Time-resolved serial crystallography to track the dynamics of carbon monoxide in the active site of cytochrome c oxidase.

Cytochrome c oxidase (CcO) is part of the respiratory chain and contributes to the electrochemical membrane gradient in mitochondria as well as in many bacteria, as it uses the energy released in the reduction of oxygen to pump protons across an energy-transducing biological membrane. Here, we use time-resolved serial femtosecond crystallography to study the structural response of the active site upon flash photolysis of carbon monoxide (CO) from the reduced heme a3 of ba3-type CcO. In contrast with the aa3-type enzyme, our data show how CO is stabilized on CuB through interactions with a transiently ordered water molecule. These results offer a structural explanation for the extended lifetime of the CuB-CO complex in ba3-type CcO and, by extension, the extremely high oxygen affinity of the enzyme.

Modeling difference x-ray scattering observations from an integral membrane protein within a detergent micelle.

Time-resolved x-ray solution scattering (TR-XSS) is a sub-field of structural biology, which observes secondary structural changes in proteins as they evolve along their functional pathways. While the number of distinct conformational states and their rise and decay can be extracted directly from TR-XSS experimental data recorded from light-sensitive systems, structural modeling is more challenging. This step often builds from complementary structural information, including secondary structural changes extracted from crystallographic studies or molecular dynamics simulations. When working with integral membrane proteins, another challenge arises because x-ray scattering from the protein and the surrounding detergent micelle interfere and these effects should be considered during structural modeling. Here, we utilize molecular dynamics simulations to explicitly incorporate the x-ray scattering cross term between a membrane protein and its surrounding detergent micelle when modeling TR-XSS data from photoactivated samples of detergent solubilized bacteriorhodopsin. This analysis provides theoretical foundations in support of our earlier approach to structural modeling that did not explicitly incorporate this cross term and improves agreement between experimental data and theoretical predictions at lower x-ray scattering angles.

Lipidic cubic phase serial femtosecond crystallography structure of a photosynthetic reaction centre.

Serial crystallography is a rapidly growing method that can yield structural insights from microcrystals that were previously considered to be too small to be useful in conventional X-ray crystallography. Here, conditions for growing microcrystals of the photosynthetic reaction centre of Blastochloris viridis within a lipidic cubic phase (LCP) crystallization matrix that employ a seeding protocol utilizing detergent-grown crystals with a different crystal packing are described. LCP microcrystals diffracted to 2.25 Šresolution when exposed to XFEL radiation, which is an improvement of 0.15 Šover previous microcrystal forms. Ubiquinone was incorporated into the LCP crystallization media and the resulting electron density within the mobile QB pocket is comparable to that of other cofactors within the structure. As such, LCP microcrystallization conditions will facilitate time-resolved diffraction studies of electron-transfer reactions to the mobile quinone, potentially allowing the observation of structural changes associated with the two electron-transfer reactions leading to complete reduction of the ubiquinone ligand.

Association Between an Increase in Serum Sodium and In-Hospital Mortality in Critically Ill Patients.

OBJECTIVES: In critically ill patients, dysnatremia is common, and in these patients, in-hospital mortality is higher. It remains unknown whether changes of serum sodium after ICU admission affect mortality, especially whether normalization of mild hyponatremia improves survival. DESIGN: Retrospective cohort study. SETTING: Ten Dutch ICUs between January 2011 and April 2017. PATIENTS: Adult patients were included if at least one serum sodium measurement within 24 hours of ICU admission and at least one serum sodium measurement 24-48 hours after ICU admission were available. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: A logistic regression model adjusted for age, sex, and Acute Physiology and Chronic Health Evaluation-IV-predicted mortality was used to assess the difference between mean of sodium measurements 24-48 hours after ICU admission and first serum sodium measurement at ICU admission (Δ48 hr-[Na]) and in-hospital mortality. In total, 36,660 patients were included for analysis. An increase in serum sodium was independently associated with a higher risk of in-hospital mortality in patients admitted with normonatremia (Δ48 hr-[Na] 5-10 mmol/L odds ratio: 1.61 [1.44-1.79], Δ48 hr-[Na] > 10 mmol/L odds ratio: 4.10 [3.20-5.24]) and hypernatremia (Δ48 hr-[Na] 5-10 mmol/L odds ratio: 1.47 [1.02-2.14], Δ48 hr-[Na] > 10 mmol/L odds ratio: 8.46 [3.31-21.64]). In patients admitted with mild hyponatremia and Δ48 hr-[Na] greater than 5 mmol/L, no significant difference in hospital mortality was found (odds ratio, 1.11 [0.99-1.25]). CONCLUSIONS: An increase in serum sodium in the first 48 hours of ICU admission was associated with higher in-hospital mortality in patients admitted with normonatremia and in patients admitted with hypernatremia.

