Page 133 - 2018_09-Mondo

Time-dependent effects on competing-risk endpoints
with a high early mortality risk in the analysis of death from other causes, an effect probably related to higher relapse rates.
The competing-risk endpoint relapse was derived from disease-free survival. Time-dependent variables that showed increased early relapse risk were intermediate and advanced disease stage, and a poor KPS.10 A poor KPS may be related to the primary disease or to any comor- bidity/toxicity.26 It can be speculated that an increased relapse risk is mainly due to patients whose primary dis- ease is the predominant reason for their poor KPS. We may, therefore, assume that, for this subset of patients, the risk may be even higher than our estimates. Interestingly, for the source of stem cells no significant differences could be detected between bone marrow and peripheral blood stem cells, which confirms the findings of a previous prospective trial.27
The analysis of non-relapse mortality (death without prior relapse) showed time-dependent effects for year of transplantation, source of stem cells, conditioning intensi- ty and poor KPS. Of these variables myeloablative condi- tioning and a poor KPS were strongly associated with early mortality after transplantation.28 These effects are related to transplantation-associated morbidity, particu- larly toxicity in the case of conditioning intensity and comorbidity or disease burden in the case of a poor KPS. Again, the first 4 months after transplantation were con- firmed to be a critical phase for conditioning toxicity and transplantation-related morbidity. The opposing effects of peripheral blood stem cell grafts on early and late mor- tality was prominent in this analysis, showing a highly significant protective effect in the first 8 months, proba- bly due to a shorter period of aplasia, and highly signifi- cant adverse effects later on, most likely caused by an increase of chronic graft-versus-host disease.19,20 Improvements in conditioning therapy and supportive care are reflected in the reduction of early mortality in the first 8 months after transplantation in the more recent periods regarding year of transplantation.24
We considered two competing-risk settings, one with time until a specific cause of death and one with time until a first event (relapse or death) (Figure 1). The results of the analyses for non-relapse mortality and transplant- related mortality are quite similar as there is a large, although not complete, overlap. Non-relapse mortality additionally includes deaths not related to transplantation or disease relapse/progression. In our study we needed both endpoints to construct different competing-risk set- tings (Figure 1). One difference between transplant-relat- ed mortality and non-relapse mortality is that the former requires the classification of deaths as transplant-related by a physician, which may be difficult in some situations, whereas the classification of non-relapse mortality does not require this assessment and is based instead on the plain and objective distinction between death in complete remission or not. Thus, the advantage of transplant-relat- ed mortality is that it is more specific for transplantation- related adverse events while the advantage of non-relapse mortality is that it is easier and more objective to classify.
Limitations of our analysis are that exact patterns of HLA-mismatches, regarding number and loci of mis- matches, were not available for the majority of the trans- plants and so only a rough stratification according to donor type was feasible (i.e. matched related, mis- matched related, matched unrelated, mismatched unre-
lated). A certain degree of heterogeneity is related to the large time span over which the transplants included in this study were performed, with changes in transplant procedures, graft source preferences, donor selection algorithms and supportive care over the years. This het- erogeneity was only partly addressed by including the time period of transplantation as a covariate. In a separate analysis in which the only transplants included were those carried out in the periods 1998-2005 and 2006- 2013, we found results comparable to those presented in this manuscript (i.e. inclusion of transplants from 1976 onwards). In order to be consistent with our previous publication, in which we used a similar approach to investigate the effect of various covariates on overall and disease-free survival in a time-dependent manner, we decided to carry out our analyses for this work based on the same data set as previously.9 Center-specific unob- served variables were included by using stratification according to frequency of transplants in the transplant center. Another highly predictive clinical variable is cyto- genetic risk.2 This predictor could not be evaluated as information regarding cytogenetic risk is sparse in the DRST/EBMT database.
Grathwohl et al. and Zwaan et al. highlighted the dis- ease-related adjustment for time from diagnosis to trans- plantation.29,30 We chose not to include this variable in our models due to difficulties in its interpretation. Several fac- tors influence time from diagnosis to transplantation. These are disease-inherent risk, pretreatment, clinical sta- tus of the patient, clinical urgency of transplantation, prompt availability of a suitable donor as well as the patient´s choice with regard to alternative treatment options. The profile of these factors differs between dis- tinct disease entities. Thus, the information provided by global estimates for this predictor is not meaningful. Several methods that allow modeling of time-dependent effects of covariables have been described.31 Here, we fol- lowed the approach of Fuerst et al.,9 and chose piecewise constant effects estimation for ease of interpretation.12,32 It is possible that the true underlying time-dependent effects are more complex, and a careful interpretation of our results would be that of averaged effects on the respective time intervals. In addition, we aimed to extend the results of Fuerst et al.9 to the more specific endpoints transplant-related mortality, death from other causes, relapse and non-relapse mortality. Consequently, prefer- ential candidate time intervals were as described in Fuerst et al.,9 and were not cross-validated.
In summary, this analysis of competing risks disentan- gles how the previously described net effect9 is achieved via the different outcomes summarized in the composite endpoint of overall survival (or disease-free survival). The description of time-dependent effects allows a better understanding of transplantation biology regarding com- peting-risk endpoints. Reasons for early mortality may be described and quantified more precisely. Predictors with ambivalent effects, such as graft source and conditioning toxicity, may be identified. These observations may sup- port treatment choices, individual patient counseling and reassurance during follow-up.
Acknowledgment
The authors would like to thank the DRST Data administra- tors F. Hanke and H. Neidlinger for providing the clinical data for this analysis
haematologica | 2018; 103(9)
1533