Table 2.

Overview of Strategies to Prevent and Manage Missing Data

Phase of research	Main strategies
Research Question and Study Design	When planning the study, consider how the research question, the target population and key variables can be defined to promote both high rates of complete data and a representative population
	Adapt the frequency of study visits, sites, and duration of participation to the capabilities of the target population
All measures	Anticipate data collection needs of participants with varying health and function.
	Anticipate the need for proxy informants. Identify potential proxies at enrollment and use key measures that have been validated for proxy use when possible
	Code reasons for missing data, especially inability to perform a test
Outcome Measures	Prespecify alternate data collection strategies to use when the primary strategy fails
	Prespecify alternate definitions and logical sequences for adjudication of major outcomes
	Anticipate need for combined outcomes
	Consider alternatives to a single fixed time point for outcome assessment
Predictor Measures	Prioritize data collection sequence
Intervention	Measure adherence and fidelity to treatment protocol
	Measure success of blinding in participants and study personnel
	Measure expectations in controls, especially if trial participants are not blinded
Pilot studies	Assess problems with data collection
	Revise study plans to reduce problems with data collection
Implementation	Plan for flexibility in schedules, sites and protocols.
	Have protocols for identifying participants at risk of missing data
	Be prepared to modify protocol if missing data problems develop
Data management	Develop and implement real time tracking and reporting system for missing data
Missing data assessment	Quantify amount of missing data (problems minor when < 5%)
	Characterize missing data rates by items, waves, and participants
	Examine potential reasons and mechanisms for missing data
	Compare participants with and without types of missing data to assess potential biases
Analysis	Weigh analytic options in the context of the limitations of each
	Determine whether imputation can be used for some missing data
	Perform sensitivity analyses to examine potential biases due to missing data