Effect of training-sample size and classification difficulty on the accuracy of genomic predictors.

Introduction: As part of the MicroArray Quality Control (MAQC)-II project, this analysis examines how the choice of univariate feature-selection methods and classification algorithms may influence the performance of genomic predictors under varying degrees of prediction difficulty represented by three clinically relevant endpoints. Methods: We used gene-expression data from 230 breast cancers (grouped into training and independent validation sets), and we examined 40 predictors (five univariate feature-selection methods combined with eight different classifiers) for each of the three endpoints. Their classification performance was estimated on the training set by using two different resampling methods and compared with the accuracy observed in the independent validation set. Results: A ranking of the three classification problems was obtained, and the performance of 120 models was estimated and assessed on an independent validation set. The bootstrapping estimates were closer to the validation performance than were the cross-validation estimates. The required sample size for each endpoint was estimated, and both gene-level and pathway-level analyses were performed on the obtained models. Conclusions: We showed that genomic predictor accuracy is determined largely by an interplay between sample size and classification difficulty. Variations on univariate feature-selection methods and choice of classification algorithm have only a modest impact on predictor performance, and several statistically equally good predictors can be developed for any given classification problem.

Sample size and statistical power in the hierarchical analysis of variance: applications in morphometry of the nervous system.

Analysis of variance is commonly used in morphometry in order to ascertain differences in parameters between several populations. Failure to detect significant differences between populations (type II error) may be due to suboptimal sampling and lead to erroneous conclusions; the concept of statistical power allows one to avoid such failures by means of an adequate sampling. Several examples are given in the morphometry of the nervous system, showing the use of the power of a hierarchical analysis of variance test for the choice of appropriate sample and subsample sizes. In the first case chosen, neuronal densities in the human visual cortex, we find the number of observations to be of little effect. For dendritic spine densities in the visual cortex of mice and humans, the effect is somewhat larger. A substantial effect is shown in our last example, dendritic segmental lengths in monkey lateral geniculate nucleus. It is in the nature of the hierarchical model that sample size is always more important than subsample size. The relative weight to be attributed to subsample size thus depends on the relative magnitude of the between observations variance compared to the between individuals variance.

A program for the computation of power and determination of sample size in hierarchical experimental designs.

We have devised a program that allows computation of the power of F-test, and hence determination of appropriate sample and subsample sizes, in the context of the one-way hierarchical analysis of variance with fixed effects. The power at a fixed alternative is an increasing function of the sample size and of the subsample size. The program makes it easy to obtain the power of F-test for a range of values of sample and subsample sizes, and therefore the appropriate sizes based on a desired power. The program can be used for the 'ordinary' case of the one-way analysis of variance, as well as for hierarchical analysis of variance with two stages of sampling. Examples are given of the practical use of the program.

Effects of sample size on the performance of species distribution models

A wide range of modelling algorithms is used by ecologists, conservation practitioners, and others to predict species ranges from point locality data. Unfortunately, the amount of data available is limited for many taxa and regions, making it essential to quantify the sensitivity of these algorithms to sample size. This is the first study to address this need by rigorously evaluating a broad suite of algorithms with independent presence-absence data from multiple species and regions. We evaluated predictions from 12 algorithms for 46 species (from six different regions of the world) at three sample sizes (100, 30, and 10 records). We used data from natural history collections to run the models, and evaluated the quality of model predictions with area under the receiver operating characteristic curve (AUC). With decreasing sample size, model accuracy decreased and variability increased across species and between models. Novel modelling methods that incorporate both interactions between predictor variables and complex response shapes (i.e. GBM, MARS-INT, BRUTO) performed better than most methods at large sample sizes but not at the smallest sample sizes. Other algorithms were much less sensitive to sample size, including an algorithm based on maximum entropy (MAXENT) that had among the best predictive power across all sample sizes. Relative to other algorithms, a distance metric algorithm (DOMAIN) and a genetic algorithm (OM-GARP) had intermediate performance at the largest sample size and among the best performance at the lowest sample size. No algorithm predicted consistently well with small sample size (n < 30) and this should encourage highly conservative use of predictions based on small sample size and restrict their use to exploratory modelling.

