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

Time-dependent study entries and exposures in medical research: sources of different and avoidable types of bias

Meeting Abstract

  • Martin Wolkewitz - Institut für Med. Biometrie und Med. Informatik, Freiburg
  • Arthur Allignol - Institut für Med. Biometrie und Med. Informatik, Freiburg
  • Stephan Harbarth - Hôpitaux Universitaires de Genève, Geneve
  • Giulia De Angelis - Universita Cattolica, Rom
  • Martin Schumacher - Institut für Med. Biometrie und Med. Informatik, Freiburg
  • Jan Beyersmann - Institut für Med. Biometrie und Med. Informatik, Freiburg

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. Doc11gmds117

doi: 10.3205/11gmds117, urn:nbn:de:0183-11gmds1171

Veröffentlicht: 20. September 2011

© 2011 Wolkewitz 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

Background: Survival bias is a common concern in medical applications of survival analysis since wrong results may directly impact patient care. We discuss two avoidable types of bias which might occur in presence of a time-dependent study entry ('length bias') or a time-dependent exposure or treatment ('time-dependent bias') [1], [2].

Methods: Using two real-data examples (one on the effect of winning the Oscar on actor survival [2], one on hospital infection on hospital stay [3], we give innovative and easy-to-understand graphical presentations of how these biases corrupt the analyses via distorted time-at-risk. Multi-state models are used to account for left-truncation and time-dependent exposure. Cumulative hazards are estimated via the Nelson-Aalen method. We performed a literature search to investigate whether the problem is relevant for various medical disciplines.

Results: Both types of bias can be displayed on an individual level and on an aggregated level with distorted risk sets and cumulative hazards. Length bias leads to an artificial underestimation of the overall hazard. Time-dependent bias has the following consequences: the hazard of the unexposed patients is always overestimated and the hazard of the exposed patients is always underestimated. The magnitude depends on the time of study entry, exposure and outcome, and it may, e.g., lead to a reversed effect.

Discussion: Survival bias might occur in many different medical disciplines and is avoidable using survival techniques.


References

1.
van Walraven, C, Davis D, Forster A, Wells G. “Time-dependent bias was common in survival analyses published in leading clinical journals.” J Clin Epidemiol. 2004;57:672-–682.
2.
Wolkewitz M, Allignol A, Schumacher M, Beyersmann J. “Two Pitfalls in Survival Analyses of Time-Dependent Exposure: A Case Study in a Cohort of Oscar Nominees.” Am Stat. 2010;64(3):205-–211.
3.
De Angelis G, Allignol A, Murthy A, Wolkewitz M, Beyersmann J, Safran E, Schrenzel J, Pittet D, Harbarth S. “Multistate Modelling to Estimate the Excess Length of Stay Associated with MRSA Colonization and Infection in Surgical Patients. J Hosp Infect. 2011, in press.