Thesis (MSc. (Faculty of Health Sciences))--University of Pretoria, 2007. / Summary in English and Afrikaans. Includes bibliographical references.

Σε αυτή την εργασία ασχοληθήκαμε με δύο μεθόδους, που σκοπός τους είναι να κατατάσσουν τις παρατηρήσεις σε γνωστές ομάδες και στη συνέχεια να κάνουν προβλέψεις για καινούριες παρατηρήσεις. Αυτές οι μέθοδοι είναι η λογιστική παλινδρόμηση (logistic regression) και η διαχωριστική ανάλυση (discriminant analysis). Στο πρώτο κεφάλαιο αναφέραμε περιληπτικά τα μη γραμμικά μοντέλα παλινδρόμησης (αφού και η λογιστική παλινδρόμηση είναι ένα τέτοιο μοντέλο). Απλά αναφέρουμε τη μορφή που έχουν αυτά τα μοντέλα, με ποιες μεθόδους μπορούμε να εκτιμήσουμε τις παραμέτρους παλινδρόμησης, ποια είναι τα διαστήματα εμπιστοσύνης για τους συντελεστές παλινδρόμησης και τη μορφή που θα έχουν οι έλεγχοι υποθέσεων. Στο δεύτερο κεφάλαιο περιγράφουμε τη λογιστική παλινδρόμηση. Η λογιστική παλινδρόμηση είναι χρήσιμη σε καταστάσεις στις οποίες επιθυμούμε να προβλέψουμε την ύπαρξη ή την απουσία ενός χαρακτηριστικού ή ενός συμβάντος. Η πρόβλεψη αυτή βασίζεται στην κατασκευή ενός μοντέλου και συγκεκριμένα στον προσδιορισμό των τιμών που παίρνουν οι συντελεστές. Αυτή η μέθοδος είναι μια γενίκευση της απλή γραμμικής παλινδρόμησης για την περίπτωση όπου η εξαρτημένη μεταβλητή είναι δίτιμη (παίρνει την τιμή 0 όταν το χαρακτηριστικό απουσιάζει και την τιμή 1 όταν υπάρχει το χαρακτηριστικό). Στο τρίτο κεφάλαιο αναλύουμε τη διαχωριστική ανάλυση, η οποία έχει δύο στόχους: να χωρίσει ένα πληθυσμό σε ευδιάκριτες ομάδες και με τη βοήθεια ενός διαχωριστικού κανόνα να κατατάσσει παρατηρήσεις στις ευδιάκριτες ομάδες. Στο τέλος του κεφαλαίου περιγράφουμε τις ομοιότητες και τις διαφορές της διαχωριστικής ανάλυσης και της λογιστικής παλινδρόμησης. Στο τέταρτο και τελευταίο κεφάλαιο απλά δίνουμε ένα παράδειγμα που το λύνουμε με τη μέθοδο της λογιστικής παλινδρόμησης και ένα παράδειγμα που το λύνουμε με τη μέθοδο της διαχωριστικής ανάλυσης. Αυτό το κάνουμε με τη βοήθεια του στατιστικού πακέτου SPSS. / In this work we dealt with two methods, that their aim are to classify the observations in known teams and afterwards to make forecasts for new observations. These methods are the accountant regression (logistic regression) and the bisector analysis (discriminant analysis).

