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


114 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 11:00-11:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust Wednesday, 27th August 2014 - 11.00-11.30 Poster session P3 P3.1 Survival analysis, multistate models and competing risks P3.1.19 Competing-risk regression model to explore risk factors associated with lost to follow-up prior to antiretroviral therapy: a multicentric observational cohort M Bastard1 , J-F Etard1,2 1 Epicentre, Paris, France, 2 UMI 233 TransVIHMI, IRD, Université Montpellier 1, Montpellier, France   In clinical research, death and lost to follow-up (LTFU) are from part of the most commonly reported outcomes. The objective here is to present the competing-risk regression model, an alternative method to Cox propor- tional-hazards model in presence of multiple outcomes that are not inde- pendent, and to apply it to explore factors associated with LTFU prior to antiretroviral therapy (ART). Standard techniques assume that the distribution of the censoring time and the time-to-event distribution are independent. In case of adminis- trative censoring, this hypothesis seems reasonable. However, if patients died, censoring is related to the time-to-event of interest (LTFU), and the independence assumption is violated leading to biased estimates of LTFU rates and hazard ratios. Fine and Gray specify a semi-parametric model which uses the hazard of the subdistribution of the event of interest to model the cumulative incidence function. With this model, patients with the competing event are kept at risk and continue to contribute person- time with the remaining time at risk weighted by the inverse probability of censoring. To illustrate the method, we apply it to assess factors associated with LTFU before ART initiation in a large multicentric cohort of HIV-infected pa- tients, treating death before ART as a competing event. We also compare it to standard Cox proportional-hazard model. Competing-risk models should be considered with interest in clinical re- search when conducting survival analysis with competing events as stan- dard methodology could provide biased estimates of both the cumulative incidence of the event of interest and hazard ratios of associated factors.   P3.1.22 Description of disease progression and relevant predictors in diabetic foot ulcer patients using a Markov chain model A Begun1 , S Morbach1,2 , G Rümenapf3 , A Icks1,4 1 German Diabetes Center at the Heinrich-Heine-University, Düsseldorf, Germany, 2 Department of Diabetes and Angiology, Marienkrankenhaus, Soest, Germany, 3 Diakonissen-Stiftungs- Krankenhaus Speyer, Mannheim, Germany, 4 Department of Public Health at the Heinrich-Heine-University, Düsseldorf, Germany   Macro- and microvascular complications among diabetic patients can lead to foot ulceration and increased risks of (minor or major) amputation and death. We have used an eight-state Markov chain model to study the course of the diabetic foot syndrome.The diabetic patients were staged in accordance with their amputation status (no, minor or major), present or previous foot ulceration (yes, no), and death as absorbing state. Potential risk factors - such as gender, age at arriving in the state, smoking habits, diabetes duration, neuropathy, peripheral arterial disease (PAD), chronic renal failure (CRF) and others - were included in the model in the form of the Cox-regression covariates. In addition, the impact of revascularization procedures was studied. We used long-term data of a patient cohort from one single specialized diabetic foot center in North Rhine-Westphalia (Germany). The estimates of unknown baseline transition intensities and Cox-regression coefficients were derived from stepwise regression with backward elimination based on the likelihood ratio test at level 5%. Amongst others, we found that established risk factors as gender, PAD or CRF were predictive for the tran- sition between some stages, while not others. For instance, male patients with diabetic foot syndrome but without previous amputations showed an increased risk of foot ulcer recurrence compared to females, while there was no gender difference regarding the risk for transition to neither minor nor major amputation. This model can help us to quantify the disease pro- gression and its predictors.   P3.1.27 Reconstructing individual patient level data: a simulation approach RH Boucher1 , KR Abrams1 , PC Lambert1,2 1 University of Leicester, Leicester, United Kingdom, 2 Karolinska Institutet, Stockholm, Sweden   There are times when the reported analysis for a time-to-event outcome may be considered to be inappropriate. For example, if a proportional haz- ards (PH) model has been used, despite the PH assumption being clearly violated. It is, therefore, desirable to re-analyse the dataset using more ap- propriate methods. However, this is only realistic if the original individual patient data (IPD) is accessible. Nevertheless, if only summary data are available, the method outlined here simulates multiple datasets that are representative of the original IPD. These can then be analysed using the desired method and the results averaged over in order to produce a more appropriate result. This approach relies on routinely reported summary information and the Kaplan-Meier curve. Coordinates are extracted from the Kaplan-Meier curve. A model is fitted to these coordinates, and then used to simulate survival times for individual patients. The censoring distribution is formed using information published on recruitment times, or from the ‘numbers at risk’table.The minimum of the survival and censoring time is then taken as the patient’s observed survival time.The last three steps are repeated to generate multiple datasets. An application of this method highlights its ability to replicate the report- ed statistics, and hence its success in representing the original IPD. In ad- dition, this example is one for which the IPD was available and, thus the comparison between the appropriate analysis on the original IPD and the averaged result over the simulated datasets can also be made.   P3.1.69 Joint modelling of multiple longitudinal markers and recurrent events of multiple types MH Hof1 , JZ Musoro1 , RB Geskus1 , AH Zwinderman1 1 Academic Medical Center of the University of Amsterdam, Amsterdam, The Netherlands   Our study was motivated by post-kidney transplantation data, where we observed four longitudinal markers and nine different recurred infection types. Moreover, as a consequence of low marker values and multiple in- fections, individuals dropped-out of the study. We used a joint modelling approach to correct for informative drop-out. Our main interests were the relations between the markers and the infection rates. We discretized the time-scale into small intervals such that individuals could experience at most one event per interval. We parameterized the sub-model for the nine competing infection risks and the drop-out risk with a multinomial regression model with subject-specific random effects. The sub-model for the marker trajectories was parameterized by a multi-

Pages Overview