Journal of Muscle Research and Cell Motility | 2019
Talks
Abstract
s Research talks ---------------------------------------------------------------------------------------------------------------Incorporation of nested frailties into multistate models, with an application to a multicenter lung cancer clinical trial Federico Rotolo, Virgini Rondeau, Catherine Legrand Institut Gustave Roussy, Service de Biostatitistique et d Epidemiologie, Villejuif, France, Università degli Studi di Padova, Departimento di Scienze Statistiche, Padova, Italy, Université catholique de Louvain, Institut de Statistique, Biostatistique et Sciensces Actuarielles, Louvain-la-Neuve, Belgium, Université de Bordeaux Segalen 2, ISPED, centre INSERM U-897-Epidémiologie-Biostatistique, Bordeaux, France, 5INSERM, ISPED, centre INSERM U-897-Epidémiologie-Biostatistique, Bordeaux, France Aims: Multistate models (MSMs) generalize proportional hazards models by jointly considering different types of events and their interrelations [1, 4], while frailty models introduce random effects to account for unobserved risk factors, possibly shared by groups of subjects [2, 5]. Integration frailty and multistate methodologies is interesting to control for unobserved heterogeneity in presence of complex event history structures, particularly appealing in multicenter clinical trials applications. Methods: In order to obtain distinct but dependent random effects for different transitions, we incorporate nested frailties in the transition-specific hazard and we propose semiparametric inference based on the EMPL algorithm originally developed for multilevel frailty models [3], which alternates the EM and PPL estimation methods. To compare parameter estimation performances, we show a simulation study with Weibull times and a treatment reducing the risk of local relapses (LR) and distant metastases (DM), with no effect on event-free death, and increasing the death risk after LR or DM; we introduce clustering by means of two nested Gamma frailties. We consider three scenarios with 10, 20 and 40 hospitals, each recruiting 20, 50 and 125 patients respectively. We generate data both under the Markov assumption and with further dependence, added by means of copula functions between the event times of different events. Finally, we present a case study relative to a cancer multicenter clinical trial aimed at evaluating the effect of adjuvant chemotherapy in 1867 patients from 26 centres, treated for non-small cell lung cancer. Results: The simulation study shows that semiparametric estimates of regression parameters are less biased and less variable when nested frailties are used, as compared to MSMs with no or shared frailties. The coverage probabilities of confidence intervals are less than the nominal 95% for the three families of models, but they are uniformly higher for the nested-frailty MSMs under the Markov assumption. Violation of this assumption does not affect estimates consistency dramatically, but it lowers coverage probabilities of confidence intervals, in particular for the models with nested frailties. Conclusion: The multistate nature of the nested-frailty MSMs allows studying the treatment effect taking into account intermediate events; at the same time, the presence of frailties parameter estimators are more consistent. [1] Andersen PK & Keiding N (2002) Multi-state models for event history analysis. Stat Methods Med Res 11(2): 91−115 [2] Duchateau L & Janssen P (2008) The Frailty Model. Springer Verlag [3] Horny G (2009) Inference in mixed proportional hazard models with K random effects. Stat Pap 50(3): 481−499 [4] Putter H etal (2007) Tutorial in biostatistics: competing risks and multi-state models. Stat Med, 26(11): 2389−430 [5] Wienke A (2010) Frailty Models in Survival Analysis. CRC Press ---------------------------------------------------------------------------------------------------------------Estimating sojourn time distributions using microsimulation Sabine Zinn (University of Bamberg) In multi-state models, expected sojourn times within states are of great interest. Commonly, three ways exist to derive them: by direct computation, by approximation, or by (micro-) simulation. In any case, preference should be given to direct computation. However, direct computation demands closed-form solutions which exist for time-homogeneous Markov multi-state models, but commonly not, for example, for time-inhomogeneous Markov and semi-Markov multi-state models. Here, approximation methods (such as linearization) might apply. Nonetheless, compared with approximation methods, microsimulation usually gives richer results. This is due to the fact that microsimulation is conceptualized to resemble the whole process described by a multi-state model. Once an initial population (of meaningful size and composition) and hazard rates for all transitions considered in a multi-state model are provided, microsimulation facilitates deriving any kind of distribution related to the model. In this talk, I show how continuous-time microsimulation can be used to derive the distribution of sojourn times for time-inhomogeneous Markov and semi-Markov multi-state models. I illustrate the capabilities of microsimulation using an application of multi-state demography where I study the transition to first child depending on an individual’s marital status and living arrangement. To parameterize my model, I use data from SHARELife which is a retrospective survey carried out in 2010 as part of the third wave of the SHARE project. ---------------------------------------------------------------------------------------------------------------Inference from multistate models from interval-censored data Daniel Commenges (ISPED) The issue of selection of the sample and of interval censoring will be considered. It is very common that pathological status is observed at discrete times, while it evolves in continuous time. This produces interval-censored observations. The most common case is that of mixed discrete-continuous time observations since death is observed in continuous time. The likelihood will be given for this type of observation. An example of application of an illnessdeath model for dementia will be detailed. The R package SmoothHazard able to manage this kind of observations will be presented. ---------------------------------------------------------------------------------------------------------------Scarring effects of remaining unemployed for long-term unemployed schoolleavers Bart Cockx (Ghent University) and Matteo Picchio (Marche Polytechnic University) This study investigates whether and to what extent further unemployment experience for youths who are already long term unemployed imposes a penalty on subsequent labour market outcomes. We propose a flexible method for analysing the effect on wages aside from transitions from unemployment and employment within a multivariate duration model which controls for selection on observables and unobservables. We find that prolonging unemployment drastically decreases the chances of finding employment but hardly affects the quality of subsequent employment. The analysis suggests that negative duration dependence in the job finding rate is induced by negative signalling and not by human capital depreciation. ---------------------------------------------------------------------------------------------------------------Implications of Demographic Changes on Familial Care Resources -With illustrative application to Hebei province of China Zhenglian Wang, Xianling Zhang and Yi Zeng ---------------------------------------------------------------------------------------------------------------Applications of multi-state models in hematology: bridging the gap Liesbeth de Wreede (LUMC Department of Medical Statistics and Bioinformatics and DKMS, Clinical Trials Unit) In methodological papers on multi-state models in biostatistical journals, their use is often motivated by applications in the field of hematopoietic stem cell (or bone marrow) transplantation (SCT). In contrast, there is a disappointing absence of multi-state models from medical journals in the field of hematology. This suggests that the gap between biostatistical theory and clinical studies in this domain is still large, and that it is a real challenge to bridge it. I’ll discuss several issues that have to be taken into account when developing a successful multi-state model that addresses clinical questions in the field of hematology. My leading example will be a multi-state model designed to analyze the impact of the treatment strategy in SCT followed by the Department of Hematology of LUMC. The focus will be on dynamic prediction of two competing failure probabilities and on the prediction of treatment success. ---------------------------------------------------------------------------------------------------------------Are Siblings’ Caregiving Decisions Efficient? Inference from One and Two Child Families Marike Knoef (University of Leiden, Dept of Economics) This paper develops a new method to identify whether siblings’ caregiving decisions are Pareto efficient. First, we estimate a structural model to reveal the preferences of children without siblings for providing care to parents, leisure, and consumption. Under the maintained assumption that – conditional on observed parental and child characteristics – children with a sibling have the same preference parameters (but different constraints) as only children, we verify whether observed allocations of siblings are efficient. We find that caregiving decisions are more likely to be inefficient when both siblings are female, are low educated, and report frequent family conflicts.