EEG-based functional brain networks: does the network size matter?

Functional connectivity in human brain can be represented as a network using electroencephalography (EEG) signals. These networks--whose nodes can vary from tens to hundreds--are characterized by neurobiologically meaningful graph theory metrics. This study investigates the degree to which various graph metrics depend upon the network size. To this end, EEGs from 32 normal subjects were recorded and functional networks of three different sizes were extracted. A state-space based method was used to calculate cross-correlation matrices between different brain regions. These correlation matrices were used to construct binary adjacency connectomes, which were assessed with regards to a number of graph metrics such as clustering coefficient, modularity, efficiency, economic efficiency, and assortativity. We showed that the estimates of these metrics significantly differ depending on the network size. Larger networks had higher efficiency, higher assortativity and lower modularity compared to those with smaller size and the same density. These findings indicate that the network size should be considered in any comparison of networks across studies.

Sensitivity of predictive species distribution models to change in grain size

Predictive species distribution modelling (SDM) has become an essential tool in biodiversity conservation and management. The choice of grain size (resolution) of environmental layers used in modelling is one important factor that may affect predictions. We applied 10 distinct modelling techniques to presence-only data for 50 species in five different regions, to test whether: (1) a 10-fold coarsening of resolution affects predictive performance of SDMs, and (2) any observed effects are dependent on the type of region, modelling technique, or species considered. Results show that a 10 times change in grain size does not severely affect predictions from species distribution models. The overall trend is towards degradation of model performance, but improvement can also be observed. Changing grain size does not equally affect models across regions, techniques, and species types. The strongest effect is on regions and species types, with tree species in the data sets (regions) with highest locational accuracy being most affected. Changing grain size had little influence on the ranking of techniques: boosted regression trees remain best at both resolutions. The number of occurrences used for model training had an important effect, with larger sample sizes resulting in better models, which tended to be more sensitive to grain. Effect of grain change was only noticeable for models reaching sufficient performance and/or with initial data that have an intrinsic error smaller than the coarser grain size.

Swain: Stata module to correct the SEM chi-square overidentification test in small sample sizes or complex models. Statistical Software Components S457617, Boston College Department of Economics.

Swain corrects the chi-square overidentification test (i.e., likelihood ratio test of fit) for structural equation models whethr with or without latent variables. The chi-square statistic is asymptotically correct; however, it does not behave as expected in small samples and/or when the model is complex (cf. Herzog, Boomsma, & Reinecke, 2007). Thus, particularly in situations where the ratio of sample size (n) to the number of parameters estimated (p) is relatively small (i.e., the p to n ratio is large), the chi-square test will tend to overreject correctly specified models. To obtain a closer approximation to the distribution of the chi-square statistic, Swain (1975) developed a correction; this scaling factor, which converges to 1 asymptotically, is multiplied with the chi-square statistic. The correction better approximates the chi-square distribution resulting in more appropriate Type 1 reject error rates (see Herzog & Boomsma, 2009; Herzog, et al., 2007).

Gait and foot clearance parameters obtained using shoe-worn inertial sensors in a large-population sample of older adults

In order to distinguish dysfunctional gait; clinicians require a measure of reference gait parameters for each population. This study provided normative values for widely used parameters in more than 1400 able-bodied adults over the age of 65. We also measured the foot clearance parameters (i.e., height of the foot above ground during swing phase) that are crucial to understand the complex relationship between gait and falls as well as obstacle negotiation strategies. We used a shoe-worn inertial sensor on each foot and previously validated algorithms to extract the gait parameters during 20 m walking trials in a corridor at a self-selected pace. We investigated the difference of the gait parameters between male and female participants by considering the effect of age and height factors. Besides; we examined the inter-relation of the clearance parameters with the gait speed. The sample size and breadth of gait parameters provided in this study offer a unique reference resource for the researchers.