Ultrafast structural changes within a photosynthetic reaction centre.

Photosynthetic reaction centres harvest the energy content of sunlight by transporting electrons across an energy-transducing biological membrane. Here we use time-resolved serial femtosecond crystallography1 using an X-ray free-electron laser2 to observe light-induced structural changes in the photosynthetic reaction centre of Blastochloris viridis on a timescale of picoseconds. Structural perturbations first occur at the special pair of chlorophyll molecules of the photosynthetic reaction centre that are photo-oxidized by light. Electron transfer to the menaquinone acceptor on the opposite side of the membrane induces a movement of this cofactor together with lower amplitude protein rearrangements. These observations reveal how proteins use conformational dynamics to stabilize the charge-separation steps of electron-transfer reactions.

The primary structural photoresponse of phytochrome proteins captured by a femtosecond X-ray laser.

Phytochrome proteins control the growth, reproduction, and photosynthesis of plants, fungi, and bacteria. Light is detected by a bilin cofactor, but it remains elusive how this leads to activation of the protein through structural changes. We present serial femtosecond X-ray crystallographic data of the chromophore-binding domains of a bacterial phytochrome at delay times of 1 ps and 10 ps after photoexcitation. The data reveal a twist of the D-ring, which leads to partial detachment of the chromophore from the protein. Unexpectedly, the conserved so-called pyrrole water is photodissociated from the chromophore, concomitant with movement of the A-ring and a key signaling aspartate. The changes are wired together by ultrafast backbone and water movements around the chromophore, channeling them into signal transduction towards the output domains. We suggest that the observed collective changes are important for the phytochrome photoresponse, explaining the earliest steps of how plants, fungi and bacteria sense red light.

Plants adapt to the availability of light throughout their lives because it regulates so many aspects of their growth and reproduction. To detect the level of light, plant cells use proteins called phytochromes, which are also found in some bacteria and fungi. Phytochrome proteins change shape when they are exposed to red light, and this change alters the behaviour of the cell. The red light is absorbed by a molecule known as chromophore, which is connected to a region of the phytochrome called the PHY-tongue. This region undergoes one of the key structural changes that occur when the phytochrome protein absorbs light, turning from a flat sheet into a helix. Claesson, Wahlgren, Takala et al. studied the structure of a bacterial phytochrome protein almost immediately after shining a very brief flash of red light using a laser. The experiments revealed that the structure of the protein begins to change within a trillionth of a second: specifically, the chromophore twists, which disrupts its attachment to the protein, freeing the protein to change shape. Claesson, Wahlgren, Takala et al. note that this structure is likely a very short-lived intermediate state, which however triggers more changes in the overall shape change of the protein. One feature of the rearrangement is the disappearance of a particular water molecule. This molecule can be found at the core of many different phytochrome structures and interacts with several parts of the chromophore and the phytochrome protein. It is unclear why the water molecule is lost, but given how quickly this happens after the red light is applied it is likely that this disappearance is an integral part of the reshaping process. Together these events disrupt the interactions between the chromophore and the PHY-tongue, enabling the PHY-tongue to change shape and alter the structure of the phytochrome protein. Understanding and controlling this process could allow scientists to alter growth patterns in plants, such as crops or weeds.

Well-based crystallization of lipidic cubic phase microcrystals for serial X-ray crystallography experiments.

