.00 to 1.04). All secondary outcomes but not pain improved significantly in the first 8 weeks after study onset.

CONCLUSIONS

We found good recovery of walking function for most patients and described the recovery of walking function of people with ICU-acquired muscle weakness.

TRIALS REGISTRATIONS NUMBER

Sächsische Landesärztekammer EK-BR-32/13-1; DRKS00007181, German Register of Clinical Trials.

Race not rural residency is predictive of surgical treatment for hepatocellular carcinoma: Analysis of the Texas Cancer Registry

BACKGROUND AND OBJECTIVES

Rural patients have poor access to specialists and are less likely to receive evidence-based cancer care. We hypothesized that hepatocellular carcinoma (HCC) patients from rural counties in Texas would be less likely to receive surgical therapy than those from urban areas.

METHODS

The Texas Cancer Registry was queried (2000-2008). County-level data included "rural or urban" designation and income variables derived by zip code. Surgical intervention included: (i) ablation, (ii) resection-partial or total lobectomy, or (iii) transplantation. A multinomial logistic regression was created to determine predictors of intervention.

RESULTS

Five thousand thirty seven HCC patients were identified (86% urban) for study. A multinomial regression demonstrated, older age, African-American race, and lower income reduced the likelihood of ablation. Younger age, female gender, Caucasian, and Asian/other race predicted surgical resection, or transplantation. Hispanic race was associated with lower likelihood of resection (RRR 0.75) and transplantation (RRR 0.74), whereas African-American race was associated with pronounced decrease for transplantation (RRR 0.48). Area of residency was not predictive of intervention.

CONCLUSIONS

Rural residency did not decrease the likelihood of surgical intervention for hepatocellular carcinoma. Race and income continue to be associated with significant treatment disparity. Additional investigation should focus on factors that govern the selection of resection or transplantation for potentially eligible patients.

Health care resource use and costs of two-year survivors of acute lung injury. An observational cohort study

RATIONALE

Survivors of acute lung injury (ALI) require ongoing health care resources after hospital discharge. The extent of such resource use, and associated costs, are not fully understood.

OBJECTIVES

For patients surviving at least 2 years after ALI, we evaluated cumulative 2-year inpatient admissions and related costs, and the association of patient- and intensive care unit-related exposures with these costs.

METHODS

Multisite observational cohort study in 13 intensive care units at four academic teaching hospitals evaluating 138 two-year survivors of ALI.

MEASUREMENTS AND MAIN RESULTS

Two-year inpatient health care use data (i.e., admissions to hospitals, and skilled nursing and rehabilitation facilities) were collected for patients surviving at least 2 years, via (1) one-time retrospective structured interview with patient and/or proxy, (2) systematic medical record review for nonfederal study site hospitals, and (3) inpatient medical record review for non-study site hospitals, as needed for clarifying patient/proxy reports. Costs are reported in 2013 U.S. dollars. A total of 138 of 142 (97%) 2-year survivors completed the interview, with 111 (80%) reporting at least one inpatient admission during follow-up, for median (interquartile range [IQR]) estimated costs of $35,259 ($10,565-$81,166). Hospital readmissions accounted for 76% of costs. Among 12 patient- and intensive care unit-related exposures evaluated, baseline comorbidity and intensive care unit length of stay were associated with increased odds of incurring any follow-up inpatient costs. Having Medicare or Medicaid (vs. private insurance) was associated with median estimated costs that were 85% higher (relative median, 1.85; 95% confidence interval, 1.01-3.45; P=0.045).

CONCLUSIONS

In this multisite study of 138 two-year survivors of ALI, 80% had one or more inpatient admission, representing a median (IQR) estimated cost $35,259 ($10,565-$81,166) per patient and $6,598,766 for the entire cohort. Hospital readmissions represented 76% of total inpatient costs, and having Medicare or Medicaid before ALI was associated with increased costs. With the aging population and increasing comorbidity, these findings have important health policy implications for the care of critically ill patients.

Estimating the risk of bowel ischemia requiring surgery in patients with tomographic evidence of pneumatosis intestinalis

BACKGROUND

Pneumatosis intestinalis (PI) presents a challenging dilemma for surgeons given its association with both benign and life threatening conditions. As such, the need for surgical intervention is oftentimes difficult to discern. We hypothesize that a clinical nomogram can be used to predict the need for surgical intervention in patients with PI.

METHODS

We performed a retrospective review of 217 consecutive cases with PI on abdominal computed tomography over a 10-year period at a tertiary care hospital. Bivariable and multivariable analysis were conducted to assess the statistical significance of the association between patient factors and need for surgical intervention, defined as positive findings at surgery.

RESULTS

There were 217 patients with PI identified during the study, of which 178 were treated with curative intent. Of these, 82 patients underwent surgical exploration, and 96 patients were managed conservatively. Forty-four percent of patients who had radiographic evidence of PI were managed conservatively and did well, whereas an additional 6% underwent nontherapeutic laparotomies. Multivariable analysis demonstrated that patients with tenderness on examination, lactic acidosis, and tachycardia had significantly higher likelihood of the need for surgical intervention, whereas patients with diabetes had a lower likelihood of surgical intervention. These and other selected patient characteristics can be used to efficiently and reliably estimate the probability of ischemic bowel at laparotomy.

CONCLUSIONS

The presence of PI does not always warrant surgical intervention. We present a nomogram to assist with clinical decision-making based on the presence of clinical factors.

