The Development and Validation of a Nomogram for Identification of Potential Donation After Cardiac Death Donors

Abstract

To the Editor: We appreciate the interest of Collins et al 1 in our recent article regarding the development and validation of a nomogram for identification of potential donation after cardiac death (DCD) donors 2. The authors emphasized some important rules that investigators should pay attention to when developing a prognostic model. We do agree with all these rules and have to point out some statistical considerations in this study to avoid misunderstanding. Firstly, the nomogram was formulated based on the results from multivariate analysis through a Cox regression model rather than a logistic regression model. Doctors in different centers accept distinct criteria of the time to death after withdrawal of life support treatment (WLST), such as 30, 60, 90, 120, or 240 min. As would be expected, it is miscellaneous and confusing to develop several specific prognostic models for all these time points based on the results from multivariate analysis through a Logistic regression model. Furthermore, it is inconvenient for health workers to use these models. Therefore, a uniform nomogram based on a Cox regression model was developed to predict the time to death after WLST in our study. In the training set, all of 175 patients expired eventually; thus, the study had 175 “events.” After the univariate analysis, only 17 variables were included in the multivariate analysis, leading to an events per variable (EPV) of 175/17 = 10.3, which is not lower than the recommended minimum of EPV = 10. Thus, our methods agreed with the rules in building a prognostic model. In addition, good performance of this model in the two validations somewhat address the concerns on “over-fitting.” What might be confusing is that we reorganized all the descriptions and the figures showing the prediction of patient “death.” Originally, the nomogram was constructed to predict patient survival and “death” was the “event.” The reviewers recommended us to “call” it patient “death” other than patient “survival” to highlight the significance of this model. We do agree with Collins et al regarding the sample size considerations for studies validating a prediction model. Getting a bigger sample size is always good for statistics but is always difficult, especially in minority issues such as DCD. Actually, the sample size for the external validation in our study was much larger than those in other studies 3, 4. Once again, all the patients in the external (201) sets ended up with the “event.” This could reduce the bias from censors to some extent. And considering the small sample size (43) of the prospective validation set, we are conducting a prospective, multicenter study with a sample size of 200. Finally, regarding the request from Collins et al, we supplemented Table 1 with all regression coefficients here for readers' reference. We are happy to share with readers these data although they were not shown because of the limitation of table number according to the manuscript preparing guideline by the journal. X. He1,*, W. Liang2, G. Xu3 and Z. Guo1,* 1Organ Transplant Center, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China 2Department of Cardiothoracic Surgery, The First Affiliated Hospital of Guangzhou Medical University, Guangzhou, China 3Department of Neurosurgery, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China *Corresponding authors: Xiaoshun He, gdtrc@126.com and Zhiyong Guo, rockyucsf1981@126.com This study was supported by the Special Fund for Science Research by Ministry of Health (201302009), the Key Clinical Specialty Construction Project of National Health and Family Planning Commission of the People's Republic of China, and the Guangdong Provincial Key Laboratory Construction Projection on Organ Donation and Transplant Immunology (2013A061401007). The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.