none

Dans le diagnostic médical, la surface ROC est l'outil statistique utilisée pour évaluer la précision d'un test diagnostic dans la discrimination de trois états d'une maladie, et le volume sous la surface ROC est l'indice utilisé pour la quantification de la performance du test. Dans certaines situations, différents facteurs peuvent affecter les résultats du test et ainsi les mesures de précision. Dans le cas des études longitudinales, le statut du patient peut changer au cours du temps. Dans ce manuscrit, nous avons développé des méthodes statistiques permettant d'évaluer les capacités discriminatoires des outils diagnostics et pronostics. Nous avons d'abord proposé une méthode semi-paramétrique pour estimer la surface ROC sous des modèles de rapport de densité. La construction de la méthode proposée est basée sur le modèle logit à catégories adjacentes et l'approche de vraisemblance empirique. Nous avons décrit la méthode bootstrap pour l'inférence des estimateurs obtenus. Ensuite, nous avons présenté une méthode d'estimation des surfaces ROC appelée famille de Lehmann des surfaces ROC. Cette méthode est basée sur la famille d'alternatives de Lehmann ou modèle à hasards proportionnels. Elle a l'avantage de prendre en compte les covariables qui peuvent affecter la précision d'un test diagnostic. En outre, nous avons développé une surface ROC covariable-spécifique basée sur la règle de Bayes. Pour cela, nous avons proposé un estimateur semi-paramétrique pour les surfaces ROC covariable-spécifique via des procédures de régression logistique polytomique et un modèle semi-paramétrique de localisation. Enfin, dans le cas où le statut du patient peut évoluer à travers différents stades d'une maladie, une méthode des surfaces ROC dépendant du temps a été développée. L'estimateur obtenu utilise l'approche "Inverse Probability of Censoring Weighting" (IPCW). Des simulations et des exemples sont fournis afin d'illustrer la performance des estimateurs proposés. / In diagnostic medical, the receiver operating characteristic (ROC) surface is the statistical tool used to assess the accuracy of a diagnostic test in discriminating three disease states, and the volume under the ROC surface is the used index for the quantification of the performance of the test. In some situations, various factors can affect the test results and subsequently the accuracy measures. In the case of longitudinal studies, the patient's status may change over time. In this manuscript, we developed statistical methods to assess the discriminatory capabilities of diagnostic and pronostic tools. We first proposed a semiparametric method for estimating ROC surface under density ratio models. The construction of the proposed method is based on the adjacent-category logit model and the empirical likelihood approach. We described the bootstrap method for inference of the obtained estimators. Next, we presented a method for estimating ROC surfaces called Lehmann family ROC surfaces. This method is based on the family of Lehmann alternatives or proportional hazards model. It has the advantage of taking into account covariates that may affect the accuracy of a diagnostic test. Moreover, we have developed a covariate-specific ROC surface based on the Bayes rule. For that, we proposed semiparametric estimator for covariate-specific ROC surfaces via polytomous logistic regression procedures and a semiparametric location model. Finally, in the case where patient's status may evolve through different stages of disease a method of time-dependent ROC surfaces was developed. The proposed estimator uses the "Inverse Probability of Censoring Weighting" (IPCW) approach. Simulations and examples are provided to illustrate the performance of the proposed estimators.

Receiver operating characteristic (ROC) analysis is one of the most widely used methods in evaluating the accuracy of a classification method. It is used in many areas of decision making such as radiology, cardiology, machine learning as well as many other areas of medical sciences. The dissertation proposes a novel nonparametric estimation method of the ROC surface for the three-class classification problem via Bernstein polynomials. The proposed ROC surface estimator is shown to be uniformly consistent for estimating the true ROC surface. In addition, it is shown that the map from which the proposed estimator is constructed is Hadamard differentiable. The proposed ROC surface estimator is also demonstrated to lead to the explicit expression for the estimated volume under the ROC surface . Moreover, the exact mean squared error of the volume estimator is derived and some related results for the mean integrated squared error are also obtained. To assess the performance and accuracy of the proposed ROC and volume estimators, Monte-Carlo simulations are conducted. Finally, the method is applied to the analysis of two real data sets.