To PTEN δρα ως ογκοκατασταλτικό γονίδιο, μέσω της δράσης του προϊόντος πρωτεΐνης της φωσφατάσης. Η φωσφατάση εμπλέκεται στη ρύθμιση του κυτταρικού κύκλου εμποδίζοντας τα κύτταρα να αναπτυχθούν και έχοντας σαν αποτέλεσμα την υπερβολικά γρήγορη διαίρεση. Το γονίδιο αυτό έχει ταυτοποιηθεί ως ογκοκατασταλτικό και σε αρκετές περιπτώσεις καρκίνων έχουν εντοπιστεί μεταλλάξεις του. Στην παρούσα διπλωματική εργασία, θα μελετηθεί η δράση του PTEN στο αδενοκαρκίνωμα του παχέως εντέρου και θα εξεταστεί η δυνατότητα χρήσης των επιπέδων έκφρασής του σαν βιοδείκτη για το συγκεκριμένο είδος καρκίνου. Σκοπός της παρούσας διπλωματικής είναι με χρήση κλινικών δεδομένων και της έντασης της έκφρασης της πρωτεΐνης του γονιδίου PTEN, να μελετηθεί με στατιστικές μεθόδους η δυνατότητα πρόβλεψης της βαθμοποίησης και της σταδιοποίησης του αδενοκαρκινώματος του παχέος εντέρου. Για τον σκοπό αυτό χρησιμοποιήθηκαν πραγματικά κλινικά δεδομένα 60 ασθενών που προήλθαν από το κυτταρολογικό εργαστήριο του Νοσηλευτικού Ιδρύματος Μετοχικού Ταμείου Στρατού 417 (Ν.Ι.Μ.Τ.Σ.). Τα δεδομένα αυτά αναλύθηκαν με εφαρμογή της περιγραφικής στατιστικής, της λογιστικής παλινδρόμησης και της πολυμεταβλητής παλινδρόμησης με χρήση του στατιστικού πακέτου R. Παρόλα αυτά κανένα από τα εξαγόμενα στατιστικά μοντέλα δεν βρέθηκε να μπορεί να συσχετίσει την ποσότητα έκφρασης του PTEN γονιδίου με τη βαθμοποίηση και τη σταδιοποίηση του αδενοκαρκινώματος του παχέος εντέρου. Για αυτό τον λόγο καταλήγουμε ότι με χρήση των συγκεκριμένων βιολογικών δεδομένων δεν μπορούμε να επαληθεύσουμε ότι το PTEN γονίδιο είναι βιοδείκτης του αδενοκαρκινώματος του παχέος εντέρου. Μελλοντικά αυτό πρέπει να διερευνηθεί περαιτέρω είτε με τη χρήση επιπλέον κλινικών δεδομένων καθώς το υπάρχον σύνολο δεδομένων περιέχει λίγα δείγματα ασθενών σε σχέση με τις ελεύθερες μεταβλητές του προβλήματος (small sample size problem), είτε με μεθόδους που αυξάνουν τεχνητά τα δείγματα του συνόλου δεδομένων. / PTEN tumor suppressor gene acts as through the action of the protein product of phosphatase. The phosphatase is involved in regulating the cell cycle by preventing cells to grow and the resulting too rapid division. This gene has been identified as tumor suppressor in several cancers identified mutations. In this paper, we studied the effect of PTEN in adenocarcinoma of the colon and consider the possibility of using the levels of expression as biomarker for this type of cancer. The purpose of this thesis is to use clinical data and the intensity of the protein expression of the gene PTEN and be studied using statistical methods for predict the grade and stage of adenocarcinoma of the colon. For this purpose we used real clinical data of 60 patients who came from the cytology laboratory of the hospital in the Army Pension Fund 417 (N.I.M.T.S.). The data were analyzed using the descriptive statistics, regression and multivariate regression using the statistical package R. However none of the extracted statistical models were found to be correlated to the amount of expression of PTEN gene with grade and stage of adenocarcinoma of the colon. For this reason, we conclude that the use of these biological data can not verify that the PTEN gene is a biomarker of adenocarcinoma of the colon. Future should be investigated further or using additional clinical data as the existing data set contains few patient samples relative to the free variables of the problem (small sample size problem), or by methods that increase artificially the samples in the data set.

