Summary
Even in clinical trials, the point in the patient’s cancer journey (and date!) on which they enter a group that is studied may differ because individual patients may wait different amounts of time before starting a study treatment. The effect of this waiting is called left-truncation and can lead to errors in the results unless appropriate statistical analysis methods are used.
Real-world data (RWD) from electronic health records (EHRs) are commonly subject to left truncation, a type of selection bias that occurs when patients need to survive long enough to satisfy certain entry criteria. Standard methods to adjust for left truncation bias rely on the idea that there is marginal independence between entry time and survival time, which may not always be the case in practice.
Researchers showed in this manuscript that relaxing the necessary assumptions for valid inference under left truncation allowed for a broader range of analyses to be conducted and that conditionally independent left truncation can be easily tested for, and result in unbiased estimates from common survival analyses.
Why this matters
Better methods such as those presented in this manuscript increase confidence in research results.
EHR-data sources are gaining in sophistication, and with them, the analytic instruments that allow researchers to gather reliable evidence from them. This methodological study furthers our understanding of the biases present in certain observational data, such as clinico-genomic data, their potential analytic impact, and proposes mitigating approaches to those issues. It is therefore a prime example of the type of conceptual work necessary to unlock the potential of EHR-derived data with analytic insight and integrity.
