Schedule
09:00 - 09:20
|
Registration
|
09:00 - 09:20
|
Registration
|
09:20 - 09:30
|
Opening session
|
09:30 - 10:30
|
Fabrizio Ruggeri
Wear of cylinder liners in ships: one dataset, many models The talk will present data about wear of cylinder liners in ships and a selection of the models which have been used so far to model such process. The major emphasis will be on two models based on jump diffusion processes, but others based on Markov chains, Nonhomogeneous Poisson and Gamma processes will be presented as well. |
10:30 - 11:00
|
Coffee break
|
11:00 - 12:00
|
Antonio Carlos Pedroso de Lima
Joint modeling of survival and longitudinal data The main motivation for this work relies on a dataset from the Heart Institute (InCor) from the Universidade de São Paulo, where patients with congestive heart failure (CHF) were followed up for a period up to 9 years. Several of them had also longitudinal measurements of a cardiac marker, potentially describing the heart condition. The main objective is to predict survival of CHF patients. Our work considered both responses (survival and longitudinal) in a joint model, that also incorporated the information of patients were only the survival information was recorded. A parametric approach was considered, based on a combination of a Gaussian linear mixed model and the Birnbaum-Saunders failure time distribution. The results based on simulations and real data application suggested that the inclusion of longitudinal data may improve the survival analysis. In particular, we were able to assess situations where this improvement would be more sensible, leading to an increase in the precision of estimates. |
12:00 - 13:00
|
Selected oral presentations
Marta Sestelo and Luis Meira-Machado condSURV: Estimation of the conditional survival function for ordered multivariate failure time data in R Luis Antunes, Denisa Mendonça, Aurelien Belot and Bernard Rachet Estimation of age-standardized net survival in sparse data using a modelling approach Inês Sousa Joint models for longitudinal and time-to-event data using INLA |
13:00 - 14:00
|
Lunch
|
14:00 - 15:00
|
Bernard Rachet
Net survival, concept and estimation using population-based data Net survival is one of the key indicators for comparing and evaluating cancer prognosis and cancer control. Net survival is defined as the survival of a group of cancer patients if they could only die from their cancer. Net survival accounts for the competing causes of death by (i) eliminating the hazard due to other causes of death and (ii) accounting for informative censoring due to these causes. When analysing population-based data, a cause-specific approach is not suitable because the cause of deaths is generally not reliably recorded. Therefore, within the relative survival setting, the mortality hazard due to competing causes of death is estimated in the general population which the patients come from. Additionally, Pohar-Perme et al described a non-parametric estimator which accounts for the informative censoring. I will first present the Pohar-Perme estimator, which is a consistent estimator of the cumulative excess hazard. Under certain assumptions, net survival of a group of cancer patients can be also estimated using a modelling approach, which consists of modelling excess hazard of key variables responsible for the informative censoring. I will present one model, how to estimate net survival using such a model and its challenges. |
15:00 - 16:00
|
Poster session
Guidelines for the analysis of multiple failure-time data with marginal models by Ivo Sousa-Ferreira and Ana Maria Abreu. Estimation of multivariate distributions for recurrent event data by Beatriz Sampaio and Luis Meira-Machado. Sequential Monte Carlo methods in Bayesian joint models for longitudinal and time-to-event data by Danilo Alvares, Carmen Armero, Anabel Forte and Nicolas Chopin. Baseline hazard functions in Bayesian Cox models for modeling virulence changes in foodborne pathogens by Elena Lázaro, Carmen Armero, Danilo Alvares, María Sanz-Puig, Antonio Martínez and Dolores Rodrigo. Survival analysis of patients with cervical cancer by Paula Lopes, Eugénio Cordeiro and Luis Meira-Machado. Joint modeling of two longitudinal biomarkers for the diagnosis of acute kidney injury by Loubna Khalifi, Maria Carmen Pardo, Teresa Pérez, Angel Candela-Toha, Alfonso Muriel and Javier Zamora. Flexible parametric models applied to the study of time to non-persistence in a chronic disease treatment by Ana Rita Godinho, Cristina Rocha and Zilda Mendes. Prognostic factors in dogs with malignant oral tumors treated with surgery by Sofia Azevedo, Gabriela Gomes and Lisa Mestrinho. Longitudinal and survival models for recurrence of breast cancer by Liliana Coelho, Ana Borges and Inês Sousa. Transthyretin familial amyloid polyneuropathy survival analysis by Mónica Inês, Teresa Coelho, Isabel Conceição, Marta Soares, Mamede de Carvalho and João Costa. Bayesian spatio-temporal modelling of survival outcomes in long-term studies by Ziyu Zheng and Benjamin Taylor. |
16:00 - 16:30
|
Coffee break
|
16:30 - 17:30
|
Jacobo de Uña-Álvarez
Sampling issues in clinical survival analysis The statistical analysis of clinical survival data requires some prudence. The naïve application of standard methods like Kaplan-Meier and Cox regression techniques may lead to inconsistent estimators and wrong predictions if the employed sampling procedure is ignored. Sampling issues here include (besides the well-known problem of right-censoring) length-bias and double truncation. Length-biased sampling appears in cross-sectional samplings or prevalence studies; it occurs in the presence of delayed entries too. This problem implies that individuals are implicitly selected on being still alive at some specific (entry) time, which results in left-truncation. Corrections for left-truncation are more or less obvious depending on the target. For example, the estimation of the disease-free survival presents in this setting some difficulties which have not been fully addressed in the literature; this also applies to the estimation of transition probabilities in multi-state models. On its hand, double truncation appears with interval sampling, which means that study cohort is made up of individuals experiencing the event within a specific calendar time interval. This second issue has received a lot of attention in the recent years, and it will be reviewed here too. |