Clinical Scorecard: Utilizing Data-Driven Simulations to Evaluate the Effects of Study Limitations in Time-to-Event Analysis
At a Glance
Category
Detail
Condition
Time-to-event outcomes in epidemiologic studies
Key Mechanisms
Quantitative bias analysis using data-driven simulations to assess impact of data imperfections such as unmeasured confounding, measurement error, and interval censoring
Target Population
Patients or subjects in real-world epidemiologic time-to-event studies, including those with time-varying exposures or covariates
Care Setting
Epidemiologic research and clinical study data analysis settings
Key Highlights
Data-driven simulations combine real-world multivariable data with simulated additional data to assess bias and coverage in time-to-event analyses.
The approach allows quantifying the impact of data imperfections (e.g., unmeasured confounders, imprecise event timing) on estimates, beyond traditional QBA methods.
Two real-world examples illustrate assessing unmeasured prognostic factors and interval-censored events with time-varying exposures.
Guideline-Based Recommendations
Diagnosis
Identify and characterize data imperfections in the available dataset, including frequency and patterns.
Management
Fit multivariable regression models to obtain initial uncorrected estimates.
Formulate plausible assumptions about true parameter values and data modifications to create 'oracle' datasets free of imperfections.
Generate oracle datasets and modify them to reflect imperfections for comparative analysis.
Monitoring & Follow-up
Perform repeated simulations (e.g., 1000 repetitions) to summarize bias, mean squared error, and coverage rates.
Compare results from oracle and imperfect datasets to assess expected impact of data imperfections.
Risks
Ignoring data imperfections such as unmeasured confounding or interval censoring may bias time-to-event estimates and affect study conclusions.
Patient & Prescribing Data
Subjects in epidemiologic studies with time-to-event outcomes, including those exposed to time-varying treatments
Data-driven simulations can inform the reliability of estimated associations between exposures (including time-varying drugs) and outcomes, accounting for data limitations.
Clinical Best Practices
Incorporate quantitative bias analysis using data-driven simulations to formally assess the impact of study limitations on time-to-event analyses.
Use real-world data combined with simulation-based oracle datasets to evaluate and correct for biases due to unmeasured confounders and measurement errors.
Apply repeated simulation approaches to quantify uncertainty and improve interpretation of survival analysis results.
by Michal Abrahamowicz, Marie-Eve Beauchamp, Anne-Laure Boulesteix, Tim P Morris, Willi Sauerbrei, Jay S Kaufman, on behalf of the STRATOS Simulation Panel