Application of Penalized Regression Techniques in Modelling Insulin Sensitivity by Correlated Metabolic Parameters

This paper aims to introduce penalized estimation techniques in clinical investigations of diabetes, as well as to assess their possible advantages and limitations. Data from a previous study was used to carry out the simulations to assess: a) which procedure results in the lowest prediction error of the final model in the setting of a large number of predictor variables with high multicollinearity (of importance if insulin sensitivity should be predicted) and b) which procedure achieves the most accurate estimate of regression coefficients in the setting of fewer predictors with small unidirectional effects and moderate correlation between explanatory variables (of importance if the specific relation between an independent variable and insulin sensitivity should be examined). Moreover a special focus is on the correct direction of estimated parameter effects, a non-negligible source of error and misinterpretation of study results. The simulations were performed for varying sample size to evaluate the performance of LASSO, Ridge as well as different algorithms for Elastic Net. These methods were also compared with automatic variable selection procedures (i.e. optimizing AIC or BIC).We were not able to identify one method achieving superior performance in all situations. However, the improved accuracy of estimated effects underlines the importance of using penalized regression techniques in our example (e.g. if a researcher aims to compare relations of several correlated parameters with insulin sensitivity). However, the decision which procedure should be used depends on the specific context of a study (accuracy versus complexity) and moreover should involve clinical prior knowledge.

A Comparison of Two Strategies for Building an Exposure Prediction Model

Cost-efficient assessments of job exposures in large populations may be obtained from models in which 'true' exposures assessed by expensive measurement methods are estimated from easily accessible and cheap predictors. Typically, the models are built on the basis of a validation study comprising 'true' exposure data as well as an extensive collection of candidate predictors from questionnaires or company data, which cannot all be included in the models due to restrictions in the degrees of freedom available for modeling. In these situations, predictors need to be selected using procedures that can identify the best possible subset of predictors among the candidates. The present study compares two strategies for selecting a set of predictor variables. One strategy relies on stepwise hypothesis testing of associations between predictors and exposure, while the other uses cluster analysis to reduce the number of predictors without relying on empirical information about the measured exposure. Both strategies were applied to the same dataset on biomechanical exposure and candidate predictors among computer users, and they were compared in terms of identified predictors of exposure as well as the resulting model fit using bootstrapped resamples of the original data. The identified predictors were, to a large part, different between the two strategies, and the initial model fit was better for the stepwise testing strategy than for the clustering approach. Internal validation of the models using bootstrap resampling with fixed predictors revealed an equally reduced model fit in resampled datasets for both strategies. However, when predictor selection was incorporated in the validation procedure for the stepwise testing strategy, the model fit was reduced to the extent that both strategies showed similar model fit. Thus, the two strategies would both be expected to perform poorly with respect to predicting biomechanical exposure in other samples of computer users.

Identification of Emergency Department Patients With Acute Heart Failure at Low Risk for 30-Day Adverse Events: The STRATIFY Decision Tool

OBJECTIVES

No prospectively derived or validated decision tools identify emergency department (ED) patients with acute heart failure (AHF) at low risk for 30-day adverse events who are thus potential candidates for safe ED discharge. This study sought to accomplish that goal.

BACKGROUND

The nearly 1 million annual ED visits for AHF are associated with high proportions of admissions and consume significant resources.

METHODS

We prospectively enrolled 1,033 patients diagnosed with AHF in the ED from 4 hospitals between July 20, 2007, and February 4, 2011. We used an ordinal outcome hierarchy, defined as the incidence of the most severe adverse event within 30 days of ED evaluation (acute coronary syndrome, coronary revascularization, emergent dialysis, intubation, mechanical cardiac support, cardiopulmonary resuscitation, and death).

RESULTS

Of 1,033 patients enrolled, 126 (12%) experienced at least one 30-day adverse event. The decision tool had a C statistic of 0.68 (95% confidence interval: 0.63 to 0.74). Elevated troponin (p < 0.001) and renal function (p = 0.01) were significant predictors of adverse events in our multivariable model, whereas B-type natriuretic peptide (p = 0.09), tachypnea (p = 0.09), and patients undergoing dialysis (p = 0.07) trended toward significance. At risk thresholds of 1%, 3%, and 5%, we found 0%, 1.4%, and 13.0% patients were at low risk, with negative predictive values of 100%, 96%, and 93%, respectively.

CONCLUSIONS

The STRATIFY decision tool identifies ED patients with AHF who are at low risk for 30-day adverse events and may be candidates for safe ED discharge. After external testing, and perhaps when used as part of a shared decision-making strategy, it may significantly affect disposition strategies. (Improving Heart Failure Risk Stratification in the ED [STRATIFY]; NCT00508638).

Hospitalized heart failure patients with preserved vs. reduced ejection fraction in Dubai, United Arab Emirates: a prospective study

AIMS

To compare the baseline characteristics, pharmacological treatment, and in-hospital outcomes across hospitalized heart failure (HF) patients with preserved LVEF (HF-PEF) and those with reduced LVEF (HF-REF).

METHOD AND RESULTS

This was a prospective analysis of consecutive patients admitted with decompensated HF at two government hospitals in the United Arab Emirates, from 1 December 2011 to 30 November 2012. Multivariate factors of HF-PEF vs. HF-REF included elevated systolic blood pressure [odds ratio (OR) 1.02; 95% confidence interval (CI) 1.01–1.03], heart rate (OR 0.98; 95% CI 0.97–0.99), age (OR 1.02; 95% CI 1.01–1.04), female sex (OR 2.38; 95% CI 1.41–4.03), angina or myocardial infarction (OR 0.42; 95% CI 0.25–0.71), AF (OR 1.82; 95% CI 1.05–3.15), COPD or asthma (OR 2.80; 95% CI 1.47–5.35), Charlson Comorbidity Index score (OR 0.75; 95% CI 0.64–0.88), and anaemia (OR 2.97; 95% CI 1.64–5.38). In-hospital outcomes were similar between the two groups. However, patients with HF-PEF were less likely to be prescribed HF medication, and used more anticoagulants and fewer antiplatelet medications.

CONCLUSION

These results suggest that patients with HF-PEF are older, more often female, and have higher prevalence of respiratory diseases and AF. Compared with developed countries, hospitalized HF patients in the Middle East are 10 years younger and have a higher prevalence of diabetes mellitus, and the majority have HF-REF.

