Time-related biases in observational studies : an illustration with COVID-19 vaccination during immune checkpoint inhibitor therapy
Preprint
Abstract
Estimating treatment effects from observational data requires careful specification of the study timeline. Investigators must determine when eligibility is assessed, when treatment is assigned, and when follow-up begins. Seemingly minor choices at this stage can introduce subtle time-related biases and compromise valid inference. We show how this is relevant to a recent article suggesting that SARS-CoV-2 mRNA (COVID-19) vaccines increase the effectiveness of immune checkpoint inhibitors in cancer patients (Grippin et al.). We first describe the original study timeline and find several sources of bias arising from its specifications. These time-related biases lead to erroneous conclusions about protective treatment effects. We then explain how time-related biases can be mitigated by careful target trial emulation, and we specify and emulate a target trial for the study of Grippin et al. In this target trial emulation, the estimated 3-year overall survival was 45.7% in the vaccinated group and 43.8% in the unvaccinated group (difference, 1.9 percentage points [95% CI, −23.6 to 25.0]), substantially attenuating the benefit of COVID-19 vaccines found in the original analysis.Keywords
This publication has 10 references indexed in Scilit:
- SARS-CoV-2 mRNA vaccines sensitize tumours to immune checkpoint blockadeNature, 2025
- Transparent reporting of observational studies emulating a target trial: the TARGET StatementBMJ, 2025
- Clinical spectrum and evolution of immune-checkpoint inhibitors toxicities over a decade—a worldwide perspectiveeClinicalMedicine, 2024
- Key concepts in clinical epidemiology: collider-conditioning biasJournal of Clinical Epidemiology, 2023
- Reflection on modern methods: trial emulation in the presence of immortal-time bias. Assessing the benefit of major surgery for elderly lung cancer patients using observational dataInternational Journal of Epidemiology, 2020
- Covariate selection strategies for causal inference: Classification and comparisonBiometrical Journal, 2018
- Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analysesJournal of Clinical Epidemiology, 2016
- Diagnosing and responding to violations in the positivity assumptionStatistical Methods in Medical Research, 2010
- The Hazards of Hazard RatiosEpidemiology, 2010
- Immortal Time Bias in PharmacoepidemiologyAmerican Journal of Epidemiology, 2007