Serial crystallography is having an increasing impact on structural biology. This emerging technique opens up new possibilities for studying protein structures at room temperature and investigating structural dynamics using time-resolved X-ray diffraction. A limitation of the method is the intrinsic need for large quantities of well ordered micrometre-sized crystals. Here, a method is presented to screen for conditions that produce microcrystals of membrane proteins in the lipidic cubic phase using a well-based crystallization approach. A key advantage over earlier approaches is that the progress of crystal formation can be easily monitored without interrupting the crystallization process. In addition, the protocol can be scaled up to efficiently produce large quantities of crystals for serial crystallography experiments. Using the well-based crystallization methodology, novel conditions for the growth of showers of microcrystals of three different membrane proteins have been developed. Diffraction data are also presented from the first user serial crystallography experiment performed at MAX IV Laboratory.

Retinal isomerization in bacteriorhodopsin captured by a femtosecond x-ray laser.

Ultrafast isomerization of retinal is the primary step in photoresponsive biological functions including vision in humans and ion transport across bacterial membranes. We used an x-ray laser to study the subpicosecond structural dynamics of retinal isomerization in the light-driven proton pump bacteriorhodopsin. A series of structural snapshots with near-atomic spatial resolution and temporal resolution in the femtosecond regime show how the excited all-trans retinal samples conformational states within the protein binding pocket before passing through a twisted geometry and emerging in the 13-cis conformation. Our findings suggest ultrafast collective motions of aspartic acid residues and functional water molecules in the proximity of the retinal Schiff base as a key facet of this stereoselective and efficient photochemical reaction.

Pain Measurement in Mechanically Ventilated Patients After Cardiac Surgery: Comparison of the Behavioral Pain Scale (BPS) and the Critical-Care Pain Observation Tool (CPOT).

OBJECTIVES: The Behavioral Pain Scale (BPS) and Critical-Care Pain Observation Tool (CPOT) are behavioral pain assessment tools for sedated and unconscious critically ill patients. The aim of this study was to compare the reliability, internal consistency, and discriminant validation of the BPS and the CPOT simultaneously in mechanically ventilated patients after cardiac surgery. DESIGN: A prospective, observational cohort study. SETTING: A 20-bed closed-format intensive care unit with mixed medical, surgical, and cardiac surgery patients in a teaching hospital in Amsterdam, The Netherlands. PARTICIPANTS: The study comprised 72 consecutive intubated and mechanically ventilated patients after cardiac surgery who were not able to self-report pain. MEASUREMENTS AND MAIN RESULTS: Two nurses assessed the BPS and CPOT simultaneously and independently at the following 4 moments: rest, a nonpainful procedure (oral care), rest, and a painful procedure (turning). Both scores showed a significant increase of 2 points between rest and turning. The median BPS score of nurse 1 showed a significant increase of 1 point between rest and the nonpainful procedure (oral care), whereas both median CPOT scores did not change. The interrater reliability of the BPS and CPOT showed fair-to-good agreement of 0.74 overall. During the periods of rest 1 and rest 2, values ranged from 0.24 to 0.46. Cronbach's alpha values for the BPS were 0.62 (nurse 1) and 0.59 (nurse 2) compared with 0.65 and 0.58, respectively, for the CPOT. CONCLUSIONS: The BPS and CPOT are reliable and valid pain assessment tools in a daily clinical setting. However, the discriminant validation of both scores seems less satisfactory in sedated or agitated patients and this topic requires further investigation.

From Macrocrystals to Microcrystals: A Strategy for Membrane Protein Serial Crystallography.

Serial protein crystallography was developed at X-ray free-electron lasers (XFELs) and is now also being applied at storage ring facilities. Robust strategies for the growth and optimization of microcrystals are needed to advance the field. Here we illustrate a generic strategy for recovering high-density homogeneous samples of microcrystals starting from conditions known to yield large (macro) crystals of the photosynthetic reaction center of Blastochloris viridis (RCvir). We first crushed these crystals prior to multiple rounds of microseeding. Each cycle of microseeding facilitated improvements in the RCvir serial femtosecond crystallography (SFX) structure from 3.3-Å to 2.4-Å resolution. This approach may allow known crystallization conditions for other proteins to be adapted to exploit novel scientific opportunities created by serial crystallography.

