How handling missing data may impact conclusions: A comparison of six different imputation methods for categorical...

Area of Science:

• Medical Research
• Statistical Analysis
• Data Science

Background:

• Missing data is a common challenge in medical research, particularly with questionnaire data.
• Incomplete datasets can compromise the validity of statistical analyses and clinical conclusions.

Purpose of the Study:

• To compare six multiple imputation methods and complete case analysis for handling missing categorical data.
• To assess the impact of different missing data handling techniques on regression model outcomes.

Main Methods:

• Evaluated complete case analysis and six imputation methods: expectation-maximization with bootstrapping, multiple correspondence analysis, latent class analysis, hot deck, and multivariate imputation by chained equations (logistic regression and random forests).
• Applied methods to real-world questionnaire data from the Norwegian opioid maintenance treatment program.

Main Results:

• All methods performed adequately with large sample sizes (n=1000).
• With smaller sample sizes (n=200) and >=20% missing data, complete case analysis, hot deck, and random forests yielded biased estimates and low coverage.
• Multiple imputation using multiple correspondence analysis demonstrated superior performance across various scenarios.

Conclusions:

• The selection of a missing data handling method significantly influences clinical interpretations derived from statistical analyses.
• Choosing appropriate imputation techniques is vital for drawing reliable conclusions from regression models with missing data.