Case study: implications of channeling bias – diabetes


Channeling bias may occur when a newly marketed medicine and an established medicine with similar therapeutic indications are prescribed to different patients according to their prognostic characteristics. This case study aimed to explore time trends for differences in confounder distributions between users of antidiabetic medicines.

What was examined in this case study?

Data on patients with type 2 diabetes who started treatment between 2006 and 2015 were extracted from the UK Clinical Practice Research Datalink (CPRD). Analyses were stratified by years since first prescription of glucagon-like peptide-1 (GLP-1) and dipeptidyl peptidase-4 inhibitor (DPP-4i). Characteristics and the relative effect of GLP-1 vs. insulin, and DPP-4i vs. sulfonylurea were compared over time in people starting treatment.

What were the findings and conclusions?

The number of patients identified starting each medicine were: 8,398 GLP-1, 14,807 insulin, 24,481 DPP-4i, and 33,505 sulfonylureas.

No major channeling was observed over time.

Considerable overlap in distributions of characteristics allowed for propensity score matched analyses of 4,072 pairs of patients starting GLP-1 or insulin, and 10,620 pairs of patients starting DPP-4i or sulfonylurea. Relative effectiveness was similar across time, and could be estimated in the first year after launch.

The overall relative effect of GLP-1 vs. insulin showed no difference for HbA1c, and a relative increase in body weight (3.76 kg, 95% CI: 3.34 to 4.18) for insulin. The overall relative effect of DPP-4i vs. sulfonylurea showed a relative decrease in HbA1c (-0.34%, 95% CI: -0.39 to -0.29) and a relative increase in body weight (1.56 kg, 95% CI: 1.28 to 1.85) for sulfonylurea.

It has been hypothesized that early after a medicine’s launch, patients starting a medicine have more co-medication and a sub-optimal medicine response, and that this affects when the medicine becomes more established. However, such a time trend was not observed in this study.

In conclusion, no major channeling was identified in the investigated glucose-lowering drugs. Relative effectiveness could be estimated already in the first year after launch and was consistent in the years thereafter.

What are the limitations?

In this study, the characteristics investigated included demographics, co-medications and comorbidities. However, channelling might be seen in covariates not available to this study. As the study is based on observational data, the analyses of relative effectiveness may be affected by confounders (unmeasured covariates that affect exposure and outcome).

For more information, see Ankarfeldt et al., 2016.

Key contributor

Mikkel Z Ankarfeldt, Novo Nordisk