Made available in DSpace on 2014-06-11T19:35:43Z (GMT). No. of bitstreams: 0 Previous issue date: 2009-07-16Bitstream added on 2014-06-13T19:05:57Z : No. of bitstreams: 1 freitas_ch_dr_rcla.pdf: 1423658 bytes, checksum: 1cf47ffae9bf2508e9b8769ad2cd9369 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / As estradas causam a mortalidade de animais, bem como a remoção de hábitats, a fragmentação da paisagem, o efeito de borda e criam uma barreira física a muitas espécies. Alguns fatores estão relacionados aos atropelamentos, como por exemplo tipo de acostamento, topografia e vegetação. O desenho da estrada (curva ou reta) tem influência na velocidade dos veículos e é um importante fator condicionante que necessita ser considerado. Além disso, os atropelamentos podem atrair animais necrófagos ou domésticos (cães e gatos) para as carcassas, aumentando a probabilidade de novos atropelamentos. Embora vários estudos sobre atropelamentos de vertebrados tenham sido publicados no mundo, poucos tem documentado a mortalidade no Brasil. Nós realizamos um estudo que focou nos fatores causais bem como no valor (Disposição A Pagar – DAP) dos animais mortos em colisões com veículos. Nossa proposta foi documentar a composição de espécies e a riqueza e compreender os fatores espaciais e temporais relacionados a mortalidade na rodovia. Nós registramos e identificamos as espécies e o número de vertebrados atropelados em dois trechos de rodovias estaduais (97 Km da MG-428 e 63 Km da SP-334) no sudeste do Brazil. As rodovias foram amostradas semanalmente, de janeiro a dezembro de 2007. Além disso medidas de variáveis explanatórias relacionadas as estradas foram incluídas: tipo de acostamento, topografia, tipo de vegetação e desenho da estrada. Nós usamos estes dados para examinar a relação entre as variáveis explanatórias e a mortalidade dos grupos de vertebrados nas estradas ao longo do tempo. Foi realizado o teste do qui-quadrado de aderência para examinar a distribuição espacial dos atropelamentos. Nós também entrevistamos 601 motoristas para obter a disposição a pagar (DAP) pela conservação da fauna. Usamos a regressão logística para estabelecer... / Roads result in mortality on the road as well as habitat removal, landscape fragmentation, edge effects, and pose a physical barrier to many species. Some roadside factors are related to road-kills, e.g., type of shoulder, topography, and vegetation. Road alignment (curved or straight) has an influence on vehicle speed and is an important conditioning factor needing consideration. Additionally, road-kills may attract scavengers or domestic animals (cats and dogs) to carcasses, increasing the probability of new road-kills. Although many studies on vertebrate road-kill have been published around the world, few have documented road mortality in Brazil. We conducted a study that focused on causal factors as well as the value (Willingness To Pay) of animals killed by vehicle collisions. Our purpose was to document species composition and richness and to understand the spatial and temporal patterns of roadrelated mortality. We recorded the number and species identification of vertebrates killed on two state roads segments (63 Km of SP - 334 and 97 Km of MG - 428) in southeastern Brazil. The roads were surveyed weekly from January to December 2007. Additional road related explanatory variables measured, included: type of shoulder, topography, vegetation type, and road design. We used that data to examine correlations between the explanatory variables and the mortality of vertebrate groups on the roads over time. We performed the chi-square goodness of fit to examines the spatial road-kills distribution. We also interviewed 601 vehicle drivers to assess their Willingness To Pay (WTP) for fauna conservation. We used logistic regression to relate the independent variables with the location and number of mortalities we recorded. We used linear correlation to examine how certain aspects of the user profile (namely, monthly income, education level, and frequency of road use) were related... (Complete abstract click electronic access below)

