M Cerou1,2,3,4,5, M Lavielle6, K Brendel7, M Chenel7, E Comets8,9,10,11. 1. INSERM, CIC 1414, 35700, Rennes, France. marc.cerou@inserm.fr. 2. University Rennes-1, 35700, Rennes, France. marc.cerou@inserm.fr. 3. INSERM, IAME, UMR 1137, 75018, Paris, France. marc.cerou@inserm.fr. 4. University Paris Diderot, IAME, UMR 1137, Sorbonne Paris Cité, 75018, Paris, France. marc.cerou@inserm.fr. 5. Division of Clinical Pharmacokinetics and Pharmacometrics, Institut de Recherches Internationales Servier, Suresnes, France. marc.cerou@inserm.fr. 6. Inria Saclay & CMAP, Ecole Polytechnique, Palaiseau, France. 7. Division of Clinical Pharmacokinetics and Pharmacometrics, Institut de Recherches Internationales Servier, Suresnes, France. 8. INSERM, CIC 1414, 35700, Rennes, France. 9. University Rennes-1, 35700, Rennes, France. 10. INSERM, IAME, UMR 1137, 75018, Paris, France. 11. University Paris Diderot, IAME, UMR 1137, Sorbonne Paris Cité, 75018, Paris, France.
Abstract
PURPOSE: Normalised prediction distribution errors (npde) are used to graphically and statistically evaluate mixed-effect models for continuous responses. In this study, our aim was to extend npde to time-to-event (TTE) models and evaluate their performance. METHODS: Let V denote a dataset with censored TTE observations. The null hypothesis (H0) is that observations in V can be described by model M. We extended npde to TTE models using imputations to take into account censoring. We then evaluated their performance in terms of type I error and power to detect model misspecifications for TTE data by means of a simulation study with different sample sizes. RESULTS: Type I error was found to be close to the expected 5% significance level for all sample sizes tested. The npde were able to detect misspecifications in the baseline hazard as well as in the link between the longitudinal variable and the survival function. The ability to detect model misspecifications increased as the difference in the shape of the survival function became more apparent. As expected, the power also increased as the sample size increased. Imputing the censored events tended to decrease the percentage of rejections. CONCLUSIONS: We have shown that npde can be readily extended to TTE data and that they perform well with an adequate type I error.
PURPOSE: Normalised prediction distribution errors (npde) are used to graphically and statistically evaluate mixed-effect models for continuous responses. In this study, our aim was to extend npde to time-to-event (TTE) models and evaluate their performance. METHODS: Let V denote a dataset with censored TTE observations. The null hypothesis (H0) is that observations in V can be described by model M. We extended npde to TTE models using imputations to take into account censoring. We then evaluated their performance in terms of type I error and power to detect model misspecifications for TTE data by means of a simulation study with different sample sizes. RESULTS: Type I error was found to be close to the expected 5% significance level for all sample sizes tested. The npde were able to detect misspecifications in the baseline hazard as well as in the link between the longitudinal variable and the survival function. The ability to detect model misspecifications increased as the difference in the shape of the survival function became more apparent. As expected, the power also increased as the sample size increased. Imputing the censored events tended to decrease the percentage of rejections. CONCLUSIONS: We have shown that npde can be readily extended to TTE data and that they perform well with an adequate type I error.
Keywords:
PSA; model evaluation; npde; time-to-event
Authors: S F Marshall; R Burghaus; V Cosson; S Y A Cheung; M Chenel; O DellaPasqua; N Frey; B Hamrén; L Harnisch; F Ivanow; T Kerbusch; J Lippert; P A Milligan; S Rohou; A Staab; J L Steimer; C Tornøe; S A G Visser Journal: CPT Pharmacometrics Syst Pharmacol Date: 2016-03-14
Authors: A Lawrence Gould; Mark Ernest Boye; Michael J Crowther; Joseph G Ibrahim; George Quartey; Sandrine Micallef; Frederic Y Bois Journal: Stat Med Date: 2014-03-14 Impact factor: 2.373
Authors: Jorie Versmissen; Daniëlla M Oosterveer; Mojgan Yazdanpanah; Joep C Defesche; Dick C G Basart; Anho H Liem; Jan Heeringa; Jacqueline C Witteman; Peter J Lansberg; John J P Kastelein; Eric J G Sijbrands Journal: BMJ Date: 2008-11-11