Fallacies of last observation carried forward analyses

Area of Science:

• Biostatistics
• Clinical Trials
• Longitudinal Data Analysis

Background:

• Last observation carried forward (LOCF) is a common imputation method for missing longitudinal data.
• LOCF replaces missing values with the last recorded observation for each subject.
• This method is widely used despite known statistical deficiencies.

Purpose of the Study:

• To provide a simple explanation of the statistical deficiencies of LOCF.
• To highlight concerns raised by the National Research Council regarding LOCF.
• To advocate for more statistically valid methods for handling missing data.

Main Methods:

• A simple repeated measures model was used to analyze quantitative observations at two time points.
• The model assumed data were missing completely at random for imputation.
• The analysis focused on the mean and variance of the resulting mixture distribution.

Main Results:

• LOCF creates a mixture distribution of observed and imputed values.
• The mean and variance of this distribution are functions of both observed and missing data distributions.
• Unbiased results from LOCF are only achieved when the distribution of observed values equals the unknown distribution of missing values.

Conclusions:

• LOCF analyses are of questionable veracity and can be specious.
• No simple expressions for mean and variance are possible when data are not randomly missing without unverifiable assumptions.
• Researchers should strive to minimize missing data and employ more valid statistical analyses, avoiding LOCF.