Made available in DSpace on 2015-03-26T13:32:11Z (GMT). No. of bitstreams: 1 texto completo.pdf: 852842 bytes, checksum: 70e1222560798c97f05cf66d73772591 (MD5) Previous issue date: 2011-02-25 / Logistic regression is the analysis usual statistical method used to verify the relationship between a dichotomous variable response and the interest explanatory variables. This work aimed to carry out a study about the factors influencing the invertebrates abundance on the soil under different management forms, using the logistic regression. This objective is that these invertebrates are considered excellent indicators of the use type and soil quality, working in several fundamental processes for maintaining the soil fertility and quality in agroecosystems and natural ecosystems, according to Brown et al. (1998), Hendrix et al. (2006), and Souza (2010). For covariates selection, the Collett (1994) proposal was used and the involved parameters estimators in each model, their interpretations, statistical properties, and some criteria for judging the suitability of the selected models were presented. The methodology presented by this work was applied to two real datasets (dry and rainy season). In the final adjusted model for the analyzed dataset in the dry season, it was verified that the covariates System Type, Calcium in litter, Soil organic matter, Potassium in litter, and the interaction between Calcium and Potassium in litter were important to explain the presence of more than nine individuals on the soil. In the final adjusted model for the analyzed dataset in the rainy season, the significant covariates to explain the presence of one hundred and one individuals on average on the soil were Magnesium in litter, Total organic carbon in the litter, Litter organic matter, and Ambient temperature. For two mentioned models, there were a good discriminatory performance and excellent areas under the ROC (Receiver Operating Characteristic) curve, thus confirming the validity of using logistic regression techniques for the models construction to describe the analyzed data. / A regressão logística é o método estatístico usual de análise utilizado com a finalidade de verificar a relação entre uma variável resposta dicotômica e variáveis explicativas de interesse. Este trabalho teve como objetivo realizar um estudo sobre os fatores que influenciam a abundância de invertebrados no solo sob diferentes formas de manejo utilizando a Regressão Logística. Tal objetivo reside no fato destes invertebrados serem considerados excelentes indicadores do tipo de uso e qualidade do solo, atuando em vários processos fundamentais para a manutenção da fertilidade e qualidade dos solos de agroecossistemas e ecossistemas naturais de acordo com Brown et al. (1998) e Hendrix et al. (2006), citado Souza (2010). Para seleção de covariáveis foi utilizada a proposta de Collett (1994) e foram apresentados estimadores dos parâmetros envolvidos em cada modelo e suas interpretações, propriedades estatísticas e critérios para se julgar a adequabilidade dos modelos selecionados. A metodologia apresentada neste trabalho foi aplicada a dois conjuntos de dados reais (período seco e chuvoso). No modelo final ajustado para o conjunto de dados analisado no período seco verificou-se que as covariáveis Tipo de Sistema, Cálcio em serapilheira, Matéria orgânica do solo, Potássio em serapilheira e a interação entre Cálcio e Potássio em serapilheira foram importantes para explicar a presença de mais de 9 indivíduos, em média, no solo. Já no modelo final ajustado para o conjunto de dados analisado no período chuvoso, as covariáveis significativas para explicar a presença de 101 indivíduos, em média, no solo foram Magnésio em serapilheira, Carbono orgânico total na serapilheira, Matéria orgânica da serapilheira e Temperatura ambiente. Para os dois modelos citados houve bom desempenho discriminatório e excelentes áreas sob a curva ROC, confirmando assim a validade da utilização de técnicas de regressão logística na construção dos modelos para descrever os dados analisados.