Predictive Factors for Type II Endoleaks after Treatment of Abdominal Aortic Aneurysm by Conventional Endovascular Aneurysm Repair

BACKGROUND

The aim of this study was to identify the predictive factors for the development of type II endoleaks (EL-II) after endovascular aneurysm repair (EVAR).

METHODS

We assessed the preoperative and postoperative computed tomography data of 308 patients who underwent EVAR between 2000 and 2012 and in 84 of whom primary or secondary EL-II occurred. The data analyzed were: demographics, number and diameter of lumbar arteries (LAs), inferior mesenteric artery (IMA), median sacral artery (MSA), accessory renal arteries (ARas), maximum diameter of infrarenal abdominal aortic aneurysm, diameter and length of proximal aortic neck. Statistical analysis was performed using Stata software (version 12). Categorical parameters were compared between groups using chi-squared or Fisher's exact tests as appropriate. Continuous variables were analyzed using Student's t-test or Mann-Whitney test as appropriate (normality studied by the Shapiro-Wilk and homoscedasticity verified using the Fisher-Snedecor test).

RESULTS

Of the 308 patients included (mean age, 73.8 ± 8.74 years), 284 (92%) were men, 61 (20%) were smokers, 113 (37%) had chronic obstructive pulmonary disease, 215 (70%) were taking antiplatelet. Respectively, 13, 51, 60, 103, 28, 40, 2, and 7 patients had 1, 2, 3, 4, 5, 6, 7, and 8 patent LAs. Before surgery, 221 IMAs and 136 MSA were patent. The sources of EL-II were: LA (n = 51), IMA (n = 22), MSA (n = 1), IMA and LA (n = 8), IMA and ARa (n = 1), and unknown (n = 1). Logistic regression models adjusting for clinically relevant covariables (age, American Society of Anesthesiologists, smoking status, dyslipidemia, and diuretics) were proposed to study morphologic EL-II predictive factors, first in the entire population, and then in the more specific population for whom IMA was patent. Risk factors of occurrence EL-II were: permeability of the IMA (70 patients [83%] vs. 155 [69%], P = 0.01), IMA diameter (3.49 mm vs. 2.71 mm, P < 0.001), number of LAs patent higher than or equal to 4 (P < 0.001), the mean LA diameter greater than 2.4 mm (P < 0.001), and MSA diameter (2.28 mm vs. 1.94 mm; P < 0.01).

CONCLUSIONS

Our results show the major role of the number and diameter of the patent aortic branches in the development of EL-II. As they can result in complications increasing the morbidity and mortality after EVAR, it is relevant to identify the risk factors of their occurrence.

Sick Leave within 5 Years of Whiplash Trauma Predicts Recovery: A Prospective Cohort and Register-Based Study

BACKGROUND

10-22% of individuals sustaining whiplash trauma develop persistent symptoms resulting in reduced working ability and decreased quality of life, but it is poorly understood why some people do not recover. Various collision and post-collision risk factors have been studied, but little is known about pre-collision risk factors. In particular, the impact of sickness and socioeconomic factors before the collision on recovery is sparsely explored. The aim of this study was to examine if welfare payments received within five years pre-collision predict neck pain and negative change in provisional situation one year post-collision.

METHODS AND FINDINGS

719 individuals with acute whiplash trauma consecutively recruited from emergency departments or primary care after car accidents in Denmark completed questionnaires on socio-demographic and health factors immediately after the collision. After 12 months, a visual analogue scale on neck pain intensity was completed. 3595 matched controls in the general population were sampled, and national public register data on social benefits and any other welfare payments were obtained for participants with acute whiplash trauma and controls from five years pre-collision to 15 months after. Participants with acute whiplash trauma who had received sickness benefit for more than 12 weeks pre-collision had increased odds for negative change in future provisional situation (Odds Ratio (OR) (95% Confidence Interval (CI) = 3.8 (2.1;7.1)) and future neck pain (OR (95%CI) = 3.3 (1.8;6.3)), controlling for other known risk factors. Participants with acute whiplash trauma had weaker attachment to labour market (more weeks of sick leave (χ2(2) = 36.7, p < 0.001) and unemployment (χ2(2) = 12.5, p = 0.002)) pre-collision compared with controls. Experiencing a whiplash trauma raised the odds for future negative change in provisional situation (OR (95%CI) = 3.1 (2.3;4.4)) compared with controls.

CONCLUSIONS

Sick leave before the collision strongly predicted prolonged recovery following whiplash trauma. Participants with acute whiplash trauma had weaker attachment to labour market pre-collision compared with the general population. Neck pain at inclusion predicted future neck pain. Acute whiplash trauma may trigger pre-existing vulnerabilities increasing risk of developing whiplash-associated disorders.

Incidence of and risk factors for severe cardiovascular collapse after endotracheal intubation in the ICU: a multicenter observational study

Introduction

Severe cardiovascular collapse (CVC) is a life-threatening complication after emergency endotracheal intubation (ETI) in the ICU. Many factors may interact with hemodynamic conditions during ETI, but no study to date has focused on factors associated with severe CVC occurrence. This study assessed the incidence of severe CVC after ETI in the ICU and analyzed the factors predictive of severe CVC.

Methods

This was a secondary analysis of a prospective multicenter study of 1,400 consecutive intubations at 42 ICUs. The incidence of severe CVC was assessed in patients who were hemodynamically stable (mean arterial blood pressure >65 mmHg without vasoactive drugs) before intubation, and the factors predictive of severe CVC were determined by multivariate analysis based on patient and procedure characteristics.

Results

Severe CVC occurred following 264 of 885 (29.8 %) intubation procedures. A two-step multivariate analysis showed that independent risk factors for CVC included simple acute physiologic score II regardless of age (odds ratio (OR) 1.02, p < 0.001), age 60–75 years (OR 1.96, p < 0.002 versus <60 years) and >75 years (OR 2.81, p < 0.001 versus <60 years), acute respiratory failure as a reason for intubation (OR 1.51, p = 0.04), first intubation in the ICU (OR 1.61, p = 0.02), noninvasive ventilation as a preoxygenation method (OR 1.54, p = 0.03) and inspired oxygen concentration >70 % after intubation (OR 1.91, p = 0.001). Comatose patients who required ETI were less likely to develop CVC during intubation (OR 0.48, p = 0.004).

