A mudança de comportamento dos consumidores através do advento da tecnologia e das redes sociais gera um grande empoderamento dos mesmos, alterando substancialmente a forma de relacionamento das empresas com seu público final. Atentas a este mercado, as empresas de mídia passam por profundas mudanças, tanto do ponto de vista da entrega de conteúdo ao seu público, quanto no seu formato administrativo, estratégico e financeiro. Sendo assim, a presente dissertação apresenta abordagens apoiadas em técnicas multivariadas para composição de equipes comerciais e de remuneração dos times de venda de uma empresa de comunicação. No artigo 1, objetiva-se gerar um modelo para estimar a premiação comercial das equipes de venda das rádios do Grupo RBS. Para tanto, inicialmente geram-se agrupamentos das emissoras de rádio do Grupo RBS no estado do Rio Grande do Sul e de Santa Catarina com base nos seus perfis de similaridades. Para cada cluster gerado, gera-se uma regressão linear múltipla da premiação comercial validado através de validação cruzada por intermédio do R2 ajustado e Mean Absolute Percentage Error (MAPE). O segundo artigo aborda a clusterização dos top clientes do Grupo RBS e o impacto na composição das equipes comerciais por meio do método da seleção de variáveis. As 7 variáveis originais foram avaliadas através do método de seleção de variáveis “Omita uma variável por vez”; o melhor Silhouette Index (SI) médio, métrica utilizada para avaliar a qualidade dos agrupamentos gerados, foi obtido quando 3 variáveis foram retidas. Os agrupamentos gerados por tais variáveis refletem o comportamento de compra de mídia dos clientes; os agrupamentos foram considerados satisfatórios quando avaliados por especialistas do Grupo RBS. / The change in the behavior of consumers with the advent of technology and social networks generates a great empowerment of themselves, substantially altering the relationship form of companies to their final audience. Attentive to this market, media companies undergo profound changes, both from the point of view of delivering content to their audience, as well as in their administrative, strategic and financial format. Thus, the present dissertation presents approaches supported by multivariate techniques for the composition of commercial and remuneration teams of the sales group of a communication company. In article 1, the objective is to generate a model to estimate the commercial awards of the sales teams of the RBS Group radios. To do this, we initially generate groupings of radio stations from the RBS Group in the state of Rio Grande do Sul and Santa Catarina based on their profiles of similarities. For each cluster generated, a multiple linear regression of the commercial award is generated, validated through cross validation through the adjusted R2 and Mean Absolute Percentage Error (MAPE). The second article addresses the clustering of RBS Group top clients and the impact on the composition of business teams through the variable selection method. The original 7 variables were evaluated through the variable selection method "Omit one variable at a time"; the best Silhouette Index (SI) average, metric used to evaluate the quality of the generated clusters, was obtained when 3 variables were retained. Clusters generated by such variables reflect customers' buying behavior of media; the clusters were considered satisfactory when evaluated by RBS Group experts.