Made available in DSpace on 2015-05-14T12:47:22Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 750727 bytes, checksum: 1d1eb8f741a35fb035d4d1eaeb11c309 (MD5) Previous issue date: 2011-02-24 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Respiratory diseases in childhood, particularly pneumonia, are a serious public health problem, since they are directly related to morbidity and mortality of children in Brazil and worldwide. World Health Organization data show that the annual incidence of pneumonia is 156 million new cases, of which 151 million occur in developing countries. Of these, 20 million require hospitalization because of the gravity. This study aims to identify risk factors associated with hospitalization for pneumonia in children between 0 and 2 years old, as well as build a decision model that can explain the occurrence of pneumonia in this age group. This is a case-control study conducted in a referral hospital in João Pessoa PB. The study included 186 children 60 pneumonia cases and 126 controls. We used the chi-square test, and then multivariate analysis using logistic regression model. In the chi-square test, the following variables were considered significant: maternal age, type of delivery, birth weight, breastfeeding, parity, previous hospitalization, reason for hospitalization, vaccination, child care, overcrowding, access to sewage treatment, health unit, type of household and maternal education. In the final decision model, were considered significant the variables previous hospitalization and maternal education as risk factors, and access to vaccination and treatment of sewage as protective factors. These study results corroborate previous studies and provide information for health managers to develop intervention programs on these factors with the aim of reducing morbidity due to pneumonia in children. / As doenças respiratórias da infância, e em especial a pneumonia, constituem-se em um grave problema de saúde pública, pois estão diretamente relacionadas à morbidade e mortalidade de crianças, no Brasil e no mundo. Dados da organização mundial de saúde indicam que a incidência anual de pneumonia é de 156 milhões de casos novos, dos quais 151 milhões ocorrem nos países em desenvolvimento e destes, 20 milhões necessitam internação hospitalar devido à gravidade. Este estudo tem como objetivo identificar fatores de risco associados à internação de crianças com pneumonia na faixa etária entre 0 e 2 anos de idade, bem como construir um modelo de decisão capaz de explicar a ocorrência de pneumonia nesta faixa etária. Trata-se de um estudo caso-controle, realizado em um hospital de referência em João Pessoa-PB. Foram incluídos no estudo 186 crianças, sendo 60 casos de pneumonia e 126 controles. Para análise estatística foram utilizados o teste qui-quadrado e o modelo de regressão logística. O teste qui-quadrado apontou como significativas as variáveis: idade materna, tipo de parto, peso ao nascer, amamentação, paridade, internação anterior, motivo de internação, vacinação, creche, aglomeração, acesso a tratamento de esgoto, unidade de saúde, tipo de domicílio, e escolaridade materna. No modelo de decisão final foram consideradas significativas as variáveis: internação anterior e escolaridade materna como fatores de risco e vacinação e acesso a tratamento de esgoto, como fatores de proteção. Os resultados deste estudo corroboram com estudos anteriores e fornecem informações para que os gestores de saúde desenvolvam programas de intervenção sobre estes fatores com o objetivo de diminuir a morbidade por pneumonia em crianças.

Rapid deployment of terrestrial wind power plants (WPPs) is a function of accurate identification of areas suitable for WPPs. Efficient WPP site prospecting not only decreases installation lead time, but also reduces site selection expenses and provides faster reductions of greenhouse gas emissions. Combining conventional predictor variables, such as wind strength and proximity to transmission lines, with nonconventional socioeconomic and demographic predictor variables, will result in improved identification of suitable counties for WPPs and therefore accelerate the site prospecting phase of wind power plant deployment. Existing and under-construction American terrestrial WPPs located in the top 12 windiest states (230 as of June 2009) plus 178 potential county level predictor variables are introduced to logistic regression with stepwise selection and a random sampling validation methodology to identify influential predictor variables. In addition to the wind resource and proximity to electricity transmission lines, existence of a Renewable Portfolio Standard, the population density within a 200 mile radius of the county center, median home values, and farm land area in the county are the four strongest nonconventional predictors (Hosmer and Lemeshow Chi-Square = 9.1250, N = 1009, df = 8, p = 0.3319, - 2LogLikelihood = 619.521). Evaluation of the final model using multiple statistics, including the Heidke skill score (0.2647), confirms overall model predictive skill. The model identifies the existence of 238 suitable counties in the twelve state region that do not possess WPPs (~73% validated overall accuracy) and eliminates 654 counties that are not classified as suitable for WPPs. The 238 counties identified by the model represent ideal counties for further exploration of WPP development and possible transmission line construction. The results of this study will therefore allow faster integration of renewable energy sources and limit climate change impacts from increasing atmospheric greenhouse gas concentrations.

