Topics in Application
Presenter: Taerim LEE
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
Clinical Decision Support System for HCC Surveillance Clinical Exam
Objectives: The purpose of this paper is to construct and optimize performance of a classification model to aid management of Clinical Decision Support System and to evaluate performance implementation effectiveness and barriers to adoption Methods: We used 8 classification models including RF, SVM, k-nearest neighbor, linear discriminant analysis, logistic regression, boosting and ANN were trained and their performance were compared for predicting patients prognosis. Variable selection was performed and only clinical variables relevant to outcomes were utilized for predicting the outcome to avoid over fitting the model and to detect the surveillance clinical exams significance. After that we plan to compare the performance of RF, ANN and SVM to other classification results using ROC curves. Using several packages of Random forest(package for Random Forest), Support vector machine(package e1071) Shrunken centroid(package pamr) LDA9package(sml), KNN(package class), and logistic regression(package stats), Boosting(package boost), ANN(Neural networks package) we can get the results of the classification for clinical support system. Results: We find that several clinical variables and SNP data were selected as important factor for clinical decision of HCC patient prognosis and manage surveillance clinical exams using 8 classification models. For the comparison of models and methods all models were run evaluation step and validation step using 10 fold cross validation and the accuracy, sensitivity, specificity, positive predictive value and negative predictive value, ROC curves and area under ROC(AUC) with Mann Whitney test. Conclusions: Random Forest seems to be overall the best performing model in terms of accuracy and balance of sensitivity and specificity. Using clinical decision support system physicians can improve their diagnosis and decision making assisted and it could be efficient and cost effective management of medical resources and surveillance clinical exam. Keyword: clinical decision support system, ROC, surveillance clinical exam
Topics in Application
Presenter: Ratchaneewan Kumphakarm
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
An alternative improvement to the Chao estimator of species richness
The Chao estimator is very popular for estimating species richness, the number of species present in a population, given a random sample from the population; it uses only the number of observed species (K), and the number of species seen exactly once (f1) and exactly twice (f2). Strictly speaking, the Chao estimator estimates a lower bound for species richness, but in practice it is often used as an estimator of species richness itself. For a homogeneous abundance model or for a large sample size, the Chao estimator is approximately unbiased. However, it is negatively biased for heterogeneous models or small sample size. A recent paper in Biometrics (Chiu et al, 2014) describes a new improved estimator, called the iChao1 estimator, which attempts to reduce the bias of the original Chao estimator by using additional data on the number of species seen three or four times in the sample (f3, f4). Simulations show that this new estimator often performs favourably in comparison to the original Chao estimator. Here, we have developed an alternative estimator that is intended to perform similarly to iChao1 but uses only the same data (K, f1, f2) as the original Chao estimator. The new estimator is evaluated using simulation and applied to real data sets in order to illustrate and compare alternative estimators in terms of bias, root mean square error and the coverage and width of confidence intervals. We find that the new estimator is generally similar to iChao1, but tends to perform slightly better for small sample sizes.
Topics in Application
Presenter: David Robertson
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
Unbiased estimation in seamless phase II/III trials with unequal treatment effect variances and adjustment for multiplicity
Seamless phase II/III clinical trials offer an efficient way to select an experimental treatment and perform confirmatory analysis within a single trial. However, combining the data from both stages in the final analysis can induce bias into the estimates of treatment effects. Methods for bias adjustment developed thus far have made restrictive assumptions about the design and selection rules followed. In order to address these shortcomings, we apply recent methodological advances to derive the uniformly minimum variance conditionally unbiased estimator (UMVCUE) for two-stage seamless phase II/III trials. Our framework allows for the precision of the treatment arm estimates to take arbitrary values; can be utilised for all treatments that are taken forward to phase III; and is applicable when the decision to select or drop treatment arms is driven by a multiplicity-adjusted hypothesis testing procedure.
Topics in Application
Presenter: Alia sajjad
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
Using Electric Resistances to Find A-Optimal Block Designs When the Concurrence Graphs are Sparse
Designs with sparse concurrence graphs are discussed for A-optimality. Explicit formulae as a function of number of blocks are drawn. The results are obtained by attaching concurrence graph to the design. The graph is taken as an electric network with resistance of one ohm along each edge. The sum of pairwise resistances are calculated to rank the designs on A-optimality. This approach has enabled us to draw several results regarding the effective resistance in the attached concurrence graphs. These results provide an effective tool to find A-optimal designs in a specific class of designs. Phase transition is observed among designs under A-optimality criteria. It has been observed that very different designs have been selected by A-optimality criterion for different values of v.
Topics in Application
Presenter: Steffen Unkel
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
Evidence synthesis for a single randomized controlled trial and observational data in small populations
We consider the scenario of a single randomized controlled trial (RCT) comparing an experimental treatment to a control in a small patient population. Whereas in large populations usually two independent RCTs are required to demonstrate efficacy and safety for marketing authorization, in small populations the conduct of even a single RCT with a sufficient sample size might be extremely difficult or not feasible. Inspired by an ongoing paediatric study in Alport syndrome (Gross et al. 2012), we consider study designs in which information external to the randomized comparison, such as data arising from disease registries, are integrated into the design and analysis of an RCT in different ways. Using a Bayesian framework, statistical models for binary, continuous and time-to-event endpoints are built. Methods of generalized evidence synthesis, in which studies from different designs are pooled in order to estimate quantities of interest, are then employed for the planning and interpretation of an RCT in a small population. The performance of the proposed methods are evaluated under different scenarios by means of experiments. This research has received funding from the EUs 7th Framework Programme for research, technological development and demonstration under grant agreement number FP HEALTH 2013 602144 through the InSPiRe project. Gross, O. et al. (2012): Safety and efficacy of the ACE-inhibitor ramipril in Alport syndrome: The double-blind, randomized, placebo-controlled, multicenter phase III EARLY PRO-TECT Alport trial in pediatric patients, ISRN Pediatrics, Volume 2012, Article ID 436046, 6 pages, doi:10.5402/2012/436046.
Topics in Application
Presenter: LUIS GRAJALES
When: Friday, July 15, 2016 Time: 11:00 AM - 12:30 PM
Room: Saanich 1-2 (Level 1)
Session Synopsis:
A proposal to select active effects in 2k-p experiments with beta response.
The variable dispersion beta regression model (VDBRM) is useful when the response is mea- sured continuously in the (0;1) or (a,b) intervals. 2k-p experiments with response in (a,b) are encountered in the literature, and its analysis presents difficulties when beta regression is not used. In this work, in order to take account of constraints on parameters, a restricted variable dispersion beta re- gression model is proposed, developed, and applied from a frequentist perspective. i) when there are not restrictions, our model coincides with the variable dispersion BRM; ii) if there are not restrictions and the dispersion parameter is assumed constant across observations, our model is the simple BRM. First, a penalized likelihood function is proposed, using La- grange multipliers for restrictions. Second, the restricted maximum likelihood estimators are obtained. Third, the respective inferential analysis is done: hypothesis tests for restrictions and goodness of t for models. Here, the application is focused active effects in 2k-p experiments with beta response. Good results were obtained for simulated and real data.
