Regression 3
Presenter: Yuko Araki
When: Thursday, July 14, 2016 Time: 2:30 PM - 4:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Sparse functional classification method with composite basis function for early detection of Alzheimer's disease based on brain MRI
Recent years have seen that functional data analysis are capable of extracting intrinsic features from recently arising complicated and high dimensional data, such as three dimensional brain sMRI, time course microarray data , or hundreds of records of human gait, for example. In this work, we introduce statistical methods for classifying individuals with such high dimensional covariates, especially for classifying Alzheimer patients based on three-dimensional MRI data.The proposed method is based on composite basis function, which is an extended version of basis expansions with the help of sparse PCA. Further, L1-type penalty constraints are imposed in the estimation of the parameters of logistic discrimination. This two-step regularization method accomplishes both covariates selection and estimation of unknown model parameters simultaneously.The proposed models are applied to real data example and Monte Carlo simulations are conducted to examine the efficiency of our modeling strategies. Among several possible classi cation techniques, our simulation study is shown to be the best in terms of sensitivity and speci city when detecting Alzheimer's disease from three-dimensional MRI data.
Regression 3
Presenter: Kenichi Hayashi
When: Thursday, July 14, 2016 Time: 2:30 PM - 4:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Modification of integrated discrimination improvement by beta divergence
Predictive performance of biomarkers is one of the important aspects of their novelty in biomedical research. This is often evaluated by comparing two statistical models: one is a "new" model with newly added biomarkers and the other is an "old" model without them. When the response variable is dichotomous, the difference between the areas under the receiver operating characteristic curve (AUC) is frequently used as a measure of prediction improvement by the additional markers. This usage of the AUCs is criticized because of its conservative decision and so on. Recently, the integrated discrimination improvement (IDI), proposed by Pencina et al. (2008), becomes popular and increasingly applied as an alternative to the AUC difference in clinical medicine. However, the IDI can falsely detect a significant improvement of the new model even if it is established without adding any information about the new biomarkers (Hilden and Gerds, 2014). This means that the IDI has the opposite problem to the AUC difference. In this study, we modify the IDI to overcome its potential problem mentioned above. The modification is based on the beta-divergence between two models. We show that our proposal has desirable properties such as Bayes risk consistency and Fisher consistency. In particular, we point out that the latter property can explain why the original IDI does not avoid false detection of apparent improvement. The performances of the proposed method are discussed via numerical experiments. References [1] Hilden, J., Gerds, T.A. (2014). A note on the evaluation of novel biomarkers: do not rely on integrated discrimination improvement and net reclassification index. Statistics in Medicine 33, 3405-3414. [2] Pencina, M.J., DAgostino, R.B. Sr., DAgostino, R.B. Jr., Ramachandran, S.V. (2008). Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Statistics in Medicine 27, 175-172.
Regression 3
Presenter: Jeremy Taylor
When: Thursday, July 14, 2016 Time: 2:30 PM - 4:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
A statistical framework for using external information in updating prediction models with new biomarkers
Models that give personalized predictions are abundant in the clinical and epidemiologic literature. There is a strong desire to improve these prediction models with new biomarkers. The information from an existing prediction model can be available in the form of coefficient estimates (with or without measures of standard error) or individual predicted probabilities. There could often be a set of different models for predicting the same outcome. We investigate different approaches to incorporating such varying types of information while building a new prediction model that adds new candidate biomarkers to the existing model. The situation is that this candidate biomarker is measured on a small group of subjects while the existing prediction models have been validated in large studies, but we do not have access to the data from the large studies. We formulate the problem in an inferential framework where the historical information is translated in terms of non-linear constraints on the parameter space of the new model. We establish an approximate relationship between the regression coefficients in the two models. We develop both frequentists and Bayesian approaches. Simulation results suggest that the information from the established model can substantially improve the predictive power of the new model of interest.
Regression 3
Presenter: Fidel Ulin
When: Thursday, July 14, 2016 Time: 2:30 PM - 4:00 PM
Room: Salon B Carson Hall (Level 2)
Session Synopsis:
Multivariate regression models to determine risk factors associated with incidence, severity and duration of chronic pain after surgery
Objective: In this study, we investigated the risk factors associated with development of chronic post-surgical pain, as well as the relationship of comorbidities (obesity, diabetes, and hypertension) with the development of chronic pain, assessing the development of chronic postoperative pain intensity of temporary transition from acute to chronic. Methods: A retrospective study was conducted with 50 patients aged 26 to 72 years who underwent amputation, mastectomy, cholecystectomy, and plasties surgery between March and June 2015. Patients were followed up by telephone in October 2015. Analysis included incidence, severity, and duration of chronic pain. Severity was categorized by means of visual analogue scale, represented on a 0-10 long line, labeled with the phrase "no pain on one end, and worst pain. Duration of chronic pain was considered by 24 hours, 2 months, and 6 months. Using Chi-square test and multivariate logistic regression analysis risk factors associated with the incidence and severity of chronic pain after surgery were also identified. Results: Chronic pain after surgery was more likely to occur in the mastectomy and cholecystectomy groups compared with the other types of surgery (P=0.005). Diabetic patients who underwent cholecystectomy surgery reported significantly more pain than non-diabetic patients (P=0.006). About risk factors between patient categories by type of surgery, it found significant risk factor on overweight and a history of hypertension and diabetes. Conclusions: Results indicate that patients with overweight and a history of hypertension and diabetes undergoing open cholecystectomy and mastectomy surgery have greater postoperative pain and incidence of severe pain scores. Therefore, analgesic treatment in those patients should consider this consideration to provide a satisfactory postoperative analgesia. Key words: Multivariate logistic regression model; diabetes, postoperative pain.