Renal transplantation following aorto-iliac reconstruction is a safe a strategy

Introduction: Renal transplantation is considered the treatment of choice for end-stage renal disease. However, the association of occlusive aorto-iliac disease and chronic renal failure is frequent and aorto-iliac reconstruction may be necessary prior to renal transplantation. This retrospective study reviews the results of this operative strategy.Material and Methods: Between January 2001 and June 2010, 309 patients underwent renal transplantation at our institution and 8 patients had prior aorto-iliac reconstruction using prosthetic material. There were 6 men and 2 women with a median age of 62 years (range 51-70). Five aorto-bifemoral and 2 aorto-bi-iliac bypasses were performed for stage II (n=5), stage IV (n=1) and aortic aneurysm (n=1). In one patient, iliac kissing stents and an ilio-femoral bypass were implanted. 4 cadaveric and 4 living donor renal transplantations were performed with an interval of 2 months to 10 years after revascularization.The results were analysed with respect of graft and patients survival. Differences between groups were tested by the log rank method.Results: No complications and no death occurred in the post-operative period. All bypasses remained patent during follow-up. The median time of post transplantation follow-up was 46 months for all patients and 27 months for patients with prior revascularization. In the revascularized group and control group, the graft and patient survival at 1 year were respectively 100%/96%, 100%/99% and at 5 years 86%/86%, 86%/94%, without significant differences between both groups.Discussion: Our results suggest that renal transplantation following prior aorto-iliac revascularisation with prosthetic material is safe and effective. Patients with end-stage renal disease and concomitant aorto-iliac disease should therefore be considered for renal transplantation. However, caution in the interpretation of the results is indicated due to the small sample size of our study.

Decision-theoretic analysis of forensic sampling criteria using Bayesian decision networks

Sampling issues represent a topic of ongoing interest to the forensic science community essentially because of their crucial role in laboratory planning and working protocols. For this purpose, forensic literature described thorough (Bayesian) probabilistic sampling approaches. These are now widely implemented in practice. They allow, for instance, to obtain probability statements that parameters of interest (e.g., the proportion of a seizure of items that present particular features, such as an illegal substance) satisfy particular criteria (e.g., a threshold or an otherwise limiting value). Currently, there are many approaches that allow one to derive probability statements relating to a population proportion, but questions on how a forensic decision maker - typically a client of a forensic examination or a scientist acting on behalf of a client - ought actually to decide about a proportion or a sample size, remained largely unexplored to date. The research presented here intends to address methodology from decision theory that may help to cope usefully with the wide range of sampling issues typically encountered in forensic science applications. The procedures explored in this paper enable scientists to address a variety of concepts such as the (net) value of sample information, the (expected) value of sample information or the (expected) decision loss. All of these aspects directly relate to questions that are regularly encountered in casework. Besides probability theory and Bayesian inference, the proposed approach requires some additional elements from decision theory that may increase the efforts needed for practical implementation. In view of this challenge, the present paper will emphasise the merits of graphical modelling concepts, such as decision trees and Bayesian decision networks. These can support forensic scientists in applying the methodology in practice. How this may be achieved is illustrated with several examples. The graphical devices invoked here also serve the purpose of supporting the discussion of the similarities, differences and complementary aspects of existing Bayesian probabilistic sampling criteria and the decision-theoretic approach proposed throughout this paper.

