(P-126) The Use of Lag Times in Cancer Pharmacoepidemiology

Background: With increasing utilization of medications, coupled with the increasing availability of long-term data, there is a growing need and opportunity for robust studies evaluating drug-cancer associations. One important methodological consideration in this setting is the application of lag times.

Objectives: The main objectives are to i) provide a detailed overview for the justification of the application of lag periods in cancer pharmacoepidemiology and ii) present practical advice for epidemiologists implementing lag periods in studies of drug-cancer associations.

Methods: We describe the main justifications for applying lag periods, namely the typically long latency period of cancer, reverse causality and detection bias. We discuss cancer latency with respect to tumor promoter or initiator effects and outline why biologically relevant lag periods will, in general, be longer than those required for reverse causation. We highlight the importance of duration response analyses alongside lagged analyses for the interpretation of drug cancer associations and provide practical advice on the implementation of lag periods in cohort and case-control settings.

Results: While the latency periods for cancer types are varied, in most instances these are long (i.e. at least several years). Variability in cancer latency can also be described in terms of drugs acting as initiators or promoters, with the later having shorter latency periods. However the mechanisms underlying carcinogenic drug effects are often unknown. The application of lag times also mitigates reverse causality and detection bias. Practically, this means that cancer events occurring shortly after drug initiation should not be regarded as exposed events. Additionally, a certain period after drug discontinuation should be considered as time at risk. Defining an appropriate lag period is not always straightforward and depends on the specific drug-cancer association under study. Researchers should acknowledge this uncertainty, presenting a range of lag times, and the removal of lag periods in sensitivity analyses. These allow readers to assess the impact of varying lags on their results and, alongside duration response analyses, enable the identification of potential confounding.

Conclusions: We recommend that researchers should always consider the issues of cancer latency, reverse causation and detection bias in studies of drug cancer associations. Researchers should consider all clinical and biological knowledge, along with the information that generated the hypothesis, to help inform a priori assumptions about lag times and for interpretation of findings.