gms | German Medical Science

MAINZ//2011: 56. GMDS-Jahrestagung und 6. DGEpi-Jahrestagung

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V.
Deutsche Gesellschaft für Epidemiologie e. V.

26. - 29.09.2011 in Mainz

Evaluating the performance of an Andersen-Gill model for time-dependent intervening events: allogeneic transplants in acute myeloid leukemia (AML)

Meeting Abstract

Suche in Medline nach

  • Richard F. Schlenk - Department of Internal Medicine III, University Hospital of Ulm, Ulm
  • Manuela Zucknick - Division of Biostatistics, German Cancer Research Center, Heidelberg
  • Axel Benner - Division of Biostatistics, German Cancer Research Center, Heidelberg

Mainz//2011. 56. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 6. Jahrestagung der Deutschen Gesellschaft für Epidemiologie (DGEpi). Mainz, 26.-29.09.2011. Düsseldorf: German Medical Science GMS Publishing House; 2011. Doc11gmds086

DOI: 10.3205/11gmds086, URN: urn:nbn:de:0183-11gmds0862

Veröffentlicht: 20. September 2011

© 2011 Schlenk et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

In acute myeloid leukemia, high-risk patients have a dismal prognosis and allogeneic hematopoietic stem-cell transplantation (allo-HSCT) is thought to be the best treatment strategy. Consequently, in younger high-risk AML patients treatment with allo-HSCT is intended. However, sometimes HSCT cannot be performed for reasons such as comorbidities frequently associated with older age or health deterioration before a suitable donor could be identified. In addition, patients get transplanted at varying times after diagnosis, depending on factors such as health status and donor source. In this situation, the effect of allogeneic HSCT on survival can be estimated in an Andersen-Gill model with allo-HSCT as a time-dependent intervening-event covariable. This model assumes that the effect is constant for all patients from the time of transplantation onwards and that there are no systematic differences between patients with and without HSCT other than through the factors included in the model. We demonstrate in a systematic simulation study how the Andersen-Gill model deals with potential model mis-specification. Simulations are based on data from the AMLSG HD98A study (Schlenk et al, JCO 2010), where allogeneic HSCT was intended for all high-risk patients. The following scenarios are investigated:

Whether (and when) a patient receives HSCT,

(1) does not depend on factors related to patient survival,

(2) depends only on known factors related to survival that are included in the Andersen-Gill model, and

(3) depends on (possibly unknown) factors relevant for survival not accounted for in the model.

We find that allogeneic HSCT effects are consistently estimated correctly in scenarios (1) and (2). Corresponding Wald tests find significant effects with sufficient power in scenarios where HSCT has a realistic effect size, and there is also correct control of the type-I error rate if HSCT is modelled to have no effect. In scenario (3), the Andersen-Gill model assumptions are violated, because HSCT status is simulated to depend on important factors related to patient survival, which are not included in the model. In this situation, the effect of allogeneic HSCT tends to be over-estimated and type-I error rates for corresponding Wald tests are not controlled at the specified significance level. However, perhaps surprisingly, estimates are still close to the true effect sizes and type-I-errors are moderately small.

Conclusion: The Anderson-Gill model is a valid approach to evaluate a time-dependent covariable such as allogeneic HSCT in the context of other important prognostic factors. Estimates and type-I-error remain quite stable in various scenarios.


References

1.
Andersen PK, Gill RD. Cox's regression model for counting processes: A large sample study. Ann Stat. 1982;10(4):1100-20.
2.
Schlenk RF, Döhner K, Mack S, et al. Prospective evaluation of allogeneic hematopoietic stem cell transplantation from matched related and matched unrelated donors in younger adults with high-risk acute myeloid leukemia: Results of German-Austrian AMLSG treatment trial AMLHD98A. J Clin Oncol. 2010;28(30):4642-8.
3.
Sylvestre MP, Abrahamowicz M. Comparison of algorithms to generate event times conditional on time-dependent covariates. Statistics in Medicine. 2008;27(14):2618-34.