Conclusions

CVC is a frequent complication, especially in old and severely ill patients intubated for acute respiratory failure in the ICU. Specific bundles to prevent CVC may reduce morbidity and mortality related to intubation of these high-risk, critically ill patients.

Trial registration

clinicaltrials.gov NCT01532063; registered 8 February 2012.

Statistical-based approach in potential diagnostic application of urinary nucleosides in urogenital tract cancer

Aim: We aimed at evaluation the potential diagnostic role of urinary nucleosides in urogenital tract cancer. Materials & methods: Concentrations of 12 nucleosides determined by LC-MS/MS were subjected to correlation, association and interaction analyses. Results: We identified six pairs of nucleosides differently correlated in the group of patients and controls (p < 0.05). N-2-methylguanosine (odds ratio: 4.82; 95% CI: 1.78–12.93; p = 0.002) and N,N-dimethylguanosine (odds ratio: 5.45; 95% CI: 1.78–16.44; p = 0.003), were significantly associated with the disease risk (p-corrected = 0.004). Interaction between N-2-methylguanosine and adenosine (p-interaction = 0.019) suggested their multiplicative effect on the outcome. Conclusion: Urinary nucleosides, namely N,N-dimethylguanosine and N-2-methylguanosine may have the potential to serve as prognostic biomarkers. Gender-specific differences in urogenital tract cancer are likely to occur.

Risk Score for Predicting Mortality in Flail Chest

BACKGROUND

Flail chest injuries are associated with high mortality and morbidity. Despite evidence that operative repair of flail chest is beneficial, it is rarely done. We sought to create a simple risk score using available preoperative covariates to calculate individual risk of mortality in flail chest.

METHODS

A logistic regression model was trained on Ontario Trauma Registry data to generate a mortality risk score. The final model was validated for calibration and discrimination and corrected for optimism.

RESULTS

The model uses five risk factors that are readily obtained during the initial assessment of the trauma patient: age, Glasgow Coma Score, ventilation, cardiopulmonary resuscitation, and number of comorbidities. It was determined that less than 6 points is consistent with 1% observed mortality, 6 to 10 points predicts 5% mortality, 11 to 15 points predicts 22% mortality, and 16 or more points predicts 46% mortality.

CONCLUSIONS

We have developed a simple model that can be easily applied at bedside to predict mortality in patients with flail chest by accessing a spreadsheet program in an application or other handheld computer device. This model has the potential to be a useful tool for surgeons considering operative repair of flail chest.

Mapping Functions in Health-Related Quality of Life: Mapping from Two Cancer-Specific Health-Related Quality-of-Life Instruments to EQ-5D-3L

Background. Clinical trials in cancer frequently include cancer-specific measures of health but not preference-based measures such as the EQ-5D that are suitable for economic evaluation. Mapping functions have been developed to predict EQ-5D values from these measures, but there is considerable uncertainty about the most appropriate model to use, and many existing models are poor at predicting EQ-5D values. This study aims to investigate a range of potential models to develop mapping functions from 2 widely used cancer-specific measures (FACT-G and EORTC-QLQ-C30) and to identify the best model. Methods. Mapping models are fitted to predict EQ-5D-3L values using ordinary least squares (OLS), tobit, 2-part models, splining, and to EQ-5D item-level responses using response mapping from the FACT-G and QLQ-C30. A variety of model specifications are estimated. Model performance and predictive ability are compared. Analysis is based on 530 patients with various cancers for the FACT-G and 771 patients with multiple myeloma, breast cancer, and lung cancer for the QLQ-C30. Results. For FACT-G, OLS models most accurately predict mean EQ-5D values with the best predicting model using FACT-G items with similar results using tobit. Response mapping has low predictive ability. In contrast, for the QLQ-C30, response mapping has the most accurate predictions using QLQ-C30 dimensions. The QLQ-C30 has better predicted EQ-5D values across the range of possible values; however, few respondents in the FACT-G data set have low EQ-5D values, which reduces the accuracy at the severe end. Conclusions. OLS and tobit mapping functions perform well for both instruments. Response mapping gives the best model predictions for QLQ-C30. The generalizability of the FACT-G mapping function is limited to populations in moderate to good health.

The determinants of efficiency in the Canadian health care system

In spite of the vast number of studies measuring economic efficiency in health care, there has been little take-up of this evidence by policy-makers to date. This study provides an illustration of how a system-level study drawing on best practice in empirical measurement of efficiency may be of practical use to health system decision makers and managers. We make use of the rich data available in Canada to undertake a robust two-stage data envelopment analysis to calculate efficiency at the regional (sub-provincial) level. Decisions about what the health system produces (the outcome to measure efficiency against) and what are the resources it has to produce that outcome were based on interviews and consultation with health system decision makers. Overall, we find large inefficiencies in the Canadian health care system, which could improve outcomes (here, measured as a reduction in treatable causes of death) by between 18 and 35% across our analyses. Also, we find that inefficiencies are the result of three main sets of factors that policy makers could pay attention to: management factors, such as hospital re-admissions; public health factors, such as obesity and smoking rates; and environmental factors such as the population’s average income.