Species distribution models for conservation-oriented studies in Switzerland : filling data and tools gaps

SUMMARYSpecies distribution models (SDMs) represent nowadays an essential tool in the research fields of ecology and conservation biology. By combining observations of species occurrence or abundance with information on the environmental characteristic of the observation sites, they can provide information on the ecology of species, predict their distributions across the landscape or extrapolate them to other spatial or time frames. The advent of SDMs, supported by geographic information systems (GIS), new developments in statistical models and constantly increasing computational capacities, has revolutionized the way ecologists can comprehend species distributions in their environment. SDMs have brought the tool that allows describing species realized niches across a multivariate environmental space and predict their spatial distribution. Predictions, in the form of probabilistic maps showing the potential distribution of the species, are an irreplaceable mean to inform every single unit of a territory about its biodiversity potential. SDMs and the corresponding spatial predictions can be used to plan conservation actions for particular species, to design field surveys, to assess the risks related to the spread of invasive species, to select reserve locations and design reserve networks, and ultimately, to forecast distributional changes according to scenarios of climate and/or land use change.By assessing the effect of several factors on model performance and on the accuracy of spatial predictions, this thesis aims at improving techniques and data available for distribution modelling and at providing the best possible information to conservation managers to support their decisions and action plans for the conservation of biodiversity in Switzerland and beyond. Several monitoring programs have been put in place from the national to the global scale, and different sources of data now exist and start to be available to researchers who want to model species distribution. However, because of the lack of means, data are often not gathered at an appropriate resolution, are sampled only over limited areas, are not spatially explicit or do not provide a sound biological information. A typical example of this is data on 'habitat' (sensu biota). Even though this is essential information for an effective conservation planning, it often has to be approximated from land use, the closest available information. Moreover, data are often not sampled according to an established sampling design, which can lead to biased samples and consequently to spurious modelling results. Understanding the sources of variability linked to the different phases of the modelling process and their importance is crucial in order to evaluate the final distribution maps that are to be used for conservation purposes.The research presented in this thesis was essentially conducted within the framework of the Landspot Project, a project supported by the Swiss National Science Foundation. The main goal of the project was to assess the possible contribution of pre-modelled 'habitat' units to model the distribution of animal species, in particular butterfly species, across Switzerland. While pursuing this goal, different aspects of data quality, sampling design and modelling process were addressed and improved, and implications for conservation discussed. The main 'habitat' units considered in this thesis are grassland and forest communities of natural and anthropogenic origin as defined in the typology of habitats for Switzerland. These communities are mainly defined at the phytosociological level of the alliance. For the time being, no comprehensive map of such communities is available at the national scale and at fine resolution. As a first step, it was therefore necessary to create distribution models and maps for these communities across Switzerland and thus to gather and collect the necessary data. In order to reach this first objective, several new developments were necessary such as the definition of expert models, the classification of the Swiss territory in environmental domains, the design of an environmentally stratified sampling of the target vegetation units across Switzerland, the development of a database integrating a decision-support system assisting in the classification of the relevés, and the downscaling of the land use/cover data from 100 m to 25 m resolution.The main contributions of this thesis to the discipline of species distribution modelling (SDM) are assembled in four main scientific papers. In the first, published in Journal of Riogeography different issues related to the modelling process itself are investigated. First is assessed the effect of five different stepwise selection methods on model performance, stability and parsimony, using data of the forest inventory of State of Vaud. In the same paper are also assessed: the effect of weighting absences to ensure a prevalence of 0.5 prior to model calibration; the effect of limiting absences beyond the environmental envelope defined by presences; four different methods for incorporating spatial autocorrelation; and finally, the effect of integrating predictor interactions. Results allowed to specifically enhance the GRASP tool (Generalized Regression Analysis and Spatial Predictions) that now incorporates new selection methods and the possibility of dealing with interactions among predictors as well as spatial autocorrelation. The contribution of different