Statins are drugs effective at lowering cholesterol and they are used in the prevention of heart attack and stroke. In addition to their favourable effect in cardiovascular diseases, some observational studies also suggested that statins might reduce the risk of developing cancer. However, it was shown that methodological problems in the design and analysis of some of these studies led to the apparent beneficial effect of statins on cancer. The challenges of comparing a treated group to untreated patients in observational studies are numerous and careful assessment of several aspects, both at the design stage and in the analysis, must be taken into account. Indeed, the complexity of studies involving the comparison of a treated group to a untreated group reside, in part, in the absence of a clear starting point in time to base the comparison for people not using the drug of interest.
We propose an approach, which provides an emulation of the classical randomized trial to compare new users of the drug under study with non-users. We illustrate this approach by assessing the effect of statin use compared with non-use on the incidence of cancer.
The new-user cohort design is an essential tool for the conduct of observational studies of the comparative effects of drugs, including effectiveness and safety, to capture potential early effects and to properly time the confounders. While this approach is commonplace, a certain level of complexity is introduced when the comparison is not against an active comparator but rather against non-use. In this context, immortal time and other time-related bias become important threats to the validity of results. The objective of this study is to describe how the prevalent new-user design can be used in this context and to illustrate its use with the example of assessing the effect of statin use compared to non-use on the incidence of cancer. We will form a cohort of all adults in the CPRD, 30 years of age or more between January 1998 and December 2015, having at least one year of up-to-standard data and the record of a LDL cholesterol level <5 mmol/L. For each statin user, we will use the prevalent new-user design to select a reference statin non-user at the same time point as the user, matched on sex and time conditional propensity scores. Matched subjects will be followed for up to 10 years or until the occurrence of a cancer. The Cox proportional hazard model will be used to perform an intention to treat analysis that assesses the effect of statin use versus non-use on the risk of cancer occurrence.
Overall cancer; 7 site-specific cancers: breast, colorectal, haematological, melanoma, non-small cell lung, prostate and urothelial (as in Dickerman (1))
Samy Suissa - Chief Investigator - Sir Mortimer B Davis Jewish General Hospital
Samy Suissa - Corresponding Applicant - Sir Mortimer B Davis Jewish General Hospital
Sophie Dell'Aniello - Collaborator - McGill University