Sleep disturbance and longitudinal risk of inflammation: Moderating influences of social integration and social isolation in the Coronary Artery Risk Development in Young Adults (CARDIA) study

Both sleep disturbance and social isolation increase the risk for morbidity and mortality. Systemic inflammation is suspected as a potential mechanism of these associations. However, the complex relationships between sleep disturbance, social isolation, and inflammation have not been examined in a population-based longitudinal study. This study examined the longitudinal association between sleep disturbance and systemic inflammation, and the moderating effects of social isolation on this association. The CARDIA study is a population-based longitudinal study conducted in four US cities. Sleep disturbance - i.e., insomnia complaints and short sleep duration - was assessed in 2962 African-American and White adults at baseline (2000-2001, ages 33-45years). Circulating C-reactive protein (CRP) was measured at baseline and follow-up (2005-2006). Interleukin-6 (IL-6) and subjective and objective social isolation (i.e., feelings of social isolation and social network size) were measured at follow-up. Sleep disturbance was a significant predictor of inflammation five years later after full adjustment for covariates (adjusted betas: 0.048, P=0.012 for CRP; 0.047, P=0.017 for IL-6). Further adjustment for baseline CRP revealed that sleep disturbance also impacted the longitudinal change in CRP levels over five years (adjusted beta: 0.044, P=0.013). Subjective social isolation was a significant moderator of this association between sleep disturbance and CRP (adjusted beta 0.131, P=0.002). Sleep disturbance was associated with heightened systemic inflammation in a general population over a five-year follow-up, and this association was significantly stronger in those who reported feelings of social isolation. Clinical interventions targeting sleep disturbances may be a potential avenue for reducing inflammation, particularly in individuals who feel socially isolated.

Predicting developmental delay in a longitudinal cohort of preschool children with single-suture craniosynostosis: is neurobehavioral assessment important?

AIM

The aim of this study was to determine whether neurobehavioral assessment before and after cranial vault surgery can improve prediction of developmental delay in children with single-suture craniosynostosis (SSC), after accounting for 'baseline' demographic and clinical variables (SSC diagnosis and surgery age).

METHOD

Children with SSC were referred by the treating surgeon or pediatrician before surgery. Neurobehavioral assessments were performed at ages of approximately 6, 18, and 36 months. Iterative models were developed to predict delay, as determined by one or more tests of cognitive, motor, and language skills at 36 months. We selected from groups of variables entered in order of timing (before or after corrective surgery), and source of information (parent questionnaire or psychometric testing).

RESULTS

Good predictive accuracy as determined by area under the receiver operating characteristic curve (AUC), was obtained with the baseline model (AUC=0.66), which incorporated age at surgery, sex, and socio-economic status. However, predictive accuracy was improved by including pre- and post-surgery neurobehavioral assessments. Models incorporating post-surgery neurobehavioral testing (AUC=0.79), pre-surgery testing (AUC=0.74), or both pre- and post-surgery testing (AUC=0.79) performed similarly. However, the specifity of all models was considered to be moderate (≤0.62).

INTERPRETATION

Prediction of delay was enhanced by assessment of neurobehavioral status. Findings provide tentative support for guidelines of care that call for routine testing of children with SSC.

Predictive value of the usual clinical signs and laboratory tests in the diagnosis of septic arthritis

OBJECTIVE

To determine the sensitivity and specificity of clinical and laboratory signs for the diagnosis of septic arthritis (SA). Patients and methods This prospective study included all adult patients with suspected SA seen in the emergency department or rheumatology department at the University Hospital, Clermont-Ferrand, France, over a period of 18 months.

RESULTS

In total, 105 patients with suspected SA were included, 38 (36%) presenting with SA (29 [28%] with bacteriologically documented SA). In the univariate analysis, chills (p=0.015), gradual onset (p=0.04), local redness (p=0.01), as well as an entry site for infection (p=0.01) were most often identified in SA. A history of crystal-induced arthritis (p=0.004) was more frequent in non-SA cases. An erythrocyte sedimentation rate (ESR)>50 mm (p=0.005), a C-reactive protein (CRP) level >100 mg/L (p=0.019), and radiological signs suggestive of SA (p=0.001) were more frequent in the SA cases. Synovial fluid appearance: purulent (p50,000/μL (p < 0.001), differentiated between SA and non-SA. In multivariate analysis, only chills (odds ration [OR]=4.7, 95% confidence interval [CI] 1.3-17.1), a history of crystal-induced arthritis (OR=0.09, 95% CI 0.01-0.9), purulent appearance of the joint fluid (OR=8.4, 95% CI 2.4-28.5), synovial WBC count >50,000/mm3 (OR=6.8, 95% CI 1.3-36), and radiological findings (OR=7.1, 95% CI 13-37.9) remained significant.

CONCLUSION

No clinical sign or laboratory test (excluding bacteriological test), taken alone, is conclusive for the differentiation between SA and non-SA, but the association of several signs, notably chills, history of crystal-induced arthritis, radiological findings, and the appearance and cellularity of joint fluid may be suggestive.

The prevalence, risk factors and short-term outcomes of delirium in Thai elderly emergency department patients

BACKGROUND

We sought to determine the prevalence of delirium in a Thai emergency department (ED). The secondary objective was to identify risk factors and short-term outcomes in delirious elderly ED patients.

METHODS

This was a prospective cross-sectional study in the ED of an urban tertiary care hospital. Patients aged ≥65 years who presented to the ED were included. We excluded patients who had severe dementia, were not responsive to verbal stimuli, had severe trauma and were blind, deaf, aphasic or unable to speak Thai. Delirium was determined using the Confusion Assessment Method for the Intensive Care Unit. We collected 30-day mortality rate, hospital length of stay and revisit rate as short-term outcomes.

RESULTS

We had a final sample size of 232 patients; 27 (12%) were delirious in the ED, of which 16 (59%) were not recognised to be delirious by the emergency physician. Multivariable logistic regression analysis showed dementia (adjusted OR (AOR) 13.1; 95% CI 2.9 to 59.6), auditory impairment (AOR 4.8; 95% CI 1.6 to 13.8) and ED diagnosis of metabolic derangement (AOR 6.5; 95% CI 1.6 to 26.8) were associated with delirium in the ED. Delirium was associated with a higher mortality rate than those without delirium (15% vs 2%, p=0.004).