sources of remotely sensed information to species distribution models was also assessed. The second paper (to be submitted) explores the combined effects of sample size and data post-stratification on the accuracy of models using data on grassland distribution across Switzerland collected within the framework of the Landspot project and supplemented with other important vegetation databases. For the stratification of the data, different spatial frameworks were compared. In particular, environmental stratification by Swiss Environmental Domains was compared to geographical stratification either by biogeographic regions or political states (cantons). The third paper (to be submitted) assesses the contribution of pre- modelled vegetation communities to the modelling of fauna. It is a two-steps approach that combines the disciplines of community ecology and spatial ecology and integrates their corresponding concepts of habitat. First are modelled vegetation communities per se and then these 'habitat' units are used in order to model animal species habitat. A case study is presented with grassland communities and butterfly species. Different ways of integrating vegetation information in the models of butterfly distribution were also evaluated. Finally, a glimpse to climate change is given in the fourth paper, recently published in Ecological Modelling. This paper proposes a conceptual framework for analysing range shifts, namely a catalogue of the possible patterns of change in the distribution of a species along elevational or other environmental gradients and an improved quantitative methodology to identify and objectively describe these patterns. The methodology was developed using data from the Swiss national common breeding bird survey and the article presents results concerning the observed shifts in the elevational distribution of breeding birds in Switzerland.The overall objective of this thesis is to improve species distribution models as potential inputs for different conservation tools (e.g. red lists, ecological networks, risk assessment of the spread of invasive species, vulnerability assessment in the context of climate change). While no conservation issues or tools are directly tested in this thesis, the importance of the proposed improvements made in species distribution modelling is discussed in the context of the selection of reserve networks.RESUMELes modèles de distribution d'espèces (SDMs) représentent aujourd'hui un outil essentiel dans les domaines de recherche de l'écologie et de la biologie de la conservation. En combinant les observations de la présence des espèces ou de leur abondance avec des informations sur les caractéristiques environnementales des sites d'observation, ces modèles peuvent fournir des informations sur l'écologie des espèces, prédire leur distribution à travers le paysage ou l'extrapoler dans l'espace et le temps. Le déploiement des SDMs, soutenu par les systèmes d'information géographique (SIG), les nouveaux développements dans les modèles statistiques, ainsi que la constante augmentation des capacités de calcul, a révolutionné la façon dont les écologistes peuvent comprendre la distribution des espèces dans leur environnement. Les SDMs ont apporté l'outil qui permet de décrire la niche réalisée des espèces dans un espace environnemental multivarié et prédire leur distribution spatiale. Les prédictions, sous forme de carte probabilistes montrant la distribution potentielle de l'espèce, sont un moyen irremplaçable d'informer chaque unité du territoire de sa biodiversité potentielle. Les SDMs et les prédictions spatiales correspondantes peuvent être utilisés pour planifier des mesures de conservation pour des espèces particulières, pour concevoir des plans d'échantillonnage, pour évaluer les risques liés à la propagation d'espèces envahissantes, pour choisir l'emplacement de réserves et les mettre en réseau, et finalement, pour prévoir les changements de répartition en fonction de scénarios de changement climatique et/ou d'utilisation du sol. En évaluant l'effet de plusieurs facteurs sur la performance des modèles et sur la précision des prédictions spatiales, cette thèse vise à améliorer les techniques et les données disponibles pour la modélisation de la distribution des espèces et à fournir la meilleure information possible aux gestionnaires pour appuyer leurs décisions et leurs plans d'action pour la conservation de la biodiversité en Suisse et au-delà. Plusieurs programmes de surveillance ont été mis en place de l'échelle nationale à l'échelle globale, et différentes sources de données sont désormais disponibles pour les chercheurs qui veulent modéliser la distribution des espèces. Toutefois, en raison du manque de moyens, les données sont souvent collectées à une résolution inappropriée, sont échantillonnées sur des zones limitées, ne sont pas spatialement explicites ou ne fournissent pas une information écologique suffisante. Un exemple typique est fourni par les données sur 'l'habitat' (sensu biota). Même s'il s'agit d'une information essentielle pour des mesures de conservation efficaces, elle est souvent approximée par l'utilisation du sol, l'information qui s'en approche le plus. En outre, les données ne sont souvent pas échantillonnées selon un plan d'échantillonnage établi, ce qui biaise les échantillons et par conséquent les résultats de la modélisation. Comprendre les sources de variabilité liées aux différentes phases du processus de modélisation s'avère crucial afin d'évaluer l'utilisation des cartes de distribution prédites à des fins