Purpose: To assess the performance of a Dual-Energy chest radiography system. Methods: A cohort of 129 patients was recruited from population referred for CT guided biopsy of a lung lesion. Digital radiography (DR) and Dual Energy (DE) images were acquired. Receiver operating characteristic (ROC) tests were performed to evaluate performance of DE images compared to DR. Five chest radiologists scored images. Performance was analyzed for all cases pooled and sub groups based on gender, nodule size, density, location, and chest diameter. Results: There was no significant difference between DE and DR for all cases (p = 0.61). There was a significant advantage for DE imaging of small nodules, and nodules located in right-upper lobe. (p = 0.02 and 0.01) Conclusions: DE imaging demonstrated significant improvement in diagnosis of sub-centimeter lung nodules and lesions in the upper lung zones which are common characteristic of early stage lung malignancy.

Purpose: To assess the performance of a Dual-Energy chest radiography system. Methods: A cohort of 129 patients was recruited from population referred for CT guided biopsy of a lung lesion. Digital radiography (DR) and Dual Energy (DE) images were acquired. Receiver operating characteristic (ROC) tests were performed to evaluate performance of DE images compared to DR. Five chest radiologists scored images. Performance was analyzed for all cases pooled and sub groups based on gender, nodule size, density, location, and chest diameter. Results: There was no significant difference between DE and DR for all cases (p = 0.61). There was a significant advantage for DE imaging of small nodules, and nodules located in right-upper lobe. (p = 0.02 and 0.01) Conclusions: DE imaging demonstrated significant improvement in diagnosis of sub-centimeter lung nodules and lesions in the upper lung zones which are common characteristic of early stage lung malignancy.

none

Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (p. 231-238).

不論是對保險監理者或是保戶來說，保險公司是否具有清償能力一直都是大家關注的焦點。這方面的議題探討不勝枚舉。在過去的文獻裡，大家所採用的模型不竟相同，但相同的是，大家焦點都是放在保險公司破產機率這方面。 本文使用Moody研發的KMV模型下針對未上市公司有顯著解釋能力的PFM模型(Private Firm Model)。並利用PFM模型來預測北美壽險業的違約風險。一開始，我們先從上市的壽險業中取得足夠的資料，進而去估計未上市壽險業的資產市值及資產波動度，並利用這些資料算出違約距離(Distance-to-Default)。 本文的另ㄧ個重點，是將過去文獻中有顯著的比率與違約距離作比較，試圖提出一個能夠代表市場資訊的新比率。因此，我們利用羅吉斯迴歸來對照不同變數下的模型，並利用ROC(Receiver Operating Characteristic Curve)曲線下的範圍來衡量模型的適合度。 本文所採用的上市北美壽險業與未上市北美壽險業資料，取自CompuStat、DataStream及NAIC。 / Insurer’s solvency has always been the primary concern of insurance regulators and policyholders. Researchers therefore have strived to develop various models to identify potentially troubled insurers. Our paper will contribute to the literature by applying a new method, the KMV’s private firm model (PFM), to predict the solvency/insolvency of life insurers.In this paper, we will apply the KMV’s PFM to estimate the default risk of life insurers. We will first apply the KMV’s public firm model to public life insurers and then use the two simple mapping methods to estimate the asset value and volatility of private life insurers. The estimated values and volatilities can then be used to calculate an insurer’s distance-to-default (DD) and default probability. The predictive power of PFM will be compared with the common ratio analysis using logistic regressions and Receiver Operating Characteristic (ROC) Curves. The data on public and private life insurers will come from CompuStat, DataStream and NAIC’s A-list data respectively. Both are readily available at our university.

Univ., Diss., 2010--Marburg.

Thesis (Ph.D.)--Virginia Commonwealth University, 2010. / Prepared for: Dept. of Biostatistics. Title from resource description page. Includes bibliographical references.