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

ISCB2014_abstract_book

124 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 11:00-11:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust = 0.004. An easy to use computer tool named PREVEXEPOC was created, whereby the clinicians could stratify patients with eCOPD by their risk of short-term mortality. P3.5.11 Assessing the prediction accuracy of cure in the Cox proportional hazards cure model J Asano1 , A Hirakawa2 , C Hamada3 1 Pharmaceuticals and Medical Devices Agency (PMDA), Tokyo, Japan, 2 Nagoya University Graduate School of Medicine, Nagoya, Japan, 3 Tokyo University of Science, Tokyo, Japan   Aims: A cure rate model is a survival model incorporating the cure rate with the assumption that the population contains both uncured and cured individuals. It is a powerful statistical tool for prognostic studies, especially in cancer. The cure rate is important for making treatment de- cisions in clinical practice. The proportional hazards (PH) cure model can predict the cure rate for each patient. This contains a logistic regression component for the cure rate and a Cox regression component to estimate the hazard for uncured patients. A measure for quantifying the predictive accuracy of the cure rate estimated by the Cox PH cure model is required, as there has been a lack of previous research in this area. Actually, we used the Cox PH cure model for the breast cancer data; however, the area under the receiver operating characteristic curve (AUC) could not be estimated due to the fact that many patients were censored. Methods: In this study, we propose to use imputation-based AUCs to as- sess the predictive accuracy of the cure rate from the PH cure model. We examined the precision of these AUCs using simulation studies. Results: The results demonstrated that the imputation-based AUCs were estimable and their biases were negligibly small in many cases, although ordinary AUC could not be estimated. Conclusion: The proposed imputation-based AUCs are useful for assess- ing the predictive accuracy of cure rates from the Cox PH cure model.   P3.5.25 External validation of a prognostic model LJ Bonnett1 , AG Marson1 , A Johnson2,3 , L Kim4 , JW Sander5,6,7 , N Lawn8 , E Beghi9 , M Leone10 , C Tudur Smith1 1 University of Liverpool, Liverpool, United Kingdom, 2 MRC Biostatistics Unit, Cambridge, United Kingdom, 3 MRC Clinical Trials Unit, London, United Kingdom, 4 London School of Hygiene & Tropical Medicine, London, United Kingdom, 5 NIHR Biomedical Research Centre, King’s College London, London, United Kingdom, 6 University College London, London, United Kingdom, 7 Epilepsy Society, Buckinghamshire, United Kingdom, 8 Royal Perth and Fremantle Hospitals, Perth, Australia, 9 IRCCS Istituto di Ricerche Farmacologiche Mario Negri, Milan, Italy, 10 Ospedale Maggiore della Carita, Novara, Italy   Background: Prognostic models can be used to stratify patients. However, before a model can be used for this purpose, it needs to be validated ex- ternally in independent data.We demonstrate methods of external valida- tion, including methods for handling a covariate missing from the valida- tion dataset, via a prognostic model for risk of seizure recurrence following a first ever seizure. Methods: Three independent datasets were obtained. External validation was evaluated for each dataset via discrimination using Harrell’s c-index. Calibration plots were also considered as was a measure of prognostic accuracy, R2 Brier. Five imputation methods were examined to handle a covariate missing from one validation dataset. These included hot deck imputation and multiple imputation. Results: Trial data for 620 people with epilepsy was used to develop the original model; the validation datasets consisted of 274, 307 and 847 trial participants respectively. The model generalised relatively well to the other datasets. All five methods of imputation performed fairly similarly. Conclusions: Prognostic models can be validated by considering cali- bration and discrimination methods and predictive accuracy measures. Although there are limitations to external validation, it is still a necessary part of modelling. Our prognostic model, used to inform driving regula- tions, has been validated and consequently has been proven as a valu- able tool for predicting risk of seizure recurrence following a first seizure in people with various combinations of risk factors. Additionally, there is evidence to support one worldwide overall prognostic model for risk of second seizure following a first. P3.5.37 Partial least square discriminant analysis of neurological outcome after cardiac arrest using bispectral index O Collignon1 , P Stammet2 , Y Devaux1 1 CRP Santé, Strassen, Luxembourg, 2 Centre Hospitalier de Luxembourg, Strassen, Luxembourg   Early prediction of neurological outcome after cardiac arrest would rep- resent a major breakthrough towards personalized medicine. This would indeed allow physicians to adapt patients‘ healthcare in the first hours of intensive care unit stay. To achieve this goal, the bispectral index (BIS), which has initially been developed as a tool to measure the depth of anesthesia, was monitored minute-by-minute during the first 24h after cardiac arrest by electroen- cephalography (EEG) in 96 patients in order to predict their cerebral per- formance category (CPC) at 6 months of follow-up. Since the predictive BIS time points were highly correlated and more numerous than the number of patients, Partial Least Square Discriminant Analysis was performed to model these functional data. Mean BIS, area under BIS curve and BIS slope over time were also evaluated. These methods were compared at each time point by plotting ROC curves and detecting the cut-offs maximizingYouden index. Sensitivity and spec- ificity were corrected for optimism using bootstrap internal validation. Finally, PLS-DA, mean BIS and area under BIS curve outperformed the slope method and all reached the same optimal prediction around 12h, with a ROC AUC of 0.89, a specificity of 89% and a sensitivity of 86% (resp. 87% and 85% after correction). Added value of mean BIS to a multivariate clinical model was evaluated using continuous Net Reclassification Index (1.44%,p=4E-12) and Integrated Discrimination Improvement (0.29, p=2E- 9). P3.5.45 Internal-external cross-validation (IECV) for prognostic model research using data from multiple studies: potential & pitfalls J Ensor1 , K Snell2 , D Moore1 , D Fitzmaurice1 , RD Riley1,2 1 University of Birmingham, Birmingham, United Kingdom, 2 MRC Midland Hub for Trials Methodology Research, Birmingham, United Kingdom   Background: Validation is a crucial step in the acceptance of any prog- nostic model within clinical practice. While the majority of models are in- ternally validated, very few are externally tested. However, the availability of individual participant data (IPD) from multiple studies can address this through IECV. We highlight strengths and weaknesses of the approach with application to a venous thromboembolism (VTE) recurrence predic- tion model. Methods: The IECV approach iteratively selects N-1 studies from the N to- tal studies available, and the prognostic model is developed within this subset of studies, leaving the remaining study for external validation of the model. A total of N different models are derived (one for each set of included studies) and each is validated in the omitted study. Model perfor-

Pages Overview