Probability prediction in multistate survival models for patients with chronic myeloid leukaemia

Abstract

In order to find an appropriate model suitable for a multistate survival experiment, 634 patients with chronic myeloid leukaemia (CML) were selected to illustrate the method of analysis. After transplantation, there were 4 possible situations for a patient: discase free, relapse but still alive, death before relapse, and death after relapse. The last 3 events were considered as treatment failure. The results showed that the risk of death before relapse was higher than that of the relapse, especially in the first year after transplantation with competing-risk method. The result of patients with relapse time less than 12 months was much poor by the Kaplan-Meier method. And the multistate survival models were developed, which were detailed and informative based on the analysis of competing risks and Kaplan-Meier analysis. With the multistate survival models, a further analysis on conditional probability was made for patients who were disease free and still alive at month 12 after transplantation. It was concluded that it was possible for an individual patient to predict the 4 possible probabilities at any time. Also the prognoses for relapse either death or not and death either before or after relapse may be given. Furthermore, the conditional probabilities for patients who were disease free and still alive in a given time after transplantation can be predicted.

Keywords

prediction / multistate / survival models / chronic myeloid leukaemia

References

[1] KaplanE L, MeierP. Non parametric estimation from incomplete observations. J Am Stat Assoc, 1958, 53457-457

[2] CoxD R. Regression models and life tables (with discussion). J Roy Statist Soc, Ser B, 1972, 34187-187

[3] GoldmanJ M. Optimizing treatment for chronic myeloid leukaemia. New Eng J Med, 1997, 337270-270

[4] GratwohlA, HermansJ, GoldmanJ M, et al. . Risk assessment for patients with chronic myeloid leukaemia before allogeneic blood or marrow transplantation. Lancet, 1998, 3521087-1087

[5] FangY, GratwohlA, van HouwelingenH C. Relapse risk assessment of transplant for patients with chronic myeloid leukaemia. Chin Med J, 2003, 116(2): 305-305

[6] CollettD. Modelling Survival Data in Medical Research, 1994, UK, Chapman & Hall

[7] ArjasE, EerolaM. On predictive causality in longitudinal studies. J Stat Plan Infer, 1993, 34361-361

[8] KleinJ P, KeidingN, CopelanE A. Plotting summary predications in multistate survival models: probabilities of relapse and death in remission for bone marrow transplantation patients. Stat Med, 1994, 132315-2315

[9] PepeM S, MoriM. Kaplan-Meier, marginal or conditional probability curves in summarizing competing risks failure time data?. Stat Med, 1993, 12737-737

[10] SasazukiT, JujiT, MorishimaY, et al. . Effect of matching of class 1 HLA alleles on clinical outcome after transplantation of hematopoietic stem cells from an unrelated donor. New Eng J Med, 1998, 3391177-1177

[11] GratwohlA, HermansJ, NiederwieserD, et al. . Bone marrow transplantation for chronic myeloid leukemia: long-term results. Chronic Leukemia Working Party of the European Group for Bone Marrow Transplantation. Bone Marrow Transplant, 1993, 12509-509