Immortal time bias in contemporary CDK4/6 inhibitor studies: Evidence from target trial emulation

Abstract

Background

Observational studies increasingly inform treatment decisions for CDK4/6 inhibitors in metastatic breast cancer, yet their validity is compromised by immortal time bias, a mechanistic bias that inflates treatment benefits when follow-up begins at treatment initiation rather than at diagnosis. However, whether historical bias estimates generalise to contemporary CDK4/6 research with rapid treatment initiation remains unknown.

Methods

We conducted a simulation-based study comparing naive observational analysis (time-zero at treatment) with target trial emulation (time-zero at diagnosis) across three synthetic cohorts reconstructed from published CDK4/6 inhibitor studies: a Danish population registry (N equals 1196), the Rugo et al. Flatiron database (N equals 9146), and a germline BRCA subgroup analysis (N equals 4050). We quantified bias magnitude and examined associations with treatment delay, patient characteristics, and healthcare system factors.

Results

Contrary to expectations, we observed minimal immortal time bias across all comparisons (mean absolute bias of 0.55 months, range 0.2 to 0.9 months; per cent bias of 0.5 to 2.1 per cent). Remarkably, all estimates showed underestimation by naive analysis, indicating that prevalent user bias (the exclusion of early deaths) dominated immortal time inflation. The duration of treatment delay accounted for only 25% of the variability in bias, indicating that context-dependent factors beyond delay duration determine bias magnitude.

Conclusions

Modern CDK4/6 inhibitor observational studies avoid historical immortal-time bias due to rapid treatment initiation and high event rates. These findings clarify that immortal time bias may be a smaller threat than historical cancer literature suggests in contemporary settings, though explicit time-zero specification remains a methodological best practice.