Grâce à l'utilisation d'un porteur liquide, la projection plasma de suspension (SPS) permet la fabrication de revêtements finement structurés. Comme pour la projection plasma conventionnelle (APS), les microstructures des revêtements SPS peuvent être adaptées en contrôlant les conditions de projection. Cependant, le procédé SPS est plus compliqué que le procédé APS par son nombre de paramètres modifiables.Cette thèse vise à apporter une compréhension plus fondamentale de la relation entre les paramètres du procédé SPS et les propriétés des revêtements YSZ en identifiant des modèles génériques basés sur l’utilisation de méthodes statistiques mathématiques pour l'étude de l'influence et de la sensibilité de paramètres individuels.Des expériences systématiques ont été menées pour étudier l'influence de six paramètres (puissance du plasma, charge massique de suspension, taille de la poudre, distance de projection, pas de projection et rugosité du substrat) sur la microstructure des revêtements qui ont aussi été analysés en terme de propriétés d’usage (mécanique, thermique, tribologique, etc.). La porosité des revêtements a fait l’objet d’une étude approfondie et les mesures ont été réalisées par trois techniques différentes : la méthode par analyse d’images, la transmission RX et la méthode USAXS (Ultra-Small Angle X-ray Scattering). Des analyses multivariées sur les données expérimentales recueillies ont été effectuées et plusieurs modèles mathématiques ont été proposés afin de prédire les propriétés des revêtements et guider ensuite vers une optimisation de la microstructure du revêtement en vu d'applications spécifiques.Dans ce contexte d'optimisation des performances mécaniques et tribologiques de ces revêtements céramiques, différentes quantités et tailles de poudre h-BN ont été ajoutées dans la suspension YSZ. Les revêtements composites YSZ / h-BN ont été fabriqués et leur analyse a montré une nette réduction du coefficient de frottement et du taux d'usure lorsque la taille et la quantité de poudre d’ajout sont optimisées l'une avec l'autre. Trois mécanismes d’usure ont finalement été identifiés et seront discutés. / Thanks to the using of liquid carrier, suspension plasma spray (SPS) enables the manufacture of finely structured coatings. As for conventional plasma spraying (APS), the microstructures of SPS coatings can be tailored by controlling the spray conditions. However, SPS is more complicated than APS due to its number of modifiable parameters.This thesis aims to provide a more fundamental understanding of the relationship between SPS process parameters and the properties of YSZ coatings by identifying generic models based on the use of mathematical statistical methods for the study of influence and sensitivity of the individual parameters.Systematic experiments were carried out to study the influence of six parameters (plasma power, suspension mass load, powder size, projection distance, projection step and substrate roughness) on the microstructure of coatings which were also analyzed in terms of the properties (mechanical, thermal, tribological, etc.). The porosity of the coatings was studied in detail and the measurements were carried out using three different techniques: the image analysis method, the X-ray transmission and the USAXS (Ultra-Small Angle X-ray Scattering) method. Multivariate analyzes of the collected experimental data were performed and several mathematical models were proposed to predict the properties of the coatings and then guide towards an optimization of the microstructure of the coating for specific applications.In this context of optimizing the mechanical and tribological performance of ceramic coatings, different amounts and sizes of h-BN powder have been added in the YSZ suspension. The YSZ/h-BN composite coatings were manufactured by SPS process and their analysis showed a clear reduction in the coefficient of friction and the wear rate when the size and the amount of addition powder are optimized together. Three wear mechanisms have finally been identified and been discussed.

Diverse native seed mixtures have many benefits for prairie restoration or seeded pastures. In natural grasslands, species naturally coexist with hundreds of other species in complex communities. Commercial seed mixtures rarely contain more than a small number of species, often with haphazard ratios of the component species. Thus there is no natural template for combining selected species into an optimally productive community and there is limited knowledge on how to compose a suitable species mixture. Identifying which features of a community drive increased productivity may aid in screening species and community compositions, leading to mixtures that are more specifically designed to be stable, and highly productive for the region. There is renewed interest native species as they have the potential to provide non-invasive, productive, and drought resistant rangelands that may prove more sustainable. Seven species with high agronomic potential and a broad native geographic distribution were selected for testing including: nodding brome [Bromus anomalus (Coult.)], blue bunch wheatgrass [Pseudoregneria spicata (Pursh)], western wheatgrass [Pascopyrum smithii (Rydb.)], side oats grama [Bouteloua curtipendula (Michx.)], little blue stem [Schizachyrium scoparium (Michx.)], purple prairie clover [Dalea purpurea (Vent.)], and white prairie clover [Dalea candida (Willd.)]. The early productivity and nutritional quality of these species was determined in simple mixtures in two field sites: Saskatoon and Swift Current. In the field sites the mixtures included all seven monocultures, 21 two-species mixtures and a mixture with all species. Productivity may be driven by the species richness, functional group richness, and species evenness of the community, the abundance and occurrence of particular species or functional groups, and average plant trait values within the community. Therefore, identifying the features of a community that drive increased productivity and applying them as predictive tools may aid in screening species and community compositions. Many complex mixtures of the species were planted in greenhouse experiments to determine the strongest drivers of productivity for communities of these species. The experimental approach was validated in a confirmatory experiment where optimum communities were tested. These results did not differ under a moderate drought treatment. Results were generally consistent between field and greenhouse studies. Western wheatgrass (WWG) had the highest overall plant density and the strongest effect on the forage yield of the mixtures and communities. In the field study, productivity and crude protein content were not reduced when other species were also included with WWG in the mixture. Dalea spp. did not establish as well as the other species, but had the highest crude protein concentrations. The strongest predictors of productivity were the presence and abundance of perennial C3 grasses. Increases in species richness, functional group richness, and the presence of C3s (more specifically western wheatgrass) also increased productivity, likely because of the high early relative growth rate and strong competitive ability of western wheatgrass. Overall, communities screened in the greenhouse reflected early establishment field results. The systematic approach for evaluating communities can be modified to consider enhancing other ecological functions in addition to high productivity, in other regions.