Brain injury has a tremendous effect on the United States. The medical system has a continuum of care available but many of these services are extremely expensive. Despite the effectiveness of residential post-acute brain injury rehabilitation (PABIR) resistance to provide adequate funding remains because of a dearth of randomized controlled trial (RCT) studies demonstrating effectiveness. Some research suggests observational trials are typically more representative of community samples and yield conclusions similar to RCT studies. This study uses a large multi-state naturalistic community-based sample of individuals who received residential PABIR. The purposes of this study were to (1) use logistic regression to identify a model that considered the relationships among the predictor variables to explain treatment outcome for individuals receiving residential PABIR and (2) better understand how self-awareness influences treatment outcome. The final model contained five independent variables (substance use at time of admit, functioning level at time of admit, change in awareness between discharge and admit, admit before or after 6 months post-injury (TPI), and length of stay (LOS) in the program less than or greater than 2 months). The model was statistically significant, ÷2 (5, N=434) = 194.751, p < .001, accounting for 36.2% (Cox & Snell R square) to 61.3% (Nagelkerke R square) of the variance in success rate, and correctly classified 89.4% of cases. Four of the five predictor variables (current substance use, change in awareness, LOS 2 months and TPI 6 months) made statistically significant contributions to the model. The strongest predictor of successful treatment outcome was change in awareness recording an odds ratio of 29.9 indicating that individuals who improved in self-awareness by at least one level were nearly 30 times more likely to be in the successful outcome group, controlling for other factors in the model. Participants were also more likely to be in the successful outcome group if they admitted within 6-months post-injury (5.5x) and stayed longer than 2-months (4.4x). Findings also suggest that active substance use at time of admission did not prevent people from being successful. Importance and implications of these findings are discussed.

[pt] Este trabalho tem como objetivo principal adaptar o modelo STR-Tree, o qual é a combinação de um modelo Smooth Transition Regression com Classification and Regression Tree (CART), a fim de utilizá-lo em Classificação. Para isto algumas alterações foram realizadas em sua forma estrutural e na estimação. Devido ao fato de estarmos fazendo classificação de variáveis dependentes binárias, se faz necessária a utilização das técnicas empregadas em Regressão Logística, dessa forma a estimação dos parâmetros da parte linear passa a ser feita por Máxima Verossimilhança. Assim o modelo, que é paramétrico não-linear e estruturado por árvore de decisão, onde cada nó terminal representa um regime os quais têm seus parâmetros estimados da mesma forma que em uma Regressão Logística, é denominado Smooth Transition Logistic Regression-Tree (STLR-Tree). A inclusão dos regimes, determinada pela divisão dos nós da árvore, é feita baseada em testes do tipo Multiplicadores de Lagrange, que em sua forma para o caso Gaussiano utiliza a Soma dos Quadrados dos Resíduos em suas estatísticas de teste, aqui são substituídas pela Função Desvio (Deviance), que é equivalente para o caso dos modelos não Gaussianos, cuja distribuição da variável dependente pertença à família exponencial. Na aplicação a dados reais selecionou-se dois conjuntos das variáveis explicativas de cada uma das duas bases utilizadas, que resultaram nas melhores taxas de acerto, verificadas através de Tabelas de Classificação (Matrizes de Confusão). Esses conjuntos de variáveis foram usados com outros métodos de classificação existentes, são eles: Generalized Additive Models (GAM), Regressão Logística, Redes Neurais, Análise Discriminante, k-Nearest Neighbor (K-NN) e Classification and Regression Trees (CART). / [en] The main goal of this work is to adapt the STR-Tree model, which is the combination of a Smooth Transition with Regression model with Classi cation and Regression Tree (CART), in order to use it in Classification. Some changes were made in its structural form and in the estimation. Due to the fact we are doing binary dependent variables classification, is necessary to use the techniques employed in Logistic Regression, so the estimation of the linear part will be made by Maximum Likelihood. Thus the model, which is nonlinear parametric and structured by a decision tree, where each terminal node represents a regime that have their parameters estimated in the same way as in a Logistic Regression, is called Smooth Transition Logistic Regression Tree (STLR-Tree). The inclusion of the regimes, determined by the splitting of the tree's nodes, is based on Lagrange Multipliers tests, which for the Gaussian cases uses the Residual Sum-of-squares in their test statistic, here are replaced by the Deviance function, which is equivalent to the case of non-Gaussian models, that has the distribution of the dependent variable in the exponential family. After applying the model in two datasets chosen from the bibliography comparing with other methods of classi cation such as: Generalized Additive Models (GAM), Logistic Regression, Neural Networks, Discriminant Analyses, k-Nearest Neighbor (k-NN) and Classification and Regression Trees (CART). It can be seen, verifying in the Classification Tables (Confusion Matrices) that STLR-Tree showed the second best result for the overall rate of correct classification in three of the four applications shown, being in all of them, behind only from GAM.

