Measurement error in continuous endpoints in randomised trials: Problems and solutions

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Statistical Inference

Background:

• Continuous endpoints in randomized trials are susceptible to measurement error.
• Ignoring measurement error can impact the accuracy of statistical inference.
• Understanding different types of measurement error is crucial for reliable trial results.

Purpose of the Study:

• To investigate the impact of ignoring measurement error in continuous endpoints.
• To propose and evaluate methods for improving statistical inference in the presence of measurement error.
• To address classical, systematic, and differential measurement error types.

Main Methods:

• Developed corrected effect estimators for different measurement error types.
• Utilized external calibration samples with error-prone and error-free measurements.
• Conducted a simulation study to test corrected estimators and confidence interval methods.
• Proposed methods implemented in a new R software package.

Main Results:

• Ignoring measurement error yields an unbiased estimator for classical error but increases Type-II error.
• Systematic or differential measurement errors can cause substantial bias in treatment effect estimation.
• External calibration samples effectively prevent bias and improve inference for systematic/differential errors.
• Required calibration sample size depends inversely on the error-prone/error-free endpoint association strength.

Conclusions:

• Measurement error correction using external calibration samples is vital for accurate inference in trials with error-prone endpoints.
• Even small calibration samples can significantly improve statistical inferences.
• The proposed methods offer a practical solution for handling measurement error in clinical trials.