de conservation.La recherche présentée dans cette thèse a été essentiellement menée dans le cadre du projet Landspot, un projet soutenu par le Fond National Suisse pour la Recherche. L'objectif principal de ce projet était d'évaluer la contribution d'unités 'd'habitat' pré-modélisées pour modéliser la répartition des espèces animales, notamment de papillons, à travers la Suisse. Tout en poursuivant cet objectif, différents aspects touchant à la qualité des données, au plan d'échantillonnage et au processus de modélisation sont abordés et améliorés, et leurs implications pour la conservation des espèces discutées. Les principaux 'habitats' considérés dans cette thèse sont des communautés de prairie et de forêt d'origine naturelle et anthropique telles que définies dans la typologie des habitats de Suisse. Ces communautés sont principalement définies au niveau phytosociologique de l'alliance. Pour l'instant aucune carte de la distribution de ces communautés n'est disponible à l'échelle nationale et à résolution fine. Dans un premier temps, il a donc été nécessaire de créer des modèles de distribution de ces communautés à travers la Suisse et par conséquent de recueillir les données nécessaires. Afin d'atteindre ce premier objectif, plusieurs nouveaux développements ont été nécessaires, tels que la définition de modèles experts, la classification du territoire suisse en domaines environnementaux, la conception d'un échantillonnage environnementalement stratifié des unités de végétation cibles dans toute la Suisse, la création d'une base de données intégrant un système d'aide à la décision pour la classification des relevés, et le « downscaling » des données de couverture du sol de 100 m à 25 m de résolution. Les principales contributions de cette thèse à la discipline de la modélisation de la distribution d'espèces (SDM) sont rassemblées dans quatre articles scientifiques. Dans le premier article, publié dans le Journal of Biogeography, différentes questions liées au processus de modélisation sont étudiées en utilisant les données de l'inventaire forestier de l'Etat de Vaud. Tout d'abord sont évalués les effets de cinq méthodes de sélection pas-à-pas sur la performance, la stabilité et la parcimonie des modèles. Dans le même article sont également évalués: l'effet de la pondération des absences afin d'assurer une prévalence de 0.5 lors de la calibration du modèle; l'effet de limiter les absences au-delà de l'enveloppe définie par les présences; quatre méthodes différentes pour l'intégration de l'autocorrélation spatiale; et enfin, l'effet de l'intégration d'interactions entre facteurs. Les résultats présentés dans cet article ont permis d'améliorer l'outil GRASP qui intègre désonnais de nouvelles méthodes de sélection et la possibilité de traiter les interactions entre variables explicatives, ainsi que l'autocorrélation spatiale. La contribution de différentes sources de données issues de la télédétection a également été évaluée. Le deuxième article (en voie de soumission) explore les effets combinés de la taille de l'échantillon et de la post-stratification sur le la précision des modèles. Les données utilisées ici sont celles concernant la répartition des prairies de Suisse recueillies dans le cadre du projet Landspot et complétées par d'autres sources. Pour la stratification des données, différents cadres spatiaux ont été comparés. En particulier, la stratification environnementale par les domaines environnementaux de Suisse a été comparée à la stratification géographique par les régions biogéographiques ou par les cantons. Le troisième article (en voie de soumission) évalue la contribution de communautés végétales pré-modélisées à la modélisation de la faune. C'est une approche en deux étapes qui combine les disciplines de l'écologie des communautés et de l'écologie spatiale en intégrant leurs concepts de 'habitat' respectifs. Les communautés végétales sont modélisées d'abord, puis ces unités de 'habitat' sont utilisées pour modéliser les espèces animales. Une étude de cas est présentée avec des communautés prairiales et des espèces de papillons. Différentes façons d'intégrer l'information sur la végétation dans les modèles de répartition des papillons sont évaluées. Enfin, un clin d'oeil aux changements climatiques dans le dernier article, publié dans Ecological Modelling. Cet article propose un cadre conceptuel pour l'analyse des changements dans la distribution des espèces qui comprend notamment un catalogue des différentes formes possibles de changement le long d'un gradient d'élévation ou autre gradient environnemental, et une méthode quantitative améliorée pour identifier et décrire ces déplacements. Cette méthodologie a été développée en utilisant des données issues du monitoring des oiseaux nicheurs répandus et l'article présente les résultats concernant les déplacements observés dans la distribution altitudinale des oiseaux nicheurs en Suisse.L'objectif général de cette thèse est d'améliorer les modèles de distribution des espèces en tant que source d'information possible pour les différents outils de conservation (par exemple, listes rouges, réseaux écologiques, évaluation des risques de propagation d'espèces envahissantes, évaluation de la vulnérabilité des espèces dans le contexte de changement climatique). Bien que ces questions de conservation ne soient pas directement testées dans cette thèse, l'importance des améliorations proposées pour la modélisation de la distribution des espèces est discutée à la fin de ce travail dans le contexte de la sélection de réseaux de réserves.