Serial femtosecond crystallography structure of cytochrome c oxidase at room temperature.

Cytochrome c oxidase catalyses the reduction of molecular oxygen to water while the energy released in this process is used to pump protons across a biological membrane. Although an extremely well-studied biological system, the molecular mechanism of proton pumping by cytochrome c oxidase is still not understood. Here we report a method to produce large quantities of highly diffracting microcrystals of ba 3-type cytochrome c oxidase from Thermus thermophilus suitable for serial femtosecond crystallography. The room-temperature structure of cytochrome c oxidase is solved to 2.3 Å resolution from data collected at an X-ray Free Electron Laser. We find overall agreement with earlier X-ray structures solved from diffraction data collected at cryogenic temperature. Previous structures solved from synchrotron radiation data, however, have shown conflicting results regarding the identity of the active-site ligand. Our room-temperature structure, which is free from the effects of radiation damage, reveals that a single-oxygen species in the form of a water molecule or hydroxide ion is bound in the active site. Structural differences between the ba 3-type and aa 3-type cytochrome c oxidases around the proton-loading site are also described.

Effectiveness and Clinical Outcomes of a Two-Step Implementation of Conservative Oxygenation Targets in Critically Ill Patients: A Before and After Trial.

OBJECTIVES: Conservative oxygen therapy is aimed at the prevention of harm by iatrogenic hyperoxia while preserving adequate tissue oxygenation. Our aim was to study the effectiveness and clinical outcomes of a two-step implementation of conservative oxygenation targets in the ICU. DESIGN: This was a before and after stepwise implementation study of conservative oxygenation targets, between July 2011 and July 2014. The primary endpoint was the proportion of PaO2 values within the target range. Secondary outcomes included ventilator-free days at day 28, length of stay, and mortality. SETTING: Three closed-format ICUs in the Netherlands. PATIENTS: We analyzed data on 15,045 eligible admissions. INTERVENTIONS: The first implementation phase consisted of providing training and feedback on new guidelines instructing for explicit targets for arterial oxygen tension (PaO2, 55-86 mm Hg) and oxyhemoglobin saturation (SpO2, 92-95%). In the second phase, bedside clinicians were additionally assisted in guideline adherence by a computerized decision-support system. MEASUREMENTS AND MAIN RESULTS: The proportion of PaO2 in the target range increased from 47% at baseline to 63% in phase 1 and to 68% in phase 2 (p < 0.0001). Episodes of hyperoxia decreased (p < 0.0001), whereas hypoxic episodes remained unchanged (p = 0.06) during the study. Mechanical ventilation time was significantly lower (p < 0.01) during both study phases. After adjustment for potential confounders, ventilator-free days in phase 1 and phase 2 were higher than baseline: adjusted mean difference, 0.55 (95% CI, 0.25-0.84) and 0.48 (95% CI, 0.11-0.86), respectively. Adjusted ICU mortality and ICU-free days did not significantly differ between study phases. Hospital mortality decreased in reference to baseline: adjusted odds ratio, 0.84 (95% CI, 0.74-0.96) for phase 1 and 0.82 (95% CI, 0.69-0.96) for phase 2. CONCLUSIONS: Stepwise implementation of conservative oxygenation targets was feasible, effective, and seemed safe in critically ill patients. The implementation was associated with several changes in clinical outcomes, but the causal impact of conservative oxygenation is still to be determined.

Self-reported attitudes versus actual practice of oxygen therapy by ICU physicians and nurses.