The primary objective of the study is to evaluate the safety performance of different freeway exit types used in current practical designs. More specific, the research objectives include the following two parts: 1) to compare the safety performance of different design types at freeway diverge areas and exit ramp sections; and 2) to identify the impact factors contributing to the crashes happening at these two specific segments. The study area includes four subjects, the freeway widely-spaced diverge areas; the freeway closely-spaced diverge areas; the left-side off-ramps and the exit ramp sections. For the freeway diverge areas, design types were defined based on the number of lanes used by vehicular traffic to exit freeways and lane-balance theory. Four exit ramp types were considered for the widely-spaced diverge area, including single-lane exit ramps (Type 1), sing-lane exit ramps without a taper (Type 2), two-lane exit ramps with an optional lane (Type 3), and two-lane exit ramps without an optional lane (Type 4). For the closely-spaced diverge areas, three types, named as Type A, Type B and Type C, are selected to compare the safety performances among the three types. For the left-side off-ramp at the freeway diverge area, this study examined the two most widely used design types at the left-side freeway diverge areas in Florida, which are defined as Type I (one-lane left-side off-ramp), and Type II (two-lane left-side off-ramp). Type I is comparable to Type 1 design type and Type II is comparable to Type 3 design type at widely-spaced freeway diverge area. For the exit ramp sections, four ramp configurations, including diamond, out connection, free-flow loop and parclo loop, were considered. Cross-sectional comparisons were conducted to compare the crash frequency, the crash rate, the crash severity and target crash types between different design groups. Crash predictive models were also built to quantify the impacts of various contributing factors. The results of this study would expectedly help transportation decision makers develop tailored technical guidelines governing the selection of the optimum design combinations at freeway diverge areas and exit ramp sections.

FDA decisions and late-stage clinical trial results regarding new pharmaceutical approvals can cause extreme moves in the share price of small biopharmaceutical companies. Throughout the clinical trial process, many potential investors are exposed to market-moving information before such information is made available to the investing public. An investor who wished to profit from advance knowledge about clinical trial results may use the publicly traded options markets in order to increase leverage and maximize profits. This research examined options data surrounding the public release of information pertaining to the efficacy of clinical trials and approval decisions made by the FDA. Events were identified for small pharmaceutical companies with fewer than three currently approved drugs in an attempt to isolate the effect of individual clinical trial and FDA-related events on the share price of the underlying company. Option data were analyzed using logistic regression models in an attempt to predict phase II and III clinical trial outcome results and FDA new drug approval decisions. Implied volatility, open interest, and option contract delta values were the primary independent variables used to predict positive or negative event outcomes. The dichotomized version of a predictor variable designed to estimate total investment exposure incorporating open interest, option contract delta values, and the underlying stock price was a significant predictor of negative pharmaceutical related events. However, none of ii the variables examined in this research were significant predictors of positive drug research related events. The estimated total investment exposure variable used in this research can be applied to the prediction of future clinical trial and FDA decision related events when this predictor variable shows a negative signal. Additional research would help confirm this finding by increasing the sample size of events that potentially follow the same pattern as those examined in this research. / text

商業知識的發展突飛猛進，其中，預測模型在眾多商業智慧中扮演重要的角色，然而，當我們從大量資料萃取隱藏、未知與潛在具有實用性的資訊處理過程時，往往會遇到許多資料品質上的問題而難以著手分析，尤其是遺失值 (Missing value)的問題在資料前置處理階段更是常見的困難。因此，要如何在建立預測模型時有效的處理遺失值是一個很重要的議題。 過去已有許多文獻致力於遺失值處理的議題，其中，Active Feature-Value Acquisition的相關研究更針對訓練資料的選填順序深入探討。Active Feature-Value Acquisition的概念是從具有遺失值的訓練資料中，選擇適當的遺失資料填補，讓預測的模型在最具效率的情況下達到理想的準確率。本研究將延續Active Feature-Value Acquisition的研究主軸，優先考量決策樹上的節點為遺失值選值填補的順序，提出一個新的訓練資料遺失值的選填順序方法─I Sampling，並透過實際的數據進行訓練與測試，同時我們也與過去文獻所提出的方法進行比較，了解不同的填值偵測與順序的選擇對於一個預測模型的分類準確率是否有影響，並了解各個方法的優缺點與在不同情境下的適用性。 本研究所提出的新方法與驗證的結果，將可給予未來從事預測行為的管理或學術工作一些參考與建議，可以依據不同性質的資料採取合宜的選值方式，以節省取值的成本並提高預測模型的分類能力。 / The importance of business intelligence is accelerated developing nowadays. Especially predictive models play a key role in numerous business intelligence tasks. However, while we extract information from unidentified data, there are critical problems of how to handle the missing values, especially in the data pre-processing phase. Therefore, it is important to identify which methods best deal with the missing data when building predictive models. There are several papers dedicated in the research of strategies to deal with the missing values. The topic of Active-Feature Acquisition (aka. AFA) especially worked on the priority order of choosing which feature-value to acquire. The goal of AFA is to reduce the costs of achieving a desired model accuracy by identifying instances for which obtaining complete information is most informative. Followed by the AFA concept, we present an approach- I Sampling, in which feature-values are selected for acquisition based on the attribute on the top node of the current decision tree. Also we compare our approach with other methods in different situations and data missing patterns. Experimental results demonstrate that our approach can induce accurate models using substantially fewer feature-value acquisitions as compared to alternative policies in some situations. The method we proposed can aid the further predictive works in academic and business area. They can therefore choose the right method based on their needs and obtain the informative data in an efficient way.

Presently, within the human resource management (HRM) genre and including the construction management discipline, the identification and development of appropriate performance measures is seen as the only viable means for validating and engendering managerial excellence. There is also a growing awareness that appropriate predictive modelling practices can help engender the identification and development of these measures. Against the background that project-based sectors of the construction industry in developing countries need to adopt a proactive approach towards recognising and embedding performance measures in HRM practices, this thesis addresses the development of a model for predicting the performance of project managers (PMs) in mass house building projects (MHBPs) in Ghana. A literature review of the significance of performance measures in the HRM genre is first presented including an evaluation of the methodologies for measuring the performance of PMs. This is followed by a review of research and development in the management of human resources in the construction industry in developing countries including Ghana. Informed by the literature, an appropriate theoretical framework is adopted which draws on the organisational psychology theory of job performance, the conventional wisdom in project success criteria and an emerging framework of project lifecycle. Subsequently, a competency-based multidimensional conceptual model is developed. The conceptual model reflects both the elements of performance behaviours and outcomes in predicting the performance of PMs at the conceptual, design, tender, procurement, construction and operational phases of the project lifecycle. Adopting positivism as an appropriate research paradigm, structured questionnaire survey is used to elicit the relevant data from property developers in Ghana for the construction phase of the project lifecycle. Subsequently the data is analysed using one-sample t-test, factor analysis and multiple regression analysis (stepwise). From a broad range of competency-based measures used as independent variables, it is found that, the best predictors of the PMs’ performance at the “construction phase” of MHBPs are: job knowledge in site layout techniques for repetitive construction works; dedication in helping works contractors to achieve works programme; job knowledge of appropriate technology transfer for repetitive construction works; effective time management practices on the house-units; ability to provide effective solution to conflicts while maintaining good relationships; ease with which the PM is approachable by works contractors; and volunteering to help works contractors solve personal problems. These independent variables explained 74.4% of the variance in the model (at p < 0.0005). Validation of the model confirmed its goodness of fit and hence predictive accuracy. The findings suggest that at the construction phase of MHBPs, PMs who exhibit these behavioural competencies are likely to achieve higher levels of performance. Accordingly, PMs who aspire to achieve better managerial performance outcome on MHBPs should strive towards developing and improving these competencies. It is contended that the developed model could be used by property developers for the selection and recruitment of potential PMs and also for developing appropriate training requirements towards best practice improvement in the implementation of MHBPs. While the study focuses on Ghana, there is the potential for the model to be adopted for use by other developing countries towards the advancement of improved HRM activities in project management practice.