CONCLUSIONS

In one middle-income country, elderly ED patients were delirious >10% of the time. Delirium was underdiagnosed and was associated with an increased 30-day mortality rate. Delirium screening needs to be improved, potentially focusing on high-risk patients.

Agreement between administrative data and the Resident Assessment Instrument Minimum Dataset (RAI-MDS) for medication use in long-term care facilities: a population-based study

Background

Prescription medication use, which is common among long-term care facility (LTCF) residents, is routinely used to describe quality of care and predict health outcomes. Data sources that capture medication information, which include surveys, medical charts, administrative health databases, and clinical assessment records, may not collect concordant information, which can result in comparable prevalence and effect size estimates. The purpose of this research was to estimate agreement between two population-based electronic data sources for measuring use of several medication classes among LTCF residents: outpatient prescription drug administrative data and the Resident Assessment Instrument Minimum Data Set (RAI-MDS) Version 2.0.

Methods

Prescription drug and RAI-MDS data from the province of Saskatchewan, Canada (population 1.1 million) were linked for 2010/11 in this cross-sectional study. Agreement for anti-psychotic, anti-depressant, and anti-anxiety/hypnotic medication classes was examined using prevalence estimates, Cohen’s κ, and positive and negative agreement. Mixed-effects logistic regression models tested resident and facility characteristics associated with disagreement.

Results

The cohort was comprised of 8,866 LTCF residents. In the RAI-MDS data, prevalence of anti-psychotics was 35.7%, while for anti-depressants it was 37.9% and for hypnotics it was 27.1%. Prevalence was similar in prescription drug data for anti-psychotics and anti-depressants, but lower for hypnotics (18.0%). Cohen’s κ ranged from 0.39 to 0.85 and was highest for the first two medication classes. Diagnosis of a mood disorder and facility affiliation was associated with disagreement for hypnotics.

Conclusions

Agreement between prescription drug administrative data and RAI-MDS assessment data was influenced by the type of medication class, as well as selected patient and facility characteristics. Researchers should carefully consider the purpose of their study, whether it is to capture medication that are dispensed or medications that are currently used by residents, when selecting a data source for research on LTCF populations.

Electronic supplementary material

The online version of this article (doi:10.1186/s12877-015-0023-2) contains supplementary material, which is available to authorized users.

Development and internal validation of a pediatric acute asthma prediction rule for hospitalization

BACKGROUND

Clinicians have difficulty predicting need for hospitalization of children with acute asthma exacerbations.

OBJECTIVE

The objective of this study was to develop and internally validate a multivariable asthma prediction rule (APR) to inform hospitalization decision making in children aged 5-17 years with acute asthma exacerbations.

METHODS

Between April 2008 and February 2013 we enrolled a prospective cohort of patients aged 5-17 years with asthma who presented to our pediatric emergency department with acute exacerbations. Predictors for APR modeling included 15 demographic characteristics, asthma chronic control measures, and pulmonary examination findings in participants at the time of triage and before treatment. The primary outcome variable for APR modeling was need for hospitalization (length of stay >24 h for those admitted to hospital or relapse for those discharged). A secondary outcome was the hospitalization decision of the clinical team. We used penalized maximum likelihood multiple logistic regression modeling to examine the adjusted association of each predictor variable with the outcome. Backward step-down variable selection techniques were used to yield reduced-form models.

RESULTS

Data from 928 of 933 participants were used for prediction rule modeling, with median [interquartile range] age 8.8 [6.9, 11.2] years, 61% male, and 59% African-American race. Both full (penalized) and reduced-form models for each outcome calibrated well, with bootstrap-corrected c-indices of 0.74 and 0.73 for need for hospitalization and 0.81 in each case for hospitalization decision.

CONCLUSION

The APR predicts the need for hospitalization of children with acute asthma exacerbations using predictor variables available at the time of presentation to an emergency department.

A simulation study of sample size demonstrated the importance of the number of events per variable to develop prediction models in clustered data

OBJECTIVES

This study aims to investigate the influence of the amount of clustering [intraclass correlation (ICC) = 0%, 5%, or 20%], the number of events per variable (EPV) or candidate predictor (EPV = 5, 10, 20, or 50), and backward variable selection on the performance of prediction models.

STUDY DESIGN AND SETTING

Researchers frequently combine data from several centers to develop clinical prediction models. In our simulation study, we developed models from clustered training data using multilevel logistic regression and validated them in external data.

RESULTS

The amount of clustering was not meaningfully associated with the models' predictive performance. The median calibration slope of models built in samples with EPV = 5 and strong clustering (ICC = 20%) was 0.71. With EPV = 5 and ICC = 0%, it was 0.72. A higher EPV related to an increased performance: the calibration slope was 0.85 at EPV = 10 and ICC = 20% and 0.96 at EPV = 50 and ICC = 20%. Variable selection sometimes led to a substantial relative bias in the estimated predictor effects (up to 118% at EPV = 5), but this had little influence on the model's performance in our simulations.

CONCLUSION

We recommend at least 10 EPV to fit prediction models in clustered data using logistic regression. Up to 50 EPV may be needed when variable selection is performed.

Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration

The TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) Statement includes a 22-item checklist, which aims to improve the reporting of studies developing, validating, or updating a prediction model, whether for diagnostic or prognostic purposes. The TRIPOD Statement aims to improve the transparency of the reporting of a prediction model study regardless of the study methods used. This explanation and elaboration document describes the rationale; clarifies the meaning of each item; and discusses why transparent reporting is important, with a view to assessing risk of bias and clinical usefulness of the prediction model. Each checklist item of the TRIPOD Statement is explained in detail and accompanied by published examples of good reporting. The document also provides a valuable reference of issues to consider when designing, conducting, and analyzing prediction model studies. To aid the editorial process and help peer reviewers and, ultimately, readers and systematic reviewers of prediction model studies, it is recommended that authors include a completed checklist in their submission. The TRIPOD checklist can also be downloaded from www.tripod-statement.org.