BACKGROUND: High inspiratory oxygen concentrations are frequently administered in ventilated patients in the intensive care unit (ICU) but may induce lung injury and systemic toxicity. We compared beliefs and actual clinical practice regarding oxygen therapy in critically ill patients. METHODS: In three large teaching hospitals in the Netherlands, ICU physicians and nurses were invited to complete a questionnaire about oxygen therapy. Furthermore, arterial blood gas (ABG) analysis data and ventilator settings were retrieved to assess actual oxygen practice in the same hospitals 1 year prior to the survey. RESULTS: In total, 59% of the 215 respondents believed that oxygen-induced lung injury is a concern. The majority of physicians and nurses stated that minimal acceptable oxygen saturation and partial arterial oxygen pressure (PaO2) ranges were 85% to 95% and 7 to 10 kPa (52.5 to 75 mmHg), respectively. Analysis of 107,888 ABG results with concurrent ventilator settings, derived from 5,565 patient admissions, showed a median (interquartile range (IQR)) PaO2 of 11.7 kPa (9.9 to 14.3) [87.8 mmHg], median fractions of inspired oxygen (FiO2) of 0.4 (0.4 to 0.5), and median positive end-expiratory pressure (PEEP) of 5 (5 to 8) cm H2O. Of all PaO2 values, 73% were higher than the upper limit of the commonly self-reported acceptable range, and in 58% of these cases, neither FiO2 nor PEEP levels were lowered until the next ABG sample was taken. CONCLUSIONS: Most ICU clinicians acknowledge the potential adverse effects of prolonged exposure to hyperoxia and report a low tolerance for high oxygen levels. However, in actual clinical practice, a large proportion of their ICU patients was exposed to higher arterial oxygen levels than self-reported target ranges.

The effect of diabetes mellitus on the association between measures of glycaemic control and ICU mortality: a retrospective cohort study.

INTRODUCTION: In critical illness, four measures of glycaemic control are associated with ICU mortality: mean glucose concentration, glucose variability, the incidence of hypoglycaemia (≤2.2 mmol/l) or low glucose (2.3 to 4.7 mmol/l). Underlying diabetes mellitus (DM) might affect these associations. Our objective was to study whether the association between these measures of glycaemic control and ICU mortality differs between patients without and with DM and to explore the cutoff value for detrimental low glucose in both cohorts. METHODS: This retrospective database cohort study included patients admitted between January 2004 and June 2011 to a 24-bed medical/surgical ICU in a teaching hospital. We analysed glucose and outcome data from 10,320 patients: 8,682 without DM and 1,638 with DM. The cohorts were subdivided into quintiles of mean glucose and quartiles of glucose variability. Multivariable regression models were used to examine the independent association between the four measures of glycaemic control and ICU mortality, and for defining the cutoff value for detrimental low glucose. RESULTS: Regarding mean glucose, a U-shaped relation was observed in the non-DM cohort with an increased ICU mortality in the lowest and highest glucose quintiles (odds ratio=1.4 and 1.8, P<0.001). No clear pattern was found in the DM cohort. Glucose variability was related to ICU mortality only in the non-DM cohort, with highest ICU mortality in the upper variability quartile (odds ratio=1.7, P<0.001). Hypoglycaemia was associated with ICU mortality in both cohorts (odds ratio non-DM=2.5, P<0.001; odds ratio DM=4.2, P=0.001), while low-glucose concentrations up to 4.9 mmol/l were associated with an increased risk of ICU mortality in the non-DM cohort and up to 3.5 mmol/l in the DM cohort. CONCLUSION: Mean glucose and high glucose variability are related to ICU mortality in the non-DM cohort but not in the DM cohort. Hypoglycaemia (≤2.2 mmol/l) was associated with ICU mortality in both. The cutoff value for detrimental low glucose is higher in the non-DM cohort (4.9 mmol/l) than in the DM cohort (3.5 mmol/l). While hypoglycaemia (≤2.2 mmol/l) should be avoided in both groups, DM patients seem to tolerate a wider glucose range than non-DM patients.

Presence of tobramycin in blood and urine during selective decontamination of the digestive tract in critically ill patients, a prospective cohort study.

