Comparison of Different LGM-Based Methods with MAR and MNAR Dropout Data

Area of Science:

• Statistics
• Biostatistics
• Data Analysis

Background:

• Missing data can introduce bias and distort results in statistical analyses.
• The missing not at random (MNAR) mechanism poses significant challenges for accurate parameter estimation.
• Comparing different methods for handling missing data is crucial for reliable study outcomes.

Purpose of the Study:

• To compare the performance of the maximum likelihood (ML) selection model (assuming missing at random - MAR) and the Diggle-Kenward selection model (handling MNAR) for missing data.
• To evaluate the influence of missingness mechanism, dropout rate, distribution shape, and sample size on parameter estimation and coverage probabilities.
• To identify the most robust method for handling MNAR data in statistical modeling.

Main Methods:

• A Monte Carlo simulation study was conducted to compare the two selection models.
• Four key factors were systematically varied: missingness mechanism (MAR vs. MNAR), dropout rate, distribution shape (skewness, kurtosis), and sample size.
• Parameter estimates and 95% confidence interval coverage probabilities (CP) were assessed for each scenario.

Main Results:

• Under MAR, the Diggle-Kenward model performed similarly to the ML approach.
• Under MNAR, the ML approach underestimated key parameters (intercept mean and slope), while Diggle-Kenward showed better, though still below-target, coverage probabilities.
• Dropout rate was the primary driver of estimation imprecision; differences between methods were negligible below a 10% dropout rate. The Diggle-Kenward model was more sensitive to non-normal distributions.

Conclusions:

• The Diggle-Kenward selection model is more appropriate than the ML approach when data are missing not at random (MNAR).
• High dropout rates significantly degrade parameter estimation precision, necessitating careful consideration in study design and analysis.
• Researchers should carefully consider the missing data mechanism and dropout rate when selecting a statistical method to ensure valid results.