The recent advent of High-Throughput Sequencing (HTS) methods has triggered a revolution in gene regulation studies. Demand has never been higher to process the immense amount of emerging data to gain insight into the regulatory mechanisms of the cell. We address this issue by describing methods to analyze, integrate and interpret HTS data from different sources. In particular, we developed and benchmarked Pyicos, a powerful toolkit that offers flexibility, versatility and efficient memory usage. We applied it to data from ChIP-Seq on progesterone receptor in breast cancer cells to gain insight into regulatory mechanisms of hormones. Moreover, we embedded Pyicos into a pipeline to integrate HTS data from different sources. In order to do so, we used data sets from ENCODE to systematically calculate signal changes between two cell lines. We thus created a model that accurately predicts the regulatory outcome of gene expression, based on epigenetic changes in a gene locus. Finally, we provide the processed data in a Biomart database to the scientific community. / La llegada reciente de nuevos métodos de High-Throughput Sequencing (HTS) ha provocado una revolución en el estudio de la regulación génica. La necesidad de procesar la inmensa cantidad de datos generados, con el objectivo de estudiar los mecanismos regulatorios en la celula, nunca ha sido mayor. En esta tesis abordamos este tema presentando métodos para analizar, integrar e interpretar datos HTS de diferentes fuentes. En particular, hemos desarollado Pyicos, un potente conjunto de herramientas que ofrece flexibilidad, versatilidad y un uso eficiente de la memoria. Lo hemos aplicado a datos de ChIP-Seq del receptor de progesterona en células de cáncer de mama con el fin de investigar los mecanismos de la regulación por hormonas. Además, hemos incorporado Pyicos en una pipeline para integrar los datos HTS de diferentes fuentes. Hemos usado los conjuntos de datos de ENCODE para calcular de forma sistemática los cambios de señal entre dos líneas celulares. De esta manera hemos logrado crear un modelo que predice con bastante precisión los cambios de la expresión génica, basándose en los cambios epigenéticos en el locus de un gen. Por último, hemos puesto los datos procesados a disposición de la comunidad científica en una base de datos Biomart.

Cette thèse traite la fabrication de pièces composites par le procédé « Resin Transert Molding » (RTM), appliquée à des tubes de protections thermiques assemblées dans des propulseurs de systèmes d’armes. Ces travaux ont pour objectif de démontrer la faisabilité d’utilisation de ce procédé pour la fabrication de ces pièces complexes. C’est le préformage, première étape du procédé de fabrication par RTM, qui est étudié dans le cadre de cette thèse. Cette étape est cruciale du point de vue de la faisabilité de l’étape d’injection qui la suit dans le procédé RTM mais aussi pour s’assurer de la qualité de la pièce composite finale obtenue. L’objectif des travaux de thèse est triple. Il faut tout d’abord développer le protocole de fabrication répétable adapté pour garantir l’obtention de préformes conformes. Ce protocole devra être viable du point de vue industriel. Pour cela, une démarche expérimentale a été mise en place. Un pilote de laboratoire puis un pilote industriel ont permis de comprendre et maitriser les phénomènes survenant en cours de préformage en faisant varier les paramètres procédé pour la fabrication de nombreux prototypes. Un modèle macroscopique prédictif de la forme globale des plis obtenus à partir des paramètres procédés a été développé à l’aide des observations expérimentales. Un modèle mésoscopique, à l’échelle de la maille élémentaire, a été écrit également. Il permet de prédire, à partir des données constitutives du matériau et d’une géométrie de pièce, la déformation de compaction et de cisaillement, modes de sollicitations prépondérants en cours de préformage, subie par le renfort en cours de la première étape du procédé de fabrication. Ces modèles mésoscopique et macroscopique couplés permettent le développement d’un outil global qui, de manière théorique et prédictive, assure la faisabilité d’une pièce de géométrie connue avec un matériau connu et fournit les paramètres « procédé » optimum pour assurer sa fabrication future. Les phénomènes de déformation en cisaillement et compaction apparaissant sur la tresse en cours de préformage sont donc identifiés et connus. Le procédé de fabrication est optimisé et l’outil prédictif permet d’envisager et tester en amont un changement