Detecting channeling bias after launch – implications for comparative effectiveness studies: a case study in anticoagulant medicines


Potential bias from the channeling of patients taking newly approved medications can pose major methodological challenges. This may occur because of patient, clinician, or system-related factors, or because of rapid changes in the characteristics of the user population during the early phase after launch. This case study aimed to compare characteristics of patients starting different oral anticoagulant medicines and the risk of ischaemic stroke, acute myocardial infarction and major bleeds over time since launch.

What was examined in this case study?

Using the US MarketScan commercial claims database and the UK Clinical Practice Research Datalink (CPRD), patients who were 18 years or older with atrial fibrillation/flutter who started an oral anticoagulant medicine were included in the case study if they were enrolled for at least 6 months and had not used an oral anticoagulant during the 6 months prior to starting treatment (index date). Hazard ratios (HR) for ischaemic stroke, acute myocardial infarction, and major bleeds were estimated in users of novel oral anticoagulants (NOACs), such as dabigatran and rivaroxaban, vs. warfarin at different time periods after launch using multivariable Cox regression and propensity scores (PS) methods. Confounder distributions among the groups were summarised as PS and time trends since launch were assessed.

What were the findings and conclusions?

  • In general, the US MarketScan population was at lower risk for stroke compared with the UK CPRD population (the MarketScan population was younger and had a lower mean CHA2DS2-VASc score), although the trend over time was similar between different oral anticoagulant medicines.
  • There was substantial overlap in PS distributions between the treatment groups in both datasets.
  • The risk of ischaemic stroke for NOACs was lower in the MarketScan population (HR 0.74, 95% CI: 0.61 to 0.90) but higher in the CPRD population (HR: 1.31, 95% CI: 1.04 to 1.65).
  • The risk for acute myocardial infarction was similar for NOACs and warfarin whereas the risk of major bleeding was higher for NOACs compared with warfarin (HR: 1.34, 95% CI: 1.11 to 1.62 in MarketScan and HR: 1.41, 95% CI: 1.06 to 1.87 in CPRD).
  • HRs were similar across different PS methods.
  • Differences between characteristics for NOAC users compared with warfarin users were small with no noticeable change over the years suggesting minimal channeling bias after launch.
  • The benefit-risk balance for treatment with NOACs seems to be better for US patients compared with UK patients.

Key contributor

Sanni Ali, UMCU