Modern modelling techniques are data hungry: a simulation study for predicting dichotomous endpoints

BACKGROUND

Modern modelling techniques may potentially provide more accurate predictions of binary outcomes than classical techniques. We aimed to study the predictive performance of different modelling techniques in relation to the effective sample size ("data hungriness").

METHODS

We performed simulation studies based on three clinical cohorts: 1282 patients with head and neck cancer (with 46.9% 5 year survival), 1731 patients with traumatic brain injury (22.3% 6 month mortality) and 3181 patients with minor head injury (7.6% with CT scan abnormalities). We compared three relatively modern modelling techniques: support vector machines (SVM), neural nets (NN), and random forests (RF) and two classical techniques: logistic regression (LR) and classification and regression trees (CART). We created three large artificial databases with 20 fold, 10 fold and 6 fold replication of subjects, where we generated dichotomous outcomes according to different underlying models. We applied each modelling technique to increasingly larger development parts (100 repetitions). The area under the ROC-curve (AUC) indicated the performance of each model in the development part and in an independent validation part. Data hungriness was defined by plateauing of AUC and small optimism (difference between the mean apparent AUC and the mean validated AUC <0.01).

RESULTS

We found that a stable AUC was reached by LR at approximately 20 to 50 events per variable, followed by CART, SVM, NN and RF models. Optimism decreased with increasing sample sizes and the same ranking of techniques. The RF, SVM and NN models showed instability and a high optimism even with >200 events per variable.

CONCLUSIONS

Modern modelling techniques such as SVM, NN and RF may need over 10 times as many events per variable to achieve a stable AUC and a small optimism than classical modelling techniques such as LR. This implies that such modern techniques should only be used in medical prediction problems if very large data sets are available.

Adverse drug reaction-related hospitalisations among patients with heart failure at two hospitals in the United Arab Emirates

BACKGROUND

Little is known about the adverse drug reaction (ADR) related admissions among heart failure (HF) patients.

OBJECTIVE

The aim of this study was to determine the rate, factors, and medications associated with ADR-related hospitalisations among HF patients.

SETTING

Two government hospitals in Dubai, United Arab Emirates.

METHODS

This was a prospective, observational study. Consecutive adult HF patients who were admitted between December 2011 and November 2012 to the cardiology units were included in this study. The circumstances of their admission were analysed.

MAIN OUTCOME MEASURES

ADRs-related admissions of HF patients to cardiology units were identified and further assessed for their nature, causality, and preventability.

RESULTS

Of 511 admissions, 34 were due to ADR-related hospitalisation (6.65, 95 % confidence interval 4.8-8.5 %). Number of medications taken by HF patients was the only predictors of ADR-related hospitalisations, where higher number of medications was associated with the odd ratio of 1.11 (95 % CI, 1.03-1.20, P = 0.005). More than one-third of ADR-related hospitalisations (35 %) were preventable The most frequent drugs causing ADR-related hospitalisation were diuretics (32 %), followed by non-steroidal anti-inflammatory drugs (15 %), thiazolidinediones (9 %), anticoagulants (9 %), antiplatelets (6 %), and aldosterone blockers (6 %).

CONCLUSION

ADR-related hospitalisations account for 6.7 % of admissions of HF patients to cardiac units, one-third of which are preventable. Number of medications taken by HF patients is the only predictors of ADR-related hospitalisations. Diuretic induced volume depletion, and sodium and water retention caused by thiazolidinediones and NSAIDs medications are the major causes of ADR-related hospitalisations of HF patients.

Augmented backward elimination: a pragmatic and purposeful way to develop statistical models

Statistical models are simple mathematical rules derived from empirical data describing the association between an outcome and several explanatory variables. In a typical modeling situation statistical analysis often involves a large number of potential explanatory variables and frequently only partial subject-matter knowledge is available. Therefore, selecting the most suitable variables for a model in an objective and practical manner is usually a non-trivial task. We briefly revisit the purposeful variable selection procedure suggested by Hosmer and Lemeshow which combines significance and change-in-estimate criteria for variable selection and critically discuss the change-in-estimate criterion. We show that using a significance-based threshold for the change-in-estimate criterion reduces to a simple significance-based selection of variables, as if the change-in-estimate criterion is not considered at all. Various extensions to the purposeful variable selection procedure are suggested. We propose to use backward elimination augmented with a standardized change-in-estimate criterion on the quantity of interest usually reported and interpreted in a model for variable selection. Augmented backward elimination has been implemented in a SAS macro for linear, logistic and Cox proportional hazards regression. The algorithm and its implementation were evaluated by means of a simulation study. Augmented backward elimination tends to select larger models than backward elimination and approximates the unselected model up to negligible differences in point estimates of the regression coefficients. On average, regression coefficients obtained after applying augmented backward elimination were less biased relative to the coefficients of correctly specified models than after backward elimination. In summary, we propose augmented backward elimination as a reproducible variable selection algorithm that gives the analyst more flexibility in adopting model selection to a specific statistical modeling situation.

A "multi-hit" model of neonatal white matter injury: cumulative contributions of chronic placental inflammation, acute fetal inflammation and postnatal inflammatory events

OBJECTIVE

We sought to determine whether cumulative evidence of perinatal inflammation was associated with increased risk in a "multi-hit" model of neonatal white matter injury (WMI).

METHODS

This retrospective cohort study included very preterm (gestational ages at delivery <32 weeks) live-born singleton neonates delivered at Hutzel Women's Hospital, Detroit, MI, from 2006 to 2011. Four pathologists blinded to clinical diagnoses and outcomes performed histological examinations according to standardized protocols. Neurosonography was obtained per routine clinical care. The primary indicator of WMI was ventriculomegaly (VE). Neonatal inflammation-initiating illnesses included bacteremia, surgical necrotizing enterocolitis, other infections, and those requiring mechanical ventilation.

