Kuramoto et al. (2008) present a very interesting overview about models for the analysis of recurrent event time data. I would like to congratulate Kuramoto and colleagues to their nice and stimulating paper. However, I have one important comment on the article. Kuramoto et al. classify five approaches as appropriate models for the analysis of recurrent event times: the mean cumulative function, the Gamma-Poisson model, the independent-increment (Anderson-Gill) model, the conditional and the marginal model.
As the authors correctly state in their article the Gamma-Poisson model is in fact a frailty model with gamma distributed frailty and exponential baseline hazard function. This is a rather restricted model, assuming a constant recurrent event rate over time. It remains unclear why the authors restrict to this model in their paper instead of considering general frailty models (Duchateau and Janssen, 2008). In this class of models similar to the Cox proportional hazards model no parametric assumption about the baseline hazard function is necessary. Furthermore, additional to the gamma other frailty distributions can be used. These models are well designed for the analysis of recurrent event time data (Duchateau et al. 2003), including total follow-up time as well as gap time model variants. Software for frailty models is freely available in form of SAS macro's and R routines.
Based on this fact I strongly recommend avoiding the restriction to the Gamma-Poisson model and suggest to consider frailty models as an appropriate tool for analysing recurrent event times.
Sincerely yours
Andreas Wienke
References:
Duchateau, L., Janssen, P., Kezic, I., Fortpied, C. (2003) Evolution of recurrent asthma event rate over time in frailty models. Applied Statistics 52, 355-363
Duchateau, L., Janssen, P. (2008) The Frailty Model. Springer, New York
Kuramoto, L., Sobolev, B.G., Donaldson, M.G. (2008) On reporting results from randomized controlled trials with recurrent events. BMC Medical Research Methodology 8, 35
Recurrent event time analysis
23 September 2009
Dear Sir,
Kuramoto et al. (2008) present a very interesting overview about models for the analysis of recurrent event time data. I would like to congratulate Kuramoto and colleagues to their nice and stimulating paper. However, I have one important comment on the article. Kuramoto et al. classify five approaches as appropriate models for the analysis of recurrent event times: the mean cumulative function, the Gamma-Poisson model, the independent-increment (Anderson-Gill) model, the conditional and the marginal model.
As the authors correctly state in their article the Gamma-Poisson model is in fact a frailty model with gamma distributed frailty and exponential baseline hazard function. This is a rather restricted model, assuming a constant recurrent event rate over time. It remains unclear why the authors restrict to this model in their paper instead of considering general frailty models (Duchateau and Janssen, 2008). In this class of models similar to the Cox proportional hazards model no parametric assumption about the baseline hazard function is necessary. Furthermore, additional to the gamma other frailty distributions can be used. These models are well designed for the analysis of recurrent event time data (Duchateau et al. 2003), including total follow-up time as well as gap time model variants. Software for frailty models is freely available in form of SAS macro's and R routines.
Based on this fact I strongly recommend avoiding the restriction to the Gamma-Poisson model and suggest to consider frailty models as an appropriate tool for analysing recurrent event times.
Sincerely yours
Andreas Wienke
References:
Duchateau, L., Janssen, P., Kezic, I., Fortpied, C. (2003) Evolution of recurrent asthma event rate over time in frailty models. Applied Statistics 52, 355-363
Duchateau, L., Janssen, P. (2008) The Frailty Model. Springer, New York
Kuramoto, L., Sobolev, B.G., Donaldson, M.G. (2008) On reporting results from randomized controlled trials with recurrent events. BMC Medical Research Methodology 8, 35
Competing interests
no