A prospective randomised multi-centre controlled trial on tight glucose control by intensive insulin therapy in adult intensive care units: the Glucontrol study.

PURPOSE: An optimal target for glucose control in ICU patients remains unclear. This prospective randomized controlled trial compared the effects on ICU mortality of intensive insulin therapy (IIT) with an intermediate glucose control. METHODS: Adult patients admitted to the 21 participating medico-surgical ICUs were randomized to group 1 (target BG 7.8-10.0 mmol/L) or to group 2 (target BG 4.4-6.1 mmol/L). RESULTS: While the required sample size was 1,750 per group, the trial was stopped early due to a high rate of unintended protocol violations. From 1,101 admissions, the outcomes of 542 patients assigned to group 1 and 536 of group 2 were analysed. The groups were well balanced. BG levels averaged in group 1 8.0 mmol/L (IQR 7.1-9.0) (median of all values) and 7.7 mmol/L (IQR 6.7-8.8) (median of morning BG) versus 6.5 mmol/L (IQR 6.0-7.2) and 6.1 mmol/L (IQR 5.5-6.8) for group 2 (p < 0.0001 for both comparisons). The percentage of patients treated with insulin averaged 66.2 and 96.3%, respectively. Proportion of time spent in target BG was similar, averaging 39.5% and 45.1% (median (IQR) 34.3 (18.5-50.0) and 39.3 (26.2-53.6)%) in the groups 1 and 2, respectively. The rate of hypoglycaemia was higher in the group 2 (8.7%) than in group 1 (2.7%, p < 0.0001). ICU mortality was similar in the two groups (15.3 vs. 17.2%). CONCLUSIONS: In this prematurely stopped and therefore underpowered study, there was a lack of clinical benefit of intensive insulin therapy (target 4.4-6.1 mmol/L), associated with an increased incidence of hypoglycaemia, as compared to a 7.8-10.0 mmol/L target. (ClinicalTrials.gov # NCT00107601, EUDRA-CT Number: 200400391440).

Multimodel inference and multimodel averaging in empirical modeling of occupational exposure levels.