de matériau, de géométrie de pièce à fabriquer ou de cahier descharges industriel. / This study deals with the manufacture of composite parts by the process "Resin Transert Molding" (RTM), applied to thermal protection tubes. This work aims to demonstrate the feasibility of using this method for the production of these complex parts. This study deals with the first step of the RTM process, the fiber performing. This is critical from the standpoint of the feasibility of injecting step that follows in the RTM process but also to ensure the quality of the final composite part obtained. The aim of the thesis is threefold. Must first develop the manufacturing protocol adapted to ensure repeatable obtaining preforms compliant. This protocol should be viable to the industrial point of view. For this purpose, an experimental approach was implemented. A pilot laboratory and an industrial pilot helped to understand and master the phenomena occurring during forming varying the process parameters for the production of many prototypes. A macroscopic model predictive of overall shape folds obtained from the process parameters has been developed with the experimental observations. A mesoscopic model, the scale of the unit cell was also writing. It can predict, based on the specifications of the material and part geometry, the deformation of compaction and shear stresses. These models mesoscopic and macroscopic allow the development of a global tool that, theoretically predictive and ensures the feasibility of a piece of known geometry with a known material parameters and provides the "process" to ensure its optimum manufacturing future. The phenomena of compaction and shear strain appearing on the braid during preforming are identified and known. The manufacturing process is optimized and the predictive tool allows to explore and test upstream change of material, part geometry in manufacturing or industrial specifications.

CONTEXTUALIZAÇÃO: A preocupação com a qualidade da assistência tem aumentado nas últimas décadas em todo o mundo. O aumento da demanda, aliado à escassez de recursos financeiros e ao desenvolvimento e incorporação de novas tecnologias, tem suscitado reflexões e pesquisas que busquem avaliar a assistência hospitalar prestada em termos de custo-efetividade. Os estudos têm utilizado, na sua grande maioria, taxas de mortalidade hospitalar, que é um indicador tradicional de desempenho hospitalar. A análise comparativa de indicadores de desempenho pressupõe que as taxas de mortalidade sejam ajustadas às características dos pacientes e ao perfil do hospital, que também contribui na probabilidade de óbito hospitalar. Muitos autores têm utilizado bases de dados administrativas para avaliar estabelecimentos de saúde, principalmente pelo baixo custo e fácil disponibilidade. Diversos estudos internacionais têm analisado a eficiência dos serviços hospitalares de forma intensa e constante. No Brasil, os estudos ainda são poucos e a maioria tem avaliado diagnósticos específicos ou faixas-etárias específicas. Além disso, são poucos os que agregam o perfil dos hospitais na análise de predição do óbito hospitalar. OBJETIVO: O objetivo desta tese é desenvolver um índice de risco para óbito hospitalar ajustado pelas características das internações e pelo perfil dos hospitais a partir dos dados disponíveis no Sistema de Informações Hospitalar (SIH-SUS), com a finalidade de comparação de desempenho entre hospitais. É também objetivo desenvolver um modelo preditivo de probabilidade de óbito hospitalar utilizando a metodologia de modelo multinível. MÉTODOS: Trata-se de um estudo transversal com dados de 453.515 Autorizações de Internação Hospitalar (AIHs) do Sistema de Informações Hospitalares do Sistema Único de Saúde (SIH-SUS) do Rio Grande do Sul no ano de 2005. Utilizou-se regressão logística tradicional a fim de desenvolver um modelo preditivo das chances de óbito hospitalar considerando as características das internações. A seguir, foi realizada modelagem multinível buscando desenvolver um modelo preditivo das chances de óbito hospitalar considerando as características das internações e o perfil dos hospitais. Após o ajuste do modelo, foi calculado o Índice de Risco (IR), que permitiu o cálculo das probabilidades de óbitos hospitalares esperados (E), que foram comparados aos óbitos observados (O). O ordenamento do desempenho dos estabelecimentos foi realizado através da razão O/E em função da incorporação das características das internações (nível individual) e do perfil dos hospitais (nível