Introduction: Obstructive sleep apnea (OSA) is a common breathing disorder with numerous health consequences, including greater risk of complications perioperatively. Undiagnosed OSA is known to place surgical patients at a higher risk of serious adverse events, including stroke and death. Polysomnography (PSG) assessment is the current gold standard test for diagnosing OSA. However, due to the significant time commitment and cost associated with PSG, a substantial number of OSA patients go undiagnosed before the perioperative period. Although the STOP-Bang questionnaire screening tool is currently used to help detect OSA patients, the low specificity to screen people without the disease is considered a major limitation. There is a clear need to develop a quick and effective prediction rule with higher overall accuracy to help streamline OSA diagnosis. Tracheal breathing sound analysis in awake patients at the bedside has shown potential to screen OSA patients with higher specificity compared to the STOP-Bang questionnaire. To date, no screening tools exist to detect OSA patients that combine the results of breathing sound analysis and STOP-Bang. Objectives: The present study aimed to develop a prediction rule, using both breathing sound analysis and variables in the STOP-Bang questionnaire, to better streamline the diagnosis of OSA. Methods: This prospective cohort study recruited patients referred for PSG at the Ottawa Hospital Sleep Centre from November 2016 to May 2017. The study conduct was approved by the Ottawa Health Science Network Research Ethics Board (#20160494-01H). After obtaining informed consent, anthropomorphic, breathing sound recordings, and STOP-Bang questionnaire data was collected from over 400 consenting patients. All patients that met the eligibility criteria were included. The breathing sound analysis and STOP-Bang results were utilized to design a prediction rule using logistic regression. Sensitivity, specificity, and likelihood ratio were used to compare the diagnostic performance of the final model. Results: Of the 439 consenting study participants, 280 study participants data were eligible for inclusion in the logistic regression analysis. Physician sleep specialists diagnosed 114 participants (41%) with moderate-to-severe OSA and 166 participants (59%) with normal-to-mild OSA. At a predicted probability of moderate-to-severe OSA greater than or equal to 0.5, breathing sound analysis had a similar sensitivity of 75.9 (95%CI; 65.4, 82.0) and higher specificity of 74.5% (95%CI; 68.5, 82.0) when compared to STOP-Bang with a sensitivity and specificity of 68.4% (95%CI; 58.9, 76.6) and 63.2% (95%CI: 55.0, 70.1), respectively. The sensitivity and specificity for the Safe-OSA rule, obtained by combining breathing sound analysis and STOP-Bang variables, were determined to be 75.4% (95%CI; 65.4, 82.0) and 74.5% (95%CI; 68.5, 82.0), respectively. A sensitivity analysis using a likelihood ratio test showed that breathing sound analysis contributed significantly to the performance of the Safe-OSA rule. The Safe-OSA rule was determined to be reasonably discriminative and well calibrated. The five-fold cross-validation showed similar results for the final model in the derivation and testing subsamples, which provides support for the internal validity of the Safe-OSA rule in our study population. Conclusion: The present study lends further support for the future testing of tracheal breathing sound analysis as a potential method to screen for moderate-to-severe OSA to help streamline patient care in the perioperative setting. Trial registration: ClinicalTrials.gov identifier NCT02987283.