Please activate JavaScript!
Please install Adobe Flash Player, click here for download

ISCB2014_abstract_book

96 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsMonday, 25th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust ping can be sensible implemented. The operating characteristics will be assessed by clinical trial simulations. Furthermore, the proposed adaptive design would allow to use different cross over designs for the first and second stage, e.g., switch from a clas- sical two-period design to a more complex replicate design. Another ap- plication of the proposed method would be for establishing biosimilarity of two products.   P1.2.89 Adaptive designs for confirmatory model based decisions using MCP-Mod S Krasnozhon1 , A Graf1 , B Bornkamp2 , F Bretz2 , G Wassmer3 , F König1 1 Medical University of Vienna, CeMSIIS, Wien, Austria, 2 Novartis Pharma AG, Basel, Switzerland, 3 ADDPLAN Inc., an Aptiv Solutions Company, Cologne, Switzerland   Adaptive seamless designs for confirmatory clinical trials have attracted a lot of attention because they offer the possibility to combine different phases of drug development into a single trial. This is of paramount in- terest in small populations, e.g., when developing drugs for rare diseases. Though the sample size is limited still an appropriate dose has to be found and sufficient evidence for its efficacy to be demonstrated. We propose adaptive clinical trial designs with multiple doses and use modelling approaches to (i) establish a positive dose-response profile, (ii) increase the power of declaring effective dose statistically signifi- cant, and (iii) support dose selection at an adaptive interim analysis. We extend MCP-Mod methodology to adaptive two-stage designs by using the closed-testing principle and applying an adaptive combination test to each intersection hypothesis. Combining the data from both stages in adaptive confirmatory designs allow for flexible interim decisions based on all (interim) data available of the ongoing trial while always ensuring strict type I error control. In particular, the MCP-Mod approach can be used to obtain model-based dose effect estimates at interim to guide early fu- tility stopping and/or re-design the second stage (e.g. choice of doses, sample size, allocation ratio) and analysis (e.g., dropping of inadequate response models). By the means of clinical trial simulations we show the operating charac- teristics (e.g., power for PoC or individual dose-control comparison, bias of effect estimates) for specific adaptations rules.   P1.2.93 Residual plots for censored data: a new approach M Law1 , D Jackson1 , S Brill2 , W Wedzicha2 1 MRC Biostatistics Unit, Cambridge, United Kingdom, 2 University College London, London, United Kingdom   All statistical models make assumptions. For the resulting statistical infer- ence to be valid, the model must describe the data well. For regression models, and when data are uncensored, residual plots are a standard tool for assessing how well the model fits the data. However, when some data are censored, standard residual plots are less appropriate for checking modelling assumptions. Various approaches have been suggested, which we summarise and compare with our own proposed approach. I will present a method for producing residual plots, analogous to those used for uncensored data, which take into account both the parameter uncertainty and the uncertainty in the location of the censored data. I will illustrate this method by examining the model fit for an analysis of bacte- rial load data from a trial for chronic obstructive pulmonary disease. Here we use a standard linear regression model but some data are cen- sored because they are only observed to be less than a threshold value. We conclude that the model fit is acceptable but is less good than might initially be thought before appropriately handling the censored observa- tions. P1.2.97 Longitudinal cluster analysis with application to identify mortality associated SOFAtrends in critical care medicine K Liu1,2 , V Cornelius1 , M Shankar-Hari1,2 , C Dangoisse2,3 , M Terblanche1,2 1 King’s College London, London, United Kingdom, 2 Guy’s and St Thomas’ NHS Foundation Trust, London, United Kingdom, 3 King’s College London Hospital NHS Foundation Trust, London, United Kingdom Background: In critically ill patients Sequential Organ Failure Assessment (SOFA) score is used as a surrogate for end organ function and is associated with mortality. In ICU literature, logistic regression analyses often used to predict patient mortality, but it does not acknowledge and are hampered by the heterogeneous nature of the patient population. We hypothesised that detecting homogenous patient groups in terms of organ dysfunction trajectory using longitudinal cluster analysis will identify mortality clusters in critically ill patients; thus improving risk prediction. Method: A two-stage approach was conducted. In stage one k-means for longitudinal data (KML) was adopted to determine homogeneity of pa- tient SOFA trajectories during their first 4 ICU days. In stage two a logistic regression model was fitted using the cluster allocation as predictor vari- able to examine the association with mortality. Model comparison was conducted using ANOVA with two conventional approaches. Result: 4438 patients admitted between April 2010 and October 2013 were included. The KML analysis identified 6 distinct patterns of organ dysfunction. These six clusters had distinct outcome patterns: mortali- ties for clusters A through F were 2.2%, 6.0% 18.2%, 22.6%, 42.1% and 54.1%, respectively. ANOVA analysis showed that our approach is signifi- cantly better than the logistic models only considering baseline or day 1-3 change in SOFA score as the regressors of mortality. Conclusion: Longitudinal cluster analysis using SOFA identified homog- enous clusters which were associated with ICU mortality. Our approach outperformed the conventional approaches and thus would be a better approach for risk stratification in clinical trials.   P1.2.100 Using simulation to examine the effects of varying cluster size on the precision of stepped wedge cluster randomised trials JT Martin1 , K Hemming1 , A Girling1 1 University of Birmingham, Birmingham, United Kingdom The use of a stepped wedge cluster randomised trial (SW-CRT) is increas- ing, especially in health service evaluations. The stepped wedge design incorporates a sequential rollout of the intervention to all clusters over multiple time periods. The SW-CRT typically includes equally spaced steps and an equal number of patients in each cluster. However, in practice, it is unlikely that an equal number of patients will be present across all clusters at each time point. Here we consider the impact of varying cluster sizes on the precision of a SW-CRT. To this end, a formula devised by Hussey & Hughes to compute precision for SW-CRTs was extended to include varying cluster size. The degree of variation in cluster sizes can be described by the coefficient of variation (CV). By altering the CV, along with the intra cluster correlation coefficient (ICC), the number of clusters and the number of randomisation points, the effects of varying cluster sizes could be determined. Simulation methods were then used to form an average precision over varying designs. For comparative purposes, the precision of a parallel CRT was also estimat- ed. We found that standard methods of adjustment in a CRT may under- estimate the precision. Simulations show that the SW-CRT is less affected by varying cluster sizes than the parallel CRT. There are also situations in which for equal cluster sizes, the parallel CRT is of higher precision, but as the CV increases, the SW-CRT becomes the trial type with higher precision.

Pages Overview