Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes

wolke@fdm.uni-freiburg.de

We congratulate the authors on an excellent work which highlights a
serious source of bias in cohort studies. The aim of this letter is to
further emphasize the importance of the work by Levesque et al.,noting
that immortal time bias is neither restricted to the field of
pharmacoepidemiology nor to specific data examples. In fact, immortal time
bias, which is sometimes also called "time-dependent bias", has
been found to be a common phenomenon in a variety of medical fields (1).
In addition, simple algebraic considerations show that immortal time bias
inevitably leads to underestimating hazard ratios (comparing exposed
versus unexposed) because of time distorted risk sets (2). If the endpoint
is death, this implies that any beneficial effect of time-dependent
treatment / exposure is overestimated. Any harmful effect will be subdued
or even reversed. Levesque et al. give an impressive example of such a
reversed effect.

Another field that is prone to immortal time bias is, e.g., hospital
epidemiology. Here, interest often focuses on the impact of some time-
dependent event such as hospital-acquired infection on hospital outcome
(3). E.g., policy makers are interested in subsequent hospital stay. Then,
the endpoint is discharge and the underestimated discharge hazard ratio
corresponds to an overestimation of extra length of stay due to the
hospital-acquired infection. Hence, policy makers often receive
exaggerative values if patients who acquire a nosocomial infection during
their hospital stay are incorrectly classified as infected between
admission and infection.

The paper by Lévesque et al. highlights an important issue, and we
join the authors in advocating a proper analysis which takes the time-
dependency of treatment / exposure into account.

References:

(1) van Walraven, C., Davis, D., Forster, A., and Wells, G. (2004),
“Time-dependent bias was common
in survival analyses published in leading clinical journals,” J Clin
Epidemiol , 57, 672–682.

(2) Beyersmann, J., Gastmeier, P., Wolkewitz, M., and Schumacher, M.
(2008), “An easy mathematical
proof showed that time-dependent bias inevitably leads to biased effect
estimation,” J Clin
Epidemiol , 61, 1216–1221.

(3) Samore M and Harbarth S., “A Methodologically Focused Review of
the Literature in Hospital Epidemiology and Infection Control”, In:
Mayhall, C.G. “Hospital Epidemiology and Infection Control”, 3th ed.
Lippincott Williams & Wilkins, 2004, chapter 93

Competing interests:
None declared