Can one assess whether missing data are missing at random in medical studies?

Area of Science:

• Statistics
• Biostatistics
• Medical Research Methodology

Background:

• Advanced statistical methods like multiple imputation improve missing data handling but often rely on the missing at random (MAR) assumption.
• The critical step of assessing the validity of the MAR assumption has been largely overlooked due to the lack of direct testing methods.
• Existing methods for missing data analysis often operate under the MAR assumption, which may not always hold true in real-world scenarios.

Purpose of the Study:

• To introduce a novel, testable assumption, MAR+, as an alternative to the untestable MAR assumption for missing data analysis.
• To evaluate the utility of MAR+ in assessing the plausibility of the MAR assumption in non-longitudinal studies.
• To underscore the importance of preventing missing data rather than solely relying on statistical correction methods.

Main Methods:

• Proposed a new assumption, MAR+ (Missing At Random Plus), which is testable and implies MAR under certain conditions.
• Developed a framework to test the MAR+ assumption, applicable primarily to non-longitudinal study designs.
• Applied the MAR+ testing approach to five illustrative medical case studies.

Main Results:

• The MAR+ assumption was found to fail in the majority of the five medical examples analyzed.
• Inability to reject MAR+ provides support for the MAR assumption, while rejection raises concerns about MAR's validity.
• Demonstrated that even sophisticated statistical techniques have limitations in correcting for missing data.

Conclusions:

• The MAR+ assumption provides a valuable, testable proxy for assessing the MAR assumption in certain study types.
• The frequent failure of MAR+ in medical examples suggests that the MAR assumption may be questionable in many practical applications.
• Emphasized the critical need for proactive strategies to minimize missing data in medical studies to ensure result integrity.