A Bartlett-type correction for likelihood ratio tests with application to testing equality of Gaussian graphical models.

This work defines a new correction for the likelihood ratio test for a two-sample problem within the multivariate normal context. This correction applies to decomposable graphical models, where testing equality of distributions can be decomposed into lower dimensional problems.

Acute Kidney Injury in Traumatic Brain Injury Patients: Results From the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury Study.

OBJECTIVES: Acute kidney injury is frequent in polytrauma patients, and it is associated with increased mortality and extended hospital length of stay. However, the specific prevalence of acute kidney injury after traumatic brain injury is less recognized. The present study aims to describe the occurrence rate, risk factors, timing, and association with outcome of acute kidney injury in a large cohort of traumatic brain injury patients. DESIGN: The Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury is a multicenter, prospective observational, longitudinal, cohort study. SETTING: Sixty-five ICUs across Europe. PATIENTS: For the present study, we selected 4,509 traumatic brain injury patients with an ICU length of stay greater than 72 hours and with at least two serum creatinine values during the first 7 days of ICU stay. MEASUREMENTS AND MAIN RESULTS: We classified acute kidney injury in three stages according to the Kidney Disease Improving Global Outcome criteria: acute kidney injury stage 1 equals to serum creatinine × 1.5-1.9 times from baseline or an increase greater than or equal to 0.3 mg/dL in 48 hours; acute kidney injury stage 2 equals to serum creatinine × 2-2.9 times baseline; acute kidney injury stage 3 equals to serum creatinine × three times baseline or greater than or equal to 4 mg/dL or need for renal replacement therapy. Standard reporting techniques were used to report incidences. A multivariable Cox regression analysis was performed to model the cause-specific hazard of acute kidney injury and its association with the long-term outcome. We included a total of 1,262 patients. The occurrence rate of acute kidney injury during the first week was as follows: acute kidney injury stage 1 equals to 8% (n = 100), acute kidney injury stage 2 equals to 1% (n = 14), and acute kidney injury stage 3 equals to 3% (n = 36). Acute kidney injury occurred early after ICU admission, with a median of 2 days (interquartile range 1-4 d). Renal history (hazard ratio = 2.48; 95% CI, 1.39-4.43; p = 0.002), insulin-dependent diabetes (hazard ratio = 2.52; 95% CI, 1.22-5.197; p = 0.012), hypernatremia (hazard ratio = 1.88; 95% CI, 1.31-2.71; p = 0.001), and osmotic therapy administration (hazard ratio = 2.08; 95% CI, 1.45-2.99; p < 0.001) were significantly associated with the risk of developing acute kidney injury. Acute kidney injury was also associated with an increased ICU length of stay and with a higher probability of 6 months unfavorable Extended Glasgow Outcome Scale and mortality. CONCLUSIONS: Acute kidney injury after traumatic brain injury is an early phenomenon, affecting about one in 10 patients. Its occurrence negatively impacts mortality and neurologic outcome at 6 months. Osmotic therapy use during ICU stay could be a modifiable risk factor.

Construction and assessment of prediction rules for binary outcome in the presence of missing predictor data using multiple imputation and cross-validation: Methodological approach and data-based evaluation.

We investigate calibration and assessment of predictive rules when missing values are present in the predictors. Our paper has two key objectives. The first is to investigate how the calibration of the prediction rule can be combined with use of multiple imputation to account for missing predictor observations. The second objective is to propose such methods that can be implemented with current multiple imputation software, while allowing for unbiased predictive assessment through validation on new observations for which outcome is not yet available. We commence with a review of the methodological foundations of multiple imputation as a model estimation approach as opposed to a purely algorithmic description. We specifically contrast application of multiple imputation for parameter (effect) estimation with predictive calibration. Based on this review, two approaches are formulated, of which the second utilizes application of the classical Rubin's rules for parameter estimation, while the first approach averages probabilities from models fitted on single imputations to directly approximate the predictive density for future observations. We present implementations using current software that allow for validation and estimation of performance measures by cross-validation, as well as imputation of missing data in predictors on the future data where outcome is missing by definition. To simplify, we restrict discussion to binary outcome and logistic regression throughout. Method performance is verified through application on two real data sets. Accuracy (Brier score) and variance of predicted probabilities are investigated. Results show substantial reductions in variation of calibrated probabilities when using the first approach.

