Measurement error and timing of predictor values for multivariable risk prediction models are poorly reported

Area of Science:

• Clinical Epidemiology
• Biostatistics

Background:

• Measurement error in predictor variables can compromise the accuracy and reliability of clinical prediction models.
• Ensuring predictors are measured at the appropriate time is crucial for valid model application.

Purpose of the Study:

• To assess the prevalence and impact of measurement error in predictors used in clinical prediction models.
• To determine if studies report the intended moment of predictor measurement relative to model use.

Main Methods:

• Systematic literature search of Medline for clinical prediction model development studies published in 2015.
• Extraction of data on predictors, measurement error control strategies, and timing of measurement.
• Classification of predictors into low and high susceptibility to measurement error.

Main Results:

• A review of 33 studies identified 151 predictors, with 33.7% classified as high risk for measurement error.
• This high risk of error was not addressed in the model development process for these predictors.
• Only 24.2% of studies clearly stated the intended moment of model use and the timing of predictor measurement.

Conclusions:

• There is a significant deficit in reporting measurement error and the timing of predictor measurements in clinical prediction model studies.
• Further research is needed to understand the consequences of ignoring measurement error and the benefits of accounting for it in model development.