RESULTS

A total of 425 live-born singleton neonates delivered before the 32nd week of gestation were included. Newborns delivered of pregnancies affected by chronic chorioamnionitis who had histologic evidence of an acute fetal inflammatory response were at increased risk of VE, unlike those without funisitis, relative to referent newborns without either condition, adjusting for gestational age [odds ratio (OR) 4.7; 95% confidence interval (CI) 1.4-15.8 vs. OR 1.3; 95% CI 0.7-2.6]. Similarly, newborns with funisitis who developed neonatal inflammation-initiating illness were at increased risk of VE, unlike those who did not develop such illness, compared to the referent group without either condition [OR 3.6 (95% CI 1.5-8.3) vs. OR 1.7 (95% CI 0.5-5.5)]. The greater the number of these three types of inflammation documented, the higher the risk of VE (P<0.0001).

CONCLUSION

Chronic placental inflammation, acute fetal inflammation, and neonatal inflammation-initiating illness seem to interact in contributing risk information and/or directly damaging the developing brain of newborns delivered very preterm.

The General Weakness Syndrome Therapy (GymNAST) study: protocol for a cohort study on recovery on walking function

INTRODUCTION

Critical illness myopathy (CIM) and polyneuropathy (CIP) are common complications of critical illness that frequently occur together. Both cause so called intensive care unit (ICU)-acquired muscle weakness. This weakness of limb muscles increases morbidity and delay rehabilitation and recovery of walking ability. Although full recovery has been reported people with severe weakness may take months to improve walking. Focused physical rehabilitation of people with ICU-acquired muscle weakness is therefore of great importance. However, although physical rehabilitation is common, detailed knowledge about the pattern and the time course of recovery of walking function are not well understood. Therefore, the aim of the General Weakness Syndrome Therapy (GymNAST) study is to describe the time course of recovery of walking function and other activities of daily living in these patients.

METHODS AND ANALYSIS

We conduct a prospective cohort study of people with ICU-acquired muscle weakness with defined diagnosis of CIM or CIP. Based on our sample size calculation, approximately 150 patients will be recruited from the ICU of our hospital in Germany. Amount and content of physical rehabilitation, clinical tests for example, muscle strength and motor function and neuropsychological assessments will be used as independent variables. The primary outcomes will include recovery of walking function and mobility. Secondary outcomes will include global motor function, activities in daily life and participation.

ETHICS AND DISSEMINATION

The study is being carried out in agreement with the Declaration of Helsinki and conducted with the approval of the local medical Ethics Committee (Landesärztekammer Sachsen, Germany, reference number EK-BR-32/13-1) and with the understanding and written consent of each patient's guardian. The results of this study will be published in peer-reviewed journals and disseminated to the medical society and general public.

Early prediction of spontaneous twin very preterm birth: a population based study 2002-2012

OBJECTIVE

The aim of this study was to establish early pregnancy risk indicators for spontaneous twin very preterm birth.

METHODS

We conducted a retrospective observational population-based study. Twenty-one potential early pregnancy risk factors were analyzed using multivariable logistic regression to determine which of them was independently associated with spontaneous twin very preterm birth.

RESULTS

Of 1815 spontaneous twin births 15.3% (277) occurred before 32 weeks. Previous preterm delivery (aOR 3.73; 95% CI, 2.52-5.52), nulliparity (aOR 2.94; 95% CI, 2.09-4.14), body mass index <18.5 (aOR 1.86; 95% CI, 1.12-3.10), body mass index ≥30 (aOR 1.87; 95% CI, 1.21-2.89), hysteroscopic metroplasty (aOR 1.63; 1.07-2.49), conization (aOR 2.05; 95% CI, 1.07-3.94) and monochorionicity (aOR 1.83; 95% CI, 1.28-2.63) were significantly associated with twin very preterm birth.

CONCLUSIONS

Pending verification in other populations, twin pregnancies at significant risk for spontaneous very preterm birth can be identified in early pregnancy using several risk indicators.

Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist

Carl Moons and colleagues provide a checklist and background explanation for critically appraising and extracting data from systematic reviews of prognostic and diagnostic prediction modelling studies. Please see later in the article for the Editors' Summary.

Diagnostic accuracy of history taking to assess lumbosacral nerve root compression

BACKGROUND CONTEXT

The diagnosis of sciatica is primarily based on history and physical examination. Most physical tests used in isolation show poor diagnostic accuracy. Little is known about the diagnostic accuracy of history items.

PURPOSE

To assess the diagnostic accuracy of history taking for the presence of lumbosacral nerve root compression or disc herniation on magnetic resonance imaging in patients with sciatica.

STUDY DESIGN

Cross-sectional diagnostic study.

PATIENT SAMPLE

A total of 395 adult patients with severe disabling radicular leg pain of 6 to 12 weeks duration were included.

OUTCOME MEASURES

Lumbosacral nerve root compression and disc herniation on magnetic resonance imaging were independently assessed by two neuroradiologists and one neurosurgeon blinded to any clinical information.

METHODS

Data were prospectively collected in nine hospitals. History was taken according to a standardized protocol. There were no study-specific conflicts of interest.

RESULTS

Exploring the diagnostic odds ratio of 20 history items revealed a significant contribution in diagnosing nerve root compression for "male sex," "pain worse in leg than in back," and "a non-sudden onset." A significant contribution to the diagnosis of a herniated disc was found for "body mass index <30," "a non-sudden onset," and "sensory loss." Multivariate logistic regression analysis of six history items pre-selected from the literature (age, gender, pain worse in leg than in back, sensory loss, muscle weakness, and more pain on coughing/sneezing/straining) revealed an area under the receiver operating characteristic curve of 0.65 (95% confidence interval, 0.58-0.71) for the model diagnosing nerve root compression and an area under the receiver operating characteristic curve of 0.66 (95% confidence interval, 0.58-0.74) for the model diagnosing disc herniation.

CONCLUSIONS

A few history items used in isolation had significant diagnostic value and the diagnostic accuracy of a model with six pre-selected items was poor.