contextual) conjuntamente no modelo preditivo. RESULTADOS: A taxa bruta de mortalidade para o conjunto dos 332 hospitais (453.515 AIHs) foi de 6,3%. A mortalidade foi maior para os homens. As doenças infecciosas e parasitárias, neoplasias, doenças do sistema nervoso, do aparelho circulatório e respiratório e, ainda, diagnósticos informados como sinais e sintomas anormais foram os que apresentaram significativamente maior número de óbitos do que o esperado através do teste Qui-quadrado. A especialidade clínica médica apresentou maior número de óbitos em comparação à especialidade cirurgia. A maioria das internações ocorreu em hospitais privados, enquanto que a taxa bruta de mortalidade foi maior nos hospitais públicos. Através da modelagem por regressão logística, utilizando o perfil das internações, obteve-se um Índice de Risco (IR) para mortalidade hospitalar. A partir do modelo preditivo foram calculados os óbitos esperados para os hospitais. Dos 206 hospitais analisados, a razão O/E (óbito observado/óbito esperado) mostrou 40 hospitais com mortalidade significativamente superior à esperada e 58 hospitais com mortalidade significativamente inferior à esperada. A partir do modelo preditivo multinível, formado por variáveis explicativas referentes à internação (primeiro nível) e variáveis explicativas referentes ao hospital (segundo nível), verificou-se que o perfil dos hospitais tem papel importante na predição do óbito hospitalar. As variáveis uso de UTI, seguida por idade foram as principais preditoras para óbito hospitalar no nível individual e porte do hospital, seguida por natureza jurídica, o foram no nível contextual respectivamente. A razão O/E baseada no modelo multinível mostrou que os hospitais de pequeno porte tem pior desempenho, os de grande porte melhoram seu desempenho e os de médio porte mantiveram-se praticamente sem modificações, quando comparados ao desempenho medido pela razão O/E obtida apenas para as características das internações. Constatou-se, ainda, um melhor desempenho dos estabelecimentos públicos, para todos os portes, e pior desempenho para os hospitais privados CONCLUSÕES: O índice de risco construído a partir das características da internação e do perfil dos estabelecimentos por modelos multinível pode ser empregado na análise de desempenho dos hospitais do SIH-SUS. O IR construído permitirá calcular a probabilidade de óbito e assim obter a taxa ajustada de mortalidade, a ser usada como um indicador de desempenho. Esta metodologia mostrou-se útil para rastrear hospitais que merecem uma atenção maior por parte de gestores, prestadores de serviços, profissionais e comunidade. A ordenação dos hospitais utilizando apenas a taxa de mortalidade bruta não é igual à ordenação quando se utiliza o ranking ajustado pelo modelo preditivo de probabilidade para o nível de internações, e esse último também não é igual quando se adiciona o nível dos hospitais. Recomenda-se que, ao comparar hospitais, seja utilizado o ajuste pelo modelo preditivo de probabilidade de risco que incorpora tanto o nível das internações, quanto dos hospitais. Estudos acrescentando outras variáveis do nível de internações, do nível hospitalar, além da região, poderão contribuir para o aprimoramento do modelo e do índice de risco. O desenvolvimento de uma série histórica de acompanhamento, bem como a discussão com representantes de várias instâncias envolvidas no processo de avaliação hospitalar poderão aumentar a eficiência do método. / CONTEXTUALIZATION: The concern with the quality of care has increased in recent decades throughout the world. Increased demand, combined with the scarcity of financial resources and the development and incorporation of new technologies, has raised debate and research that seek to evaluate the hospital care provided in terms of costeffectiveness. Studies have mostly used hospital mortality rates, which is a traditional indicator of hospital performance. Comparative analysis of performance indicators means that mortality rates are adjusted to the characteristics of patients and to the hospital profile, which also contributes to the risk of death in hospital. Many authors have used administrative databases to assess health institutions, especially for their low cost and easy availability. Several international studies have analyzed the efficiency of hospital services in intense and constant way. In Brazil, studies are still few and most have evaluated specific diagnoses or specific age ranges. Moreover, few studies add the profile of hospitals to the analysis of prediction of