INTRODUCTION: Tobramycin is one of the components used for selective decontamination of the digestive tract (SDD), applied to prevent colonization and subsequent infections in critically ill patients. Tobramycin is administered in the oropharynx and gastrointestinal tract and is normally not absorbed. However, critical illness may convey gut barrier failure. The aim of the study was to assess the prevalence and amount of tobramycin leakage from the gut into the blood, to quantify tobramycin excretion in urine, and to determine the association of tobramycin leakage with markers of circulation, kidney function and other organ failure. METHODS: This was a prospective observational cohort study. The setting was the 20-bed closed format-mixed ICU of a teaching hospital. The study population was critically ill patients with an expected stay of more than two days, receiving SDD with tobramycin, polymyxin-E and amphotericin-B four times daily in the oropharynx and stomach. Tobramycin concentration was measured in serum (sensitive high performance liquid chromatography - mass spectrometry/mass spectrometry (HLPC-MS/MS) assay) and 24-hour urine (conventional immunoassay), in 34 patients, 24 hours after ICU admission, and in 71 patients, once daily for 7 days. Tobramycin leakage was defined as tobramycin detected in serum at least once (> 0.05 mg/L). Ototoxicity was not monitored. RESULTS: Of the 100 patients with available blood samples, 83 had tobramycin leakage. Median highest serum concentration for each patient was 0.12 mg/L; 99% of the patients had at least one positive urinary sample (> 0.5 mg/L), 49% had a urinary concentration ≥ 1 mg/L. The highest tobramycin serum concentration was significantly associated with vasopressor support, renal and hepatic dysfunction, and C-reactive protein. At binary logistic regression analysis, high dopamine dose and low urinary output on Day 1 were the significant predictors of tobramycin leakage. Nephrotoxicity could not be shown. CONCLUSIONS: The majority of acute critically ill patients treated with enteral tobramycin as a component of SDD had traces of tobramycin in the blood, especially those with severe shock, inflammation and subsequent acute kidney injury, suggesting loss of gut barrier and decreased renal removal. Unexpectedly, urinary tobramycin was above the therapeutic trough level in half of the patients. Nephrotoxicity could not be demonstrated.

External validation of Acute Physiology and Chronic Health Evaluation IV in Dutch intensive care units and comparison with Acute Physiology and Chronic Health Evaluation II and Simplified Acute Physiology Score II.

PURPOSE: The aim of this study was to validate and compare the performance of the Acute Physiology and Chronic Health Evaluation (APACHE) IV in the Dutch intensive care unit (ICU) population to the APACHE II and Simplified Acute Physiology Score (SAPS) II. MATERIALS AND METHODS: This is a prospective study based on data from a national quality registry between 2006 and 2009 from 59 Dutch ICUs. The validation set consisted of 62,737 patients; the 3 models were compared using 44,112 patients. Measures of discrimination, accuracy, and calibration (area under the receiver operating characteristic curve (AUC), Brier score, R(2), and C-statistic) were calculated using bootstrapping. In addition, the standardized mortality ratios were calculated. RESULTS: The original APACHE IV showed good discrimination and accuracy (AUC = 0.87, Brier score = 0.10, R(2) = 0.29) but poor calibration (C-statistic = 822.67). Customization significantly improved the performance of the APACHE IV. The overall discrimination and accuracy of the customized APACHE IV were statistically better, and the overall C-statistic was inferior to those of the customized APACHE II and SAPS II, but these differences were small in perspective of clinical use. CONCLUSIONS: The 3 models have comparable capabilities for benchmarking purposes after customization. Main advantage of APACHE IV is the large number of diagnoses that enable subgroup analysis. The APACHE IV coronary artery bypass grafting (CABG) model has a good performance in the Dutch ICU population and can be used to complement the 3 models.

Mean glucose during ICU admission is related to mortality by a U-shaped curve in surgical and medical patients: a retrospective cohort study.

INTRODUCTION: Lowering of hyperglycemia in the intensive care unit (ICU) is widely practiced. We investigated in which way glucose regulation, defined as mean glucose concentration during admission, is associated with ICU mortality in a medical and a surgical cohort. METHODS: Retrospective database cohort study including patients admitted between January 2004 and December 2007 in a 20-bed medical/surgical ICU in a teaching hospital. Hyperglycemia was treated using a computerized algorithm targeting for glucose levels of 4.0-7.0 mmol/l. Five thousand eight hundred twenty-eight patients were eligible for analyses, of whom 1,339 patients had a medical and 4,489 had a surgical admission diagnosis. RESULTS: The cohorts were subdivided in quintiles of increasing mean glucose. We examined the relation between these mean glucose strata and mortality. In both cohorts we observed the highest mortality in the lowest and highest strata. Logistic regression analysis adjusted for age, sex, Acute Physiology and Chronic Health Evaluation II (APACHE II) score, admission duration and occurrence of severe hypoglycemia showed that in the medical cohort mean glucose levels <6.7 mmol/l and >8.4 mmol/l and in the surgical cohort mean glucose levels < 7.0 mmol/l and >9.4 mmol/l were associated with significantly increased ICU mortality (OR 2.4-3.0 and 4.9-6.2, respectively). Limitations of the study were its retrospective design and possible incomplete correction for severity of disease. CONCLUSIONS: Mean overall glucose during ICU admission is related to mortality by a U-shaped curve in medical and surgical patients. In this cohort of patients a 'safe range' of mean glucose regulation might be defined approximately between 7.0 and 9.0 mmol/l.

