A mean score method for sensitivity analysis to departures from the missing at random assumption in randomised trials

Area of Science:

• Biostatistics
• Clinical Trials
• Psychiatric Epidemiology

Background:

• Randomized trials often have incomplete outcome data, necessitating sensitivity analyses.
• Current sensitivity analysis methods are not widely adopted due to complexity or untestable assumptions.

Purpose of the Study:

• To propose a novel, accessible mean score approach for global sensitivity analysis of incomplete outcome data in randomized trials.
• To provide a method with interpretable sensitivity parameters for exploring departures from missing at random assumptions.

Main Methods:

• A mean score approach is developed for a single outcome analyzed under a generalized linear model.
• Sensitivity parameters quantify deviations from the missing at random assumption within a pattern mixture model framework.
• The method is non-stochastic, computationally efficient, and yields results consistent with standard methods under specific parameter values.

Main Results:

• The proposed sensitivity parameters are interpretable and can be elicited from subject matter experts.
• The method demonstrates computational speed and a non-stochastic nature.
• Point estimates, standard errors, and confidence intervals align with standard analyses when assumptions are met.

Conclusions:

• The mean score approach provides a practical and interpretable tool for sensitivity analyses in randomized trials with incomplete data.
• This method enhances the robustness of trial findings by systematically exploring the impact of missing data assumptions.
• Application to a mental health trial illustrates the method's utility in real-world research settings.