hospital death. OBJECTIVE: The objective of this thesis is to develop a risk index for hospital death adjusted by characteristics of hospital admissions and by the profile of hospitals, using the available data in the SIH-SUS, for the purpose of comparison of performance between SUS hospitals. It also aims to develop a multilevel model of hospital risk of death. METHODS: This is a cross-sectional study with data from 453.515 Authorization Form for Hospital Admittance (AIHs) of the Hospital Information System of the Unified Health System (SIH-SUS) in Rio Grande do Sul in 2005. A traditional logistic regression was used to develop a predictive model of the chances of hospital death considering the characteristics of hospital admissions. Additionally a multilevel modeling was employed to develop a predictive model of the chances of death considering the characteristics of hospital admissions and hospital profiles. After fitting the model, the risk index (IR) was calculated, which allowed for the calculation of the likelihood of hospital expected deaths (E), which were then compared to the observed deaths (O). The performance ranking of the establishments was conducted through the ratio O/E depending on the incorporation of characteristics of hospital (individual level) and the profiles of hospitals (contextual level) together in the predictive model. RESULTS: The crude death rate for all 332 hospitals (453.515 AIHs) was 6.3%. Mortality was higher for men. Infectious and parasitic diseases, neoplasms, diseases of the nervous system, of the circulatory and of the respiratory apparatus, and also informed diagnoses as abnormal signs and symptoms were those that had significantly more deaths than expected by the chi-square test. Higher number was observed for the speciality medical clinic of deaths compared to surgery. Most hospitalizations occurred in private hospitals, while the crude death rate was higher in public hospitals. Through the RL model, by using the profile of hospitalizations, a Risk Index (IR) was obtained for hospital mortality. From the predictive model were calculated expected deaths for hospitals. In 40 out of the 206 hospitals studied, the ratio O/E (observed deaths / expected deaths) showed mortality rates significantly higher than expected and, in 58 hospitals the mortality rates were significantly lower than expected. As for the multilevel predictive model, consisting of explanatory variables related to hospitalization (first level) and explanatory variables for the hospital (second level), the profiles of hospitals had an important role in prediction of hospital death. The variable use of Intensive Care Unit (UTI), followed by patient age, were the main predictors for hospital death at the individual level and size of the hospital, followed by a legal nature were the more important variables for the contextual level. The ratio O/E based on the multilevel model showed that small hospitals had a worse their performance, large institutions had better performances and those of medium size virtually unchanged when compared to the ratio O/E only for the characteristics of admissions It was also verified an improvement of performance of the public hospitals, for all sizes, and worsening of performance for private hospitals. CONCLUSIONS: The risk index constructed from the characteristics of hospitalization and the profile of establishments by multilevel models can be used in the analysis of performance of the SIH-SUS hospitals. The presently developed IR will yield a probability of death and thereby an adjusted rate of mortality, to be used as an indicator of performance. This methodology proved to be useful to track hospitals that deserve greater attention from managers, providers, professionals and community. The ordering of the hospitals using only the crude mortality rate is not equal to the ordering that uses the ranking set by the predictive model of probability for the level of admissions, and the latter is not equal when it adds the level of hospitals. When comparing hospitals, it is recommended the use of adjustment of the predictive model of probability of risk that incorporates both the levels of admissions and of the hospitals. Studies adding other variables in the level of admissions, the hospital level, as well as the region, could contribute to the improvement of the model and the risk index. The development of a historical series of monitoring and discussion with representatives of various groups involved in hospital evaluation will add validity to the assessment method.