On the 18th of September, Malcolm Gillies of the University of South Wales presented a paper describing the challenges that arose from target trial emulation of PCSK9 inhibitors (Cars et al, 2026). Over 50 people joined this session, representing the full breadth of expertise and resulting in a lively discussion covering the strengths, limitations and impact of the paper.
This paper examines how researchers can emulate a target trial to assess the effectiveness of initiating PCSK9 inhibitor treatment after myocardial infarction. Rather than presenting new clinical evidence on PCSK9 inhibitor effectiveness, the authors used the case study to demonstrate the methodological challenges that arise when treatment initiation may occur during follow up. In particular, the paper showed how decisions about time zero, grace periods, artificial censoring, weighting and comparator groups can shape the validity and interpretation of findings from real-world data.
Malcolm explained how the authors used a clone censor weight approach: duplicating each eligible person into treatment and no treatment strategies, censoring clones when their observed behaviour no longer matched the assigned strategy, and applying inverse probability of censoring weights to address the artificial censoring introduced by this design while adjusting for confounders.
Malcolm also highlighted the authors’ use of negative control outcomes, including fractures, glaucoma, joint replacement, non-melanoma skin cancer, kidney stones and other cancers, to assess whether the method still produced associations where no causal effect was expected.
The presentation showed that, even with the target trial emulation framework and covariate adjustment, some negative control outcomes suggested residual bias. Malcolm used this to underline the paper’s broader message: clone censor weight methods can help address immortal time bias, but they do not by themselves resolve confounding by indication or the instability caused by extreme weights.
As with every Journal Club session, the presentation led to a lively discussion with the group. Participants pointed out challenges in specifying and fitting the model for the inverse probability of censoring weights. For time-dependent models, such as that used in the untreated group, event times must be accumulated into time windows. Overly broad windows may lose important temporal detail or introduce immortal time bias, while very narrow windows can create unstable models and extreme weights. The discussion also touched on measurement bias in secondary-use data, including the possibility that treatment may influence healthcare-seeking behaviour and therefore outcome ascertainment.
An alternative design was also suggested. By restricting the cohort to people who initiate PCSK9 inhibitors and then comparing those who continue with those who discontinue treatment, researchers can ensure that both groups have an indication for treatment. However, as noted during the discussion, discontinuation may itself be related to patient preferences, side effects or other factors.
Participants also discussed active comparator new user designs and prevalent new user cohort designs as possible alternatives. These approaches may help identify more comparable treatment groups, depending on the clinical context and available data.
Further questions considered whether clone censor weight methods answer a different causal question from ordinary propensity score adjustment or sequential trial approaches. The discussion highlighted that the chosen grace period is not merely a technical detail. It actually defines the treatment strategy being compared. A 12-month grace period, for example, compares initiation within 12 months with non-initiation during that period. As several participants noted, the appropriate grace period should reflect the clinical decision-making process and the expected timing of outcomes.
Congratulations to Malcolm for a successful and interesting Journal Club! His session illustrated that target trial emulation requires careful specification of the causal question, thoughtful choice of comparators, clinically meaningful timing decisions and close attention to residual bias. While clone censor weight methods can be a powerful tool for avoiding immortal time bias and estimating the effect of clearly defined treatment strategies, they are not a substitute for a well-chosen comparator group or robust control of confounding.
Interested in exploring more about target trial emulation? The next VAC4EU Journal Club session will also address this topic. On the 29th of October, Anna Girardi from ARS Toscana will present on the target trial emulation framework and its role in improving evidence generation on medication effectiveness and safety during pregnancy. Register at the link below!