Hypoglycemia is associated with intensive care unit mortality.

OBJECTIVE: The implementation of intensive insulin therapy in the intensive care unit is accompanied by an increase in hypoglycemia. We studied the relation between hypoglycemia on intensive care unit mortality, because the evidence on this subject is conflicting. DESIGN: Retrospective database cohort study. SETTING: An 18-bed medical/surgical intensive care unit in a teaching hospital (Onze Lieve Vrouwe Gasthuis Hospital, Amsterdam, The Netherlands). PATIENTS: A total of 5961 patients admitted to from 2004 to 2007 were analyzed. Readmissions and patients with a withholding care policy or with hypoglycemia on the first glucose measurement were excluded. Patients were treated with a computerized insulin algorithm (target glucose range, 72-126 mg/dL). INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: All first episodes of hypoglycemia (glucose < or =45 mg/dL) were derived from 154,015 glucose values. Using Poisson regression, the incidence rates for intensive care unit death and incidence rate ratio comparing exposure and nonexposure to hypoglycemia were calculated. Patients were considered to be exposed to hypoglycemia from the event until the end of intensive care unit admittance. We corrected for severity of disease using the daily Sequential Organ Failure Assessment score. Age, sex, cardiothoracic surgery, sepsis, and diabetes mellitus were also included as possible confounders. Two hundred eighty-eight (4.8%) patients experienced at least one episode of hypoglycemia. Median age was 68 yrs (range, 58-75 yrs), 66% were male, and 6.4% died in the intensive care unit. The incidence rate of death in patients exposed to hypoglycemia was 40 per 1000 intensive care unit days compared with 17 per 1000 intensive care unit days in patients without exposure. The adjusted incidence rate ratio for intensive care unit death was 2.1 (95% confidence interval, 1.6-2.8; p < .001). CONCLUSIONS: Hypoglycemia is related to intensive care unit mortality, also when adjusted for a daily adjudicated measure of disease severity, indicating the possibility of a causal relationship.

Learning predictive models that use pattern discovery--a bootstrap evaluative approach applied in organ functioning sequences.

An important problem in the Intensive Care is how to predict on a given day of stay the eventual hospital mortality for a specific patient. A recent approach to solve this problem suggested the use of frequent temporal sequences (FTSs) as predictors. Methods following this approach were evaluated in the past by inducing a model from a training set and validating the prognostic performance on an independent test set. Although this evaluative approach addresses the validity of the specific models induced in an experiment, it falls short of evaluating the inductive method itself. To achieve this, one must account for the inherent sources of variation in the experimental design. The main aim of this work is to demonstrate a procedure based on bootstrapping, specifically the .632 bootstrap procedure, for evaluating inductive methods that discover patterns, such as FTSs. A second aim is to apply this approach to find out whether a recently suggested inductive method that discovers FTSs of organ functioning status is superior over a traditional method that does not use temporal sequences when compared on each successive day of stay at the Intensive Care Unit. The use of bootstrapping with logistic regression using pre-specified covariates is known in the statistical literature. Using inductive methods of prognostic models based on temporal sequence discovery within the bootstrap procedure is however novel at least in predictive models in the Intensive Care. Our results of applying the bootstrap-based evaluative procedure demonstrate the superiority of the FTS-based inductive method over the traditional method in terms of discrimination as well as accuracy. In addition we illustrate the insights gained by the analyst into the discovered FTSs from the bootstrap samples.