Noninvasive Intracranial Pressure Estimation With Transcranial Doppler: A Prospective Observational Study.

BACKGROUND: Transcranial Doppler (TCD) ultrasonography has been described for the noninvasive assessment of intracranial pressure (ICP). This study investigates the relationship between standard, invasive intracranial pressure monitoring (ICPi) and noninvasive ICP assessment using a simple formula based on TCD-derived flow velocity (FV) and mean arterial blood pressure values (ICPTCD). MATERIAL AND METHODS: We performed a prospective observational study on 100 consecutive traumatic brain injury patients requiring invasive ICP monitoring, admitted to the Neurosciences and Trauma Critical Care Unit of Addenbrooke's Hospital, Cambridge, UK. ICPi was compared with ICPTCD using a method based on the "diastolic velocity-derived estimator" (FVd), which was initially described for the noninvasive estimation of cerebral perfusion pressure but subsequently utilized for ICP assessment. RESULTS: Median ICPi was 13 mm Hg (interquartile range: 10, 17.25 mm Hg). There was no correlation between ICPi and ICPTCD (R=-0.17; 95% confidence interval [CI]: -0.35, 0.03; P=0.097). Bland-Altman analysis demonstrated wide 95% limits of agreement between ICPi and ICPTCD (-27.58, 30.10; SD, 14.42). ICPTCD was not able to detect intracranial hypertension (ICPi >20 mm Hg); the area under the receiver operating characteristic curve for prediction was 34.5% (95% CI, 23.1%-45.9%) with 0% sensitivity and 74.4% specificity for ICPTCD to detect ICPi>20 mm Hg. CONCLUSIONS: Using a formula based on diastolic FV, TCD is an insufficiently accurate method for the noninvasive assessment of ICP. Further studies are warranted to confirm these results in a broader patient cohort.

Incidence, Risk Factors, and Effects on Outcome of Ventilator-Associated Pneumonia in Patients With Traumatic Brain Injury: Analysis of a Large, Multicenter, Prospective, Observational Longitudinal Study.

BACKGROUND: No large prospective data, to our knowledge, are available on ventilator-associated pneumonia (VAP) in patients with traumatic brain injury (TBI). RESEARCH QUESTION: To evaluate the incidence, timing, and risk factors of VAP after TBI and its effect on patient outcome. STUDY DESIGN AND METHODS: This analysis is of the Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury data set, from a large, multicenter, prospective, observational study including patients with TBI admitted to European ICUs, receiving mechanical ventilation for ≥ 48 hours and with an ICU length of stay (LOS) ≥ 72 hours. Characteristics of patients with VAP vs characteristics of patients without VAP were compared, and outcome was assessed at 6 months after injury by using the Glasgow Outcome Scale Extended. RESULTS: The study included 962 patients: 196 (20.4%) developed a VAP at a median interval of 5 days (interquartile range [IQR], 3-7 days) after intubation. Patients who developed VAP were younger (median age, 39.5 [IQR, 25-55] years vs 51 [IQR, 30-66] years; P < .001), with a higher incidence of alcohol abuse (36.6% vs 27.6%; P = .026) and drug abuse (10.1% vs 4.2%; P = .009), more frequent thoracic trauma (53% vs 43%; P = .014), and more episodes of respiratory failure during ICU stay (69.9% vs 28.1%; P < .001). Age (hazard ratio [HR], 0.99; 95% CI, 0.98-0.99; P = .001), chest trauma (HR, 1.4; 95% CI, 1.03-1.90; P = .033), histamine-receptor antagonist intake (HR, 2.16; 95% CI, 1.37-3.39; P = .001), and antibiotic prophylaxis (HR, 0.69; 95% CI, 0.50-0.96; P = .026) were associated with the risk of VAP. Patients with VAP had a longer duration of mechanical ventilation (median, 15 [IQR, 10-22] days vs 8 [IQR, 5-14] days; P < .001) and ICU LOS (median, 20 [IQR, 14-29] days vs 13 [IQR, 8-21] days; P < .001). However, VAP was not associated with increased mortality or worse neurological outcome. Overall mortality at 6 months was 22%. INTERPRETATION: VAP occurs less often than previously described in patients after TBI and has a detrimental effect on ICU LOS but not on mortality and neurological outcome. CLINICAL TRIAL REGISTRATION: ClinicalTrials.gov; No.: NCT02210221; URL: www.clinicaltrials.gov.