Empirical modeling of exposure levels has been popular for identifying exposure determinants in occupational hygiene. Traditional data-driven methods used to choose a model on which to base inferences have typically not accounted for the uncertainty linked to the process of selecting the final model. Several new approaches propose making statistical inferences from a set of plausible models rather than from a single model regarded as 'best'. This paper introduces the multimodel averaging approach described in the monograph by Burnham and Anderson. In their approach, a set of plausible models are defined a priori by taking into account the sample size and previous knowledge of variables influent on exposure levels. The Akaike information criterion is then calculated to evaluate the relative support of the data for each model, expressed as Akaike weight, to be interpreted as the probability of the model being the best approximating model given the model set. The model weights can then be used to rank models, quantify the evidence favoring one over another, perform multimodel prediction, estimate the relative influence of the potential predictors and estimate multimodel-averaged effects of determinants. The whole approach is illustrated with the analysis of a data set of 1500 volatile organic compound exposure levels collected by the Institute for work and health (Lausanne, Switzerland) over 20 years, each concentration having been divided by the relevant Swiss occupational exposure limit and log-transformed before analysis. Multimodel inference represents a promising procedure for modeling exposure levels that incorporates the notion that several models can be supported by the data and permits to evaluate to a certain extent model selection uncertainty, which is seldom mentioned in current practice.

RTOG 0525: Exploratory Subset Analysis from a Randomized Phase III Trial Comparing Standard (std) Adjuvant Temozolomide (TMZ) with a Dose-dense (dd) Schedule for Glioblastoma (GBM)

Purpose/Objective(s): RTwith TMZ is the standard for GBM. dd TMZ causes prolongedMGMTdepletion in mononuclear cells and possibly in tumor. The RTOG 0525 trial (ASCO 2011) did not show an advantage from dd TMZ for survival or progression free survival. We conducted exploratory, hypothesis-generating subset analyses to detect possible benefit from dd TMZ.Materials/Methods: Patients were randomized to std (150-200 mg/m2 x 5 d) or dd TMZ (75-100 mg/m2 x 21 d) q 4 weeks for 6- 12 cycles. Eligibility included age.18, KPS$ 60, and. 1 cm2 tissue for prospective MGMTanalysis for stratification. Furtheranalyses were performed for all randomized patients (''intent-to-treat'', ITT), and for all patients starting protocol therapy (SPT). Subset analyses were performed by RPA class (III, IV, V), KPS (90-100, = 50,\50), resection (partial, total), gender (female, male), and neurologic dysfunction (nf = none, minor, moderate).Results: No significant difference was seen for median OS (16.6 vs. 14.9 months), or PFS (5.5 vs. 6.7 months, p = 0.06). MGMT methylation was linked to improved OS (21.2 vs. 14 months, p\0.0001), and PFS (8.7 vs. 5.7 months, p\0.0001). For the ITT (n = 833), there was no OS benefit from dd TMZ in any subset. Two subsets showed a PFS benefit for dd TMZ: RPA class III (6.2 vs. 12.6 months, HR 0.69, p = 0.03) and nf = minor (HR 0.77, p = 0.01). For RPA III, dd dramatically delayed progression, but post-progression dd patients died more quickly than std. A similar pattern for nf = minor was observed. For the SPT group (n = 714) there was neither PFS nor OS benefit for dd TMZ, overall. For RPA class III and nf = minor, there was a PFS benefit for dd TMZ (HR 0.73, p = 0.08; HR 0.77, p = 0.02). For nf = moderate subset, both ITT and SPT, the std arm showed superior OS (14.4 vs. 10.9 months) compared to dd, without improved PFS (HR 1.46, p = 0.03; and HR 1.74, p = 0.01. In terms of methylation status within this subset, there were more methylated patients in the std arm of the ITT subset (n = 159; 32 vs. 24%). For the SPT subset (n = 124), methylation status was similar between arms.Conclusions: This study did not demonstrate improved OS for dd TMZ for any subgroup, but for 2 highly functional subgroups, PFS was significantly increased. These data generate the testable hypothesis that intensive treatment may selectively improve disease control in those most likely able to tolerate dd therapy. Interpretation of this should be considered carefully due to small sample size, the process of multiple observations, and other confounders.Acknowledgment: This project was supported by RTOG grant U10 CA21661, and CCOP grant U10 CA37422 from the National Cancer Institute (NCI).